UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

An expert system to diagnose quality problems in the continuous casting of steel billets Kumar, Sunil 1991

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
831-ubc_1992_spring_kumar_sunil.pdf [ 7.62MB ]
Metadata
JSON: 831-1.0078575.json
JSON-LD: 831-1.0078575-ld.json
RDF/XML (Pretty): 831-1.0078575-rdf.xml
RDF/JSON: 831-1.0078575-rdf.json
Turtle: 831-1.0078575-turtle.txt
N-Triples: 831-1.0078575-rdf-ntriples.txt
Original Record: 831-1.0078575-source.json
Full Text
831-1.0078575-fulltext.txt
Citation
831-1.0078575.ris

Full Text

AN EXPERT SYSTEM TO DIAGNOSE QUALITY PROBLEMS IN THE CONTINUOUS CASTING OF STEEL BILLETS  by SUNIL KUMAR B.Tech. ( Metallurgical Engineering ) Banaras Hindu University, India, 1987  A THESIS SUBMITTED IN PARTIAL FULFILMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF APPLIED SCIENCE in THE FACULTY OF GRADUATE STUDIES Department of Metals and Materials Engineering  We accept this thesis as conforming to the required standard  THE UNIVERSITY OF BRITISH COLUMBIA October 1991  © Sunil Kumar, 1991  In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission.  (Signature)  Department of ti tE TAG S & 1.0 A 7 CIQ. is^(^2 N 6) -  The University of British Columbia Vancouver, Canada  Date^  DE-6 (2/88)  /  9 c?/  ABSTRACT  Quality problems, such as cracks, rhomboidity and breakouts are often observed in continuously cast steel billets. Although information is available in the literature on the causes of, and solutions to, these problems, the transfer of this knowledge to operating personnel is not always an easy task. To address this issue, an expert system has been developed. The important features of knowledge engineering an expert system to diagnose quality problems in the continuous casting of steel billets are described in this thesis. The process used to extract knowledge from the Experts, the experiments employed to represent knowledge in different ways, the strategies adopted to accumulate evidence and the methodologies to present coherent recommendations to both experienced and inexperienced operating personnel are discussed. Expertise was derived from diverse sources. On the one hand fundamental knowledge about heat transfer, solidification and mechanical behaviour of steel was applied to identify the origin of a quality problem in the process; on the other, heuristic knowledge associated with billet characteristics, machine design and operating factors was required to focus on the problem causes. Integration of these two different types of knowledge was essential in developing a useful system. From the Experts' viewpoint, this exercise provided them with a formal representation of their knowledge to solve quality problems and identified new areas for future research. Moreover, the Experts actually generated new ideas about the domain which previously were unrecognized. For the users, this system serves two important functions - as a diagnostic tool for analysing quality problems and as a teaching tool for new operating personnel.  ii  TABLE OF CONTENTS ABSTRACT ^  ii  TABLE OF CONTENTS ^  iii  LIST OF TABLES ^  vii  LIST OF FIGURES ^  viii  LIST OF SYMBOLS ^  x  ACKNOWLEDGEMENTS ^  xi  CHAPTER 1 INTRODUCTION ^  1  CHAPTER 2 SCOPE AND OBJECTIVES ^  3  CHAPTER 3 EXPERT SYSTEMS TECHNOLOGY ^  5  3.1 Expert Systems - An overview ^  5  3.2 The Development Process ^  6  CHAPTER 4 KNOWLEDGE DOMAIN - Part I ^  12  4.1 Billet Casting Machine ^  12  4.2 Fundamental Knowledge on Billet Casting ^  15  4.2.1 Heat Transfer ^  15  4.2.1.1 Mould Heat Transfer ^  15  4.2.1.2 Spray Heat Transfer ^  19  4.2.1.3 Heat-transfer Mathematical Model ^  21  4.2.2 Mechanical Properties of Steel ^  21  4.2.3 Stress Generation during the Casting Process ^ 25  iii  4.2.4 Solidification of Steel ^  27  4.2.5 Operating Parameters ^  30  4.2.5.1 Mould ^  30  4.2.5.1.1 Mould Distortion ^  30  4.2.5.1.2 Mould Taper ^  39  4.2.5.1.3 Mould Oscillation ^  40  4.2.5.1.4 Mould Lubrication ^  42  4.2.5.2 Sprays ^  44  4.2.5.3 Liquid steel ^  47  4.2.5.4 Pinch-rolls ^  48  CHAPTER 5 KNOWLEDGE DOMAIN - Part II ^  49  5.1 Specificity Issues ^  49  5.2 Quality Problems ^  51  5.2.1 Cracks ^  51  5.2.1.1 Internal Cracks ^  53  5.2.1.1.1 Off-Corner Cracks ^  55  5.2.1.1.2 Midway Cracks ^  61  5.2.1.1.3 Diagonal Cracks ^  64  5.2.1.1.4 Centreline Cracks ^  66  5.2.1.1.5 Pinch-roll Cracks ^  68  5.2.1.2 Surface Cracks ^  69  5.2.1.2.1 Transverse Cracks (and depressions) ^ 69 5.2.1.2.2 Longitudinal Midface Cracks ^  75  5.2.1.2.3 Longitudinal Corner Cracks ^  76  iv  5.2.1.2.4 Craze Cracks ^  79  5.2.2 Rhomboidity ^  81  5.2.3 Breakouts ^  85  5.3 Other Observations ^  88  CHAPTER 6 KNOWLEDGE ENGINEERING ^ 6.1 Knowledge Acquisition ^  91 91  6.1.1 Identification/Definition of Problem Domain ^ 91 6.1.2 Background Knowledge on Continuous Casting ^ 91 6.1.3 Expert System Development ^  92  6.2 Knowledge Representation ^  100  6.2.1 Knowledge units ^  100  6.2.1.1 Keyword Triplets ^  102  6.2.1.2 Rules ^  104  6.2.1.3 Procedures ^  105  6.2.1.4 Meta-knowledge ^  105  6.2.2 Search Techniques and Conflict Resolution ^  107  6.2.3 Information used in the analysis ^  109  6.2.4 A novel inferencing strategy for the analysis of midway cracks ^ 111 6.3 Structure Of The Expert System ^  115  6.4 Justification of Knowledge Engineering Decisions ^ 117 CHAPTER 7 TESTING AND EVALUATION OF THE EXPERT SYSTEM ^ 123 7.1 Evaluation Procedure ^  123  7.2 System Operation - Results of the feedback ^  124  v  7.3 Analysis of User Feedback ^  127  7.3.1 System's Successes ^  127  7.3.2 System's Drawbacks ^  129  7.3.3 Areas of Future Expansion ^  130  7.4 Implementation Of Feedback ^  131  CHAPTER 8 CONSULTATION SESSIONS - CASE STUDIES ^ 133 CHAPTER 9 SUMMARY AND CONCLUSIONS ^  146  REFERENCES ^  148  APPENDIX A ^  156  APPENDIX B ^  157  APPENDIX C ^  158  APPENDIX D ^  168  APPENDIX E ^  174  GLOSSARY ^  178  vi  LIST OF TABLES Table 5.1^Internal Cracks in Billet Casting.  ^54  Table 5.2^Midway Cracks in Billet Casting.  ^55  Table 5.3^Surface Cracks in Billet Casting.  ^70  Table 5.4^Rhomboidity Problem in Billet Casting.  ^81  Table 5.5^Breakout Problem in Billet Casting.  ^86  Table 6.1^Initial scheme proposed by the Experts to analyse midway cracks originating in the radiation zone.  ^93  Table 6.2^Procedural knowledge units : Library functions in ^ COMDALE/X used for system development. 106 Table 7.1^Feedback implementation strategy.  ^132  Table 8.1^Final report.  ^138  vii  LIST OF FIGURES Fig. 3.1^Stages in expert system development.  ^9  Fig. 4.1^A schematic diagram of a billet casting machine.  ^12  Fig. 4.2^Billet casting mould [67]  ^14  Fig. 4.3^Schematic representation of thermal resistance encountered in the mould [15].  ^16  Fig. 4.4^Mould heat-fluxes for various steel grades [19].  ^18  Fig. 4.5^Variation of heat transfer coefficient versus billet surface temperature [23].  ^20  Fig. 4.6^Schematic representation of temperature zones of reduced hot ductility of steel related to embrittling mechanisms [42]. Fig. 4.7^Mechanism of oscillation mark formation [19].  ^22 ^41  Fig. 5.1^Schematic diagram of a rhomboid billet showing various types of cracks.  ^52  Fig. 5.2^Transverse section of a billet showing off-corner internal cracks.^56 Fig. 5.3^Schematic diagram showing generation of an internal crack due to bulging of the billet shell and a hinging action in the off-corner region [73]. ^58 Fig. 5.4^Schematic diagram illustrating the mechanism generating off-corner internal cracks due to bulging and hinging of the shell^59 [68]. Fig. 5.5^Transverse section of a billet showing midway cracks.  ^62  Fig. 5.6^Transverse section of a billet showing centreline cracks.  ^67  viii  Fig. 5.7^Photograph of a billet showing transverse depressions.  ^71  Fig. 5.8^Longitudinal section of a billet showing an internal crack below a depression  ^73  Fig. 5.9^Schematic diagram showing the formation of transverse depressions and cracks due to sticking or binding in the mould [20]. ^74 Fig. 5.10^Schematic diagram showing the formation of subsurface, longitudinal crack on diagonal at obtuse-angle corners of rhomboid billet [22]. Fig. 5.11^Photograph of craze cracks.  ^  78  ^80  Fig. 5.12^Schematic diagram showing a billet with non-uniform shell thickness being distorted into rhomboid shape by spray cooling  ^  [64•  83  Fig. 6.1^A typical output of the mathematical model showing predicted solid shell thickness and billet surface temperature profiles.  ^96  Fig. 6.2^Scheme for diagnosing quality problems requiring detailed mould and/or spray analysis.  ^98  Fig. 6.3^A general outline of mould-related factors contributing to quality problems in billet casting.  ^99  Fig. 6.4^Knowledge units available in COMDALE/X Development Tool. ^101 Fig. 6.5^A schematic diagram illustrating the calculation of degree of belief in a spray-related tensile strain problem from belief in high reheat in the radiation zone, high temperature and composition problems. ^113 Fig. 6.6^A schematic diagram illustrating the calculation of final belief in a spray-related tensile strain problem following the detection of a mould disorder.  ^114  Fig. 6.7^A flow-sheet representing the structure of the expert system  ix  ^116  LIST OF SYMBOLS tN^Negative strip time ( s ) n^3.142  f^Oscillation frequency ( Hz ) v,^Casting speed ( m/s )  S^Oscillation stroke length ( m )  ACKNOWLEDGEMENTS I would like to express my sincere gratitude to my supervisors Dr. Keith Brimacombe, Dr. Indira Samarasekera and Dr. John Meech, for providing excellent guidance, valuable help and encouragement throughout this work. It was indeed an intellectually stimulating experience for me. This work would not have been possible without the knowledge acquired by researchers in the field of continuous casting. I take this opportunity to thank all the individuals who contributed to this knowledge base. During the development, I had useful discussions with Ian Bakshi, Sanjay Chandra, Neil Walker and Bob Hapke. I thank them all. I am also grateful to Harold Ng for preparing the photographs used in this thesis. I am greatly indebted to Stelco Steel, Ivaco Rolling Mills, Slater Steel, Sidbec-Dosco Inc., Courtice Steel, Western Steel, Hatch Associates and the Natural Sciences and Engineering Research Council of Canada for support of this study. Gratitude also is expressed to Comdale Technologies Inc. for their assistance with the software. I am grateful to Dr. Amit Chatterjee, Assistant General Manager (Research), Tata Steel, for his support and encouragement. Sanjay Chandra and his family need special thanks for hosting me and helping me settle down comfortably in Vancouver. The two years were exciting thanks to the company of fellow graduate students at UBC and excellent room-mates at home. Lastly, I would like to thank my parents, for supporting my decision to pursue higher studies in Canada.  xi  CHAPTER 1 INTRODUCTION  The continuous casting of billets, slabs and blooms has been accepted worldwide by steelmakers in favour of the traditional ingot (casting and rolling) route. Today, emphasis on quality control is a prerequisite in the competitive environment to meet customer specifications and to reduce operating costs. The goal of high quality cannot be realized, however, without knowledgeable operating and maintenance personnel who understand their casting system. Similarly, the expertise these individuals possess should be considered invaluable and a method to retain and transfer this expertise within an organization can be of great advantage. There has been significant research into the process control of billet casting which has related operating parameters to quality problems. Although well-documented in the literature, transfer of this fundamental knowledge to operating personnel has been difficult despite the fact that numerous short courses on the subject have been given. There is a need then, for an expert system to guide operators in analysing quality-related problems and to provide them with a ready source of fundamental knowledge related to the operation of the casting machine This thesis describes such a system that has been developed with this purpose in mind. It is a diagnostic tool for analysing quality problems in continuously cast steel billets, specifically designed for use by operating personnel. The interface provided by the development tool, COMDALE/X, helps a user with explanations of terminology in the system, justification of questions asked and mapping of the strategy followed to arrive at final conclusions and recommendations. In this way, the system is able to train inexperienced operators.  1  The thesis consists of nine chapters. Chapter 2 outlines the scope and objectives of the project. Chapter 3 presents a brief overview of expert systems technology. Chapter 4 reviews the fundamental knowledge required to analyse quality problems in billet casting. The types of problems (cracks, rhomboidity and breakouts) are described in Chapter 5 with emphasis on  mechanisms of formation and influencing factors. Chapter 6 deals with the major stages of knowledge engineering involved in the development of this system - knowledge acquisition and representation; the structure of the expert system is described and benefits of the knowledge engineering methodology are outlined. This chapter also justifies various decisions made during structuring and coding of the system. Chapter 7 describes the procedure used to evaluate the expert system and the results of this exercise. Chapter 8 presents several case studies and discusses the consultation sessions. Chapter 9 presents the final conclusions and recommendations for future work.  2  CHAPTER 2 SCOPE AND OBJECTIVES  This research work was undertaken to develop an expert system to meet the following objectives: [1]  diagnosing quality problems in billet casting - cracks, shape defects and breakouts; and  [2]  training less experienced operating personnel The knowledge domain used to build the expert system has been generated over two decades  of research work and industrial experience in this field. Domain Experts^J.K.Brimacombe I.V.Samarasekera Knowledge Engineers ^S.Kumar J.A.Meech System Development Tool^COMDALE/X - version 3.0 Comdale Technologies Inc. 833, The Queensway, Toronto, Ontario, Canada M8Z 5Z1 The steps followed to achieve the above objectives are outlined below: (a) The knowledge base was reviewed and a practical understanding of the problem domain was established. Numerous consultation sessions with the Experts were held during the development of the system.  3  (b)  The knowledge base was structured. Flow-charts of the analysis schemes were constructed at various stages and modified as recommended by the Experts.  (c)  Coding of the knowledge was done using COMDALE/X as the development tool.  (d)  A two-dimensional heat-transfer mathematical model for billet solidification was developed to predict billet solid shell thickness and surface reheat temperature profiles. The alternating direction implicit fmite-difference method was used to solve the unsteady state heat-conduction equation.  (e)  The expert system was evaluated at five Canadian steel companies. The main driving force for this project was the need to ease the transfer of Experts'  knowledge to the industries. A system was desired that could be used as a consultant by the users for solving quality problems in billet casting. The initial development focussed on the major quality problems in billet casting- cracks, rhomboidity and breakouts.  4  CHAPTER 3 EXPERT SYSTEMS TECHNOLOGY  3.1 Expert Systems - An overview Expert Systems evolved from the field of Artificial Intelligence (A.I.) when researchers realized the power of "expert" or "knowledge-based" systems in solving problems. An Expert System is a computer program that embodies human knowledge and understanding about a particular domain and uses this information to simulate human thought processes in the analysis of a problem for end-users [1]. It consists of equations, models, rules of thumb, do's and don'ts, etc., all based on expert's knowledge required for solving a particular problem. Expert Systems are preferred over conventional computer programs for three main reasons: *  Knowledge is separated from the inference mechanism whereas in conventional programs data and reasoning are integrated.  *  Both quantitative and qualitative information can be handled; conventional programs can handle only quantitative information.  *  Incomplete and uncertain data can be accommodated in the operation whereas conventional programs simply cannot function without all the facts. Expert Systems are classified on the basis of the performed task. The major classes are  interpretation, diagnosis, monitoring, prediction, design, planning, and control [2].  5  3.2 The Development Process The power of Expert Systems derives from the knowledge they possess, not from the particular formalisms and inference schemes they employ [3]. Three main reasons can be identified for the need to focus on knowledge as the primary component in the development process [2]: •  Many problems do not have well-defined algorithmic solutions. Planning, reasoning and diagnosis are examples where the solutions can be extremely complex making it difficult to define precisely a rigorous scheme for analysis.  •  Human experts have achieved tremendous success with their knowledge in solving problems. If a computer program can be designed that utilizes the problem-solving abilities of the experts, then it is possible to attain high performance levels in making decisions comparable to the experts, with the help of computers. A number of systems based on expert knowledge have proven to be very successful [2]. These areas include mineral prospecting [4], medical diagnosis [3, 5 - 8] and computer configuration [9].  •  Expert knowledge is often a scarce resource and its capture can be of great benefit to an organization or a group. Once this knowledge has been extracted from the experts and the information stored in a computer, lesser experts or novices can have easy access to it. This enhances transfer of high-level knowledge to users who require it. Researchers in the field of expert system development have coined the term "knowledge  engineering" to describe their activities and the title "knowledge engineer" for a worker involved in this area. Knowledge engineering investigates methodologies and techniques necessary to code expert knowledge (or expertise) into computer programs for solving problems. The expertise generally consists of knowledge and skills needed for solving problems related to a  6  particular domain. Knowledge is made up of two kinds: fundamental and experiential [2]. Fundamental knowledge consists of published information, facts and theories in textbooks and journals. Experiential knowledge is possessed by the human experts and involves "heuristics" - largely comprised of "rules-of-thumb" or "do's" and "don'ts". This knowledge enables human experts to make intelligent guesses, wherever necessary, and to deal effectively with incomplete, incorrect or conflicting information during problem solving. An important task in knowledge engineering is to integrate these two very diverse sources of expertise. In addition to expert knowledge, these systems have two other important components: an inference engine and a user interface. The inference engine is the thinking part of the expert system. It consists of an interpreter to apply the rules in the knowledge base, a scheduler to control the order of rule processing, a knowledge accumulator to adjust belief in previous conclusions when new data become known, and a justifier to rationalize and explain the system's behaviour [2]. The user interface allows communication between the user and the computer. The user can ask "why" a question is being asked or may need to know more about some fact in the knowledge base. In addition, when the final conclusions are presented, there may be a need to obtain justification for each piece of advice. With the help of this facility, expert systems can help in the training of an inexperienced user. There are numerous ways in which data can be requested: simple Boolean Yes/No, multiple choice selections, numerical inputs, string inputs, mutually exclusive concepts or form-field inputs. The different techniques must be carefully chosen to ease the process of data input for different types of users. Similarly, the style of data-output presentation must be flexible. Some data must be presented as text messages and others as icons or graphs. The ability to customize output reports is an important feature of a truly intelligent system.  7  Stages in Expert System Development The development of expert systems can be broken down into the following principal stages: definition of problem domain, acquisition of background knowldege about the domain, development of the expert system, testing and evaluation, delivery and acceptance, and system maintenance. Figure 3.1 shows the different stages involved and the various linkages. Definition of Problem Domain: During this stage a problem is identified and its scope is defined. In addition, participants (experts, knowledge engineer and end-users) are also identified.  Acquisition of Domain Knowledge: This involves the extraction of experts' knowledge. The experts and the knowledge engineers identify key concepts, relationships and information-flow patterns needed for problem solving. Subtasks, strategies and constraints involved in the analysis are also outlined.  Development of the Expert System: Important decisions related to knowledge representation are made during this stage. A programming language or development tool/shell is selected for the project and evaluated. The knowledge acquired is programmed into the computer and a prototype system is developed. The knowledge base is refined and expanded.  Testing and Evaluation : Expert Systems, like any other conventional software, has to be tested for accuracy and user-friendliness. In fact, the need for validation is much greater in the case of expert systems since they are based on far less certain theories such as heuristics and rules of thumb, all of which have been utilized in problem solving but were never subjected to any detailed examination. Expert systems are evaluated by the Knowledge Engineers, the Domain Experts and the Users. During development, the system is constantly reviewed and tested by the Knowledge Engineers. Evaluation by the Domain Experts determine the accuracy  8  PROBLEM DEFINITION  y  ACQUISITION OF DOMAIN KNOWLEDGE  Redefinition  Reformulation  ^ Refinement ^EXPERT SYSTEM ^ ^DEVELOPMENT Redesign  Y SYSTEM EVALUATION  Feedback  SYSTEM DELIVERY AND ACCEPTANCE  SYSTEM MAINTENANCE  Feedback  Fig. 3.1^Stages in expert system development.  9  of the knowledge used and the advice/conclusions provided by the system whereas those by the users helps to assess the system with respect to its user-friendliness, efficiency, speed and reliability.  Delivery and Acceptance : The system is delivered to the users and installed for actual use. User groups examine and test the system, often discovering "failures" or "bugs" or "areas of improvement" to be modified by the developers.  System Maintenance : The system is updated and modified according to the future requirements. New knowledge is added as required. There are no clear-cut boundaries between the above stages, which are closely linked and heavily dependent on each other. The term, "knowledge acquisition", is used by workers in this area to characterize the complex processes involved in the development process. Knowledge acquisition has been defined as the transfer and transformation of problem-solving expertise from some knowledge source to a computer program [2]. It is an iterative process where each of the above stages interact with each other. The knowledge engineer redesigns, reformats, reformulates and refines the system in an on-going, never-ending process. Figure 3.1 shows the different stages and the various interactions involved. The last two stages are concerned with installation of the system at the place of the user and its maintenance. It must be noted that once the system is put into practice by the users, the process may or may not continue in the maintenance mode.  Knowledge representation methods Knowledge representation involves the development of structures that assist in coding knowledge into the system so that intelligent behaviour is exhibited. The main knowledge  10  items used in this expert system can be grouped into "structural", "procedural", "external" and "meta-knowledge" [10]. Structural knowledge consists of rules (statements and procedures), and facts (classes, objects and their description). These units provide the various links among  different knowledge elements as the system searches for final conclusions. Procedural knowledge units are employed to direct the inferencing process for a number of knowledge-specific tasks. External knowledge provides an interface for the system to interact with external programs. Meta-knowledge or "knowledge about knowledge" gives useful and appropriate explanations and justifications to the users. Some workers in this field, for example Turban [11], use the term "meta-knowledge" to distinguish rule structures that help direct the thought process from rules that deal specifically about the domain. In this work, the term  "procedural knowledge" is used for this while "meta-knowledge" is reserved for describing user-support information supplied by the system to assist users.  Future trends Expert systems technology is expanding rapidly and numerous applications have been developed in the last decade. Commercialization of expert system development tools has further enhanced the growth of this technology. Future advancements include development of automated methods for knowledge acquisition and of learning systems (neural networks). These improvements will considerably ease the knowledge engineering process.  11  CHAPTER 4 KNOWLEDGE DOMAIN - Part I  4.1 Billet Casting Machine Figure 4.1 is a schematic diagram of a billet casting machine. The important units of a billet casting machine are tundish, copper mould, water sprays, pinch rolls and shear/torch.  TUNDISH  MOULD  4 SPRAYS ^W L 7^PINCH-ROLLS^SHEAR/TORCH IL  Fig 4.1 A schematic diagram of a billet casting machine.  12  Tundish - Molten steel from the steelmaking shop is brought to the casting shop in a refractory lined ladle. Steel from the ladle is transferred to another vessel called a "tundish" which is an important unit in the casting machine. Apart from acting as a reservoir of liquid steel, it helps in controlling metal flow and also in floating out inclusions. Details have been published earlier [12 - 14].  Mould - Figure 4.2 shows a typical billet casting mould [15]. The mould generally consists of two types- vertical and curved in which a copper tube is held concentrically inside a steel jacket. Cooling water flows in an upward direction through the gap between the mould tube and the steel liner, steel spacers or set screws are used to maintain a uniform gap. The annulus is always full of water. The mould is oscillated either mechanically or hydraulically, with the help of a cam arrangement to provide the necessary stripping action for the newly formed solid shell. The mould is lubricated with the help of oil (or more recently, powders). Oil is pumped and distributed through a splitter to channels in an oiling plate from which it is transmitted uniformly around the mould wall. In case of powders, the lubricant is fed manually by operators and a submerged entry nozzle to deliver the steel to the mould.  Sprays - Below the mould the strand is cooled by banks of pressure-atomized water sprays. Water from a common header is sprayed onto the strand via nozzles.  Pinch rolls - Pinch rolls are used to provide the pulling action necessary for the withdrawal of the strand from the mould.  ,Shear/Torch - A torch (or shears) is used to cut the strand into the desired lengths.  13  i Mould 2 Steel jacket 3 Housing 4 Support plate 5 Lubricator plate 6 Cover plate 7 Water channel  Fig 4.2 Billet casting mould [67].  14  4.2 Fundamental Knowledge of Billet Casting In billet casting, the generation of quality problems stems from the nature of the process where rapid cooling of the steel results in steep temperature gradients in the solid shell that change rapidly and generate thermal strains as the shell differentially expands or contracts. These problems have been linked unequivocally to the operation and design of the casting machine and also to the mechanical behaviour of steel at continuous casting temperatures. Therefore, at the heart of most quality problems in this process is the nature of cooling (intensity and uniformity) and at the same time, the mould/shell interaction. The analysis of quality problems in billet casting, like in any other process, requires a thorough understanding of the fundamental principles governing it. In this case, it is essential to study the basics of heat transfer, mechanical properties of steel at high temperatures, sources of stress generation in the machine (thermal and mechanical), and solidification of the steel [121 Once possible mechanisms are identified, links between quality problems and related operating and/or design variables can be established. Then, these quality problems can be eliminated by altering related process parameters or design conditions, identified as important in the analysis.  4.2.1 Heat Transfer 4.2.1.1 Mould Heat Transfer In the mould, heat is transferred from the liquid steel to the cooling water through four media namely, the solidfying shell, the air gap between the mould and the strand,  15  the mould wall, and the mould / cooling water interface. Figure 4.3 is a schematic representation of the thermal resistance encountered in the mould [15]. The air gap constitutes the largest resistance to heat flow [16]. In the upper region of the mould, conduction through the air gap constitutes nearly 84 percent of the total and is the dominant mode of heat transfer whereas in the lower part of the mould, conduction through the solid shell is the major component [15]. The amount of heat extracted is inversely proportional to the width in each case, in accordance with Fourier's Law.  I^1^1  I^1^I 1^I^I I^I^1 I water I^I^I I^I^I 1I^1 1^i^ I I I  gap  Fig 4.3 Schematic representation of thermal resistance encountered in the mould[l 5].  16  The mould/shell air gap is a complex function of design and operating variables namely, shell shrinkage, mould distortion, oscillation marks and ferro static pressure. The gap also varies in width in both the longitudinal and transverse directions [17]. Thus, it is difficult to predict the width of the air gap accurately as a function of distance below the meniscus. For these reasons, measurement of heat-transfer boundary conditions at the mould/billet interface is necessary. The heat fluxes at the billet/mould interface have been calculated from time-averaged responses of thermocouples embedded in the mould wall a set distance away from the hot face during normal casting [18]. Details about the scheme of calculation of the heat-extraction profiles for single- and double-tapered mould tubes have been published earlier [19 - 22]. In addition, heat transfer in the mould is extremely sensitive to the shape of the copper tube which during operation changes due to differential heating and expansion. Samarasekera et al.[22], with the aid of a finite element stress-strain model, have studied the thermo-mechanical behaviour of the mould during operation. Heat transfer in the mould is influenced by operating parameters such as steel composition, superheat and cooling water velocity [19]. The effect of steel carbon content on mould heat transfer is shown in Figure 4.4. In the upper regions of the mould, the magnitude of heat flux decreases with decreasing carbon level in the 0.14-0.36 percent range. The peak and the average mould heat fluxes are a minimum for 0.1 percent carbon steels but increases with increasing carbon content up to 0.20 percent and are fairly constant for higher carbon grades. This behaviour has been attributed to the peritectic 8 — yreaction. In the range of 0.016 to 0.028 percent phosphorus, for medium carbon steels, higher phosphorus levels lead to a reduction in heat transfer especially at the meniscus. A greater reduction in the upper mould heat-flux is seen with increasing sulphur levels. This  17  behaviour is attributed to the deleterious effect of sulphur and phosphorus on the mechanical properties of steel which influences the resistance of the steel to deformation close to the meniscus during mould oscillation.  5000 se., No  1  4  C (V  P (%)  L44 LK)  S  Ito  So MI  AT  ra  K.  (m/s1  V ImAr41  — 24268  014  0020 0 55 0039 014  19  97  24  - - -^24259 — - 24248  046  0024 O51 0022 100  0033 042 0036 0 48  (28)  028  23  036  0026 086 0028 012  (32)  92 93 97  — 24254  (27)  49 21  3000  41 Z r 2000 --  1000  0  50^00^50^100 Time ( s)  150  200  Fig 4.4 Axial profiles of mould heat flux for various steel grades [19].  18  The effect of mould cooling water velocity is quite significant [19]. For low carbon grades, the magnitude of peak heat flux is maximum at a water velocity of 4.0 m/s whereas for medium carbon ranges, the peak is observed to be a maximum at a velocity of 5.0 m/s. Heat transfer in the mould is also a strong function of oscillation parameters [19]. In the upper part of the mould, the presence of oscillation marks on the billet surface influences the width of the mould/shell air gap. Hence, deep oscillation marks reduce heat extraction considerably in the mould. Also, varying depths of oscillation marks across the billet surface leads to non-uniform heat transfer conditions in the mould.  4.2.1.2 Spray Heat Transfer Cooling of the strand beneath the mould is achieved by banks of pressure-atomized water sprays. Figure 4.5 shows the variation in the heat-transfer coefficient with surface temperature during spray cooling for a 1/4 GG10 nozzle operating at 90 psi and set back approximately 200 nun from the billet surface [23]. As shown in the figure, above 550°C, the heat-transfer coefficient changes very little with temperature, which is characteristic of film boiling [24]. At lower temperatures, the heat-transfer coefficient increases sharply as the heat transfer mechanism changes from film boiling to transition boiling (breakdown of the steam barrier). The critical temperature at which this occurs is called the Liedenfrost temperature which increases with increasing water flux [25]. Numerous laboratory experiments and in-plant trials have established relationships between the spray heat-transfer coefficient and spray water flux [23, 25 - 39].  19  1.0  Nozzle -  ,4 GG 10  1  Spray Distance - 8 In. Spray Pressure - 90 psig j„, .......... ,  ** N..  Critical Temperature .2 ..""ll' 4  .•  o o  1^I  ^  'Nfir•r •frir  4re"..411...w**1....*•••P ir*"  J  200^400^600^800^1000 1200 1400 1600 1800 2000 SURFACE TEMPERATURE, uF  Fig 4.5 Variation of heat transfer coefficient with billet surface temperature [23].  20  4.2.1.3 Heat-Transfer Mathematical Model In the continuous casting of steel billets, the mathematical model employed for predicting temperature distribution, the solid shell profile and the liquid pool depth is based on the unsteady-state, two-dimensional heat conduction equation:  a(  aT) a (al=pCpaT aT +  k ax  (1)  Details of the mathematical model i.e. assumptions, formulation scheme, boundary conditions and the predictions have been adequately described in earlier works [41,42]. Correlations which relate the working mould length, casting speed and shell thickness at the mould exit have been established by Brimacombe [40]. In the same publication, design curves of spray length, spray heat-transfer coefficient and water flux distribution were presented. The boundary conditions for the model are derived from the heat-transfer measurements made in the mould and sprays.  4.2.2 Mechanical Properties of Steel There are three distinct high-temperature zones where steel exhibits low strength and poor ductility and, therefore, is highly susceptible to cracking. Figure 4.6 is a schematic representation of the reduced hot ductility of steel based on a review of the literature by Thomas et al. [42]. As shown in the diagram, a high-temperature zone of low ductility occurs within 30-50 °C of the solidus temperature, the intermediate temperature zone of low ductility lies between the Ar3 temperature and about 1200 °C while a low temperature zone extends below the Ar3 temperature. In the continuous casting of steel billets, only the high-temperature zone is critical from the viewpoint of crack formation.  21  Crock  600  ^  900  ^  1200  ^  1500  Temperature (°C)  Fig 4.6 Schematic representation of temperature zones of reduced hot ductility of steel related to embrittling mechanisms [42].  22  High Temperature Zone : 1340 ° C to Solidus At temperatures well above 1340 °C, there is significant deterioration of the strength and ductility of steel. In a review conducted by Brimacombe and Sorimachi [43], strength data measured at various strain rates in four different studies were extrapolated to a common strain rate of 10 -3 s"' (commonly obtained in the continuous casting process) and were compared. The strain-to-fracture of steel in the high temperature zone of low ductility appears to be of the order of 0.20 to 0.30 percent according to Vom Ende and Vogt [44]. The low strength and poor ductility of steel close to the solidification front is due to the presence of liquid films in the interdendritic regions which remain molten until temperatures well below the solidus are reached [45]. These liquid films are rich in solutes such as sulphur and phosphorus which have segregation coefficients (CJC 1) less than unity. Scanning electron micrographs of the interior surface of a crack that originated close to the solidus temperature have shown a smooth surface, with no signs of solid fracture, indicating the presence of a liquid film at the time of crack formation [46].  Effect of Phosphorus : Sopher [47] studied the influence of phosphorus on fracture strength and ductility of SAE 4340 steel at different temperatures. It was found that steel with a phosphorus concentration of 0.039 percent has drastically reduced strength and ductility as compared to steel containing 0.017 percent phosphorus.  Effect of sulphur : The effect of sulphur on the strain-to-fracture and the UTS of steel near the solidus temperature was studied by Morozenskii et al.[48]. Two points can be noted from their study. Firstly, the presence of sulphur in steel adversely affects both strain-to-fracture and the UTS; and secondly, an increase in Mn/S ratio in steel improves the strain to fracture but has no significant effect on the UTS. Similar results were obtained by Kinoshita and  23  Kuroki [49] in another study related to steel at 1350 °C. This study also investigated the effect of Mn concentration on the proportion of MnS and the FeS in sulphide inclusions. It was shown that FeS content decreased with increasing Mn concentration. The authors also found that in 0.2 percent C steel, for Mn concentrations less than 0.7 percent, sulphides have a tendency to exist in liquid form below the solidification temperature of 1480 °C. A Mn/S ratio greater than 20 is effective in minimizing the cracking tendency by preventing the liquid film formation [48].  Effect of carbon : The effect of carbon on mechanical properties was studied by Morozenskii et al.[48]. The strain-to-fracture, S E , and its plastic component, 5, 1„stic , were found to be minimum for steel containing 0.17 to 0.20 percent C. Similar results were reported by Guessier and Castro [50].  Effect of tin and copper : Increased contents of Sn[51] and Cu[52,53] deleteriously affect the ductility of steel. This behaviour is attributed to the presence of the above mentioned low-melting impurities at the grain boundaries. Failure occurs due to liquid metal embrittlement. The high temperature zone of low ductility has also been examined by numerous other authors [55 - 60]. It has been established that this zone of low ductility is operative at temperatures within 30 to 70°C of the solidus temperature, and the associated strain-to-fracture is virtually zero [61]. Thus, the presence of even a small tensile strain in the solidfying shell in this temperature region will initiate a crack which will propagate outward from the solidification front between dendrites.  24  4.2.3 Stress Generation during the Casting Process During the casting process, the strand is subjected to varying thermal conditions and changing mechanical loading which generate stresses and strain in the strand. In order for a crack to originate at a given location, two conditions must be present [12]: (i) the stress must be tensile in nature (ii) the fracture strain must be exceeded  Thermal Stresses/Strains - The rapid rate of cooling in continuous casting, results in steep temperature gradients in the solid shell that can change rapidly and generate thermal strains as the shell expands or contracts differentially. Thermal stresses can be generated either when free expansion or contraction of the material is constrained or when the thermal gradient in the material is non-linear and changing [12]. In the continuous casting of steel, the thermal stress condition in the strand can be approximated to that of a generalized plane-strain condition [12]. In the process, there is some allowance for expansion in the casting direction which helps minimize generation of longitudinal stresses and strains. Further, thermal gradients in the longitudinal direction are quite small and unlikely to cause any cracking. Thus, transverse cracks, which require a longitudinal stress component, simply cannot originate due to adverse thermal gradients in the billet but instead are generated mechanically. In the transverse plane however, conditions are present which generate high thermal stresses. The free thermal expansion is restrained and also the temperature gradients are steep and frequently non-linear [12]. In addition, sudden changes in heat extraction rates, which could occur at the boundary between any two consecutive cooling zones, may cause the thermal gradients to shift, particularly at the surface; the resulting expansion or  25  contraction of this region also generates stresses in the transverse plane. If this strain is tensile in nature and its value exceeds the strain-to-fracture in a zone of low ductility, particularly close to the solidification front, then longitudinal cracks may form. Reheating of the surface of the billet below the mould or the sprays causes expansion of the surface layers which imposes tensile stresses at the solidification front where the steel has the lowest ductility [62]. Non-uniform cooling in the mould and the sprays imposes tensile stresses at the obtuse angle corners or off-corners regions of billets [22,61]. Also, during binding at the corners of the billet in the mould, excessive cooling at some locations of a face may cause localized tension [12]. Van Drunen et al.[46] have suggested another source of thermal stress generation. At the point of complete solidification, when the final trace of latent heat has been extracted from the centre region, the drop in centreline temperature is considerably more rapid than the decrease in surface temperature with the result that the central region contracts more than the surface. The centre, however, is constrained from contraction by the surrounding colder steel and is therefore, put into tension. Depending on the magnitude of the stress and the  strength of the steel, centreline cracks may form. Segregation of elements in the central region of the billet adversely affects the strength and ductility of steel and increases the susceptibility to cracking. Mechanical Stresses - Sticking in the mould due to inadequate lubrication and/or oscillation conditions or binding of the strand in the mould due to excessive taper, imposes resistance to smooth withdrawal of the strand. This  generates axial tensile stresses and strains that  concentrate on locally thin regions of the shell and may cause transverse depressions with transverse cracks [20]. The nature of this transverse crack, whether it appears as an internal  26  crack or a surface crack, depends on the magnitude of the stress and the shell thickness at that location. If the taper is insufficient in the lower portion of the mould, bulging of the solid shell can occur and may lead to hinging of the strand at the off-corner sites which generates a tensile strain at the solidification front and internal cracks [63]. Straightening or bending operations, when carried out on a section with a liquid centre or when the centre is solid but within 50 to 70 °C of the solidus temperature, may also lead to the generation of a tensile stress at the solidification front [43]. Tensile stresses are also generated when excessive pinch-roll pressures are applied on the strand with a liquid core [43].  4.2.4 Solidification of Steel In continuous casting, two aspects of solidification are important from the viewpoint of quality: cast structure (columnar / equiaxed) and the nature of shell growth around the liquid pool.  Cast Structure - Cast structure influences internal crack formation and macrosegregation. Dahl and Hengstenberg have proposed a mechanism of crack formation based on dendritic separation in the columnar zone [63]. The columnar structure is more susceptible to crack formation as compared to the equiaxed structure because it provides an easy path for crack propagation. There are a number of operating parameters that influence the size of the central equiaxed zone relative to that of the surrounding columnar zone in a continuously cast billets. These factors are superheat, steel composition, section size, casting speed and machine design (straight or curved mould-tubes).  Effect of superheat: The effect of superheat (usually measured in the tundish) on the cast structure was investigated by Van Drunen et. al. [46]. It was found that the columnar zone  27  is favoured at the expense of the equiaxed zone with increasing steel temperature (liquidus + superheat). A superheat level of 30 °C or below is desired to maximize the equiaxed structure which makes the strand more resistant to cracking and minimizes macrosegregation.  Effect of carbon conteni : The effect of carbon content on the length of the columnar zone in a continuously cast billet has been examined by Bommaraju et al.[64]. Equiaxed structure is favoured in the medium-carbon range of 0.17 to 0.38 percent carbon. Increasing the phosphorus content from 0.008 to 0.02 percent in 0.13 to 0.20 percent C steel also causes the columnar zone to shrink [64].  effect of section size : A large section size favours the growth of an equiaxed zone [12]. Hence, crack formation by dendritic separation in the columnar zone is less likely to be serious as compared to smaller sizes.  Effect of machine design : The design of the casting machine (i.e. straight or curved) also  influences the cast structure. Van Drunen et al.[46] observed that the length of the columnar zone adjacent to the inner-radius face was greater than that adjacent to the outside-radius face. Lait et al.[65] postulated that this non-symmetrical structure results from the settling of crystallites (dendritic debris) against the solidification front advancing from the outside radius face. In this way, the crystallites interfere and prevent further growth of the columnar grains. On the other hand, the columnar structure adjacent to the inside radius experiences unimpeded growth.  Solid Shell Formation - The initial shell growth in the mould is governed by the mould  heat-flux distribution. Watanabe et al.[16] observed that in the upper part of the mould, the mould/shell air gap constitutes a large fraction of the total thermal resistance encountered. Thus, the nature of shell growth in the upper part of the mould is largely dependent on the  28  dynamics of air gap formation. Hence, variables influencing air-gap formation, and therefore the heat-transfer in the mould, directly impact on the nature of shell growth. In the lower part of the mould and further down the casting machine, after a reasonable amount of solidification has occurred, conduction of heat through the solid steel shell is critical in controlling heat transfer and solid shell formation [15]. Two aspects of solid shell formation that are important from the viewpoint of billet quality, are the magnitude of the shell thickness and the uniformity of the shell around the liquid core [12]. The solid shell at the mould exit must be strong enough to counteract the ferrostatic pressure exerted by the liquid core. If the shell thickness is inadequate and the mould taper insufficient, bulging of the solid shell in the mould may occur [64]. Bulging can also occur near the mould exit, before the spray zone, if the shell is thin. In an adverse situation, a thin shell at the mould exit may be incapable of withstanding the ferrostatic pressure and could rupture. Breakouts can also occur if there is a local reduction in shell thickness due to surface depressions, cracks or deep oscillation marks. Therefore, uniformity of the shell around the liquid core is also a critical parameter for quality and operations. Non-uniform shell thickness may result from deep and non-uniform oscillation marks, wrinkling of the surface especially in low-carbon grades, or the presence of surface depressions. Singh and Blazek [66] showed the influence of the carbon content of steel on the uniformity of shell growth. Non-uniformly deep oscillation marks on the billet surface locally reduce the heat-extraction rate in the mould, relative to other areas around the periphery of the billet, as a result of an increased air gap width [64]. Thus, the billet leaves the mould with a non-uniform shell. The severity of this non-uniformity depends on the extent of variation in the heat-extraction rate around the billet periphery in the mould. Links will be established later between this event and rhomboidity.  29  4.2.5 Operating Parameters  4.2.5.1 Mould Mould/shell interaction at the meniscus is by far the most important phenomenon occurring in the mould as it directly impacts on the formation of oscillation marks which affect the width of the mould/shell air-gap and, hence, influence the mould heat-extraction [19]. Mould/shell interaction away from the meniscus is also important because it influences binding in the mould which occurs whenever the size of the billet exceeds the mould tube dimensions. A mould/shell gap that is too large, is also not desirable. In the upper part of the mould, a large mould/shell gap leads to significant reduction in the heat extraction capability of the mould while in the lower part of the mould, coupled with a thin solid shell, it causes bulging and cracking at off-corner locations [64].  4.2.5.1.1 Mould Distortion Mould-tube distortion affects the shape of the mould during operation and is directly linked to the thermo-mechanical behaviour of the mould [21]. As shown in Figure 4.4, the heat flux is a maximum very close to the meniscus due to the small air gap across which heat flows essentially by conduction. Below the meniscus, the heat flux falls as the gap increases in width due to steel shrinkage. The second maximum seen in the mould heat-flux profile is due to mould distortion [19]. Above the meniscus where the steel does not contact the mould, the heat flux declines rapidly and causes relatively steep temperature gradients to be set up in the axial as well as the through-thickness direction in the mould wall.  30  In the study by Samarasekera and Brimacombe [22], it was found that although the maximum heat flux is at the meniscus, the peak temperature is not. The maximum temperature is observed at a location slightly below the meniscus. This behaviour is attributed to significant heat conduction vertically up the mould wall. The axial temperature distribution observed in the mould determines the expansion of the mould-tube. The hottest region of the mould wall expands the most and causes bulging of the mould tube. The peak bulge in the mould appears about 75 mm below the meniscus. As a result, the mould has a negative taper above the peak bulge and a positive taper below the maximum distortion point. The second peak in the mould heat-flux profile, according to S amarasekera and Brimacombe, is due to twisting of the mould as a result of non-uniform mould tube support [22].  Classification of Mould Distortion - Profiles of distance between opposite faces for a large number of used moulds were examined by Brimacombe et al.[67]. Used moulds  were classified into three main categories based on the extent of permanent deformation observed: Minimal distortion : In this category, moulds exhibited relatively little distortion in the  meniscus area. Small permanent deformation, less than 0.50 mm or less between the opposite faces, was observed. It was found that these moulds were operated with deep meniscus levels, high water velocities and good water quality. Outward bulging near the meniscus : Moulds in this group exhibited an outward bulge  toward the water channel, very close to the meniscus region. Permanent distortion was within 0.50 mm and 2.00 mm between the opposite faces; the constrained sides were deformed more than the unconstrained sides because the former are constrained against  31  expansion and undergo greater plastic strains, and therefore have larger permanent distortion. These moulds had been operated with shallow meniscus levels, poor water quality or low water velocity.  Abnormal / severe distortion : In this case, moulds showed permanent deformation which exceeded a value of 2.0 mm between opposite faces. This strongly indicated problems such as poor water quality, low water velocity, shallow meniscus and poor centering of the mould tube in the liner. Regular distortion measurements of both new and used mould tubes is important in the analysis of mould disorders. Measurements made on used moulds reflect the nature of operating parameters employed in the process whereas, checks on the dimensions of new moulds are necessary to ensure that mould-tube specifications are being met.  Factors Influencing Mould Distortion (i) Cooling Water Quality : Water quality strongly influences the thermal field and the  distortion of the mould-tube. This is linked to the deposition of scale from the cooling water on to the cold face of the mould-tube. The presence of a layer of scale introduces a large resistance to heat flow which raises the mould temperature [21]. It is clear that even with a relatively thin layer of deposit, the local cold-face temperature is hotter than the hot-face temperature in the absence of scale [21]. Fouling nearly doubles the bulging of the mould-tube below the meniscus from 0.20 mm to 0.35 mm [21]. The presence of scale deposits on the cold face of the mould is a clear indication of poor water quality. The severity of fouling was quantified by Brimacombe et al.[67]  32  with an index, F, which was defined on the basis of the ratio of the deposit length, 1, to the effective mould length, L. The higher the value of F, the poorer is the water quality. Poor water quality may be due to a high hardness level in the cooling water or due to the presence of corrosion products, oil, grease or as a result of biological fouling stemming from algae and moss. Inferences on the source of the contamination may be drawn from the colour and properties of the scale deposit. A red deposit, similar to rust, suggests oxide of iron and corrosion in the water delivery system. A black deposit may be due to magnetite (Fe3 04) or carbonaceous matter such as oil, grease or biological matter. The presence of iron oxide suggests corrosion in the water delivery system while carbon-rich matter indicates petroleum products or biological fouling. With the help of chemical analysis, it is possible to distinguish between the two contaminants. In addition, if the loss on ignition of the scale deposit is large, the scale is expected to be rich in carbon. The black deposit can be tested with an ordinary magnet to check for Fe 304. A light-coloured scale deposit suggests excessive hardness in the cooling water. (ii) Cooling Water Velocity : Cooling water velocity is directly linked to the thermal distortion of the mould [21]. A high water velocity is recommended to prevent uncontrolled nucleate boiling in the cooling channel. It has been shown that nucleate boiling, which occurs on the hottest region of the mould very close to the meniscus, could be both asynchronous and intermittent. Water velocity in excess of 12 m/s is required to prevent boiling [68]. In addition, it is also important that the mould cooling water velocity is uniform around the periphery of the mould.  33  (ii) Mould-Tube Alignment and Tolerances : It is not only important to maintain a high cooling water velocity but also a uniform water flow around the periphery of the mould. If the flow is non-uniform then one or more walls of the mould may operate hotter than the others and cause non-uniform distortion of the mould tube which in turn may lead to uneven shell growth in the billet. Hence, it is essential to ensure that the width of the annulus between the steel jacket and the mould tube, through which the cooling water flows, is constant around the mould periphery. The water gap is typically 4.8 mm (3/16 inch), although some plants employ gaps of 3.2 mm (1/8 inch). It is important to maintain a close tolerance on the outside dimensions of the water jacket due to the small width of the water gap [68]. For example, in the case of 4.8 mm wide gaps, a loss of dimension of 1 mm can change the water velocity by up to 20 percent. In addition, the tolerances on the outside dimension of the mould wall and the inside dimension of the water jacket are also important. Typically values for each is about 0.5 mm. On its own, this may give rise to a variation of 1.0 mm on different faces. In addition, the mould tube could translate a further 1 mm from the concentric position. because of the tolerances on the support plates and flanges. In the worst case, the water gap may differ by as much as 2 mm (roughly 40 percent of gap width) between the opposite faces of the mould. Local increase in the gap width may also occur due to non-uniform distortion or bulging or misalignment or even off-centring of the mould tube. An increase in the gap width leads to local reduction in water velocity which causes a marked increase in the mould temperature [21]. Frequently, at the corners of the water jacket, the gap is not equal to the gap at the midface [68]. This occurs when the jacket has square corners which is typical of jackets fabricated from steel plates which are welded at the corners. When the two are  34  assembled, the water gap in the corners is larger than that in the midface [68]. A larger gap at the corners leads to a reduction in the resistance to water flow through it. This causes the water to flow preferentially up the corners thereby reducing the water velocity on the faces. A large gap at the corners may also isolate one face from another leading to severe non-uniformity in the heat transfer conditions around the mould periphery [68]. The material of the water jacket is also important [68]. A mild steel jacket corrodes easily and affects the tolerances. A material, such as stainless steel, that is resistant to corrosion is desired to maintain the dimenional integrity of the water jacket during the operation. (iii) Mould-wall Thickness : Thick mould walls give rise to a smaller negative taper at the meniscus [68]. This is because thicker walls have larger thermal gradients and hence bending of the wall is a significant mode of deformation. Thus, the midface bows inwards relative to the corners at the meniscus. The effect of thick wall mould tubes have been observed on the depth of oscillation marks [68]. It has been found that thicker walled tubes generate shallower oscillation marks as compared to thinner ones. This is expected because smaller negative taper of the thick wall tube at the meniscus results in reduced mould/shell interaction during negative strip. A thick wall mould tube also enhances heat transfer in the meniscus region [68]. It was found that the heat flux to the thick wall tube is greater and remains at a higher level close to the meniscus as compared to thin wall tube. The rapid drop in the heat extraction rate for the thin wall tube is associated with greater negative taper which leads to quicker formation of a wider mould/shell gap below the meniscus. A thicker mould wall is desired even for small  35  billets. From the viewpoint of mould distortion, wall thickness must be greater than one-tenth of the largest dimension of the billet cross-section and should not be less than 12.5 mm. (iv) Mould-Copper : The selection of a copper grade for the mould-tube depends on three critical parameters, namely, thermal conductivity, yield stress and softening resistance [68]. A high thermal conductivity is required to minimize mould temperatures and resulting thermal distortion. A high yield stress and an adequate softening resistance are necessary to minimize permanent distortion of the mould tube. Pure copper is better than other metals for the mould owing to its high thermal conductivity. Unfortunately, it lacks the softening resistance and cannot maintain its shape at peak operating temperatures for more than one or two heats. Therefore, it is necessary to use an alloy of copper having high yield strength and good softening resistance. DHP-122, a phosphorus deoxidized, high residual phosphorus copper with only a small residual silver content, is commonly used as a mould material for billet casting. It has adequate softening resistance and does not undergo plastic deformation early in the campaign. The phosphorus content should be greater than 0.03 percent to ensure good softening resistance. DLP-120 is a phosphorus-deoxidized, low residual phosphorus grade of copper. The phosphorus content is below 0.005 percent and hence its softening resistance is lower than that of DHP-122 grade. However, a silver content greater than 0.1 percent enhances the softening resistance of the alloy.  36  The STP grade is a tough-pitch alloy containing some silver. It lacks adequate softening resistance required to maintan its shape at peak operating temperatures for more than one or two heats, a behaviour very similar to that of pure copper. Copper-Silver alloys with a silver content exceeding 0.10 percent is frequently used as a mould material. Adequate silver content is necessary to ensure a good softening resistance. New alloys of copper containing elements such as chromium, zirconium, nickel and aluminum are also being introduced as mould materials [68]. The alloying elements such as chromium, nickel and aluminium strengthen the copper by solid solution hardening, whereas zirconium is a precipitation hardening agent. These elements improve the softening resistance as well as the yield strength of the copper beyond that of phosphorus and silver. These alloys substantially improve the mould life. However, it is important to ensure that the thermal conductivity of the alloy is greater than 80 percent of pure copper. Three conditions are essential in a mould material: *  Thermal conductivity  > 300 W/m°C  (W/m°C)  (80 percent of pure copper or more)  *  Half-softening point (°C)  > 350 °C  *  Yield stress ( MPa )  > 250 MPa at 20 °C  An alloy satisfying the above conditions is unlikely to contribute to mould distortion provided that the cooling water velocity exceeds 12 m/s, cooling water quality is controlled carefully and the metal level is at least 100 mm below the top of the mould.  37  (v) Mould-tube support system : The mould tube is held inside a steel jacket or baffle by steel plates that fit into slots in the upper part of the tube. The slots may appear only on two opposite faces (usually the straight sides in a curved mould machine) or on all the four faces of the mould tube. There is yet another design in which the mould tube is held on the top and bottom by expansion plates with gaskets for sealing the mating surfaces. The two-sided constraint system is undesirable because it gives rise to non-uniform mould distortion and greater negative taper at the off-corners relative to the midface [68]. The constrained sides deform more than the unconstrained sides. Non-uniform distortion results in greater mould/shell interaction at the off-corners during negative strip, causing deeper and more non-uniform oscillation marks at these locations. A four-sided constraint system results in more uniform mould distortion [68]. The negative taper at the off-corner and the midface is comparable. Mould/shell interaction and hence, the depth of oscillation marks is uniform around the mould periphery. However, these advantages may be lost if the keeper plates do not fit snugly into the bottom of the mould slots and also if mould tube is not positioned concentrically within the steel liner. A top and bottom support system for the mould-tube has been found to be a safe design since it does not lead to non-uniform mould distortion, provided the supports are parallel with the top and bottom of the mould [67].  (vi) Position of the meniscus : Metal level has a significant effect on the permanent distortion of side-supported mould tubes. Permanent bulging was observed in moulds operated with the meniscus very close to the top of the mould [67]. This happens because  38  the point of maximum expansion of the mould tube which is very near the meniscus comes closer to the support plates which oppose the outward movement. Hence, the stresses in the mould wall are large. The metal level and the type of mould support system are closely related especially in the generation of mould distortion. A metal level which is closer than 100 mm to the top of the mould is likely to create adverse conditions for mould distortion in the case of both two-sided and four-sided mould constraint systems. Hence, in moulds having a side-supported system, the distance of the meniscus from the top of the mould is critical and must not be less than 100 mm. However, in the case of moulds with a top and bottom support system this is not a criterion and therefore operating with shallow meniscus levels is less likely to be a serious concern.  4.2.5.1.2 Mould Taper Dippenaar et al.[69] have found that a single taper of 0.6 to 0.7 % per metre is inadequate to compensate for shrinkage of the cooling steel and gives rise to a large mould/shell air gap. Also, it is insufficient to counter the development of negative taper and resulting mould/shell interaction. A double tapered mould, with an upper taper of 3.5 to 4.0 % per m and a lower taper of 0.5 % per m, is considered to be better than a single tapered mould. The advantages of a double tapered mould has been discussed in earlier publications [68,69]. Recent work at UBC has examined the use of multiple tapered moulds. Taper calculations have been made for different carbon grades [70]. Caution must be exercised in employing multiple tapered mould tubes since the mould heat-flux profile and billet shrinkage profile are strongly dependent on casting  39  speed, mould-tube distortion and metal level. Since the upper taper is based on a constant residence time between the meniscus and the break-point to the next taper, it is mandatory to control the above mentioned parameters when using multiple tapered mould tubes.  4.2.5.1.3 Mould Oscillation Mechanism of Oscillation-Mark Formation Oscillation of the mould is essential to -  provide a stripping action which aids withdrawal of the strand; but it also generates oscillation marks. The mechanism of oscillation mark formation has been discussed by Samarasekera et al.[71] and is shown schematically in Figure 4.7. During negative strip, (the period when the mould moves downward faster than the strand), due to the negative taper at the meniscus, the mould squeezes inward and downward and jams on the newly formed solid shell. Due to this action, a downward compressive force is generated, the shell deforms and buckles to generate an oscillation mark on the billet surface.  Factors Influencing Mould Oscillation (i) Negative strip time : Negative strip time is expressed in terms of stroke length, oscillation frequency and the casting speed. I. k=-- cos — itfS  (2)  The effect of negative strip time on the depth and uniformity of oscillation marks on the billet surface was studied by Samarasekera et al. [71]. A reduction in the negative strip time from 0.21 to 0.12 s, reduces the depth and improves the uniformity of  40  oscillation marks on the billet surface. The average depth of oscillation marks and the difference in their depth between the midface and the off-corner has been shown to increase with increasing negative strip time. By reducing the time of this mechanical interaction, it is possible to decrease the depth and improve the uniformity of oscillation marks. On the other hand, operating with extremely low negative strip time causes sticking of the strand in the mould. An optimum range for negative strip time to minimize sticking as well as problems related to deep and non-uniform oscillation marks is 0.12  to 0.15 seconds.  Fig 4.7 Mechanism of oscillation mark formation [19].  41  (ii) Mould Lead : Mould lead is the downward displacement of the mould relative to that of the strand during negative strip. It is related to negative strip time, oscillation frequency, stroke length and the casting speed by the following mathematical formula:  MouldLead = S sin(nftN ) — VC tN  (3)  The stripping action appears to be related to the mould lead generated during the oscillation of the mould. Operating with small mould leads increases the tendency of sticking in the mould. The stripping action improves as the displacement of the mould during negative strip time is increased.  (iii) Other variables : In addition to negative strip time and mould lead, stroke length and oscillation frequency must be within specified limits. The desired range for the stroke is 9 to 16 mm. The upper limit imposed on oscillation frequency is around 4 Hz. and is based on vibration problems (of the structures) encountered at higher frequencies. Also, the oscillation frequency should not be linked to casting speed. Regular maintenance of the oscillation system is necessary to avoid wobbling and vibration of the oscillation system.  4.2.5.1.4 Mould Lubrication Lubrication in the mould is necessary to prevent sticking of the steel to the mould wall at the meniscus. Sticking hinders smooth withdrawal of the strand from the mould and leads to the generation of quality problems in billet casting.  42  Factors Influencing Mould Lubrication  (i) Oil - flow and distribution : Insufficient lubrication increases the tendency for sticking. On the other hand excessive supply of oil for lubrication leads to entrapment of hydrogen and pinhole formation in the billet. Thus, it is extremely important to have the proper amount of oil in the mould. Flow rates within 0.10 to 0.20 ml of oil per min per mm of mould tube inside perimeter are considered to be an optimum range. In addition, the oil must be uniformly distributed around the mould periphery. Regular checks on the distribution of oil are necessary. A simple experiment can be done on a cold mould when casting is not in progress to assess the adequacy of the oil distribution system. Oil is allowed to flow down the inside mould wall and is collected in containers placed below each face at the bottom of the mould tube. The volume of oil collected in each container is compared. A difference exceeding 10 percent is considered to be unsatisfactory.  (ii) Oil - Physical Properties : Viscosity, boiling point and flash point of the lubricant are important parameters. Oil viscosity determines the resistance to oil flow down the mould wall. A high flash point of oil is desired to prevent the oil from burning off before reaching the metal level. Similarly, the boiling point of oil must also be high such that the oil does not vapourize before reaching the metal level. Flash point and boiling point greater than 300 °C are considered to be adequate provided there are no scale deposits on the cold face of the mould. If scale deposits are present on the cold face of the mould, the mould operates hotter than normal and therefore, the lubricating oil vaporizes or burns before reaching the metal level. In this situation, the mould lubrication is likely to be poor.  43  (iii) Cleanliness of oil distribution system : Regular cleaning is essential to prevent blockage of oil-distribution slots. Blockage of these slots disturbs the flow and distribution of oil.  (iv) Removal of oil-pyrolysis products from the meniscus : The meniscus is the most important region of the mould from the viewpoint of heat extraction and initial solidification. If there is a build up of oil pyrolysis products at the meniscus during the process, the debris interferes with heat transfer and introduces non-uniformity in the rates of heat extraction. Cleaning of this area is therefore essential and can be done in two stages. Firstly, the solid loose oil particles and residual oil are removed with a rag soaked in solvent. Secondly, the mould surface is rubbed with emery paper to remove adherent materials and uncover the chromium layer on the mould surface.  4.2.5.2 Sprays From the viewpoint of quality products, an ideal spray cooling operation is one that generates minimum surface reheat, provides uniform cooling conditions around the billet periphery and ensures a high solidification rate. (i) Spray Design and Operation : The important parameters are spray length, specific water flow, water pressure, nozzle configuration, nozzle positions and stand-off distances from the strand. Reheating is minimized by preventing sharp reduction in heat extraction rates as the billet moves from one cooling zone to another. Therefore, the cooling rate in the upper portions of the sprays must match the heat flux at the mould exit, and the gap between the mould and the sprays must be as small as possible. The presence of a large gap may  44  cause a significant rebound in the billet surface temperature. In addition, the presence of support rolls at the mould exit hinders cooling of the billet and may cause the surface to reheat. At the exit, the spray cooling rate must be comparable to the radiative heat flux. The spray water flux must be reduced with axial distance to minimize drastic transitions in the cooling conditions. Brimacombe et al.[62] proposed a methodology for spray design by combining the understanding of the origin of thermal stresses with a mathematical model capable of simulating the thermal history of steel (profiles of solid shell thickness and surface temperature profiles along with the prediction of liquid pool depth). Based on experience, it is felt that surface reheating of the billet can be minimized by operating with a specific water flow less than 1.0 litre per kg of steel cast and a spray length greater than 3 m. The optimum operating conditions can be obtained with the help of the mathematical model which takes into account cooling in the mould, casting speed, section size and grade of steel being cast. The significance of nozzle configuration has been discussed by many authors. Brimacombe et al.[62] have outlined the methodology of translating the distribution of spray heat-transfer coefficients into a spray water flux distribution which can be linked to the operating spray-nozzle variables. In this study, the authors maintained the surface temperature at 1100 °C and predicted the heat transfer coefficients required to achieve this condition. Higher values of heat transfer coefficients are required for the central regions of the face in the upper spray zone than towards the bottom. Cooling requirements decrease towards the corners because of the effect of two-dimensional heat flow. The coefficients  45  are smaller for the lower regions of the sprays because the rate of conduction of heat to the surface is lower due to increased thickness of the solid shell. The effect of the spray length on the magnitude of reheating of the billet surface was also investigated. The stand-off distance, which is the distance of the spray nozzles from the billet surface, is important. The distance must be such that the edge of the water sprays coincide with the billet corners. In addition, the alignment of the nozzles must ensure uniform cooling around the billet. The distance between the nozzles in each spray zone is dictated by the need to maintain a uniform water-flux over the entire spray length. Mizikar [23] has shown that the water flux between two sprays is about 20 to 30 percent greater than the algebraic sum of the individual contributions of each nozzle. The calculation was based on the knowledge about water flux resulting from overlapping spray patterns. Spray pressure and spray angle are other variables that influence the spray water flux. During the operation the pressure drop over each header must be small as compared to that through each nozzle. Therefore, careful manipulation of the spray variables is necessary to optimize the spray heat flux profile for each zone. Brimacombe et al. [62] have evaluated numerous empirical correlations for different spray configurations proposed by various workers. Van Drunen et al. [46] proposed that a sudden decrease in the centreline temperature at the point of complete solidification generates stresses and strains that can give rise to centreline cracks. The mathematical model is used to predict the location of the point of complete solidification and the amount of cooling required to suppress the stress generation. (ii) Spray Maintenance : Regular maintenance of the spray cooling system is necessary to eliminate problems such as blocked or bent spray nozzles and also issues related to the  46  alignment of the spray nozzles. The blockage of spray nozzles due to poor spray water quality not only introduces non-uniformity in the cooling conditions but also can cause surface reheating of the billet. Bent or damaged spray nozzles also disturb the uniformity in the cooling conditions around the billet and thus generates rhomboidity and diagonal cracks. The only remedy in this situation is the immediate replacement of the damaged spray nozzles. However, while replacing, care must be taken to ensure that the new nozzle selected is appropriate for that location to avoid non-uniformity in the cooling conditions.  4.2.53 Liquid steel The important parameters from the viewpoint of billet quality are the high temperature strength of the steel and its cast structure. The high temperature strength of steel is related to Mn/S ratio in the steel. The severity of cracking problem reduces as the Mn/S ratio is increased above 25 [64]. The cast structure is linked to the steel superheat. The formation of columnar structure is favoured at high superheat levels. A columnar structure is undesirable since it provides easy paths for crack propagation. Hence, operation with low superheat levels, preferably below 25 °C, is recommended provided the flowability of steel is not adversely affected. The influence of higher superheat levels is more significant in low and high carbon steels as compared to the medium carbon grades [64]. The presence of low-melting impurities such as copper, tin, antimony, etc. which cause liquid-metal embrittlement is not desired. Copper is by far the most commonly  47  found contaminant in liquid steel. Copper levels greater than 0.2 percent are considered to be undesirable. Nickel in steel is beneficial and a Ni/Cu ratio of 1.0 minimizes the severity of this problem [72].  4.2.5.4 Pinch-rolls The pinch rolls provide the necessary withdrawal action during casting. The important parameters for the pinch rolls are the magnitude of pressure exerted on the strand and also the point of application. The pressure exerted on the strand is based on the force needed to pull the strand. However, if this pressure is excessive, it can generate tensile strains close to the solidification front, and cause internal cracks.  48  CHAPTER 5 KNOWLEDGE DOMAIN - Part II Analysis of Cracks, Rhomboidity and Breakouts in Billet Casting  5.1 Specific Issues A particular quality problem may be strand- or grade-specific, may occur at specific times during a heat and may also have a specific orientation. These trends provide important information about the causes of the problem that help identify related operating parameters. (a) Quality problem is predominant on certain specific strands : If a quality problem is predominant on certain specific strands as compared to the others, then the problem is likely to be influenced by factors that differ from one strand to another. Assuming the same design and operating parameters for all the strands, factors that could differ are those related to maintenance of the casting machine - mould, sprays and other accessories. Improper tolerances, misaligned mould tube, presence of scratch or gouge marks on the mould wall, wobbling oscillation system and plugged oil distribution slots are some of the major mould-related maintenence issues. For the sprays, the condition of the nozzles (blockage or any physical damage) as well as the overall alignment with respect to the strand are important. When the quality problem(s) are more frequent on the inboard strands as compared to the outer ones, the problem is likely related to the condition of the metal stream flowing into the mould. Thus, design and operation of the tundish are critical. On the other hand, if the quality problem appears on all the strands, then operating variables which are common to all the strands such as liquid steel quality, cooling water quality and design parameters are likely to be the cause of the problem.  49  (b) Quality problem is predominant at certain specific locations on the strand: When a quality problem appears on some specific region (corners or midface) of the billet, strong non-uniformity in the operating conditions in the mould around the billet periphery, is suggested. Non-uniformity may exist in the heat-extraction rate, lubrication and mould distortion. The presence of dark and bright patches on the billet surface is a clear indication of non-uniform cooling.  (c) Quality problem is predominant at a specific orientation: A quality problem may appear at a specific orientation more frequently as compared to the others. This is related to non-uniform cooling conditions arising from some abnormal operating conditions of a permanent nature such as misalignment or mould distortion that does not vary with time. On the other hand, when its orientation changes with time, the governing phenomena is transient in nature. Intermittent boiling in the cooling water jacket which varies with time, is likely to be the cause of the problem. (d)  Quality problem is predominant at a specific time in the heat: Depending on its cause, a  quality problem may occur more frequently at some specific time in the heat. The higher frequency of the problem in the initial stage of a heat could be related to excessive steel superheat or to poor start-up practice - improper linkage between the dummy bar and the steel strand. Faulty engagement drastically reduces the negative strip time (sometime the stripping action may be totally be absent) and can lead to sticking of the solid shell in the mould. Higher superheat favours the formation of a columnar structure in the billet. If the frequency of the problem, especially internal cracks, is higher in the initial stage of heats, it possible that high superheat may be responsible. In addition, with thermal stratification of the steel in the ladle, it is also possible to have higher superheat levels at the end of heats and  50  therefore, problems could be more severe in the later stage of a heat as well. (e) Ouality problem is predominant in certain specific grades of steel: The columnar-zone length is a minimum for steels having carbon content in the range 0.20 to 0.38 percent and increases sharply at carbon levels greater than 0.38 percent [64]. Also, the columnar-zone length increases with decreasing carbon content below about 0.15 percent. Thus, quality problems influenced by a large columnar zone are less frequent in steel grades with carbon content in range of 0.20 to 0.38 percent. In low carbon steels having carbon content in range 0.10 to 0.14 percent, billet surface is rougher and the oscillation marks, deeper than in other grades [61]. Thus, binding is likely to be worse in this grade because less heat is transferred to the mould because of a larger mould/shell air gap as a result of which cooling and shrinkage of the shell are less than in the case of other grades. In addition, steels having 0.17 to 0.24 percent carbon exhibits a higher frequency of defects due to reduced ductility at elevated temperatures [61].  5.2 Quality Problems  5.2.1 Cracks Crack formation is a major problem in continuously cast steel billets. Cracks have been observed at almost every conceivable location in cast steel as shown in Figure 5.1. In the interior, cracks may be seen near the corners, midway between the centre and the surface, at the centreline or diagonally between the obtuse corners. On the surface, transverse and longitudinal cracks may appear in both the midface and the corner regions.  51  TRANSVERSE MIDFACE CRACK LONGITUDINAL MIDFACE CRACK  CRAZE CRACKS  RHOMBOID BILLET  TRANSVERSE CORNER CRACK LONGITUDINAL CORNER CRACK  DIAGONAL CRACK  CENTRELINE CRACK MIDWAY CRACKS PINCH-ROLL CRACKS OFF-CORNER INTERNAL CRACKS  Fig. 5.1^Schematic diagram of a rhomboid billet showing various types of cracks.  52  Another kind of surface cracking problem is craze cracks which are linked to high copper levels in steel. The surface cracks pose a more serious problem than the internal cracks because, they do not reweld during rolling and the crack surfaces oxidize giving rise to oxide-rich seams in the rolled products. The internal cracks can also be a problem  particularly if during rolling, they do not close leaving voids in the steel products. The mechanism for the formation of different cracks has been established in numerous studies, most of which were reviewed by Brimacombe and Sorimachi [43]. Off-corner cracks were studied by Brimacombe et al.[73] and Bommaraju et al.[64].  5.2.1.1 Internal Cracks The internal cracks result from high tensile strains and stresses acting on regions of the solid shell that are in the high temperature zone of low strength and ductility. They are influenced by sulphur and phosphorus content of steel and the superheat level. Van Drunen et al.[46] showed that the internal cracks are hot tears that initiate close to the solidification front. Hence, the depth of the outer tip of an internal crack from the billet surface directly reflects the local shell thickness at the time of its formation. On comparing this with the solid shell thickness profile predicted by the heat-transfer mathematical model for billet solidification, it is possible to ascertain where in the machine the crack originated. The orientation of the crack gives the direction of tensile stress causing it as this stress is normal to the direction of crack orientation. The internal cracking problems in billet casting have been summarized in Tables 5.1 and 5.2.  53  'n Table 5.1 Internalaacksasug  QUALITY PROBLEMS Off-corner Crack  CAUSES  INFLUENCING FACTORS  Bulging of solid shell and hinging at off-corners  Thermo-mechanical behaviour of the mould; Adverse mould/shell interaction; Deep and non-uniform oscillation marks; Steel composition and superheat  Non-uniform shell generated by the mould  Thermo-mechanical behaviour of the mould; Adverse mould/shell interaction; Asynchronous intermittent boiling in the mould; Deep and non-uniform oscillation marks; steel composition and superheat  Asymmetric spray cooling  Poor spray design and maintenence; Steel composition and superheat  LOCATION Lower part of mould or very close to mould exit  Diagonal Crack Spray zone  Centreline Crack Near the point of Sudden decrease complete in the centreline solidification temperature at the point of complete solidification  Inadequate spray cooling near the point of complete solidification; Steel composition and superheat  Pinch-roll Crack Close to the pinch rolls  Excessive pinch roll pressure; Steel composition and superheat  Unbending Cracks  Squeezing on a strand with liquid core  Close to the point Unbending on a Excessive bending strains; of unbending strand with Steel composition and liquid centre superheat; High casting speed  54  Table 5.2^Midway Cracks in Billet Casting  QUALITY PROBLEM  LOCATIONS  CAUSES  INFLUENCING FACTORS  Midway Crack  Mould exit or in the gap between the mould and the sprays  Reheating of the billet surface  Mismatch between the mould and the sprays: due to design or maintenance problem; Poor design of cooling jacket near the mould exit; Steel composition and superheat  Upper portion of the sprays  Reheating of the billet surface  Poor spray maintenance : bent or plugged spray nozzles; Steel composition and superheat  Lower portion of the sprays or the radiation cooling zone  Reheating of the billet surface due to the sprays  Sprays : design and maintenance issues; Steel composition and superheat  Reheating of dark overcooled patches generated by the mould  Thermo-mechanical behaviour of the mould; Adverse mould/shell interaction; Deep and non-uniform oscillation marks; Steel composition and superheat  5.2.1.1.1 Off-Corner Cracks Location - Off-corner internal cracks can be observed in the transverse section of the  billet shown in Figure 5.2. They occur roughly at a distance of about 15 mm from a  55  Fig. 5.2^Transverse section of a billet showing off-corner internal cracks.  56  given corner and 4 to 6 mm from the surface. These cracks appear randomly on any one, or several, of the eight off-corner sites. The cracks may be seen together with surface depressions and deep oscillation marks.  Mechanism of off corner crack formation -  -  The position of the outer tip of an  off-corner crack indicates whether the crack formed in the lower part of the mould or near the mould exit. Brimacombe et al. [67] postulated that off-corner internal cracks are generated by bulging of a given face of the billet in the lower part of the mould. It was also proposed that bulging of the solid shell may not extend to the corner but can hinge about the off-corner region. This hinging action generates a tensile strain at the solidification front and leads to off-corner cracks. A schematic diagram illustrating this mechanism is shown in Figure 5.3. Very often, the outer tips of the off-corner cracks overlap with white bands. This suggests that both phenomena originate at the same time as bulging of the shell. The bulging of the shell disturbs the interdendritic liquid and creates these white bands [64]. Bommaraju et al.[64] linked the generation of off-corner cracks to the depth and the uniformity of oscillation marks on the billet surface. In another study, Samarasekera et al.[71] observed that the oscillation marks at the off-corner locations were deeper than at the midface. Deep oscillation marks in the off-corner regions locally reduce the heat extraction rate and the shell growth as the billet passes through the mould. In the lower part of the mould where the mould/shell gap is large and the billet contraction has virtually stopped, bulging can occur, particularly on faces having thin and weak shells in the off-corner regions. The resultant hinging action then generates tensile strains  57  Mould wall  Off - corner crack Cold corner^  (II  Fe rrostati c preSsure  Liquid pool  Solid shel l  Fig. 5.3^Schematic diagram showing generation of an internal crack due to bulging  of the billet shell and a hinging action in the off-corner region [73].  58  strains at the solidification front near the off-corners. A schematic diagram illustrating the significance of deep oscillation marks in the generation of off-corner cracks is shown in Figure 5.4 [681  Deep oscillation mark  Mould wall  Hinging and Crack Formation  t Ferrostatic Pressure  .111.■••••  Fig. 5.4 Schematic diagram illustrating the mechanism generating off-corner internal cracks due to bulging and hinging of the shell. Significance of deep oscillation marks in the generation of this problem is also evident [68].  59  Factors influ encing ner cracking problem - Off-corner cracks result from bulging of the solid shell. Hence, factors that increase the mould/shell gap will influence bulging and affect the cracking problem. Off-corner cracking is also linked to the formation of oscillation marks. Therefore, any variable that influences the moulcYshell interaction near the meniscus will directly impact on the nature of oscillation marks, and the cracks. Mould taper that does not follow the billet shrinkage profile leads to a large mould/shell gap which may cause a thin shell to bulge. Bulging is exacerbated by thermally generated reverse taper and wear in the lower part of the mould. At the mould exit, if the solid shell is thin and weak, bulging can occur. This can be prevented by increasing the water flux in the upper sprays which will cool the strand and increase the solid shell thickness. The formation of deep oscillation marks is linked in a complex way to the thermo-mechanical behaviour of the mould tube during the process [71]. The distortion of the mould (both elastic and plastic) together with the lubrication conditions and the oscillation characteristics, govern the mould/shell interaction near the meniscus and influence the nature of oscillation marks - depth and uniformity across the billet surface. Off-corner cracks were randomly seen on the eight off-corner sites [64]. This observation suggests that bulging of the shell is not localized to a single face. The behaviour may be due to intermittent boiling in the mould cooling jacket, wobbling of the mould system, poor setting of the foot-rolls or misalignment of the casting machine. Casting speed is important as it influences the solid shell thickness. Increasing the casting speed reduces the shell thickness in the mould, thereby exacerbating the tendency for bulging of the shell. Since, the cracking phenomena is linked to dendritic separation, a low Mn/S ratio will increases the severity of this problem [64].  60  5.2.1.1.2 Midway Cracks Location - Midway cracks are also called radial streaks, halfway cracks and ghost lines. They can be seen in the transverse section of the billet shown in Figure 5.5. They appear as dark lines running normal to a given face in a region midway between the surface and the centreline. As described earlier, the distance of the outer tip of the crack from the billet surface can be utilized to determine where in the machine the crack formed. It can occur close to the mould exit, in the gap between the mould and the sprays, in the spray zone or in the radiation cooling zone.  Mechanism of midway crack formation - Midway cracks are caused by reheating of the surface of the strand which causes the billet surface to expand [47]. This expansion imposes a tensile strain on the interior, hotter regions of the solid shell which being weaker, may crack. The stress distribution that results from reheating was calculated using finite-element analysis by Grill et al.[74]. The stress distribution in the billet clearly indicated relatively large tensile stresses parallel to the wide face near the solidification front. In an another study, Brimacombe and Sorimachi [43] found that the location of the high stress region corresponds closely to that of midway cracks seen in the billet. Van Drunen et al.[46] postulated that midway cracks are hot tears which form in the high temperature zone of low ductility. An examination of the inside surface of an open midway crack viewed with a scanning electron microscope showed smooth dendrite branches with no signs of deformation. This suggests the presence of a liquid during the dendritic separation. This liquid is rich in sulphur and other elements with a positive segregation coefficient.  61  Fig. 5.5^Transverse section of a billet showing midway cracks.  62  E Actors influencing midway cracks - Three conditions appear necessary for midway crack formation: excessive reheating of the billet surface, the presence of a long columnar zone and a low Mn/S ratio in the steel. Reheating of the billet surface generates tensile stresses acting on the solid steel close to the solidification front. A columnar structure provides easy paths for crack propagation. The Mn/S ratio directly impacts on the high temperature strength and ductility of steel and the formation of liquid film between the dendrites. Surface reheating occurs when the billet experiences a sudden reduction in the heat extraction rate after a period of excessive cooling. This may happen at the mould exit, in the spray zone or at the end of the sprays. Reheating in the radiation cooling zone occurs due to spray- and/or mould- related issues. If spray cooling is excessive over a short length, the billet experiences a sudden reduction in the heat extraction rate as it exits the sprays and enters the radiant cooling zone. A reheating problem in the radiation zone has its origin in the mould if the billet leaving the mould has dark and bright patches. The dark patches cool very rapidly in the spray zone due to nucleate boiling. When the billet enters the radiation cooling zone, these over-cooled regions on the billet surface begin to reheat excessively leading to midway cracks. These patches indicate non-uniform cooling in the mould which is related to the thermo-mechanical behaviour of the mould during operation. Hence, factors related to mould distortion, oscillation mark formation, taper and lubrication are important. Reheating can occur near the mould exit if the distance of the first spray nozzle from the bottom of the mould is large or if there is a mismatch between the heat fluxes  63  at the bottom of the mould and the first spray zone. Therefore, spray cooling at the mould exit and the position of the first spray nozzle are critical parameters. When the mould cooling jacket does not extend to the bottom of the mould, a drastic reduction in the heat extraction rate is likely to occur. Maintenance issues contribute to reheating if the spray nozzles are bent or blocked as a result of which there is a sudden reduction in the heat extraction rate. Midway cracks in the upper spray region occur due to this problem. This conclusion follows from the logic that the mathematical model assumes ideal spray conditions and therefore cannot predict a high reheat for the upper spray zone. The only cause, for the midway cracks in this region are issues related to maintenance. Similarly, if midway cracks are seen in the region between the mould and the sprays and if the reheat predicted by the model for this zone is not high, it is very likely that issues related to spray maintenance are a problem. Factors such as superheat and carbon content of steel, machine design (straight vs curved mould design) and section size are critical parameters influencing the columnar structure in the billet. The presence of a sulphur-rich liquid film between the dendrites leads to easy separation of the dendritic arms. A Mn/S ratio greater than 25 minimizes the cracking problem by preventing the formation of a liquid film.  5.2.1.1.3 Diagonal Cracks Location - Diagonal cracks run between the obtuse corners of a rhomboid section [43]. Sometimes these cracks may grow outwards towards the corners of the billet to form longitudinal cracks depending on the magnitude of the strain and the thickness of the  64  solid shell.  Mechanism of diagonal crack formation - When two adjacent faces of a billet cool more rapidly than the others, the billet contracts and generates a diagonal strain between the colder faces. If this strain is large, the billet distorts and take on a rhomboid shape with an acute angle between the colder faces and an obtuse angle between the hotter ones. A crack may then initiate near the solidification front at right angles to the strain axis which runs along the diagonal joining the obtuse angled corners.  Factors influencing diagonal cracks  -  Diagonal cracks result from the distortion of  the billet due to unsymmetrical cooling. This is an extreme case of rhomboidity and its presence clearly indicates severe asymmetric cooling in the sprays and/or the mould. Diagonal cracks are linked to mould cooling but the sprays are also definitely involved. Non-uniform shell growth in the mould generates cracking strains along the diagonal joining the hotter corners, even with a perfectly aligned and a uniform spray cooling system. Cooling conditions in the mould are very important from the viewpoint of shell growth. The thermo-mechanical behaviour of the mould which influences the mould shell interaction, directly impacts on the uniformity of heat extraction rates around the billet periphery. Asynchronous intermittent boiling in the mould is also an important phenomena leads to varying shell thickness around the billet periphery. Thus, mould-distortion, oscillation, lubrication and taper are critical. The design tolerances of the mould tube, liners (baffle-tubes) and support system are important since they influence the augment of the mould-tube.  65  The spray cooling may also be asymmetrical. If this is the case, maintenance issues such as bent or plugged nozzles and their physical alignment with respect to the strand are certain causes of diagonal cracks. According to Mori [75,76], the rhomboidity problem is generally greater in smaller size billets and also with higher superheat levels. Based on the same logic, diagonal cracks will also tend to be more frequent under these conditions.  5.2.1.1.4 Centreline Cracks location - Centreline cracks or core cracks appear in the central region of a cast section and form towards the end of solidification. Figure 5.6 shows a photograph of a transverse section of a billet showing centreline cracks.  Mechanism of centreline crack formation - A sudden decrease in the centreline  temperature at the point of complete solidification generates strains which could cause cracking [46]. An abrupt decrease in the centreline temperature occurs when the latent heat of fusion has been completely removed from the central region of the billet. At this point, the centreline cools faster than the surface as a result of which, the centre region contracts. The centre is constrained from contraction by the surrounding colder steel and is therefore put into tension. If reheating of the billet surface coincides with the bottom of the pool, then the resulting tensile strains acting on the centreline will be the sum of the two stresses and then the effect is likely to be large. factors influencing centreline cracks  -  The abrupt decrease in the centreline  temperature, occurring as a result of complete removal of latent heat, is a natural phenonema and will happen in any casting process.  66  Fig. 5.6^Transverse section of a billet showing centreline cracks.  67  The severity of this problem, however, can be reduced by minimizing the difference in the cooling rates between the surface and the centre. Brovman et al.r77] suggested that centreline cracks could be suppressed by applying spray cooling at the point of complete solidification. When surface reheating coincides with the point of complete solidification, it is first necessary to isolate the two phenonema and then take appropriate steps to eliminate them. Reheating can be suppressed, as was suggested for midway cracking, by adequate spray design and also by eliminating the dark and bright patches on the billet surface at the mould exit. Spray design to minimize the stresses below the liquid pool has been studied by Grill et al.f78J.  5.2.1.1.5 Pinch-roll Cracks Location - Pinch roll cracks can be seen on a sulphurprint of a longitudinal or a transverse  section of a billet. They are very similar to midway cracks in their appearance except that they always appear normal to the faces parallel to the axis of the pinch-rolls. Mechanism of pinch-roll crack formation - The application of excessive pinch-roll  pressure on a strand having a liquid core or with its centreline above 1340 °C, causes cracking by dendritic separation in the high temperature zone of low ductility [43]. factors influencing pinch-rolls cracks - The positioning of the pinch-rolls is a critical  factor from the viewpoint of crack formation as it is necessary to avoid squeezing on a liquid centre. A safe location for the pinch-rolls can be predicted based on the model calculation of liquid pool depth. The cracking severity depends on the magnitude of the pressure exerted by the pinch-rolls on the liquid core. Therefore, reduction of the  68  roll pressure decreases the tensile stresses and hence minimizes cracking. Like any internal cracks, pinch-roll cracks are hot tears and therefore are influenced by the extent of the columnar structure and also by the high temperature mechanical properties of steel. Hence, superheat and steel composition are important.  5.2.1.2 Surface Cracks Surface cracks discussed in this section include transverse cracks (and depressions), longitudinal midface and corner cracks and craze cracks. A summary of these cracks are provided in Table 5.3.  5.2.1.2.1 Transverse Cracks (and depressions) mcation - Transverse cracks are usually seen at the midface or near the corners on the billet surface. Figure 5.7 is a photograph of a billet showing transverse depressions. Transverse depressions may or may not have a crack at their base. In the case of a depression without a surface crack, if a longitudinal section of the billet across the depression is observed, an internal crack will be seen below the depression. Figure 5.8 is a photograph of a longitudinal section of a billet across a depression showing an internal crack below it.  Mechanism of transverse crack (and depression) formation  -  Samarasekera and  Brimacombe [20] postulated that transverse cracks and depressions originate in the mould. Due to sticking or binding of the billet in the mould, the withdrawal system mechanically pulls the strand. The pulling action imposes a longitudinal tensile stress on the shell as a result of which a transverse crack can form close to the solidification  69  Table 5.3^Surface Cracks in Billet Casting  QUALITY PROBLEMS  ORIGINS  CAUSES  INFLUENCING FACTORS  Transverse Crack In the mould (and depression)  Pulling action on the strand as a result of binding or sticking in the mould  Longitudinal Corner Crack  In the mould  Reheating of the billet corner due to a large mould/shell gap  Longitudinal Midface Crack  In the mould  Excessive reheating of a localized portion of the billet surface; Stream impingement on a face  Craze Crack  In the sprays  Grain boundary High level of Cu, Sn or embrittlement due to low melting impurity in the presence of Cu, steel Sn or low melting impurities in steel.  Thermo-mechanical behaviour of the mould; Adverse mould/shell interaction; Deep and non-uniform oscillation marks; Steel composition and superheat  Large corner radius; Presence of corner "key holes"; Mould tube alignment; Asynchronous and / or intermittent boiling in the mould; Deep and Presence of thin and non-uniform oscillation weak shell at the marks; hotter corners Thermo-mechanical behaviour of the mould; Adverse mould/shell interaction; Steel composition  70  Presence of scratch or gouge marks on the inner surface of the mould wall; Misalignment of metal stream  Fig. 5.7^Photograph of a billet showing transverse depressions.  71  front where the ductility is extremely small. The surface of the billet is in the temperature range 1150 to 1430°C and therefore has a high ductility [42,43]. Depending on the magnitude of the stress, the solid shell may flow plastically and form a depression on the billet surface. This mechanism is similar to necking in tensile testing. Figure 5.9 is a schematic diagram illustrating this mechanism. The internal crack, depending on the solid shell thickness at the time of its formation, as well as the extent of necking, may penetrate to the surface. Transverse crack formation is linked to the nature of oscillation marks. In an earlier study [64], it was found that the obtuse angle corners of rhomboid billets have deeper oscillation marks than the acute angled corners. This suggests that the obtuse-angled corners are hotter than the acute angled corners due to lower heat transfer arising from a larger mould/shell gap. The colder corners having shallower oscillation marks tend to be in closest contact with the mould. The billet can bind in the mould at these corners and develop large axial stresses during the withdrawal of the strand. Depending on the magnitude of the force, transverse cracks may form at these corners.  Factors influencing transverse cracks (and depressions)  -  Transverse defects are  linked to sticking and/or binding of the billet in the mould. They are also related to the nature of the oscillation marks. Hence, the thermo-mechanical behaviour of the mould and the mould/shell interaction at the meniscus as well as away from it are critical. Hence, design and operating variables influencing mould distortion, lubrication, oscillation characteristics and taper are important parameters.  72  Fig. 5.8^Longitudinal section of a billet showing an internal crack below a depression.  73  Binding (High Friction) High Ductility  Liquid Steel  Necking of Ductile Shell to Form Depression  — Crack  Zone of Low Ductility  N/ .  Withdrawal Force  Fig. 5.9^Schematic diagram showing the formation of transverse depressions and cracks in billet due to sticking or binding in the mould [20].  74  Mould taper is an important design variable that significantly affects heat extraction and binding in the mould. It is calculated based on a designed dwell time in the mould. The successful implementation of a mould taper requires a strict control of metal level and casting speed. If the casting speed is reduced, the dwell time increases and the influence of the thermal resistance of the shell increases with a decrease in the heat flux and the cooling rate. This reduces billet shrinkage and may influence binding. By the same logic, increasing the casting speed will reduce the tendency of the billet to bind. If the upper taper is too steep, then the metal level can be lowered to decrease the dwell time and minimize the binding problem. The dwell time in the mould must be sufficiently long to prevent problems related to thin and weak solid shell- bulging of the shell and generation of off-corner internal crack and perhaps, breakouts. In low carbon steels containing 0.10 to 0.14 percent carbon, the surface of the billet is rougher due to shrinkage associated with the 8 — ?phase transformation and less heat is transferred to the mould as a result of which cooling and shrinkage of the shell is small. Therefore, binding is expected to be worse in this grade as compared to the others [61]. The frequency of transverse cracking was also found to be higher in steels having carbon in the range 0.17 to 0.24 percent due to poor ductility at high temperatures [61].  5.2.1.2.2 Longitudinal Midface Cracks Location A longitudinal midface crack is located on the midface of a billet surface -  and is parallel to the withdrawal direction.  75  Mechanism of longitudinal midface crack formation - The mechanism of longitudinal midface cracking problem in billet casting is related to the presence of some scratch or gouge marks on the mould wall which locally increases the mould/shell gap adjacent to it. This leads to a drastic drop in the heat extraction rate over a small localized region on the billet surface. This area reheats and expands as the surface temperature approaches the high temperature zone of low ductility. The colder steel surrounding it prevents this expansion and results in the generation of tensile stresses at the solidification front which ultimately leads to cracking. Longitudinal midface cracks tends to form more easily where the air-gap resistance dominates the heat extraction rate (i.e. near the meniscus). Longitudinal midface cracks can also form due to stream impingement on a face due to cocked or misaligned stream.  Factors influencing longitudinal midface cracks - Scratches, gouge marks or any defect present on the inside wall of the mould that increase the air gap resistance over a localized area on the billet surface will influence the formation of longitudinal midface crack. Misalignment of the metal stream is also an important issue.  5.2.1.2.3 Longitudinal Corner Cracks Location - These cracks directly appear on the hotter, obtuse corners of the billet. Mechanism of longitudinal corner crack formation - Longitudinal corner cracking originates in the mould. Two mechanisms have been proposed for this problem. The first mechanism is based on the formation of a large mould/shell gap near the corners which forces the billet surface to reheat to temperatures approaching the high temperature zone of low ductility. The tensile stresses generated due to thermal  76  shrinkage and ferrostatic pressure lead to cracking if the ductility of steel above 1340°C  is exceeded [79 - 81]. Figure 5.10 is a schematic diagram illustrating this mechanism of formation of longitudinal corner cracks. The second mechanism is linked to rhomboidity and thus, to the formation of oscillation marks on the billet surface [61]. It was observed that corners with deeper oscillation marks are hotter at the mould exit. The presence of deep oscillation marks lead to the formation of a large mould/shell air gap which reduced the rates of heat extraction and solidification. Therefore, at the hotter corners, the shell tends to be thin and weak. Further, contraction of the cooling faces generates tensile strains in the transverse direction. Depending on the magnitude of this strain, cracks may initiate near the solidification front and penetrate the solid shell as a longitudinal corner crack.  Factors influencing longitudinal corner cracks Longitudinal corner cracking has -  been observed in moulds with large corner radius, usually more than 4 mm. Sometimes key-hole effects at the mould corners can also generate conditions for the formation of longitudinal corner cracks. Asynchronous intermittent boiling in the cooling water channel can also generate longitudinal corner cracks on the billet surface as it leads to rhomboid mould distortion, which varies with time as boiling events change asynchronously on different faces and causes non-symmetrical cooling of the billet due to the changing air-gap width [22]. In addition, since the longitudinal corner crack is also linked to the nature of oscillation marks on the billet surface, its formation is related to the thermo-mechanical behaviour of the mould and to the nature of mould/shell interaction. Factors related to the mould (distortion, oscillation, lubrication and taper) are critical.  77  strain  strain  Fig. 5.10 Schematic diagram showing the formation of subsurface, longitudinal crack on diagonal at obtuse-angle corners of rhomboid billet [22].  78  5.2.1.2.4 Craze Cracks Craze cracks appear as a fine interwoven network of cracks as shown in Figure 5.11. These are caused by grain boundary embrittlement. It is difficult to spot craze cracks on the billet surface during normal inspection. However, etching of the surface with a hot acid can reveal the presence of craze cracks. Craze cracks appear as fine striations on final rolled products such as wire-rods and ribbed-bars. It is believed that craze cracks are related to the presence of impurities with low melting point such as copper, tin and antimony in the steel. The mechanism of craze cracking, though not well established, appears to be linked in some way to the enrichment of the low melting point elements as a result of surface oxidation of the billet [72]. The high temperature oxidation of steel results in the formation of iron oxide scales. Copper being nobler than iron remains unoxidized and is concentrated at the steel/oxide interface [82]. Since the oxidation rate of iron is high and the time for back diffusion of copper is insufficient, the solubility of copper in austenite is exceeded. A molten copper-rich phase forms at the scale/steel interface and penetrates into cracks at the surface and along the grain boundaries [82]. In the sprays, cooling of the billet results in thermal strains which generate cracks along the grain boundaries embrittled by the molten phase. It has been observed that the presence of nickel in the steel reduces the frequency of craze cracking. Copper levels below 0.2 percent [72] or Ni/Cu ratio greater than 1.0 are helpful in reducing the severity of this problem [82]. Two theories have been proposed for the beneficial role of nickel in steels containing copper [82]. The first is related to the fact that nickel increases the solubility of copper in austenite. The second  79  states that nickel alloys with the copper-rich phase and reduces its melting point. In both, the formation of the low melting copper-rich phase is prevented and thus, craze crack formation is avoided.  1  Fig. 5.11^Photograph of craze cracks.  80  5.2.2 Rhomboidity Severely rhomboid billets pose problems in pusher-type reheat furnaces and also during rolling where the corners may fold over and generate seams in the final product. In square billets, the term "off-squareness" is often used whereas in rounds, this problem is called "ovality". The difference between the lengths of the two diagonals is a measure of the severity of this problem. In rounds, the difference between the longest and the shortest diameters is used to indicate ovality. An upper limit imposed by the rolling mills for this difference is in the region of 6 mm. Table 5.4 is a summary of rhomboidity- mechanisms and influencing factors.  Table 5.4^Rhomboidity Problem in Billet Casting.  QUALITY PROBLEM Rhomboidity  CAUSES  ORIGINS In the mould and/or the sprays  Non-uniform shell generated in the mould  .. In the sprays  Asymmetric spray cooling  INFLUENCING FACTORS Thermo-mechanical behaviour of mould; Adverse mould/shell interaction; Asynchronous intermittent boiling in the mould; Deep and non-uniform oscillation marks; Mould-tube alignment; Steel superheat Poor spray design and maintenence; Steel superheat  Mechanism of rhomboidity - The mechanism for the generation of rhomboidity, like  off-corner cracking involves oscillation mark formation and non-uniform heat extraction in  81  the mould and the sprays. The problem usually begins with the formation of deep and non-uniform oscillation marks around the billet periphery. In the vicinity of deep oscillation marks, due to a wide billet/mould air gap, the rate of heat removal is low. On the other hand, regions of the billet having shallow oscillation marks experience higher rates of heat extraction. Thus, the presence of non-uniform oscillation marks on the billet surface gives rise to widely differing heat extraction rates around the billet periphery which ultimately leads to a solid shell having non-uniform thickness. The situation of non-uniform heat extraction is further exacerbated by rapid cooling of some portions of the solid shell which leads to shrinkage as a result of which the hotter corners of the billet get pulled farther away from the mould wall. The billet exiting the mould, although reasonably square, has a non-uniform solid shell as shown in Figure 5.12 (a schematic representation of this idea) [64]. In the sprays, the colder portions of the billet, having a thicker solid shell, tend to cool faster than the hotter regions because of the greater thermal path and effects of unstable boiling which leads to non-uniform shrinkage of the billet and generates rhomboidity. The above mechanism is supported by the following observations made on square billets during the casting process : (i) the obtuse angle corners of rhomboid billets usually have the deepest oscillation marks. (ii) the billets emerging from the mould, when observed through the peep-hole, showed that, of the two corners in view, one was cold (dark) and the other was hot (bright). Subsequent billet inspection on the cooling bed indicated that the acute-angle corners in the billet corresponded to the colder corners whereas the hot corners formed the obtuse angle of the billet.  82  .■•^ .."^  ■  ..0"^  ■  I'^  ■  ■  Upper Sprays  5%,,,^ -.....  ■  ■  ...."  ."  ei  'N  Off-square billet containing off-corner internal cracks  Fig. 5.12 Schematic diagram showing a billet with non-uniform shell thickness being distorted into rhomoid shape by spray cooling [64].  83  Asynchronous intermittent boiling in the cooling water channel has also been linked to the generation of rhomboidity in the billet [22]. It is proposed that mould distortion varies with time as the boiling events change asynchronously on different faces and causes non-symmetrical cooling of the billet due to the changing air-gap width. The orientation and severity of rhomboidity varies during the heat due to this phenomenon and so, supports the above mechanism. Further, this observation is useful as it can be used to distinguish rhomboidity due to asynchronous intermittent boiling from that caused by other factors such as machine alignment, wobbly mould oscillation, or poor spray cooling. In the absence of any mould effects, rhomboidity must be caused by asymmetric spray cooling.  Factors influencing rhomboidity Since rhomboidity is linked to deep and non-uniform -  oscillation marks on the billet surface, factors related to the thermo-mechanical behaviour of the mould and the nature of mould/shell interaction are critical. Hence, mould distortion, taper, oscillation characteristics and lubrication parameters are important. Factors such as tolerances with respect to the water gap, mould tube alignment and the nature of the mould constraint system also need special attention. Even though the mould generates rhomboidity, the difference between the lengths of billet diagonals at the mould exit is small. This is expected since the rhomboidity in the billet at the mould exit cannot exceed the rhomboidity of the mould. The degree of rhomboidity can then increase significantly as the billet moves through the spray cooling zone. However, the contribution of the mould to generate a non-uniform shell and therefore rhomboidity, is quite significant. Hence, mould parameters (operating, design and maintenance) are extremely critical. When rhomboidity is caused by asynchronous intermittent boiling and mould-distortion, simple suppression of the boiling phenomena can help to minimize the  84  problem. In this case, factors influencing asynchronous boiling such as cooling water velocity, non-uniform scale deposition on the cold wall of the mould, surface roughness and mould design parameters such as wall thickness and support system are critical. Spray cooling influences rhomboidity in two ways. First, it acts on the non-uniform shell generated by the mould and further aggravates non-uniformg heat extraction, uneven shell growth and the rhomboidity problem. Thus, when the mould effects are present, the sprays influence rhomboidity under all circumstances- whether symmetric or asymmetric. However, when the spray cooling is symmetric, the mould is the only cause of rhomboidity. Second, in the absence of any adverse mould effects, asymmetric spray cooling gives rise to rhomboidity. Non-uniform spray cooling is usually a maintenance problem and occurs due to plugged or bent nozzles and poorly positioned spray risers.  5.2.3 Breakouts Breakouts occur when there is a rupture of the solid shell as a result of which molten metal flows out. This is a serious problem as it not only hampers productivity but is also a hazard. Table 5.5 is a summary of factors influencing breakouts.  Mechanism of breakouts Breakouts are linked to the presence of a thin and weak solid -  steel shell at the mould exit. The shell may be thin at localized regions or all around the billet periphery. The presence of cracks may also lead to breakouts, particularly if the cracks act as stress concentration sites and aid the rupture of the shell. The mechanism by which thin shells are generated over localized regions in the billet is based on the formation of a large mould/billet air gap that reduces the rate of heat extraction and retards the shell growth. These thin regions of the shell tend to be hotter as compared  85  to the areas with thicker shell. The presence of depressions or deep oscillation marks on the billet surface increases the mould/billet air gap which in turn influences the generation of thin shells especially in those regions of the mould where thermal resistance of the air gap dominates the heat transfer.  Table 5.5^$reakout Problem in Billet Casting.  QUALITY PROBLEM Breakouts  ORIGINS  CAUSES  INFLUENCING FACTORS  Close to obtuse-angle corners of the strand  Thin shells generated at the obtuse angle corners due to asymmetrical cooling in the mould  Thermo-mechanical behaviour of the mould; Adverse mould/shell interaction; Deep and non-uniform oscillation marks; Mould-tube alignment; Steel superheat  Close to transverse depressions or deep oscillation marks on the billet surface  Local reduction in the shell thickness due to the presence of depressions or deep oscillation marks  Adverse mould/shell interaction; Deep and non-uniform oscillation marks; Thermo-mechanical behaviour of the mould  Close to a weak Inadequate shell spot in the shell thickness at the mould exit  Insufficient dwell time in the mould; Entrapment of slag or scum between mould and billet  Mould overflow Operator error  Poor metal control  86  Breakouts in billet casting have been evaluated by Samarasekera and Brimacombe [83]. The results of this study indicate that breakouts occur close to transverse depressions (with cracks) and also very close to the obtuse corners which are hotter and have a thinner solid shell. This finding supports the above mechanism quite adequately. Breakouts have also been related to the entrapment of a scum between the steel and the mould-wall. This leads to the formation of an extremely thin shell adjacent to it. Once outside the mould system, it is possible that this region of the shell may rupture and cause breakouts. Mould-overflow which is mainly an operator error may also cause breakouts.  Factors influencing breakouts Breakouts are directly linked to the nature of shell growth -  which are further related to the characteristics of the oscillation marks. The formation of oscillation marks is related to the thermo-mechanical behaviour of the mould and to the nature of mould/shell interaction. In the above study, Samarasekera and Brimacombe [83] related breakouts to sticking in the mould, caused by poor mould lubrication, which results in the formation of transverse depressions (and cracks) and also to non-uniform cooling which cause the development of hot and thin corners. Therefore, factors influencing mould distortion, taper, lubrication parameters and oscillation characteristics are important issues. Factors causing breakouts can be identified by examining the breakout shells. When breakouts occurs close to transverse depressions (with cracks), it suggests strong linkages with the mechanisms involving transverse depressions. Hence, sticking in the mould due to poor lubrication or due to poor stripping action are important. In other situations when breakouts occurs close to the obtuse-angled corners, non-uniform cooling becomes a key issue. Non-uniform cooling arises from mould-tube misalignment, poor tolerances or design  87  parameters. It can also be related to deep and non-uniform oscillation marks. Higaki et al.[84] suggested that in billet casting, breakouts can be minimized by effective lubrication and by maintaining proper mould alignment. The dwell time in the mould is an important parameter which must be selected so that the shell thickness at the bottom of the mould is sufficient to counter the ferrostatic pressure exerted by the liquid core. Dean [85] has emphasized the importance of using an optimum mould length. This is because a mould that is too short will result in a thin shell, whereas excessively long moulds give rise to a higher frequency of breakouts from corners, through cracks which occur during resetting by the roller apron. The dwell time in the mould can be also be controlled by changing the position of the metal level or by changing the casting speed. Metal level fluctuations, by influencing mould lubrication, dwell time and casting speed, indirectly affect the nature of oscillation marks and therefore, breakout problems. Proper metal level control has been proposed to minimize breakouts [84,86].  5.3 Other Observations Observations made during casting can be used to focus the diagnosis, verify predictions from the mathematical model and also keep a check on conflicting inputs provided for analysis. Thus, they provide a means to integrate two kinds of, very diverse, expertise: fundamental knowledge about the casting process and operating experience in the plant. (a) Dark and bright patches on the billet surface at the mould exit : In the case of rhomboidity, the billet emerging from the mould has bright and dark corners. The bright corners which are thin and hot, become the obtuse-angled corners of the rhomboid billet. On the other hand, the dark corners are colder and form the acute-angled corners. This indicates varying amounts of  88  solidification, due to non-uniform cooling around the billet periphery. Non-uniform cooling conditions arise due to misaligned mould tube and/or non-uniform mould distortion. It could also be related to non-uniform oscillation marks- deeper oscillation marks at the hotter corners. In addition, the presence of dark and bright patches on the billet surface at mould exit can lead to the formation of midway cracks. The dark patches which are overcooled regions on the billet surface, tend to cool very rapidly in the spray zone due to nucleate boiling and later, reheat excessively in the radiation zone.  (b) Scale deposits on the cold face of the mould : The presence of scale deposits on the cold face of the mould wall indicates that the cooling water is contaminated. The colour of these deposits is indicative of the nature of contamination. The uniformity of scale deposits around the mould periphery is a check for misalignment and can be employed to assess the nature of the cooling.  (c) Change in billet colour in the radiation zone : A change in the billet colour from dark red in the spray zone to bright red in the radiation zone, is indicative of reheating of the billet surface and is very likely to cause midway cracks in the billet. This information can be used to verify the magnitude of reheat predicted by the mathematical model.  (d) Breakout shell : Breakouts can occur close to transverse depressions, cracks, deep oscillation marks, laps, slag patches and even thin and hot billet corners. Examination of breakout shells can be used to infer about the possible cause(s) of breakout.  (e) Condition of metal stream flowing into the mould : A ropey metal stream leads to excessive gas entrainment in the liquid steel. In the mould, these gases escape from the metal surface  89  as bubbles creating turbulence and therefore, metal level variation. This is also indicative of some deficiency in the tundish design with regards to controlling flow of metal to the inboard and the outboard strands.  (f) Distribution of pinholes on the billet surface : The presence of more pinholes in some regions of the billet as compared to others indicates hydrogen pick-up due to excessive lubricating oil on that face. If this problem was related to liquid steel, it would have resulted in a uniform distribution of pinholes on the billet surface. A deficiency in the lubrication system is indicative.  (g) Position of chromium layer discolouration in a used mould : This indicates the position of the meniscus during the operation. This information can be used to cross check the metal level during the operation and can be used to assess the efficiency of the metal level control system in use.  90  CHAPTER 6 KNOWLEDGE ENGINEERING 6.1 Knowledge Acquisition Knowledge acquisition consisted of four principal stages: identification/definition of the problem, acquisition of background knowledge on continuous casting, development of the expert system and testing and validation in the field. The system has not been put into actual use and therefore the last two stages described in Chapter 3 are not applicable at the present moment.  6.1.1 Identification/Definition of Problem Domain The problem identification stage was driven by the experts who wanted a system to help transfer their knowledge to industry. The experts defined the problem domain and the need for an expert system to diagnose quality problems in continuously cast steel billets and to train less experienced operating personnel. It was envisaged that the system could be used as a consultant by users and so provide the experts with more time for other activities. Thus, the problem modality was well-defined and the experts were ready and willing to participate in the development process.  6.1.2 Background Knowledge on Continuous Casting The experts provided the knowledge engineers with literature on continuous casting of steel billets - operations, design and product quality. One of the knowledge engineers attended a short course provided by the experts and their colleagues in Vancouver. This established a practical understanding of the problem domain and helped considerably in  91  subsequent discussions with the experts during development. Buchanan et al. [2] have pointed out that communication problems during knowledge acquisition can be severe when the knowledge engineer has far less knowledge about the domain than the experts. In this work, adequate steps were taken to ensure that this was not an issue. An outline of quality problems, comprising the origin of defects, their causes of formation and suggested remedies for prevention was prepared by the Knowledge Engineers. This preliminary categorization have been summarized in Tables 5.1 to 5.5. Although these tables contain the major elements of the knowledge domain, the methodology for diagnosis is much more complex than a simple "look-up" table. With a desire for a system possessing human-like thought and communication processes, considerable efforts was expended to structure the actual approach used by the Experts in the analysis of quality problems.  6.1.3 Expert System Development The problem domain consisting of cracks (internal and surface), rhomboidity and breakouts was initially broken down into individual modules- each of which could be worked on independently. Accordingly, the development process began with an examination of midway cracking as this problem was considered to be simple and well-understood. The Experts initiated the knowledge acquisition process by identifying that midway cracks were caused by reheating of the billet surface. The focus then, was aimed at analysing spray-related issues that are less complex than mould disorders. It quickly became apparent, however, that poor mould operation could also contribute to midway cracks. In the end, midway cracking turned out to be the most complicated problem from the viewpoint of diagnosis and knowledge representation.  92  The experts outlined the steps followed in dealing with midway cracks and explained the significance of operating parameters involved at each stage. The initial scheme to analyse midway cracks originating in the radiation zone is summarized in Table 6.1. Factors such as spray length, water flow rate, steel temperature and composition were considered to be important.  Table 6.1  h  in the Radiation Zone. DETAILS  STAGES 1  Input  Distance of the outer tip of the crack from the billet surface  2  Compare  Measured distance with solid shell thickness profile  3  Establish  Location of crack origin in the machine  4  Input  Model-predicted billet surface reheat for radiation zone Spray parameters - zone length and water flow rate High temperature strength of steel - Mn/S ratio Cast structure - Superheat level  5  Analyse  Compare input parameters with values specified by the experts  6  Correlate  Generation of midway crack with the operating parameters  7  Conclude  Conclusions and justification  Based on this information, a small prototype was developed. Almost immediately, questions were raised about the solid-shell thickness profile and the surface temperature distribution. In the expert system, it was assumed that the users would have prior information about these predictions. At this point, the system simply informed the users when operating  93  parameters were high, low, or acceptable, and therefore, its usefulness was nothing more than a check-list with input based on intuition rather than measurements or analysis. Thus, the first stage of implementation was not very successful. The knowledge base was reviewed and emphasis was re-focused on the consultation process used to deal with a client rather than the actual details of the data. The system was modified to provide conclusions and suggest remedial actions in the same way as the experts. The knowledge base was re-structured to facilitate examination of all parameters in the analysis. In addition, information for the users was incorporated into the system in the form of explanations, screen-messages and rule-descriptions, using the user-interface facilities of COMDALE/X. This approach was acceptable to the experts. At this stage, the expert system was capable of handling only typical situations which cause midway cracks to form - high reheat, high superheat and poor strength at high temperature. The Experts were consulted on other situations where midway cracks might be seen such as; when both spray and steel parameters were satisfactory, when only steel parameters were satisfactory and when only spray parameters were satisfactory. The Experts felt that it was impossible for midway cracks to form when both spray and steel parameters were satisfactory. However, when a problem related to steel quality was evident but spray parameters were satisfactory, the Experts believed that reheating must be related to a mould disorder. Under these circumstances, the presence of dark and bright patches on the billet surface at the mould exit would confirm a mould-related problem. These overcooled regions can reheat excessively in the radiation zone, leading to midway cracks. Therefore, a major expansion in the system was necessary to include knowledge dealing with mould parameters (distortion, oscillation, taper and lubrication).  94  It was pointed out that the Users would not be in a position to predict surface reheat levels and the solid shell thickness profile. A heat-transfer mathematical model for billet solidification was developed to predict these profiles. The model uses operating parameters as inputs and predicts solid shell thickness and maximum reheat temperatures at key locations in the casting machine. The details about the mathematical model have been discussed earlier. A typical output profile for a test situation is shown in Figure 6.1. Here, one-quarter of the billet section was modelled using a 30 x 30 mesh and a time increment of 0.10 second. Until this stage, the focus was on midway cracks originating in the radiation zone. But these cracks can also originate in the spray zone and in the region between the mould and the sprays (submould). It is difficult to visualize the origin of midway cracks in the spray zone because cooling rates in this region are high and therefore, the billet surface is unlikely to reheat. If midway cracks do originate in the spray zone, the Experts surmised that the only factor which could influence surface reheating is poor spray maintenance- damaged or blocked spray nozzles. The system was expanded to include the above situations. This completed the prototype for midway cracks and the next problem was selected for development. It was pointed out by the Knowledge Engineer that more than one quality problem might be observed at the same time. This meant that the plan to have a separate module for each problem was inappropriate. There was need for a general strategy to handle any combination of quality problems.  95  1550  80  1/  1325  0  1100 0^5^10^15^20  25  Distance from the meniscus ( m ) Fig 6.1^A typical output of the mathematical model showing predicted solid shell thickness and billet surface temperature profiles. (Casting speed- 25 mm per second; Section size- 150 mm square)  Development of Strategy for Handling Combinations of Quality Problems Based on knowledge about factors that influence the formation of quality problems in billets, the domain was divided into two broad groups. The first group consisted of those problems that require a detailed analysis of the mould and/or the spray cooling system. Off-corner cracks, transverse cracks and depressions, midway cracks, rhomboidity, diagonal cracks and breakouts are included in this category. The second group deals with very specific  96  issues; a detailed analysis of the mould and/or the sprays is not needed. Longitudinal corner cracks, longitudinal midface cracks, pinch-roll cracks, centreline cracks and craze cracks are in this group. The knowledge base was organized into three modules as shown in Figure 6.2 to handle combinations of quality problems and to accommodate the memory limitations of the DOS operating system. This classification is based on three different approaches identified for the diagnosis of quality problems that require detailed mould and/or spray analysis. Module-1 is designed to deal with situations where no midway cracks are present but any combination of off-corner cracks, transverse cracks and depressions, breakouts, rhomboidity and diagonal cracks are observed. In this situation, only a detailed analysis of the mould is necessary, although the system can provide warnings about possible spray maintenance problems. Module-2 examines cases that include midway cracks with any other type of quality problem. Mould issues are examined first followed by a detailed spray analysis to address midway cracking problem. Module-3 deals with midway cracks as the sole quality problem. In this case, the sprays are examined first. If a spray problem is uncertain, then mould issues are examined. As well, the mould may be examined if required by inference of the system or by request of the user. If such a request is not desired but the system detects certain mould disorders, a severe warning message is issued. In all three modules, there are rules dealing with quality problems that do not require detailed analysis of the mould or the sprays. Adverse mould-shell interaction is the root of all mould related problems. A thorough "breadth-search" of factors related to adverse mould-shell interaction, both at the meniscus and away from it, is conducted. Figure 6.3 shows the factors examined.  97  QUALITY PROBLEMS REQUIRING DETAILED ANALYSIS  Off-corner Crack  Midway Cracks  Transverse Defects  ASCERTAIN CRACK  ORIGIN  Rhomboidity Breakouts  Midway Cracks  DETAILED SPRAY ANALYSIS NO  DOES SPRAY PROBLEM EXIST ?  High Disbelief  YES illeb Belief  Vacertale  DETAILED MOULD ANALYSIS  DETAILED MOULD ANALYSIS  DETAILED SPRAY ANALYSIS  DETAILED MOULD ANALYSIS  MOULD ANALYSIS NEEDED  MODIFICATION OF BELIEF IN A SPRAY PROBLEM  I  CONCLUSION  Fig 6.2^Scheme for diagnosing quality problems requiring detailed mould and/or spray analysis.  98  ADVERSE MOULD-SHELL INTERACTION  AWAY FROM MENISCUS  MOULD TAPER  AT MENISCUS  MOULD LUBRICATION  MOULD DISTORTION  1: 011 Flowrate 2: 011 Distribution 3: OU-Phydcal Properties 4: OU System Cleanliness 5: Cleanliness ((Meniscus Region 6: Mould Cooling Water Temperature  1: Cooling Water Velocity 2: Cooling Water Quality 3: Mould Tube Alignment 4: Type of Mould Constraint 5: Mould Wall Thickness 6: Position of Metal Level 7: Mould Copper 8: Mould Design Tolerances  OSCILLATION CHARACTERISTICS  1: Negative Strip Time 2: Mould Lead 3: Nature of Oscillation Marks 4: Metal Level Variation  Fig 6.3^A general outline of mould-related factors contributing to quality problems in billet casting.  The important issues are mould-tube distortion, oscillation characteristics, mould lubrication and taper. Mould-tube distortion is related to cooling water quality and velocity, the nature of the mould constraint system, copper composition, wall thickness, metal level, tube alignment and design tolerances [68]. The analysis of oscillation characteristics involves evaluation of negative strip time, mould lead and the nature of oscillation marks - depth and uniformity across the billet surface. Important parameters dealing with lubrication in  99  the mould are oil flow conditions (flowrate and distribution) , physical properties of oil  (boiling point, flash point and viscosity), and the maintenance of the oil-distribution system. Finally, an analysis of the type of taper and its magnitude is conducted. Taper can influence the formation of both off-corner cracks and transverse depressions and cracks [68]. A detailed spray analysis is done to analyse the cause of midway cracks. In this methodology, positions of the outer tip of the midway cracks is compared with the predicted solid shell thickness profile and the crack origin sites in the machine are determined. This development had a major impact on the project as it presented to the Experts a structure that was very similar to their own. The Experts were able to observe links between various quality problems that had not been considered previously in the analysis. This strategy was also beneficial as it significantly reduced the amount of coding necessary as per the earlier plan since only three modules were needed.  6.2 Knowledge Representation  6.2.1 Knowledge units The development tool used for this expert system is COMDALE/X version 3.0. The main knowledge units available in this tool are shown in Figure 6.4. Here, facts are stored as keyword triplets. Procedures are used to control the inference process - breadth search versus depth search, forward chaining versus backward chaining and they are an important feature of this tool. The tool provides considerable flexibility in knowledge accumulation which are domain-specific [87]. It also allows specific facts to be represented as exclusive sets, fuzzy sets, restrictions, default values, etc.  100  KNOWLEDGE Classes I--- Objects — Attributes — Values — Degree of Belief  Keyword Triplets ^ Logical  E  Exclusive sets ^ Multichoice sets Fuzzy sets  String Numeric  Rules — Condition Statements  —  Conclusion Statements  —  Else Statements  E  Single Double Integer  Keyword Triplets Logical Connectives Predicates / Operators Functions Keyword Triplets Certainty Factors Assignments Functions  Procedures  Search Strategies Inference Strategies Input/Output Strategies Communication Control  Meta-Knowledge  Triplet Representations Customized Questions Customized Rule Descriptions Customized Explanations  Fig. 6.4^Knowledge units available in COMDALE/X Developent Tool.  101  6.2.1.1 Keyword Triplets A keyword triplet combines an object together with a particular attribute and value. Objects are physical entities in the real world such as sprays for cooling the billet. Objects have attributes which describe properties or characteristics such as "length" and "flow" for sprays. A value is attached to the object-attribute pair such as "short" and "satisfactory" for "spray length". Classes represent hierarchical relationships between objects. All objects in a class automatically inherit the attributes of that class. Examples of Keyword triplets are as follows: OBJECT ATTRIBUTE VALUE spray^length^short spray^length^satisfactory spray^flow^excessive spray^flow^satisfactory  A degree of certainty, which is a number ranging from 0 to 100, is attached to each keyword-triplet and this number determines whether the keyword triplet is significant or not at this stage of instantiation. Initially, the degree of certainty of the keyword triplet is "not known". When information about a keyword-triplet is needed but the degree of certainty is "not known", the system is forced to search through rules in the knowledge base. Keyword triplets together with degrees of certainty fonn the fundamental basis for representing knowledge as well as providing a driving force for the system in its search for knowledge during system-execution. Mutually Exclusive Sets: In the knowledge base, there are numerous keyword-triplets  which have common states of an Object's Attribute.  102  For example, the following keyword triplets describe different materials used as mould material: mould^copper^Pure Copper mould^copper^DHP_122 mould^copper^DLP 120 mould^copper^Copper Silver Alloy mould^copper^STP_Grade mould^copper^Cr Zr Copper mould^copper^Special Alloys mould^copper^Other Grades  Here, when one value has been instantiated with definite certainty, there is no need to search for the degree of certainty of the other possibilities. This is prevented in the system by defining these keyword-triplets as mutually exclusive sets. By doing this, the remaining keyword-triplets in the set are automatically assigned a certainty of 0. Fuzzy sets: There are occasions, when keyword-triplets with common object-attribute  pairs, but different values, are not mutually exclusive. Also, there can be a need to allow flexibility in defining multiple sets with varying degree of certainty depending upon a numerical value assigned to the keyword triplet. For example, three keyword-triplets are set up to characterize the negative strip time based on the operating parameters : calculated negative strip time^high calculated negative strip_time^okay calculated negative strip_time^low  Each numerical negative strip time value is assigned a membership percentage in each of these sets which is further interpreted as the degree of certainty of the respective  103  keyword-triplet. Fuzzy sets have been created to represent extremely subjective concepts in the knowledge base. The fuzzy sets in the knowledge base have been defined based on the experts' definition of the subjective concepts. The advantage of fuzzy sets is that it allows the knowledge engineer to program the entire set of conditions, associated with the subjective concept, using a much reduced rule set.  6.2.1.2 Rules In COMDALE/X, keyword triplets are linked through rules. A rule is a statement composed of a premise and a conclusion. A rule may contain more than one premise and conclusion statement. Each condition of a premise or conclusion consists of a keyword-triplet in which the attribute and value are separated by a predicate function. For example: IF^midway_crack origin^IS^radiant zone AND^radiant model_reheat^IS^high AND^billet colour^MIGHT BE bright red THEN radiant zone model reheat^IS^high CF = 100 # When a rule is fired, the inference engine examines each condition in the premise to establish the "degree of truth" about each statement. The degree of truth is calculated as a function of the degree of certainty of the keyword-triplet and the predicate function being used. After examining all the conditions in the premise, a net degree of truth is determined. The product of net degree of truth and the certainty factor associated with each statement of the rule conclusion is assigned to the degree of certainty of the keyword-triplet in the conclusion of the rule. COMDALE/X also has the facility to  104  establish complex mathematical relationships between the degrees of truth of each premise statement and the degree of belief in a conclusion fact. This unit is known as an "inference" knowledge unit.  6.2.1.3 Procedures Procedures refer to instructions which perform special functions during the execution of the expert system. These are similar to conventional programming where the sequence of tasks to be performed, as is the case with batch-processing, can be controlled through a series of statements. Table 6.2 is a list of library functions of COMDALE/X that were used in developing this system and the tasks each perform.  6.2.1.4 Meta-knowledge Meta-knowledge or knowledge about facts have been extensively used in the knowledge base with the purpose of providing technical assistance as well as training to the users. The user interface feature of COMDALE/X, which provides features like customized questions, rule descriptions and explanations, have been extensively employed in the knowledge base. An example of each related to the presence of dark and bright patches on the billet surface at the mould exit is given in the following paragraphs.  Customized Question : In the analysis of midway cracks, before the examination of mould-related factors, the system asks the user whether any dark patches are seen on the billet surface at the mould exit. The customized question posed by the system is:  Do you see any dark patches on the billet surface as it comes out of the mould?  105  Table 6.2 procedural Knowledge Units: Library Functions in COMDALE/X used for System Development ACTIVATE  Executes external programs or MS-DOS commands.  APPLYRULE  Conduct a breadth-search in which every rule containing a specific fact in its premise is fired.  ASNCERTAINTY  Assigns the degree of certainty of a keyword triplet to a numerical keyword triplet.  AS NVALUE  Assigns the value held by a numerical keyword triplet to another numerical keyword triplet.  DISPLAY  Displays the content of a document file of COMDALE/X.  EXPORT  Creates a file and writes data to it.  FIND  Conduct backward search for a particular fact. Here, every rule containing this fact in its conclusion is fired.  FORGET  Forgets a fact (key-word triplet) that was instantiated during the consultation.  FREERULE  Makes already evaluated rules available for future examination.  GOTO  Ditcates the next rule to be examined.  HALT  Terminates the consultation.  IGNORE  Ignores rules or rulesets in the knowledge base.  IMPORT  Reads data from a file.  LOAD  Loads a new knowledge module.  MACRO  Dictates the next rule to be examined and then returns control to the present rule.  TEXT  To present textual material.  VALUECERTAINTY  Assigns the degree of certainty of a keyword triplet to the value of a numeric keyword triplet.  106  Customized Explanation : The user may request information on dark patches before answering the above question. The customized explanation facility of the system provides the user with a text message containing knowledge about this subject:  These are dark regions present on the billet surface as the billet exits the mould. They are dark because of excessive localized cooling in the mould as a result of a mould disorder. Their presence suggests that there are problems in the mould cooling system. These dark patches cool very rapidly in the spray zone due to nucleate boiling. When the billet enters the radiation cooling zone, these over-cooled regions on the billet surface begin to reheat excessively leading to midway cracks. It should be noted then, that poorly designed sprays are not the only source of midway cracks.  Customized Rule description : The customized rule description has been used to provide the user an explanation of "why" a particular question is being asked. This is the english version of the rule currently being examined by the system. In response to a "why" from the user for the above question related to the presence of dark patches, the following text message is presented:  If dark patches are seen on the billet surface as it leaves the mould, then an adverse mould-shell interaction is confirmed.  6.2.2 Search Techniques and Conflict Resolution By default, COMDALE/X uses a forward chaining depth-first strategy that allows backward-chaining to interrupt the search for a fmal conclusion whenever sub-goal information is required. However, the tool allows backward-chaining and breadth-search control strategies to occur from rule conclusions through the use of library functions such as  107  "FIND" and "APPLYRULE". A breadth search has been used for the analysis of mould-related parameters, where all possible factors such as distortion, taper, oscillation and lubrication need to be examined to arrive at the correct diagnosis. A backward-chaining approach has been selected to modify belief in a spray-related problem for midway cracks based on reheat temperature, superheat and high temperature strength of steel. The accuracy of the heat-transfer model is limited only by certain assumptions made about the physical characteristics of the casting machine and the associated cooling conditions. For example, significant blockage of spray nozzles could lead to localized reheating when the model is suggesting no reheating. Mould design and operating factors could also contribute to reheating in the radiation zone by producing a non-uniform solid shell in the billet. These non-uniformly cooled regions appear as dark and bright patches on the billet surface at the mould exit. As pointed out earlier, these dark patches reheat significantly in the radiation cooling zone. The model cannot predict reheating in these situations. Measurement estimates and the large subjectivity associated with visual observations puts increased doubt on the reliability of these sources of information. Therefore, facts obtained in these ways are cross-checked by the system wherever possible. When contradictions do occur, the system resolves them in a manner comparable to the experts. An example is the conclusion reached about the severity of a rhomboidity problem. At the start of the consultation, the expert system asks the user to select his choice ("yes" or "no") from a list of quality problem(s) present in the billet. Let's say, the user said "no" to rhomboidity as well as "no" to diagonal cracks.  The system verifies these inputs by asking  the user to enter the difference in the lengths of billet diagonals. If the user enters a value which is large as per the experts' definition, the system concludes that rhomboidity is severe.  108  entered a small value for the difference in the lengths of billet diagonals, the system informs the user that rhomboidity is severe and that the measurements is incorrect. This is done because the presence of diagonal cracks in the billet is directly indicative of a severe rhomboidity problem. The following example illustrates a conflicting situation encountered during knowledge processing and the approach adopted for its resolution. Let's assume that the midway crack origin corresponds to the upper part of the spray cooling zone. This clearly suggests reheating of the billet surface in this zone. However, the model does not predict excessive reheating for this region because it is assumed that all parameters are satisfactory. This prediction therefore does not conform to the conditions prevalent in the operation. This situation is resolved by concluding that spray nozzles, particularly those lying on the same side of the billet as the crack, are plugged or damaged and that the cooling is insufficient.  6.2.3 Information used in the analysis There are three different sources of information used by the system namely; user-input, calculations from external programs and facts inferred from rules in the knowledge base.  Information from users  : Users enter the majority of data through an external program. This  information relates to the operating parameters such as steel composition, casting temperature, casting speed, oscillation frequency, oscillation stroke, billet dimensions, mould cooling water flow rate, spray cooling water flow rate and lubrication oil flow rate. Other parameters such as machine radius, position of spray nozzles and pinch-roll/shear from the mould exit, mould parameters - length, metal level, corner radius, wall thickness, water channel gap, type of copper and nature of constraints, spray length and physical  109  properties of lubrication oil are also included. In addition, during the consultation, the system requests observations needed for the analysis. The user is asked about the presence of dark and bright patches on the billet surface at the mould exit, colour change in the billet in the radiation cooling zone and nature of scale deposits - severity and colour. Questions are asked about measurements such as mould tube distortion, mould cooling water velocity and oil distribution. This data provide the main source of information for the expert system. Also, users are asked to examine a macro-etched transverse section or a sulphur print of a billet and enter information about the outer tip position of any internal cracks seen.  Calculated information : The heat-transfer finite-difference model calculates solid shell thickness profile and billet surface reheat temperature. It is impossible for the users to provide an estimate of these prediction. The mathematical model, with the help of additional subroutines, also calculates mould cooling water velocity (linear) based on mould-design and cooling water flow rate (volumetric), negative strip time, mould lead and specific spray water flow rate. This program creates a data-file which interfaces with the expert system.  Inferred information : During consultation, the system concludes about facts based on rules in the knowledge base. For example, deep and non-uniform oscillation marks are inferred from knowledge about the uniformity of the solid shell around the billet periphery. The presence of off-corner cracks and also dark and bright patches on the billet at the mould exit also suggests that the oscillation marks are deep and non-uniform. Frequently, off-corner cracks have been found together with deep oscillation marks. In case, the system is unable to infer about "deep and non-uniform oscillation marks", it asks the user for measurements of the oscillation marks and then tries to conclude about "deep and non-uniform oscillation marks" based on the Experts' definition of "deep" and "non-uniform".  110  6.2.4 A novel inferencing strategy for the analysis of midway cracks The appearance of midway cracks in a billet suggests that solid steel close to the solid-liquid interface has been subjected to stresses and strains during processing. The generation of such stress is frequently due to reheating of the strand surface due to a sudden reduction in heat extraction. Reheating causes surface expansion and imposes a tensile strain in the interior region of the solid shell, which is weak and brittle at temperatures within 50°C of the solidus temperature [42]. The level of reheat determines the amount of stress generated and therefore the intensity of a cracking problem. It is assumed therefore that belief in a high stress level is equivalent to the degree of belief in a "high" reheat. Belief in a "high" reheat level is determined by defining a fuzzy set for the adjective "high":  Radiation Zone Reheat Temperature (°C)  0  50  75  100  150  Degree of belief in a high reheat (%)  0  25  50  75  100  It is known that a low Mn/S ratio and a high superheat are two operating factors that can aggravate the cracking susceptibility of steel when the reheat is high [40]. Although the presence of a high stress level is mandatory for crack formation, problems with composition and superheat can lead to cracking at lower stress levels and at lower reheat levels. On their own, superheat and composition problems are not of concern since cracks cannot initiate in the absence of a stress. Figure 6.5 depicts how belief in a high stress level is inferred. In the methodology adopted, belief in a composition and/or superheat problem must be higher than belief in a high reheat in order to increase belief in a stress problem when a high reheat "might" exist. If a high reheat "might not" be present, then belief in a stress problem can only be increased  111  if the composition or superheat problems are known with a certainty greater than 50 percent. Thus, when the fact "reheat temperature is high" is true or false with equal certainty (degree of belief = 50 percent) the absolute presence of one or both of these problems will increase belief in a stress problem to 70 or 90 percent respectively. Belief in a spray-related problem may be affected when a mould problem is also identified. If the system discovers a mould problem, belief in a spray-related problem will be reduced, provided belief in the mould-related problem is greater than belief in the spray-related problem. Thus, when the fact "midway cracking problem is spray-related" is true or false with equal certainty (degree of belief = 50 percent) the absolute presence of a mould-related problem will reduce belief in a spray related problem to 30 percent. Figure 6.6 illustrates this reduction in belief in a spray problem when a mould problem is identified with 100 percent certainty. When there is uncertainty associated with the fact "reheat temperature is high", the absolute presence of problems associated with both superheat and high temperature strength, cannot increase belief in a spray related stress problem to 100 percent. This is logical since "a high reheat temperature" is known with less than full certainty and the modified belief in  a spray-related stress problem must reflect this uncertainty. Thus, when there is equal uncertainty about the fact "reheat temperature is high" being true or false (degree of belief = 50 percent), the maximum possible belief is 90 percent. Using the same logic, the absolute presence of either problem ( high superheat or poor high temperature strength of steel ) will increase belief in a spray-related stress problem to 70 percent. Similarly, when the fact "midway cracking problem is mould-related" is true with a 100 percent certainty, belief in spray related problem is reduced to 30 percent.  112  100  D  ----_----^..----- -A /---___- — — .- ,-/ ,......---/ / __..^--- ..-/ ...B7 I / / / / I/ / / 1 t.. /^, / / / / / / / / / 1 / / // OCD Only Reheat // -  / / / / / /  .  Problem OBD - Composition^Superheat "OR" OAD - Composition "AND" Superheat Problem  // // //  0^  50  ^  100  BELIEF IN HIGH REHEAT DUE TO THE SPRAYS ( % )  Fig 6.5^A schematic diagram illustrating the calculation of degree of belief in a spray-related tensile strain problem from belief in high reheat in the radiation zone, high superheat and composition problems.  113  C  / / , /  t/  , /  / / / /  / /  / /  / /  .  .. ..  B / . A  ...-"  ■  OBC-NohloWdProbkm  ■ .."'  OAC-NloWdProblem  . ..  ■ ■  0  ^  50  ^  100  INITIAL BELIEF IN SPRAY-RELATED STRAIN PROBLEM ( % )  Fig 6.6^A schematic diagram illustrating the calculation of fmal belief in a spray-related tensile strain problem following the detection of a mould disorder.  114  Thus, the contribution of belief in these effects on increasing or decreasing belief in a spray-related stress problem is considered to be similar. The influence of a mould problem on belief in a spray problem is only taken into account after the effects of superheat and high temperature strength of steel have been considered.  6.3 Structure Of The Expert System Figure 6.7 outlines the methodology adopted in this system for the analysis of quality problems. At the start of a consultation, the system acquires operating parameters with the help of an external program. The heat transfer model is executed to calculate the solid shell thickness and the surface temperature profiles. An output file is created to interface with the expert system shell, COMDALE/X. The user is then prompted to describe the quality problems and to select the most important one. The outputs from the mathematical model as well as the quality problems entered by the user are "imported" into the "main" module of the system. In the "main" module, the system verifies the quality problem(s) entered by the user. In the case of midway cracks and pinch-roll cracks for example, the user is asked questions on the distances of the outer tip of the cracks from the billet surface. This establishes initiation sites for midway cracks and confirms or denies the observation about pinch roll cracks. For the most important quality problem, the system addresses issues related to the frequency of the problem with regard to particular strands, grade of steel, time in the heat, location and orientation. One of the three analysis modes is then used depending on the combination of quality problems observed. Prior to presenting a summary of the analysis in the "report" module, the system provides the user with an opportunity to examine all final conclusions and to probe into the knowledge base to obtain justification for each piece of advice.  115  OPERATING PARAMETERS INPUT HEAT TRANSFER MODEL  QUALITY PROBLEMS INPUT  MAIN MODULE Analysis of specificity associated with the most important problem  1  I  MODULE 1 I 1 MODULE 2 I I MODULE 3  I  REPORT I  Fig 6.7 A flow-sheet representing the structure of the expert system.  116  The following are some of the highlights of this system: * RULES Main Module Module - 1 Module - 2 Module - 3 Report Module  310 300 384 362 94  * KEYWORD TRIPLETS  552  * FUZZY-SETS  68  * MUTUAL-SETS  9  * EXPLANATIONS  73  * QUESTIONS  73  * INPUT DATA ELEMENTS  44  * DISPLAY FILE PAGES  271 pages in 10 files  * FINAL CONCLUSIONS  6 ( 30 )  * EXTERNAL PROGRAMS  3  - Data Input - Quality Problem Input - Heat Transfer Model for Billet Solidification  6.4 Justification of Knowledge Engineering Decisions During the development of the expert system, various decisions concerning knowledge engineering were made to accommodate constraints related to computer software (COMDALE/X version-3.0 and the operating system, MS-DOS version-4.01) and the complexities involved in the analysis of quality problems. Wherever possible, the input of the  117  user is tested for consistency and correctness. Sometimes, this is irritating for the users. For example, one user pointed out that the system always asked for the difference in the lengths of the diagonals, irrespective of whether a rhomboidity problem was selected or not. (a) Due to the large size of this knowledge base and also the memory limitations imposed by the operating system, it was decided to break the knowledge base into five modules, four for analysis and one for report-generation. The details about these modules have been discussed earlier. The modules are linked to each other through data-files. There exists a potential problem during justification of the final conclusions - it is difficult to chain back for a particular fact present in a different module. At this point, the system ends the justification process by issuing a statement saying "information was imported from data file". Thus, a complete justification is not possible with the approach taken in constructing this system. (b) A batch-file has been used to provide connections between the various programs used in this expert system because of memory restrictions. Due to memory limitations, it was not possible to "ACTIVATE" external programs from within the knowledge base and also it was difficult to "LOAD" one module from another. Batch-processing also allowed the users to "REDO" a consultation with a new set of input data and quality problems. (c) According to the Experts, mould parameters are linked to each other in a complex manner and therefore, when a quality problem is mould-related, all parameters need to be analysed. Therefore, decision was made to adopt a "breadth-search" strategy for mould analysis using a COMDALE/X function, "APPLYRULE". This slows down the consultation considerably since this function examines all rules dealing with a particular keyword triplet, whether successful or not. However, with the help of "IGNORE-RULE and RANGE" functions, it  118  was possible to enhance speed of the consultation to a great extent. Here, once a rule in a rule-set is successful, all remaining rules are ignored. However, if no rule is fired, then all possible members of the set must be exhausted during the search.  (d)In the case of a midway crack analysis, the system examines the possibility of a mould disorders even when there exists a high belief in a spray problem. The user is asked about the presence of dark and bright patches on the billet surface at the mould exit. If the response is "no", the system asks if the user wishes to go through a detailed mould analysis. There is a large emphasis on the mould because the experts strongly believe that mould issues must also be considered. In fact, when the user does not want to go through the mould analysis, the system issues a warning message- "be forewarned - a mould analysis may be essential for complete diagnosis of the problem cause". (e)During the development, it was decided to transfer all the "TEXT-MESSAGES" in the knowledge base into external "DISPLAY-FILES". This was beneficial because it enhanced the speed of the compilation process. Maintenance of the knowledge base was easier since corrections to the text could be made independently of the knowledge base. However, the justification feature (how the system found out about certain facts presented in the text-message) associated with "TEXT-MESSAGES" was lost. The "DISPLAY-FILE" does not allow the users to have immediate justification. (f) The system summarizes its final conclusions under one or more of the following: *  Quality problem is mould-related  *  Quality problem is spray-related  *  Quality problem is related to steel  119  *  Quality problem is related to pinch-rolls  *  Abnormality in certain operating parameters is present  *  Additional information is needed  This approach was taken to prevent cluttering of the conclusions. A user can justify each of them as usual. The following is the rule structure used for the conclusion "quality problem is mould-related": RULE 289 IF^mould problem is due to distortion OR^mould taper is a potential problem OR^mould problem is related to oscillation system  OR^mould problem is related to lubrication system OR^metal level control in the mould is a problem OR^longitudinal midface crack is mould-related OR^longitudinal corner crack is mould-related THEN quality problem is mould-related cf=100  (g) In the above approach, a fmal conclusion is made up of different facts, some of which may or may not be instantiated during the consultation. In case, a particular fact is not instantiated by the system, the fact is "considered to be "unknown". During justification, the system indicates that the "degree of certainty of the fact is "unknown". This information is not required during justification and therefore required elimination. In order to do this, additional rules  120  were created to "FORGET" these facts before the final conclusions were presented. Since, the number of such facts was large, the knowledge base increased in size. This also slowed down the consultation process. (h) In addition, in this approach of summarizing final conclusions, all facts associated with the conclusion have been given equal weightage. This is again based on the complexities associated with mould analysis and also on the Experts' view that all mould parameters must be within limits for good operation. (i) In the inference strategy for midway crack analysis, upper limits of 70 and 90 percent and lower limit of 30 percent were arbitrarily selected. This may need some refining based on users' feedback. In addition, the variation is assumed to be linear - an asymptotic variation may be better, although the author believes this to be only of academic interest. (j) Reheat predicted by the mathematical model for the radiation zone is verified based on users' observation of billet colour. There are two conflicting situations encountered and they are resolved as follows: * When the model predicts a high reheat and no colour change is observed, the degree of belief in high reheat is reduced by a factor of 0.90. When the model does not predict a high reheat and a colour change is observed, the degree of belief in high reheat is increased from high disbelief (below 50 percent) to 80 percent. Again, these limits are arbitrary and only reflect the decision of the Experts and their approach.  121  (k) At the end of the consultation, important points in the analysis are summarized dynamically in a report where values of concerned operating parameter are listed and compared with those recommended by the Experts. Against each parameter, comments of the Experts are provided. Comments have been created with the help of embedded keyword triplets. These keyword triplets are strings which are instantiated during the consultation process. In the end, these triplets are "EXPORTED" to a data-file which is subsequently "IMPORTED" by the report module. For creating the report, a "WRITE" function was used. This function was developed at UBC and linked to the inference engine of COMDALE/X. It creates an ASCII file into which concerned information is written. With the help of another external program, this ASCII file is displayed. This program also provides user-friendly features such as, "quit", "page-up" and "page-down".  122  CHAPTER 7 TESTING AND EVALUATION OF THE EXPERT SYSTEM  7.1 Evaluation Procedure In this work, the system was evaluated in two stages: informal evaluation by the Knowledge Engineers and the Domain Experts and a formal evaluation by the End-users. (a) Informal Evaluation: This stage formed an integral part of the knowledge acquisition and  representation where the system's performance was continuously assessed by the Experts and the Knowledge Engineers. Recommendations and feed-back were incorporated immediately into the system. (b)Formal Evaluation: At a later stage in the development, the system was installed at several  Canadian steel companies for formal evaluation. The procedure followed was: (i)  Five Canadian steel companies were selected. The operating personnel in these plants were willing to participate in the evaluation process.  (ii)  A list of operating parameters for input needed in the analysis (as shown in Appendix-A) was prepared and despatched to the users before evaluation. This provided the operators with sufficient time for collecting data. A list of basic computer (hardware/software) requirements necessary for the implementation of the system was also sent to them. The list is provided in Appendix-B.  (iii)^One of the Knowledge Engineers installed the system in each of the companies. A brief demonstration of the system was presented to the operating personnel.  123  (iv)  One person was identified in each company for co-ordinating the evaluation process. The Knowledge Engineer explained the features of the system in detail. The system was consulted for quality problems using operating plant data.  (v)  Preliminary feedback was obtained at this stage. Errors in the computer code were discovered and corrected immediately. A summary report (shown in Appendix-C) was prepared by the Knowledge Engineer.  (vi)  A questionaire was distributed to the Users for providing detailed feedback after using the system for a longer period. Completed questionaire were obtained from three companies. The responses/comments obtained from them have been summarized in Appendix-D.  (vii)  The feedback obtained was analysed and decisions for implementation were made by the Experts and the Knowledge Engineers.  7.2 System Operation - Results of the feedback This section summarizes feedback associated with various aspects of the expert system based on the preliminary feedback and those provided in the questionaire. (a) Operating Parameter - Input Module * This module serves its purpose of providing a way to enter data quite adequately. However, some Users found it a little slow to operate as only one parameter can be changed at a time. In addition, the units assigned to these parameters do not match with those used in the plant.  124  •  The terminology used is easy to follow for the "educated" Users such as metallurgists or process-control engineers. But there may be a problem with operators on the shop-floor. The Users feel that they are aware of the role of each input in the analysis and that the data is available easily.  •  Even though additional information on measurement techniques is adequate, some Users feel the need to include a glossary of terms used in the casting business.  (b) Quality Problem - Input Module *  The terms used in this module are easy to understand. However, Users also feel the need to have some background information on quality problems in the cast product.  *  There is a need to include more quality problems in the list such as laps, bleeds, centreline segregation.  *  There is no problem in handling the issue of selecting the most important problem.  (c) Mathematical Model *  Additional information is needed on the mathematical model- formulation of governing equations, underlying assumptions and its role in the analysis of quality problems in billet casting.  •  One User felt that the time-delay associated with the model was significant. However, other did not find this issue to be a problem.  *  The output format is satisfactory. However, Users feel that graphical outputs of the model predictions will be more useful.  *  One user pointed out that the shell thickness predicted by the model did not match with those measured in the billet (based on his judgement). There exists doubts about the  125  boundary conditions used in the model. The Users pointed out that the spray configuration as well as the mould heat-flux differs from plant to plant and this must be accounted for in the model. (d) *  Customized Questions. Explanations and Rule Description Questions asked by the system and associated rule-descriptions and explanations are clear and easy to follow.  *  The technical level of information provided through the customized explanations and rule-descriptions is adequate. However, there still exists a need to account for different user expertise-levels in these presentations.  *  The customized features of COMDALE/X have a great potential in training novices.  *  No discrepancy was noted in the information provided through the customized explanations and rule-descriptions. The Users agreed with the contents.  (e) *  System's Recommendations The recommendations are accurate, complete, clear and easy to understand. They are easy to implement. (one User was neutral).  *  No discrepancy was encountered in the recommendations provided. The Users agreed with these recommendations. (one User did not answer questions related to discrepancy and disagreement)  *  In addition, the Users feel the need to know about other potential problems which may start coming up based on the existing operating conditions.  (f)  System's Final Conclusions and Justification  *  Final conclusions are complete, clear and easy to understand (one User was neutral).  *  The justification scheme is useful, clear and easy to understand (one User was neutral).  126  * The justification feature is a potential training tool for novices. * The logic adopted by the system for analysis was mostly satisfactory. Only one deficiency was pointed out. The explanations provided by the system for transverse depression in AISI 1018 grade steel was found to be inadequate by one User. (g) Consultation Session * The system is adequate at diagnosing and for training. However, some Users feel that its operation may be a little complicated for personnel on shop-floor. This may hamper the issue of training these operators. * The consultation sessions were found to be long for some cases. * The system informs the Users about its course of action at each stage in the analysis. Changes in the system's thought process are smooth. Thus, the User is aware of the analysis procedure and this feature of the system makes it very easy to understand. * Report is useful, satisfactory and complete. * Problems related to insufficient computer memory were encountered. * One of the Users disagreed with the approach adopted for concluding about a severe rhomboidity problem. His comment was, "rhomboidity is always a problem". The experts pointed out that their definition of rhomboidity is satisfactory.  73 Analysis of User Feedback  7.3.1 System's Successes (a) The system proved to be successful in linking dominant problems at several sites with  expected causes. Some examples are as follows:  127  Company A The system related a problem with rhomboidity to plugged spray nozzles. The operators solved this problem by improving the spray cooling water quality.  Company B Off-corner cracks were found to originate near the mould exit. The system suggested a mismatch between spray and mould heat extraction rates due to a large gap between the mould exit and the first spray nozzle.  Company  The heat transfer model calculated a high reheat in the radiation zone. Midway cracks were seen in the radiation zone. The cause of this high reheat was a short spray zone.  (b) The following features of the system were useful: * The User is informed about what the system is doing at each stage during the consultation. * The mechanisms involved in the generation of the quality problems are explained at relevant stages. * Text-messages, explanations and rule descriptions are important tools for training. * The system not only points out flaws in design and operation but also informs the Users about those parameters that are satisfactory. (c) The knowledge structure provides the Users with a consistent interpretation of the domain knowledge. (d) The system serves as a valuable consulting tool for operators during brain-storming sessions for solving quality problems.  128  7.3.2 System's Drawbacks (a) During an analysis on transverse depressions (and cracks), the system reached the limit of its knowledge. When consulted, the Experts explained that the problem was related to the effects of oscillation frequency on heat-transfer in the mould (a recent finding). Expansion of the knowledge base is necessary to include this knowledge. (b) It was pointed out that the system cannot be used for day-to-day quality-problem because of the approach adopted. Some of the recommendations require large scale modification to the operation. However, inexperienced operators can use it for learning the subject. (c) There is a need to recognize different levels of expertise among the possible Users since the system considers all the Users to be the same. (d) There are severe computer memory limits imposed by COMDALE/X version-3.0 and MS-DOS version-4.01. Hence, expansion of the knowledge base under the present constraints is difficult. * RAM (computer memory) available for application is around 580 Kb. The size of the inference engine "COMDX.EXE" is about 385 Kb. The knowledge base can only be expanded to 160 to 170 Kb. Problems of memory are encountered when MS-DOS commands are activated from within the COMDX program. * External programs such as "DATA-INPUT' program or the "HEAT-TRANSFER MODEL" are difficult to "LOAD" from within the COMDX program because of their large size. (e) Errors in the code- omission, misprints, incorrect units, etc. were present. These were corrected.  129  (f)  At the present moment, the system is incapable of handling different spray configurations. The inputs to the mathematical model are number of spray zones, zone length and water flow rate. As well, the data-base of mould heat-flux for different casting conditions is not complete.  (g)  The Users found the data input stage to be very cumbersome.  (h)  The units assigned to operating parameters in the data-input module does not match with those followed in the plant. The Users are uncomfortable with this. The major problem in most plant is the inconsistency in the use of unit-systems, FPS, METRIC or SI.  (i)  The system does not attract the attention of the Users. As a starting point, the focus of this project was the knowledge base. Efforts will have to be made to enhance the user-interface.  733 Areas of Future Expansion (a)  Expansion of the present knowledge base is needed to include centreline segregation, laps, bleeds, blowholes and pinholes.  (b)  There is a need to enhance the user-interface such as graphical output profiles of billet surface temperature and solid shell thickness.  (c) Mould heat-transfer boundary conditions are not available for different companies. The system uses a general heat-flux profile. Also, the system does not take into  130  account the effects of oscillation frequency on mould-heat transfer (recent findings). Data base of heat-fluxes for different grades, casting speeds, oscillation frequencies and mould-taper has to be developed. (d) There is a need to recognize different expertise levels for the Users while presenting explanations, rule-descriptions or text-messages.  7.4 Implementation Of Feedback Due to the constraints, a decision has been made to change to a newer version of COMDALE/X-version-5.00. Important features of this new tool are as follows:  COMDALE/X version 5.0 * Runs under WINDOWS/3 environment which is easy-to-use and does not have memory limitations. * Handles different expertise levels for the Users. * Has a form-generator which can considerably ease the data-input process. * Hypertext is an inherent feature of this environment which can enhance the presentation of "help", "text-messages" and "display-files". * Has enhanced graphics features, much better than the previous version. * The multi-tasking feature of WINDOWS/3 can be utilized to reduce waiting time during the execution of the mathematical model.  131  * There are also new features for the developer: - Incremental Compiler - Rule and Class Editors - Rule and Class Browsers - Dynamic Objects  implementation Scheme The decisions made to implement the feedback are summarized in Table 7.1.  Table 7.1^Feedback Implementation Strategy FEEDBACK  COMMENTS  *  Include other quality problems  The knowledge base will be expanded and modified to include these problems.  *  Enhance user-friendliness and graphics.  Will be implemented in version 5.0 of COMDALE/X.  *  Need to recognize different user levels  Will be implemented in version 5.0 of COMDALE/X.  *  Ease data input process  Use of form generator provided in version 5.0 of COMDALE/X will be useful. Also, explanations will be provided for each input parameter.  *  Add spray configurations in the mathematical model  Mathematical model will be expanded to include a subroutine dealing with spray configuration.  *  Establish boundary conditions for the mathematical model  Data bases will be created for a variety of conditions and for each company through industrial trials.  132  CHAPTER 8 CONSULTATION SESSIONS - CASE STUDIES  In this chapter, three cases have been presented to highlight the consultation process. The operating parameters for the three cases have been provided in Appendix E. After data input, the heat transfer model for billet solidification is activated to calculate shell thickness and billet surface temperature profiles. With the help of an additional subroutine in this program, mould cooling water velocity, negative strip time and mould lead are calculated. The outputs are written to a data file which is subsequently acquired by the expert system module. At this stage, quality problems are selected by the user from a list. CASE 1 Midway Cracks  Let's assume midway cracks were seen in the billet (at depths between 22 to 25 mm from the surface of a 150 mm square billet). No other problem was indicated by the user. The system begins the consultation by presenting the following predictions of the mathematical model: Solid Shell Thicknesses  Maximum Reheat Temperatures  Mould exit  = 7 mm  Submould  = 25 °C  Spray entry  = 9 mm  Radiation zone  = 158 °C  Spray midpoint  = 12 mm  Spray exit  = 18 mm  Pinch roll  = 31 mm  The user did not select rhomboidity. The system verifies this information by requesting the difference in the lengths of the two diagonals. The following is the dialogue that takes place between the system and the user.  133  System The difference in the length of the diagonals of the billet is a measure of rhomboidity.  What is the difference in the length of the two diagonals in mm? User Why are you asking me this question? System You have indicated that the billet is not rhomboid. This suggests that rhomboidity is  not a problem. However, if the difference between the lengths of the two diagonals is large, then a rhomboidity problem is confirmed. User Please explain the meaning of this question. System The presence of rhomboidity indicates that the heat extraction rates are not uniform  on all foursides of the billet. Non-uniform heat removal suggests that oscillation marks are probably unevenly deep. We can confirm rhomboidity from the difference in the length of the two diagonals. A difference greater than 6 mm is unacceptable to the rolling mills. A difference greater than 4 mm strongly suggests problems related to non uniform cooling. User The difference between the lengths of the diagonals is 2 mm. Thus, rhomboidity is not a problem. The session continues with a series of questions to establish the location of cracks in the casting machine. Since the outer tips of the midway cracks are located between 20 to 25 mm from the surface of the billet, the system concludes that these cracks originated in the radiation zone. The system also cross-checks the observation about pinch-roll cracks. It confirms that pinch-roll cracks are truly absent since no cracks originated at depths between 28 mm to 34 mm from the billet surface. Since midway cracking is the only  134  quality problem, specific issues related to midway crack(s) are analysed. These include observations about the frequency of the problem with respect to strand location, time in the heat and steel grade. To confirm belief in high reheat as predicted by the model, the system looks for information on the colour of the billet. System A change in the colour of the billet from red ( in the sprays ) to bright red ( in the  radiation cooling zone ), if observed, will help confirm or deny the belief in a high radiation zone reheat as per the model prediction. Colour change should be observed if reheating is severe. Do you see any such colour change? User No Since a colour change was not observed, the system reduces its degree of belief in the model-predicted high reheat from 100 percent to 90 percent. The following message warning the user of this contradiction is generated to suggest that the input data may be faulty. System Our belief that the radiation zone reheat is high based on the model prediction is 100  percent. Since you did not observe a colour change in the billet in the radiation zone, the model prediction is not confirmed. Belief in a high radiation zone reheat must be reduced to 90 percent. The belief in a strain problem is then modified based on the presence of a composition problem or a superheat problem as shown in Figure 6.5. The system examines the sprays first since belief in a spray-related strain problem is high. Spray parameters include spray length and specific spray water flow. The levels of these are assessed with comments and recommendations presented by the system as follows:  135  Radiant Zone Reheat is 160 °C Spray Water flow is 127 litre water / kg steel. Spray Length^is 2.07 m. We are 91 percent sure that faulty sprays have generated a high stress level. This is due to problems with the spray water parameters. We are 100 percent sure that the spray flow is excessive and 69 percent certain that the spray zone is short. RECOMMENDATIONS: 1. Reduce the specific spray water flow to 1.0 liters/kg of steel produced. 2. The length of the spray zone should be increased to at least 5 m. Even though a problem related to the sprays was present, the system still examines the possibility of a mould disorder. The system inquires about the presence of dark and bright patches on the billet at the mould exit. Let's say that no such patches were seen. The system then asks if the user wishes to go through a detailed mould analysis. Let's say the answer is "no". A detailed mould-analysis is not done. However, the following warning is presented.  Dark patches were not seen on the billet surface at the exit of the mould. Since you do not want to conduct a detailed mould analysis, we will not examine the mould parameters. However, BE FOREWARNED - a mould analysis may be essential for complete diagnosis of the problem cause. The analysis is then complete and the system presents the following final conclusions for justification:  * Quality Problem is Spray-related. * Quality Problem is Related to Steel Parameters. Let's evaluate the conclusion related to steel parameters.  136  Quality problem is related to steel parameters because of a rule that states.. Quality problem related to steel parameters can occur if problems exist with superheat and/or steel composition. And it is known that.. 1.It appears that high temperature strength of steel is poor 2. I am certain that a large columnar zone is present Let's examine reasons for a large columnar zone.  I am certain that a large columnar zone is present because of a rule that states.. A high superheat leads to an enlarged columnar zone which provides an easy path for crack propagation. The effect of superheat is more pronounced in high and low carbon grade steels as compared with medium carbon grades. And it is known that.. 1.I am certain that cast superheat is high 2. Cast grade is high carbon. Let's see why cast superheat is high.  I am certain that cast superheat is high because cast superheat = 33  r  and a high superheat  level is described by the following fuzzy set:  Superheat (°C)  5  10  15  20  25  30  35  40  Degree of belief (%)  0  5  10  45  80  90  95  100  137  In the same way the user can justify each of the remaining conclusions. At the end of the consultation, the system summarizes the important points of the analysis and compares existing operating parameters with those recommended by the experts. Table 8.1 shows the final report.  Table 8.1^Final Report  SUMMARIZED REPORT ON BILLET QUALITY PROBLEMS QUALITY PROBLEM - Midway Cracks LIQUID STEEL  YOUR  RECOMMENDED  PARAMETERS  VALUE  VALUE  Mn/S Ratio  27  above 25  satisfactory  Superheat (°C)  58  below 25  too high  Copper Level  0.15  below 0.20  satisfactory  Length (m)  2.1  above 5  should be increased  Specific Flow (1/kg)  1.3  below 1.0  should be decreased  Nozzle Blockage  NO  NO  no problem  Colour change observed  NO  NO  check observation  COMMENTS  SPRAY PARAMETERS  138  CASE 2 Off-corner cracks and Transverse defects (cracks and depressions).  Most Important Problem : Transverse Depressions The system begins the analysis by presenting the model predictions. The system then verifies the information about rhomboidity. Let's say the difference in the lengths of the diagonals is 5 mm. The following warning message informing the user about a severe rhomboidity problem is presented. You did not observe rhomboidity in the billet. However, the difference in the lengths of the diagonals is 5 nun. We are 75 percent sure that this is large. This strongly suggests that the billet is rhomboid and the problem is severe. Please note : RHOMBOIDITY PROBLEM IS SEVERE !!  The system then includes rhomboidity in the list of quality problems to be analysed. Since transverse depression is the most important quality problem, specific issues related to it are analysed. The following message is presented before the start of the consultation. You have selected TRANSVERSE DEFECTS as the most important important quality problem. Pertinent observations ( i.e. if the transverse cracks andl or depressions occur more on some specific occasions ) provide important clues related to its formation. Some of these are as follows: If the transverse cracks andl or depressions are predominant : * on some specific strand(s) * on some specific region(s) of the billet * in some specific grades * at some specific time in the heats  System Do you see transverse defects on all the strands?  139  User^No System Are the transverse defects more on the inboard strands as compared to the outboard  strands? User^Yes System Are the transverse defects more frequent in certain specific regions on the billet  surface? User^No System You indicated that the transverse defect is present on the inboard strands. You also  indicated that it does not appear on some specific portions of the billet surface. We feel that this problem is affected strongly by the following factors: 1. Metal level variation 2. Mould-tube alignment 3. Mould lubrication These factors will be analysed now. System Would you describe the surface of the metal stream flowing into the mould as being  "ropey" or "rough"? User^Yes System WARNING : Metal level variation is significant !!  You have observed that the condition of the metal stream coming out of the tundish is rough. It is certain that this could be leading to excessive gas entrainment in the liquid steel. In the mould, these gases escape from the liquid pool surface as bubbles  140  creating turbulence and metal level variation. The situation is worse in the absence of an adequate tundish design for flow control. We advise you to localize turbulence from the ladle stream by the use of dams and weirs in the tundish. We strongly re commend water modelling of the tundish to establish optimum tundish design. System Let's examine issues related to mould-tube misalignment. Misalignment of the mould  tube can be deduced from the following: 1. Uniformity of scale deposits on the cold face of the mould 2. Cooling water velocity measurements System Is the scale deposits seen on the cold face of the mould? User^Yes System Would you describe the scale deposits as extensive? User^Yes System Is the scale uniformly deposited on all faces of the mould? User^No System Non-uniform scale deposits indicate that the water gap around the mould periphery  is not constant and this leads to non-uniform cooling water velocity. The water gap may differ by 1 mm or more around the mould periphery due to poor tolerances on the outside dimensions of the mould wall and the inside dimensions of the water jacket or  141  due to poor tolerances on the mould assembly, especially support plates, which result in non-concentricity of the mould tube in the baffle. We are certain that the tolerances of the mould assembly are inadequate. The system continues the consultation and examines the following aspects of mould lubrication system: * Lubrication oil flow and distribution in the mould * Lubrication oil physical properties * Cleanliness of the oil slot system * Cleanliness of the meniscus region in the mould System Are the transverse defects more frequent in the initial stage of the heats? User^Yes System Does the frequency of transverse defects reduce after the dummy bar has crossed the  pinch-rolls? User^Yes System You have indicated that this problem appeared more in the initial stage of the heat as  compared to the other parts of the casting cycle. It could be related to poor start-up practice. We feel that the higher frequency of defects in the initial stage of the heat, might be due to the dummy bar practice. Since the frequency of transverse defects reduces considerably after the dummy bar has crossed the pinch-rolls, it is certain that this problem is related to the dummy bar practice. Our experience has shown that the transverse defects may also be related to poor casting speed control at the beginning of the heats.  142  The system conducts a detailed analysis of the mould system to ascertain the causes of the quality problems observed. Other issues related to the mould (distortion, oscillation and taper) are analysed in detail. Spray maintenance issues as well as steel quality parameters are also examined. In addition, for off-corner cracks, the following analysis is also done: System Please let me know whether the outer tip of any of the off-corner cracks is close to 7.0  mm from the billet surface so that we can establish whether there exists a problem near the mould exit. User^Yes System Off-corner crack depth, when compared with the model predictions, suggests that  bulging of the solid shell occurred in the last stages of mould cooling or just at the start of the sprays. It is very likely that one or more of the following factors may also be contributing to off-corner cracks: * Wear in the lower part of the mould * Distortion in the lower part of the mould * Mould cooling jacket arrangement * Insufficient spray cooling at the mould exit The above issues are analysed. The analysis ends with the following final conclusions:  * Quality Problem is Mould-Related * Quality Problem is Spray-Related * Quality Problem is Related to Steel * Abnormality in Certain Operating Parameters is Present * Additional Information is Needed  143  Justification can be obtained for each of the above conclusions. The system ends the consultation by presenting the summarized report.  CASE 3 Craze Cracks Only craze cracks were seen. The system considers this problem to be the most important problem. The following is the consultation: System Do you see craze cracks on all the strands? User^No System You indicated that craze cracking problem occurs on certain specific strands. Since  this is a steel related issue, it is very unlikely that your observation is correct. We strongly feel that the problem is present on the other strands as well. Please examine the billets from other strands once again. We will now examine steel quality parameters. System The copper levels in the steel is 030 percent. This is high.  The ratio of Ni/Cu is 033. This is low. You may reduce the severity of this problem by reducing the level of copper in the steel to below 020 percent or by increasing the Ni/Cu ratio in steel to more than 1.0. The consultation is over. The following final conclusion is presented by the system:  * Quality Problem is Related to Steel Justification can be obtained for this conclusion in the usual manner. The system then generates the final report.  144  The system is capable of analysing other problems occuring in any combinations. The above examples exhibit that the analysis procedure can be very straightforward and short or extremely detailed and long. The nature of the analysis solely depends on the kinds of problems being analysed.  145  CHAPTER 9 SUMMARY AND CONCLUSIONS  An expert system has been developed to guide operators in analysing quality-related problems and in providing them with a ready source of knowledge related to the operation of the casting machine. Outcomes of Knowledge Engineering Knowledge engineering this expert system was beneficial in many ways. Fundamental and heuristic knowledge required for problem solving were integrated during system development. This exercise lead to a formal representation of the Experts' knowledge which had a major impact on the problem domain. A structure evolved that was very similar to that of the Experts and was easy to understand and explain to others. It provided the users with a consistent interpretation of the domain knowledge. A new classification scheme for quality problems in billet casting was developed. The problems are divided into groups; those that have complex origins and those where very specific issues need to be examined. With this strategy, all combinations have been handled. A new knowledge link was established between mould cooling water quality and mould lubrication. When scale deposits are present on the cold face of the mould, the mould operates hotter than normal and therefore, the lubricating oil vaporizes or burns before reaching the metal level. In this situation, mould lubrication is likely to be poor.  146  A novel inferencing strategy was developed in the analysis of midway cracks to integrate the influence of important parameters such as surface reheat temperature, steel superheat and composition. The effect of mould disorders on belief in a spray-related strain problem was also included in this scheme. Future Work The knowledge base will be modified to a WINDOWS/3 environment. User-feedback related to expansion of the problem domain, enhancement of user-interface and refining of the knowledge base will be incorporated into the system. Future efforts will also examine the application of this knowledge base to a real-time on-line monitoring and process-control system. By providing this assistance to the quality-control process, significant improvements in productivity and product quality are likely to occur.  147  REFERENCES  1^COMDALE/X version 3.0- Users' Manual, Comdale Technologies Inc., Toronto, Canada. 2^F.Hayes-Roth et al. : Building Expert Systems. Reading, Massachusetts, USA: Addison-Wesley, 1983. 3^E.A.Feigenbaum: Proceedings of the International Joint Conference on Artificial Intelligence, vol.5, 1977, pp.1014-1029. 4^J.Gaschnig: Proceedings of the National Conference of the American Association for Artificial Intelligence, von, 1980, pp.295-297. 5^E.H.Shortliffe: Computer-based Medical Consultation - MYCIN, New York, American Elsevier, 1976. 6^H.E.Pople: Proceedings of International Joint Conference on Artificial Intelligence, vol.5, 1977, pp.1030-1037. 7^R.K.Lindsay et al.: The Dendral Project, New York, McGraw-Hill, 1980. 8^S.M.Weiss and C.A.Kulikowski: Proceedings of International Joint Conference on Artificial Intelligence, vol.6, 1979, pp.942-947. 9^J.M.McDetmott: Proceedings of the National Conference of the American Association for Artificial Intelligence, void , 1980, pp.269-271. 10^J.A.Meech: Minerals Engineering, vol.3, 1990, pp.129-136.  148  11^E.Turban: Decision Support and Expert Systems, New York, NY, Macmillian, 1988. 12^J.K.Brimacombe and I.V.Samarasekera: Proceedings of Indo-US Workshop on Principles of Solidification and Materials Processing, Hyderabad, India, January 1988, Oxford and IBH Publishing Company, New Delhi, India, pp.179-222. 13^A.W.Cramb and M.Byrne: Iron and Steel Maker, May 1986, pp.27-36. 14^A.Mclean: Proceedings of 1988 AIME Steelmaking Conference, Toronto, Canada, pp.3-23. 15^I.V.Samarasekera and J.K.Brimacombe: International Metals Review., 1978, vol.23, No 6, pp.286-300. 16^S. Watanaba et al.: Tetsu-to-Hagne, 1972, Vol. 58, pp.S393-S394. 17^A.Grill et al.: Metallurgical Transactions, 1976, Vol.7B, No.2, pp.177-189. 18^J.K.Brimacombe et al.: ISS Transactions, 1984, Vol.5, pp.71-78. 19^I.V.Samarasekera et al.: ISS Transactions, 1984, Vol.5, pp.79-94. 20^I.V.Samarasekera and J.K.Brimacombe: W.O.Philbrook Memorial Symposium Proceedings, 1988, AIME Steelmaking Conference, Toronto, Canada. 21^I.V.Samarasekera and J.K.Brimacombe: Canadian Metallurgical Quarterly, 1979, Vol. 18, pp.251-266.  149  22^I.V.Samarasekera and J.K.Brimacombe: Ironmmaldng and Steelmaking, 1982, Vol. 9, pp.1-15. 23^E.Mizikar: Iron and Steel Engineer, 1970, Vol.47, pp.53-60. 24^F.Kreith: Principles of Heat Transfer, 3rd Ed., Intext Educational Publishers, 1976. 25^C.J.Hoogendorn and R.den Hond: 5th Int. Heat Transfer Conf., Paper B 3.12, Tokyo, 1974. 26^R.Jeschar et al.: Steelmaking Proceedings, ISS, Warrendale, PA, 1986, Vol.69, pp.511-521. 27^M.Mitusutsuka: Tetsu-to-Hagane, 1968, Vol. 54, pp.1457-1471. 28^H.Junk: Neue Heutte, 1972, Vol.17, pp.13-18 [HB Trans'. No.8740 ]. 29^H. Muller and R.Jeschar: Arch. Eisenhuttenwes, 1973, Vol. 44, pp.598-594. 30^R.Alberny et al.: Circulaire d'Informatio-techniques, 1973, Vol.3, No.315, pp.763-776. 31^R.Alberny: Committee of European Communities, Luxembourg, 1977, Voll, IPC Science and Technology Press, pp.278-335. 32^Y.Sugitani et al.: Tetsu-to-Hagane, 1973, Vol.59, pp.S388-S389. 33^M.Ishiguro et al.: Tetsu-to-Hagane, 1974, Vol.60, pp.S464-S465 [HB Trans'. No. 8753 ].  150  34^T.Kawakazu et al.: Tetsu-to-Hagane, 1974, Vol.60, pp.S 103-S 104 [[HB Transl. No. 9319 J. 35^P.Nilles et al.: Proceedings of Open Hearth Conference, Chicago, ISS-AIME, 1978, Vol.61, pp.399-410. 36^E.Bolle and J.C.Moureau: International Conference on Heat and Mass Transfer Metallurgical Processes, Dubrovnikm, Yugoslavia, 1979. 37^E.Bolle and J.C.Moureau: Proceedings of Two Phase Flows and Heat Transfer, NATO Advanced Study Institute, 1976, Vol.III, pp.1327-1346. 38^K. Sasaki et al.: Tetsu-to-Hagane, 1979, Vol.65, pp.90-96. 39^S.G.Hibbins and J.K.Brimacombe: ISS Transactions, 1983, Vol.3, pp.37-51. 40^J.K.Brimacombe: Canadian Metallurgical Quarterly., 1976, Vol.15, pp.163-175. 41^S.Kumar: Unpublished Report, UBC, 1989. 42^B.G.Thomas et al.: ISS Transactions., 1986, Vol.7, pp.7-18. 43^J.K.Brimacombe and K.Sorimachi: Metallurgical Transactions B, 1977, Vol.8B, pp.489-505. 44^H.Vom Ende and G.Vogt: Journal of Iron and Steel Institute, 1972, Vol.210, pp.889-894. 45^C.J.Adams: Open Hearth Proceedings, 1971, Vol.54, pp.290-302. 46^G. Van Drunen et al.: Ironmaldng and Steelmaking, 1975, Vol.2, pp.125-133.  151  47^R.P.Sopher: Welding Journal, 1958, Vol.37, pp.481S-492S. 48^L.I.Morozensldi et al.: Stal, 1965, No.4, pp.272-276. 49^T.Kinoshita and S.Kuroki: Imono, 1972, Vol.44, pp.1080-1089. 50^A.Gueussier and R.Castro: Revue Met., 1960, Vol.57, pp.117-134. 51^E.T.Stephenson: Journal of Metals, 1974, Vol.20, No.3, pp.48-51. 52^R.Cooper and T.H.Burns: Metals Eng Quart., August 1974, pp.41-44. 53^L.Ericson: Scandinavian Journal of Metallurgy, Vol.6, No.3, 1977, pp.116-124. 54^K.Kinoshita et al.: Tetsu-to-Hagane, 1976, Vol.62, No.S505, pp.43/1-26 55^F.Weinberg: Metallurgical Transactions B, 1979, Vol.10B, pp.219-227. 56^J.Miyazaki et al.: 100 th ISU Meeting, October 1980, No.S806. 57^U.K.Bhattacharya et al.: Transactions AFS, 1952, Vol.60, pp.675-686. 58^C.F.Christopher: Transactions AFS, 1956, Vol.64, pp.293-310. 59^H.F.Bishop et al.: Transactions AFS, 1957, Vol.65, pp.247-258. 60^H.C.Suzuki et al.: 100th ISIJ Meeting, October 1980, No.S805 61^J.K.Brimacombe et al.: Steelmaking Proceedings, ISS, Warrendale, PA, 1986, Vol.69, pp.409-423.  152  62^J.K.Brimacombe et al.: Steelmaking Proceedings, NOH-BOF Conference, ISS-AIMS, Warrendale, PA, 1980, Vol.63, pp.235-252. 63^W.Dahl and H.Hengstenberg: Z. Meta11k, 1969, Vol.60, pp.340-350. 64^R. Bommaraju et al.: ISS Transactions, 1984, Vol.5, pp.95-105. 65^J.E.Lait and J.K.Brimacombe: ISS Transactions, 1982, Vol.1, pp.1-13. 66^S.N.Singh and K.E.Blazek: Open Hearth Proceedings, AIME, 1976, Vol.59, pp•264-283. 67^J.K.Brimacombe et al.: ISS Transactions, Vol.1, 1982, pp.29-40. 68^J.K.Brimacombe et al.: Proceedings of 69th Steelmaking Conference, 5th Inemational Iron & Steel Congress, April, 1986, Washington D.C., USA. 69^R.J.Dippenaar et al.: Electric Furnace Conference Proceedings, Atlanta, Vol.43, 1985. 70^The Centre for Metallurgical Process Engineering, UBC, Vancouver, B.C., Canada. Unpublished work. 71^I.V.Samarasekera et al.: Proceedings of International Symposium on Continuous Casting of Billets, Vancouver, Canada, CIM, 1985, pp.33-58. 72^R.Hapke: Private Communication. 73^J.K.Brimacombe et al.: Canadian Metallurgical Quarterly., 1980, Vol.19, pp•215-227.  153  74^A.Grill et al.: Ironmaking and Steelmaking, 1976, Vol.3, pp.38-47. 75^H.Mori: Tetsu-to-Hagane, 1972, Vol.58, pp.1511-1534 [HB Transl. No.9000-  76^H.Mori: Tetsu-to-Hagane, 1974, Vol.60, pp.784-806 [HB Transl. No. 9355 I&II]. 77^M.Ya.Brovman: Stal Eng., 1967, Vol.1, p.26. 78^A.Grill et al.: Ironmaking and Steelmaking, 1976, No.1, pp.38-47. 79^I.V.Samarasekera and J.K.Brimacombe: Proceedings of 3rd PTD Conference on Applications of Mathematical and Physical Models in Iron and Steel Industries, ISS -AIME, 1982, pp.76-91. 80^I.V.Samarasekera and J.K.Brimacombe: Metallurgical Transactions B, Vol.13B, 1982, pp.105-116. 81^I.V.Samarasekera and J.K.Brimacombe: Proceedings of 2nd PTD Conference on Continuous Casting of Steel, ISS-AIME, 1981, pp.2-21. 82^G.L.Fisher: Journal of Iron and Steel Institute, July 1969, pp.1010-1016. 83^I.V.S amarasekera and J.K.Brimacombe: Report to Stelco Edmonton Steel Works, unpublished work, 1987. 84^T.Higaki et al.: Continuous Casting of Steel, 1977, London, The Metals Society, pp.177-181.  154  85^C.W.Dean: Open Hearth Proceedings, AIME, 1974, Vol.57, pp.253-257. 86^H.Ichikawa et al.: Continuous Casting of Steel, 1977, London, The Metals Society, pp.304-308. 87^J.A.Meech: Processing Complex Ores, R.Rao and G.Dobby (Eds.), London, Pergamon Press, 1989, pp.575-585.  155  APPENDIX A - OPERATING PARAMETERS REQUIRED IN THE ANALYSIS  *  Steel Composition  *  Pouring Temperature (three times in the heat)  *  Section Size  *  Casting Speed (minimum, maximum and average values)  *  Mould Design (length, wall thickness, corner radius, type of constraint, taper, width of cooling channel)  *  Mould Copper (type, chemical composition, thermal conductivity, half-softening point, yield strength)  *  Mould Operation (metal level, cooling water flow rate, cooling water quality- hardness)  *  Mould Oscillation (stroke length and frequency- minimum, maximum and average values)  *  Mould Lubrication (oil flow rate, physical properties of oil- flash point, boiling point and viscosity)  *  Spray Parameters (number of spray zones, zone length, water flow rate in each zone and water temperature)  156  APPENDIX B - EXPERT SYSTEM REQUIREMENTS  1. HARDWARE AND SOFTWARE REQUIREMENTS  - IBM PC 386 with MS-DOS ver 3.3 - 640 Kb RAM - Free hard disk space - not less than 5.0 Mb 2. USERS REQUIREMENTS  - Familiarity with computers ( IBM-PC ) - Little knowledge about metallurgical terminologies employed in casting - Some knowledge about the continuous casting process 3. RECOMMENDED USERS  - At least three different levels of operating personnel - An inexperienced operator - to assess training capabilities  APPENDIX C  PRELIMINARY REPORT ON USERS' FEEDBACK  158  1. COMPANY A 1. Hardware / Software :1st computer^(a)^286 COMPAQ machine  Operating system : MS-DOS version 3.30 (b) High density disk drive (51/4") was not available (c) High density disk drive (31/2") was not available (d) Expert system program could not be loaded (e) 2nd computer^(a)^High density disk drive (51/4") was available High density disk drive (31/2") was not available (b) Operating system : MS-DOS version 3.20 (c) Expert system program was loaded and started (d) Math co-processor was not installed in the computer and therefore (e) it was not possible to run FORTRAN programs because of floating point error arising due to the absence of a math co-processor programs were compiled for machines having math co-processor. (0^FORTRAN programs were re-compiled for a machine with no math co-processor. The system was restarted. The mathematical model took about 20 -25 minutes to run compared to 5 to 50 seconds for conventional machines with a co-processor. In addition, version 3.20 of DOS did not support the "CALL" (g) function used in the batch file. (h) Manuals for the computer were not available and therefore it was not possible to ascertain the type of the machine being used. Based on the above experience, the following requirements (hardware and software) for running the EXPERT SYSTEM using COMDALE/X version 3.0, have been identified as mandatory: IBM-PC - 386 or 286 with math co-processor (i) (ii) MS-DOS version 3.30 or above (iii) 640 kb of RAM Hard-disk space of 5.0 Mb or more (iv) 2. Error(s) in the code :  (a)^Unit of metal level in the mould was being incorrectly reported. This was pointed out by the operating personnel and was later corrected. 3. Quality problem analysis :  (a) (b)  The expert system was demonstrated to the operating personnel using the TOSHIBA 286 machine. A "TEST" file was created and used as an example for the demonstration. The program was run with quality problems : transverse depression, rhomboidity and centreline cracks.  159  (c) (d) (e)  (f) (g) (h)  Transverse section of the billet was not available and therefore it was impossible to identify internal cracks in the billet. It is impossible to observe dark and bright patches on the billet surface as it comes out of the mould because of design restrictions. The expert system suggested that rhomboidity problem could be due to blocked nozzles as a result of poor spray water quality. It was immediately pointed out by the operators that they had problems of blocked spray nozzles and they minimized rhomboidity problem by incorporating chemical treatment for their spray cooling water. The expert system highlighted the mechanism of formation of transverse depressions and centreline cracks. This information was useful to the operating personnel in learning about the source of the problem. The operating personnel felt that centreline segregation problems must be included in the list. In general, measurements of mould tube distortion, oscillation mark depths, cooling water velocity, oil distribution and the chemical analysis of the scale deposits were not available. In some cases, lack of the above information could be a potential threat to successful analysis.  160  •  2. COMPANY B 1. Hardware / Software :-  (a) (b)  A 386-computer with a math co-processor and MS-DOS version 3.30 was available. There was no problem during the installation of the expert system program. Option for exiting in the middle of a consultation session was not available in the expert system. Thus, the user could exit only at the end of the session by selecting the option "STOP". The operator requested this option. It was incorporated through the batch file but there was a memory problem (due to insufficient memory) while "ACTIVATING" the "COPY" command in COMDALE/X. This problem was later solved by modifying the "CONFIG.SYS" file (to increase available memory).  2. Error(s) in the code :  (a)^Missing Fuzzy sets:^position_of^chromium_layer_discolouration^close to_the_mould_top * distortion in_the_lower_part_of the_mould large 3. Quality_problem analysis :  The term "distance of the first spray nozzle from the mould exit" was not very clear to the operating personnel. Although this distance was 7 cm, the spray water reached points very close to the mould exit and therefore the actual gap between the mould and the spray was negligible. This term needs to be modified in the data input program. This point will be taken care of in the mathematical model by including a spray-configuration subroutine. (b) The mathematical model does not take into account the spray configuration - gap between the spray zones, nozzle types and the stand-off distance. It is necessary to incorporate this in the model to be more accurate. The operators were doubtful about the calculation of heat transfer coefficient for the spray. (c)^Units employed in the expert system: • preferred units for shell thickness- mm • conversion for SUS (unit for oil viscosity) was not available • in general, users felt that they must be comfortable with the units employed in the expert system- English, SI or even mixed. It is very inconvenient to switch from the units commonly used to SI or to any others. (d)^Nb needs to be included in the list of elements in the steel composition. Also, the effect of Nb on the high temperature ductility of steel is to be included in the knowledge base, particularly with an emphasis on the generation of transverse corner cracks. It is assumed in the expert system that mechanisms involved in the generation of transverse (e) cracks (midface and corner) are the same and therefore are treated in the same way. However, the operators felt that transverse midface and corner cracks arise due to different reasons. Expert's view on this issue is required. Are the operators correct ? In the expert system, the recommended value for superheat is below 25°C. The operators (f) felt that this is "too low" and therefore, not a practical recommendation. According to them, it is very difficult to cast steel with a superheat of 25-30°C. In company B, superheat levels were generally in the range 50-80°C. Therefore, a need exists to re-examine the definition of "high" superheat. The Experts' resolved this issue by saying that their definition of "high" superheat is correct and a redefinition is not necessary. (a)  161  (g)  (h)  (i)  (j)  Here again, it is difficult to observe dark and bright patches on the billet surface as it comes out of the mould. It is possible that in some situations the subsequent analysis strategy may critically depend on this information. Therefore, ways must be devised to get around this problem. Some of the measurements are available in the database and very often these are used as answers to questions in the analysis. In some cases, it is possible that these may not correlate with the quality problems in question. Thus, it is necessary to ensure that the data corresponds to the actual situation. The Experts pointed out that they also face this problem. In the expert system, metal level variation in the mould has been related only to the condition of the metal stream. It was pointed out by the operating personnel that another important factor that could be added to this is metal level control system - bad signal. This point will be emphasized through a text message. Water velocity measurement data was not available. Operators felt that most companies will not have this facility for regular process control. The Experts felt that this option must stay in the knowledge base.  162  3. COMPANY C 1. Hardware / Software  (a) (b)  :-  386-computer with math co-processor and MS-DOS version 3.30. There was no problem in the installation of the expert system program. Operator printed the report file. There were some errors in the report generator. These were corrected immediately.  1 Error(s) in the code :  (a)  (b)  IGNORE RANGE or RULE statements were missing. The system became unstable at the cross-over point (fuzzy set for large and small distortion cross over at 0.2 mm distortion). This was corrected by incorporating IGNORE RANGE/RULE statements in the concerned rules. A misprint was discovered in one keyword triplet - "boiling problem present_in_the_mould" was printed as "boiling problem persent_in_the_mould". This was noticed during the justification session.  3. Quality problem analysis :  (a)  (b)  (c)  (d)  The model predicted a high reheat in the radiation zone. Midway cracks were seen in the radiation zone. A direct correlation was observed. Mn:S ratio was also low and the superheat was high. This was pointed out by the expert system. A sulphur print having midway cracks were observed. The above prediction was verified - midway crack in the radiation zone. Frequency of midway cracks were usually higher at the end of heats as compared to the rest. According to the operators, this happened because the withdrawal rate increased substantially at the end of heats - purpose to finish off fast. The superheat at the end of the heat was the same as the start. No correlation could be established between the frequency of the problem and the superheat level. Does this mean that the problem is mould related - reheating associated with dark patches ? Superheat at the start and the end of heats were lower than that in the middle. This was a general observation in all the plants. The effect of thermal stratification in the ladle was not very obvious. In addition, the superheat at the start and the end were valid for smaller time as compared to the superheat in the middle of the heats which covered a larger period. The problem of transverse depression (and cracks) was analysed. The permanent mould distortion was 0.2 mm (50 percent sure that this is small based on the fuzzy set for "small"). Dark patches were not seen. According to the expert system, distortion was not a problem and therefore analysis of parameters related to mould distortion was not necessary. Oscillation parameters- mould lead (5-7 mm) and negative strip time (0.16-0.17 s) were satisfactory. Oil flow was satisfactory. No jerking was observed. Oil distribution was measured and was uniform. Meniscus area and the oil distribution slot were cleaned regularly. Double taper was used. Oscillation marks were not deep (visual). Everything seemed to be satisfactory. The only other reason could be improper taper. The expert system reached its limit.  163  (e)  (f)  The definition of severe rhomboidity - a difference of 4 mm or more between the length of the two diagonals is considered to be a problem. Is this range valid for larger section sizes such as 190 mm 9 Is there a need of a factor say (DIFFERNCE/SECTIONSIZE) that takes into account section size as well ? The Experts said that the present definition of "large" rhomboidity is satisfctory. The heat-flux profile used for the mould was general and not specific to the steel company "C". There were some doubts in the minds of the operating personnel about the heat flux profile being used. It was later pointed out to them that this was an assumption in the model. It was also clarified that mathematical model was a heat transfer solidification model using finite difference formulations and that its prediction was much better than a kt" model for shell thickness profile. This was appreciated by them. It is necessary to inform the users about the formulation scheme. It was again pointed out by the users that the mathematical model does not take into account the spray configuration - gap between the spray zones, nozzle types and the stand-off distance. It is therefore necessary to incorporate this in the model to be more accurate. Knowledge about the width of oil slot has not been incorporated in the expert system. The users wanted to about its significance. There is a need to refine the definition of "high" and "okay" superheat level based on practical limitations. There is a need to incorporate the effects of Nb and B on the severity of cracking - transverse cracks. Craze cracks are not a problem. No complaints from the company "C". Is it because they are not aware of the problem and also how to locate them in the billet - etching technique used by R. Hapke at UBC ? The knowledge related to taper has to be incorporated in the system and also the strategy in a situation where large mould distortion has been measured or inferred. If mould distortion is a problem, how would one handle the taper issue ?  164  4. COMPANY D 1. Hardware / Software :  (a)^286-computer with math co-processor and MS-DOS version 3.30 was available. There was no problem in the installation of the expert system program. 2. Errors) in the code :-  (a)^The system was demonstrated to the operating staff. No errors were noticed. 3. Quality problem analysis :  (a) (b) (c) (d)  Operating parameters were not easily available. Completed data sheet was not available. The operators did not have sufficient time to spare. The spray parameters (zone length and flow) were not clearly known to the operators. It may be necessary to provide the users with some help on converting spray pressures to flow based on some data base on spray nozzles -type and configuration. A consultation session involving transverse depressions and cracks was demonstrated using the limited data available from company "D". Problem related to dummy bar practice was identified by the expert system - analysis of specificity issues. The users agreed that there was a lot of information in the expert system. They were interested in the information.  165  $, COMPANY E 1. Hardware / Software :  (a)^286-computer with math co-processor and MS-DOS version 3.30 was available. There was no problem in the installation of the expert system program. 2. Error(s) in the code  :-  (a)^The system was demonstrated to the operating staff. No errors were noticed. 3. Quality problem analysis : Eventhough there are three strands, only two may be used at a given time. There is a need (a) to include a question on "number of strands in operation". (b) There was some problems in the model predicted shell thickness for high carbon grades (carbon > 1.0 percent) - model predicted a small shell. It was later discovered that the data base did not have the heat-flux for this carbon range. The heat-flux being used may not be valid. Also, large mushy-zone associated with these grades was influencing the prediction. Data-base of heat-fluxes for different grades will be developed for each plant. (c) Distortion measurements involves checking the distance between two opposite faces. This obviously does not take into account twist. The operators did not observe any correlation between distortion measurements and mould-related quality problems. (d) Problems such as laps and bleeds needs to be included.  166  GENERAL COMMENTS  1^The system will be useful during "brain-storming" sessions for trouble-shooting quality problems. Because of the "general" nature of the knowledge base, the system is unlikely to be successful in day-to-day problem solving. 2^English language will be a problem in non-english speaking areas. 3^Computer literacy is very low in the steel companies. This may be a potential problem. One way to reduce this threat is to increase the user-friendliness of the system. 4^The system should attract the users - like computer games. Version 5.0 of Comdale/X will handle part of this problem. 5^Graphical images must be introduced - pictures of shell thickness and temperature profiles should appear on the screen dynamically while the mathematical model is running. 6^According to the officials, users particularly on the shop floor will use the system during coffee breaks and some other free time. This may be a potential problem from the viewpoint of teaching/training. The company must incorporate training schemes or encourage the users on the shop floor. 7^The text message provided at various stages during the consultation must take into account the expertise/level of the users. This is important because the system appears to be a "TEACHER" all the time and a lot of information is supplied at inappropriate places/time. The system treats all the users in the same way. This aspect of different levels of users will be dealt with in the version 5.0 of Comdale/X. 8^The "JUSTIFICATION" feature of the expert system, though useful, did not interest the users very much. The main concern of users was to solve the problem quickly. However, the elimination of this feature is not a consideration. 9^It is not easy to establish the limit for "high pinch-roll pressure". Usually, it is limited by the force required to pull the billet out of the mould. Operators do not control this parameter. Hence, it is very difficult to judge this parameter. One of the operators suggested that by examining the deformation due to the pinch rolls - comparing the billet size at the mould exit with the final cold billet dimension, it is possible to conclude about the pinch roll pressure. In case of pinch-roll cracks, the expert system recommends a decrease in pinch-roll Pressure. What happens if the pinch pressure is at the lowest set-point ? 10^Some operators suggested that rhomboidity can also be minimized by increasing the spray water flow. This point has not been included in the recommendation. 11^One user felt that the concepts highlighted in the expert system made sense to him because he had attended the Casting Course in Vancouver. It may be necessary to give a short course in the concerned company not only on the expert system but also on continuous casting during the installation. An overview on continuous casting will get them started. The concepts in the expert system will be easy to follow after this "start-up".  167  APPENDIX D : RESULTS OF FORMAL EVALUATION  The following is the key to the rating:Strongly Disagree  1  Mildly Disagree  2  Neutral  3  Mildly Agree  4  Strongly Agree  5  The Companies are indicated by the letters "A" to "E". Completed questionaire were received only from three companies namely: "B", "C" and "E". 1. OPERATING PARAMETERS - INPUT MODULE RATING  STATEMENTS  COMMENTS  1  Easy and convenient to operate  1*  C 5  2  Terminology used is easy to understand. Need to find out why certain operating parameters are requested.  5*  5  4  3  1  3  Requested operating parameter data are readily available. Need for more information on measurement techniques for operating parameters. Need a glossary of basic terms in continuous casting.  5  5  4  3  1  1  4*  1  1  B  3 4 5 6  168  E 4  * Too  slow and cumbersome to operate. Also, units are different from those followed in the plant. * Difficult for operators on the shop-floor. Not necessary to provide explanation on "why" certain parameters are requested. Input data for the system is not a problem Not necessary to provide explanation on the measurement techniques. * Hyper-Text feature of COMDALE/X version 5.0 will be beneficial.  2. QUALITY PROBLEMS - INPUT MODULE RATING  STATEMENTS  COMMENTS  B 5*  C 5  E 4  1  Terms used are easy to understand.  2  Need for information on where and how to look for the quality problems in the cast product is necessary. Background information on quality problems in billet casting is necessary.  5  1  2  5  1  2  4  Other quality problems need to be included in the list of options.  3  1  4  5  It is difficult to select "the most important problem". The "most important problem" needs to be defined better.  3  1  1  3  1  1  3  6  * Not sure about operators on the shop-floor. Some operators may require additional information Some operators may require additional information Need to include laps, bleeds, centreline segregation. No problem with the most important problem No problem with the most important problem  3. MATHEMATICAL MODEL RATING  STATEMENTS  1 2 3 4 5  Additional information on the heat-transfer mathematical model would be useful. Access to information on the formulation of equations in the mathematical model would be useful. It would be useful if the underlying assumptions in the mathematical model were available for the users to examine. The time delay associated with running the mathematical model is too long. The output format from the mathematical model is satisfactory.  169  COMMENTS  B 5  C 5  E 2  5  5  2  Definitely required  5*  5  2  *Need to incorporate spray design into the model  3  1  4  3  5  -  Not sure about operators on the shop-floor.  4. QUESTIONS ASKED BY THE SYSTEM  1 2a 2b 2c  3a 3b 4 5  6  7  8  COMMENTS  RATING  STATEMENTS  Questions asked are clear and easy to answer. The "EXPLANATIONS" provides useful and additional information about facts The"EXPLANATIONS" are clear and easy to understand. It would be better to have an explanation such that excess information for experienced personnel is avoided and at the same time, more detail for novices are provided. The "WHY" feature was useful in explaining the purpose behind questions asked. "WHY" messages are clear and easy to understand. The "EXPLANATION" and "WHY" features provided in the system would be useful in the training of a novice. The technical level of information provided through "EXPLANATIONS" and "WHY" features is : (a) very high (b) too elementary The guidelines provided by the expert system for collection of data and for maintenance of the caster are easy to follow. Discrepancies were encountered in the information provided by the "EXPLANATION" or the "WHY" feature. There exists some disagreement with information provided through the "EXPLANATION" and/or the "WHY" feature of the expert system.  170  B 5  C 5  E 5  4  1*  5  4  5  4  3  5  4  4  4  4  4  5  5  5  5  5  3 3 4  1 3 4  4* 2 5  3  1  2  3  1  1  * Explanations  needed for the "data-input" module  Definitely need to address different user-levels  The present level is adequate. *High for operators on the shop-floor  5. RECOMMENDATIONS PROVIDED BY THE SYSTEM STATEMENTS  1 2 3 4 5 6  RATING  System's recommendations are clear and easy to understand. System's recommendations are easy to implement (they are feasible). Discrepancies were encountered in the recommendations provided by the expert system. There exists some disagreement with some of the recommendation(s) made by the system. It would be useful to know about other possible problem areas during the consultation. System's recommendations are accurate and complete.  COMMENTS  4  5  E 5  3  5  4  -  1  1  -  1  1  5  5  3  3  5  4  6. FINAL CONCLUSIONS AND JUSTIFICATION STATEMENTS  RATING  1  The conclusions reached by the system are complete.  2  The conclusions are clear and easy to understand. The justification scheme provided at the end of the consultation is useful. The justification scheme is clear and easy to understand. The justification of the conclusions at the end of the consultation will be extremely useful in the training of new operators. Discrepancies were encountered during the justification process. There exists some disagreement with the logic used by the system in analysing data.  3 4 5 6 7  171  COMMENTS  B 3  C 4  E 2*  4  5  5  5  3  4  4  5  4  5  5  5  -  1  4*  * same  as 1  -  1  4*  * same  as 1  * Explanation  for transverse depressions in AISI- 1018 grade was found to be inadequate  7. CONSULTATION SESSION COMMENTS  RATING  STATEMENTS  1  The consultation session is too long.  4  C 1  2  The system is adequate at diagnosing quality problems. The system is adequate for training operating personnel. System instabilities were encountered during the consultation.  4  4  4*  5*  5  4  51  1  52  -  2*  4  5  5  4  4  5  2*  4  5  5  3  5  5  -  1  2  5  5  4  B  3 4  5  The ability of the system to handle conflicting information or input is satisfactory.  6  The system informs the user about what it is doing at different stages of the analysis. 7 Changes in the system's thought process, during the analysis, are smooth and comfortable. 8 The final report generated by the system is satisfactory and complete 9 The information contained in the report is adequately summarized. 10 There is a need to include some more information in the report. 11 A computer print-out of the report for record purposes would be useful.  172  E 4  Session is too long for some cases *Updating of knowledge is needed * May be complicated for operators on shop-floor 'Problems related to insufficient memory 2 Problem related to corrupted files * Rhomboidity is always a problem. For a large section size (190 mm), 4-5 mm is large as per the present definition  * Sometimes difficult to understand why certain modules are loaded.  Report can be printed out feature already exists  8. SYSTEM OPERATION RATING  STATEMENTS  1  The system is easy and convenient to learn and operate.  2  The user interface provided by COMDALE/X is sufficiently "friendly".  3  System problems were encountered during the consultation session(s).  21,2  B  C 5  4  5 1  173  COMMENTS  E 22 'Need more places to exit from the program; 2Not sure about operators; easy for metallurgists 1 * *Not very easy for operators with no computer background 5 2 'Problems related to insufficient memory; Need explanations for error messages 2Problems related to boundary conditions used in the model  APPENDIX E : OPERATING PARAMETERS FOR CASE STUDIES  1. Steel Composition  CASE "A"  CASE "B"  CASE "C"  Carbon  0.40  0.17  0.51  Manganese  0.80  0.80  0.88  Sulphur  0.03  0.030  0.035  Phosphorus  0.02  0.010  0.010  Copper  0.15  0.14  0.30  Nickel  0.12  0.07  0.10  Silicon  0.30  0.22  0.25  Molybdenum  0.04  0.02  0.02  Chromium  0.12  0.12  0.12  Aluminium  0.001  0.001  0.001  Titanium  0.001  0.001  0.001  Vanadium  0.001  0.001  0.001  174  Unit  CASE "A"  CASE "B"  CASE "C"  mm  150 x 150  150 x 150  150 x 150  °C  1560  1580  1548  °C  1550  1570  1540  °C  1540  1560  1532  mm/s  25.4  25.4  25.4  Oscillation Frequency  Hz  2.33  2.33  2.33  Casting Speed (max)  mm/s  42.3  42.3  42.3  Oscillation Frequency  Hz  2.67  2.67  2.67  mm/s  31.8  31.8  31.8  Oscillation Frequency  Hz  2.50  2.50  2.50  Oscillation Stroke  mm  12.7  12.7  12.7  Section size Pouring Temperature (start of heat) Pouring Temperature (middle of heat) Pouring Temperature (end of heat) Casting Speed (min)  Casting Speed (avg)  175  3. Mould Parameters Units  CASE "A"  CASE "B"  CASE "C"  Mould length  mm  740  740  740  Metal level  mm  125  125  125  Corner radius  mm  3.175  3.175  3.175  4-sided  2-sided  4-sided  mm  3.175  3.175  3.175  Cooling water flow rate  m3/s  0.0395  0.0300  0.0395  Oil flow rate  mVs  1.33  1.25  1.33  Oil- boiling point  °C  240  240  240  Oil- flash point  °C  320  300  320  Poise  0.50  0.60  0.50  Type of constraint Gap of cooling water channel  Oil- viscosity  at 25°C  176  4. Machine Parameters CASE "A"  CASE "B"  CASE "C"  6  6  6  m  6.0  6.0  6.0  Distance of the first spray nozzle from the bottom of the mould  mm  10.0  10.0  10.0  Distance of pinch-rolls from the meniscus  m  9.0  9.0  9.0  Distance of shear from the meniscus  m  26.5  26.5  26.5  CASE "A"  CASE "B"  CASE "C"  3  3  3  Units Number of strands Machine radius  5. Spray Parameters Units Number of spray zones ZONE 1 Length Water Flow-rate  m m3/s  12 16.2  12 16.2  12 16.2  ZONE 2 Length Water Flow-rate  m m3/s  44 131.7  44 131.7  44 131.7  ZONE 3 Length Water Flow-rate  m m3/s  151 263.4  151 263.4  151 263.4  °C  23  23  23  Water Temperature  177  GLOSSARY  Algorithm A step-by-step approach for solving a problem with the help of a precisely defined group of rules or processes that leads to a desired output from a given set of inputs.  Arc The lines interconnecting nodes in a search tree. Artificial Intelligence (AI) The branch of computer science that is devoted to the study of how computers can be used to simulate or duplicate functions of the human brain. It deals with hardware and software techniques that make it appear as though a computer is thinking, reasoning, making decisions, storing or retrieving knowledge, solving problems, and learning.  Backtracking A technique used in tree searches by which the system works backwards from a failed objective or an incorrect result in order to examine unexplored alternatives. It is simply the process of backing up to a point where a previous choice point was made in a computation and then trying again. Backward Chaining A "goal-driven" method of reasoning technique used in tree searches. It starts with a given goal and works backwards, examining facts and rules that support the desired outcome. It is also called "expectation driven", top-down reasoning" and "backward reasoning".  Belief A hypothesis about some unobservable situation; a measure of the believer's confidence in an uncertain proposition. Blind Search A general search technique that does not utilize any knowledge or heuristics to help accelerate or simplify the search process. Thus, it is a time-consuming process that attempts to exhaust all possibilities in searching instead of relying upon information that can help narrow down the search.  178  Breadth-First Search A search strategy where all the nodes on one level of the search tree are examined before any of the nodes of the next lower level.  Certainty Factor A number ranging from 0 to 100 percent that indicates the possibility that a conclusion reached by the system is true. It is also the degree of belief of a domain expert about a conclusion being true when the premise of a rule is evaluated with a net degree of truth of 100 percent. Class A grouping of similar objects in a hierarchical and schematic manner.  Cognition The process of knowing. Confidence Level A number lying in the range of 0 to 100 percent which represents a threshold value that must be overcome in order for a rule to be successful. If a keyword triplet is to be utilized in forward chaining or to be displayed as a final conclusion, its degree of belief must exceed the confidence level.  Conflict Resolution (of rules) The priority-ordering process which is employed by an inferencing system to decide which rule to recognize or "fire" when more than one rule's IF statement(s) matches the current focus of the system.  Consultation System A general term used for expert and knowledge-based systems that provide advice or diagnose problems.  Control (of expert systems) Any procedure, explicit or implicit, that determines the overall order of problem-solving activities; the temporal organization of subprocesses.  Control Strategy A method of reasoning in a search space such as forward- or backward-chaining, depth- or breadth-first search etc..  Data Base Information stored in a computer.  179  Data Driven Data-directed inferencing strategy used in tree searches. Decisions Tree A graphical structure of nodes and arcs that presents alternative paths for various decisions or outcomes.  Declarative Knowledge Information about objects, events or situations in the form of facts. Default Values A value given to a symbol or variable automatically when no other value is defined by the programmer or user. Degree of Belief A measure of the believer's confidence in an uncertain proposition.  Degree of Certainty A number in the range of 0 to 100 percent that represents the truth about the value of a keyword triplet. True equals 100 percent while False is 0 percent.  Depth-First Search A search strategy that explores each branch of a search tree to its full vertical length. It is a general search technique that is used when no guiding information is available. Here, each branch is searched for a solution and if none is found, a new branch is searched to its depth. This procedure is carried on until a solution is found.  Domain A field of knowledge or expertise; a problem area of interest in expert systems application.  Domain Expert A person with expertise in the domain in which the expert system is being developed.  DOS Disk Operating System for IBM-compatible machines. Expertise The set of capabilities that underlies the high performance of human experts, including extensive domain knowledge, heuristic rules that simplify and improve approaches to problem solving, metaknowledge and metacognition, and compiled forms of behaviour that afford economy in skilled performance.  180  Expert System A computer program that embodies facts, knowledge, rules of thumb, and other elements of heuristic expertise in a particular domain and then uses this information to imitate the human thought process to solve a problem.  Explanation Facility That part of an expert system which explains why a question is being asked or how the system reached a particular conclusion. Fact A proposition or datum whose validity is accepted. Firing a Rule Initiating the action specified by a rule if the conditions are met. Focus The keyword triplet being used by the system in its forward chaining search strategy or the keyword triplet being looked for by the system in its backward chaining search strategy. Forward Chaining A problem-solving technique used in production and rule-based systems in which conclusions are drawn or decisions are made by starting with known facts. A search procedure or reasoning process that uses known facts to produce new facts and to reach a final conclusion. It is also known as "data driven" and "inductive " or "antecedent reasoning". Frame A knowledge representation scheme that associates one or more features with an object in terms of various slots and particular slot values (similar to classes). Fuzzy Logic A well defined reasoning system that is based on the use of fuzzy sets rather than binary values associated with the traditional on/off logic. The development of fuzzy logic was pioneered by L.Zadeh at The University of California at Berkeley in the 1960's. Fuzzy Reasoning A method of dealing with inexact or imprecise information. It involves techniques for avoiding complexities when dealing with subjective information or poorly understood processes. It is a method of determining an adequate solution from imprecise information.  181  Fuzzy Set A mapping of discrete numerical concepts into a set of ranking numbers that describe membership in a particular concept or the degree of belief of a particular linguistic expression.  Goal Driven A method of reasoning that begins with the goal or a conclusion and works backwards through the rules and facts of a knowledge base searching for the path that will achieve the desired goal.  Heuristic Consists of empirical information, rules of thumb, tricks, procedural tips and other experiential knowledge that help to guide, limit, and speed up the search process. Anything that helps a human or a computer to discover or learn. Hierarchy A ranked or graded series of persons or things; a body of knowledge or information organized into successive ranks or grades. Most hierarchies can be illustrated with the use of tree graphs.  High-Level Language A programming language in which the instructions are closer to natural language or algebraic notations than to machine language. IF-THEN The form of rules used in many AI systems, programs and expert systems; a conditional rule in which a certain action is taken only if some condition is satisfied; decision-making tests that initiate an action if a specific condition is met. Inference The process of drawing a conclusion from given evidence; to reach a decision by reasoning. Different types of evidence associations may produce different degrees of belief in a particular conclusion.  Inference Engine The "thinking part" or the "brain" of an expert system that actually performs the reasoning function; that part of an AI program which analyses the information or  knowledge base using rules to make decisions or reach conclusions.  182  Instantiation Applied to a keyword triplet when the degree of certainty of a keyword triplet is known. Intelligence The ability to acquire and apply knowledge through thought and reason. Interface That portion of a computer system or program that links two other portions of the system and then allows them to communicate; a portion of a computer program that interfaces with both the computer user and the remainder of the program or system.  Keyword Triplet Consists of an object-attribute-value combination. An object is an actual or conceptual entity. Attributes are properties associated with objects. Each attribute can take on different values.  Knowledge Understanding, awareness, or familiarity acquired through education or experience; anything that has been learned, perceived, discovered, inferred, and understood.  Knowledge Acquisition The term used by workers in the area of expert systems to characterize the complex activities involved in the development process. It is defined as the transfer and transformation of problem-solving expertise from some knowledge source to a computer program.  Knowledge Base That part of an expert system that is made up of data, rules, inferences, and procedures organized into frames, blackboard, semantic networks, scripts, rules, and other formats; the assembly of all the information and knowledge of a specific domain that forms the basis of an intelligent computer system. Knowledge-Based Systems Al programs that use a knowledge base as their source for problem solving in a particular field of interest.  183  Knowledge Engineer A person who designs and builds expert systems; computer science or AI specialist who acquire knowledge from all sources including human experts and organize it into a knowledge base.  Knowledge Engineering The process of acquiring and formatting knowledge to form a knowledge base.  Knowledge Representation A vocabulary and syntax of symbols and conventions used to describe and present knowledge and information; the structure and organization of information (or knowledge) used to solve a problem.  Learning The process of improving performance by changing or control. LISP Acronym for List Processor. The most widely used AI programming language developed by John McCarthy at MIT in the 1950's. Logic A system of reasoning based on the study of propositions and their analysis in making deductions; a system developed by philosophers and mathematicians for the process of making inferences and facts. Logical Connectives The words "AND" and "OR" that are used to link conditions in rules. Meta A prefix of a term that designates a self-reference to the given term. Meta-Knowledge Knowledge about knowledge. Net Degree of Truth A number in the range of 0 to 100 percent that represents the degree of belief that the premise of any rule is true. This number is ascertained by combining the degrees of truth of each condition statement, and depends on the logical connectives (AND, OR) used to combine condition statements. "ANDing" will assign the minimum degree of truth while "ORing" will assign the maximum.  184  Path A route through a search tree. Predicate A statement about the subject of a proposition. An assertion that denotes the relationship among two or more objects or elements. Premise The condition or the "IF" part of an "IF-THEN" rule. Problem Solving The purpose of most AI programs. This includes the process of answering a question, seeking a solution to an issue, resolving a conflict, and making a decision. Procedural Knowledge Information about the course of action. Production Rule An IF-THEN rule. PROLOG Acronym for Programming in Logic; an AI language that was developed in France using the concepts of predicate calculus. Reasoning The mental process of drawing conclusions from facts, observations, or hypotheses; making inferences from arguments. Rule A regulation or statement defining a particular conduct, habit, or behaviour; in AI, a two-part direction consisting of a condition and a consequent action. An IF-THEN rule states that if a given condition is true, then a specific action should be taken. Rule-Based Any program or system that uses a set of rules to draw conclusions, make decisions, and solve problems. Rule of Thumb A heuristic, principle, technique, trick, or method often used to simplify,  accelerate, or facilitate a process; a technique that does not always work or that is not always accurate. Ruleset A collection of rules that constitute a module of heuristic knowledge.  185  Search The basic process involved in AI applications; to explore or examine in order to discover or learn something; to seek a particular object or goal. Search Tree A graph that looks like an inverted tree and is used to illustrate the search of all the various alternatives in a search space; a hierarchical structure showing all the goals and subgoals (nodes) interconnected by arcs discovered during the search process. Shell An expert system generator; a software package that allows the users to create an expert system without programming Software A collection of programs and routines that support the operation of the computer system. Thinking The process of thought or reasoning; to formulate mentally, decide, judge, consider, and otherwise reflect upon a subject. Tool A software package, such as an expert system shell, that makes it easier to create other softwares; includes high level languages such as LISP and PROLOG.  Top-Down Reasoning A structured method of working backward from a desired goal to determine subgoals and a path through the search tree that will yield a suitable solution to a problem.  Uncertainty In the context of expert systems, it refers to an intermediate truth value. Many expert systems can accommodate uncertainty, i.e. they allow the users to indicate their degree of belief in a concept. User A computer operator.  User-Friendly The term used to describe a facility designed to make interaction with a computer system easy and comfortable for the user.  186  User-Interface That portion of a computer program which communicates with the operator; a portion of the program that accepts inputs and generates outputs with such techniques as natural language and menus.  REFERENCES  Course notes on "Expert System for Metallurgists", 1989, The University of British Columbia, Vancouver, B.C, Canada. E.Turban, Decision Support and Expert Systems, 1990, Macmillan Publishing Company, New York, NY, USA. Building Expert Systems, Eds.F.A.Hayes-Roth, D.A.Waterman, D.B.Lenat, 1983, Addison-Wesley Publishing Company Inc., Reading, MA, USA.  187  

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.831.1-0078575/manifest

Comment

Related Items