International Construction Specialty Conference of the Canadian Society for Civil Engineering (ICSC) (5th : 2015)

Adaptive control of bulldozer's workflows Bulgakov, Alexey; Bock, Thomas; Tokmakov, Georgy Jun 30, 2015

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

Item Metadata


52660-Bulgakov_A_et_al_ICSC15_148_Bulldozer_Workflows.pdf [ 1.28MB ]
JSON: 52660-1.0076477.json
JSON-LD: 52660-1.0076477-ld.json
RDF/XML (Pretty): 52660-1.0076477-rdf.xml
RDF/JSON: 52660-1.0076477-rdf.json
Turtle: 52660-1.0076477-turtle.txt
N-Triples: 52660-1.0076477-rdf-ntriples.txt
Original Record: 52660-1.0076477-source.json
Full Text

Full Text

5th International/11th Construction Specialty Conference 5e International/11e Conférence spécialisée sur la construction    Vancouver, British Columbia June 8 to June 10, 2015 / 8 juin au 10 juin 2015   ADAPTIVE CONTROL OF BULLDOZER’S WORKFLOWS Alexey Bulgakov1,4, Thomas Bock2 and Georgy Tokmakov3 1 South West State University, Russian Federation 2 Technical University Munich, Germany 3 South Russian State Polytechnic University, Russian Federation 4 Abstract: The most important task for bulldozer’s traction mode control is to use its traction capacity in full by means of its end-effectors control. To keep traction mode at maximum or at a given resistance value applied to end-effectors automatically is difficult due to a great number of stochastic factors affecting the bulldozer. Bulldozer is taken as a mechatronic system [1, 2]. The study presents analytic dependences for the sub-processes where analytic modeling based on bulldozer’s parameters correlation knowledge is applicable. Models of the sub-processes are included into the general structure of bulldozer’s workflow simulation model. Simulation technique is demonstrated through model development of the bulldozer as a universal machine operating in modes of soil movement and subgrade surfacing. In developing the models mathematical apparatus of the theory of random processes, transfer functions, table interpolation, numerical solution of algebraic equations and ordinary differential equations in the Cauchy form was used. A dynamic model of the drawing prism formation was developed describing the dependence of the volume of prism on the variable digging depth and variable bulldozer speed. A general structure of the model of bulldozer’s workflows [3] due to the working process control objectives was developed.     1 INRODUCTION Bulldozers equipped with modern navigation and information systems are mobile mechatronic objects, and they can be integrated into general process of intellectual construction. The integration will provide optimal efficiency of the construction cycle and will ensure lean production process. On the basis of bulldozer’s workflow dynamics modeling and analyses described in a variety of works, we have concluded that the models to describe kinematics and dynamics of its working equipment, hydraulic and transmission features tend to be analytical formulas derived from well-known laws of physics and from information on bulldozer’s structure and mechanisms. If some parameters of the workflow are unknown or constantly changing, the models are either statistical tables or empiric dependences summarizing experimental data. The models depict interaction of end-effectors, engines and environment as well as statistic features of bulldozer’s complex units. Application of regulators based on classical control theory is difficult due to the frequent changes in workflow conditions. Thus, it is necessary to develop adapted control systems to eliminate the difficulties described. The system includes both the bulldozer’s dynamics modeling and bulldozer’s workflow control method to take into consideration the complex non-linear dependencies between workflow parameters and incomplete information on its working conditions changes. 148-1 Having reviewed adaptive and intellectual control methods [4, 5], we propose to create an adaptive control system for technological processes to increase efficiency of bulldozer’s control in comparison with traditional control methods. 2 MOBILE MECHATRONIC OBJECT - MATHEMATICAL DESCRIPTION TO PERFORM EXCAVATION WORKS ON THE BASIS OF A DOZER When researching a dozer’s working process usually a number of design schemes are considered – straight line, thread milling, wedge and exponential cutting. Meanwhile, a dozer moves along the surface that is formed by its blade. Therefore, when driving onto any surface roughness resulting from the dozer blade control or the change in its position due to any reason, causes position changes of the machine frame and along with the cutting edge that is any face deviation from a straight line in some extent is copied by the dozer. Observations [1] show that quite often while designing a face its roughness is progressing, reaching a size at which the control over the workflow is lost. In this case, the operator has to align the face deliberately, trying to ensure its "tranquil" profile that allows doing excavation works smoothly, without frequent control system switching and reducing the dozer’s operating speed that causes a slowdown and shows inferiorities of the blade control system. Obviously, if the control system operates in the antiphase towards deviations of the tractor frame with sufficient accuracy, the initial face roughness will not evolve and will be gradually cut. One of the most likely causes of the opposite phenomenon observed in practice, is the disparity between the velocity of the dozer Vp and actual conveying speed of the working body Vot required in certain areas Si of the digging operating cycle, where i- is the number of the speed change Vot. Speed ratio depends on the dozer’s geometrical dimensions (Figure 1) and its control system.  Figure 1: Dozer’s geometrical dimensions Mathematical model of the dozer’s movement on a straight line tracking (frame alignment) is built using the Lagrange equations of the 2nd kind, under the assumption that the contribution to the dynamics of the drive gears and a track is small, compared with the contribution of the remaining parts of the dozer.   𝑑𝑑𝑑𝑑𝑑𝑑�𝜕𝜕𝜕𝜕𝜕𝜕?̇?𝑥� −𝜕𝜕𝜕𝜕𝜕𝜕𝑥𝑥= 𝑄𝑄𝑥𝑥  𝑑𝑑𝑑𝑑𝑑𝑑�𝜕𝜕𝜕𝜕𝜕𝜕?̇?𝜑� −𝜕𝜕𝜕𝜕𝜕𝜕𝜑𝜑= 𝑄𝑄𝜑𝜑   where kinetic energy: G C 4 C 1 C 2 B C 3 l 2 C 5 [1]  148-2 [2] 𝜕𝜕 = 12𝑚𝑚1?̇?𝑥2 +12𝑚𝑚2�?̇?𝑥2 + (𝑙𝑙2𝑙𝑙𝑐𝑐2?̇?𝜑)2 + 2?̇?𝑥𝑙𝑙2𝑙𝑙𝑐𝑐2?̇?𝜑𝑠𝑠𝑠𝑠𝑠𝑠(𝜑𝜑)� +12𝐽𝐽𝑐𝑐2?̇?𝜑2 + +12𝜎𝜎ℎ𝑥𝑥�?̇?𝑥2 + +(𝑙𝑙2?̇?𝜑)2 + 2?̇?𝑥𝑙𝑙2?̇?𝜑𝑠𝑠𝑠𝑠𝑠𝑠(𝜑𝜑)� +12𝜎𝜎ℎ𝑥𝑥𝑠𝑠г𝑐𝑐𝑧𝑧2 ?̇?𝜑2 generalized forces acting on a dozer: [3]𝑄𝑄𝑥𝑥 = −𝜎𝜎ℎ𝑔𝑔𝑙𝑙2𝑠𝑠𝑠𝑠𝑠𝑠𝜑𝜑 + 𝐹𝐹т − 𝐹𝐹сопр; [4]𝑄𝑄𝜑𝜑 = −(𝑚𝑚2𝑙𝑙𝑐𝑐2 + 𝜎𝜎𝑥𝑥ℎ)𝑔𝑔𝑙𝑙2𝑐𝑐𝑐𝑐𝑠𝑠𝜑𝜑 +𝑀𝑀; m1--tractor mass; m2- blade frame mass; 𝜎𝜎 - soil surface density; ℎ - depth of the soil cutting;  𝑙𝑙𝑐𝑐2 - center of the blade mass; irz - gyration radius of the dumping soil. [5]𝑚𝑚1?̈?𝑥 +𝑚𝑚2?̈?𝑥 +𝑚𝑚2𝑙𝑙2𝑙𝑙𝑐𝑐2?̈?𝜑𝑠𝑠𝑠𝑠𝑠𝑠𝜑𝜑 +𝑚𝑚2𝑙𝑙2 𝑙𝑙𝑐𝑐2?̇?𝜑2𝑐𝑐𝑐𝑐𝑠𝑠𝜑𝜑 + 𝜎𝜎ℎ?̇?𝑥2 + 𝜎𝜎ℎ𝑥𝑥?̈?𝑥 + 𝜎𝜎ℎ?̇?𝑥𝑙𝑙2?̇?𝜑𝑠𝑠𝑠𝑠𝑠𝑠𝜑𝜑 + 𝜎𝜎ℎ𝑥𝑥𝑙𝑙2?̈?𝜑𝑠𝑠𝑠𝑠𝑠𝑠𝜑𝜑 +𝜎𝜎ℎ𝑥𝑥𝑙𝑙2?̇?𝜑2𝑐𝑐𝑐𝑐𝑠𝑠𝜑𝜑 −12𝜎𝜎ℎ?̇?𝑥2 −12𝜎𝜎ℎ𝑠𝑠г𝑧𝑧2 ?̇?𝜑2 −12𝜎𝜎ℎ(?̇?𝑥2 + 𝑙𝑙22?̇?𝜑2 + 2?̇?𝑥𝑙𝑙2?̇?𝜑𝑠𝑠𝑠𝑠𝑠𝑠𝜑𝜑) = −𝜎𝜎ℎ𝑔𝑔𝑙𝑙2𝑠𝑠𝑠𝑠𝑠𝑠𝜑𝜑 + 𝐹𝐹т − 𝐹𝐹сопр; [6] 𝑚𝑚2𝑙𝑙22𝑙𝑙𝑐𝑐22 ?̈?𝜑 +𝑚𝑚2?̈?𝑥𝑙𝑙2𝑙𝑙𝑐𝑐2𝑠𝑠𝑠𝑠𝑠𝑠𝜑𝜑 +𝑚𝑚2?̇?𝑥𝑙𝑙2𝑙𝑙𝑐𝑐2𝑐𝑐𝑐𝑐𝑠𝑠𝜑𝜑?̇?𝜑 + 𝐽𝐽𝑐𝑐2?̈?𝜑 + 𝜎𝜎ℎ?̇?𝑥𝑙𝑙22?̇?𝜑 + 𝜎𝜎ℎ𝑥𝑥𝑙𝑙22?̈?𝜑 + 𝜎𝜎ℎ?̇?𝑥𝑙𝑙2𝑠𝑠𝑠𝑠𝑠𝑠𝜑𝜑 + 𝜎𝜎ℎ𝑥𝑥𝑙𝑙2𝑐𝑐𝑐𝑐𝑠𝑠𝜑𝜑?̇?𝜑 +𝜎𝜎ℎ?̇?𝑥𝑠𝑠г𝑧𝑧2 ?̇?𝜑 + 𝜎𝜎ℎ𝑥𝑥𝑠𝑠г𝑧𝑧2 ?̈?𝜑 − 𝑚𝑚2?̇?𝑥𝑙𝑙2𝑙𝑙𝑐𝑐2?̇?𝜑𝑐𝑐𝑐𝑐𝑠𝑠𝜑𝜑 − 𝜎𝜎ℎ𝑥𝑥?̇?𝑥𝑙𝑙2?̇?𝜑𝑐𝑐𝑐𝑐𝑠𝑠𝜑𝜑 = −(𝑚𝑚2𝑙𝑙𝑐𝑐2 + 𝜎𝜎𝑥𝑥ℎ)𝑔𝑔𝑙𝑙2𝑐𝑐𝑐𝑐𝑠𝑠𝜑𝜑 − (𝑚𝑚2𝑙𝑙𝑐𝑐2 + 𝜎𝜎𝑥𝑥ℎ)𝑔𝑔𝑙𝑙2𝑐𝑐𝑐𝑐𝑠𝑠𝜑𝜑 +𝑀𝑀; The system (1)  solution allows getting the differential equations (4) and (5) that describe the dozer’s movement on a straight line track, and determining control actions through the parameters of the machine in areas Si of the digging operating cycle as the coefficients ai in the dependence Vot=aiVp. Such a dependence is typical for dozers with a single-motor drive with a hard pump hydraulic drive connection to the motor shaft.   Figure 2: The movement of the tractor frame the beginning of digging. At the beginning of digging (Figure 2), the frame of the tractor makes a strictly forward movement over a distance of S1+S2 without hesitation relatively its mass center. The blade cutting edge in the area S1 dives into the soil to a depth equal to a predetermined cutting thickness h. Thus, the control action а1 may be determined by the formula:  S 2 V ot = 0 G C 2 S2 S1 S 1 + V ot  148-3 [7] a1=30itrm l2πrkFziprC5n; where itr, ipr - tractor transmission and hydraulic pump ratios;  n - number of hydraulic cylinders;  m - fluid mass in the hydraulic cylinders; In the area S2 the movement is made with а2=0 until the mass center of the tractor won’t move to the buttonhole edge.  Figure 3: Movement the dozer "dives" in the drawn buttonhole On further movement the dozer "dives" in the drawn buttonhole (Figure 3), so in the area S3 it is necessary to lift the blade at a rate of Vot, determined by the coefficient a3: [8] 𝑎𝑎3 = tg𝛽𝛽 �𝑒𝑒𝑎𝑎𝑉𝑉п𝑡𝑡𝐶𝐶1+𝑉𝑉п𝑡𝑡 �1 +𝑎𝑎𝐶𝐶1𝐶𝐶1+𝑉𝑉п𝑡𝑡� − 1� ; The area S3 ends after the dozer’s back gear hits the edge of the face and reverse alignment of tractor frame starts. Length of the alignment area is S4≈ S1. Obviously, during this period it is necessary to start dropping the blade. The а4 determines the rate of dropping the blade in the given area: [9] 𝑎𝑎4 =С3𝑆𝑆1(𝐶𝐶4+𝑆𝑆3+𝑉𝑉п𝑡𝑡)2; To implement control actions аi=f(Si, t, h) the dozer must be equipped with a vertical blade control system. 3 NEURAL NETWORK MODEL OF BULLDOZER WORKFLOW The Autoregressive model structure with external inputs (Figure 4) is a dynamic two-layer recurrent neural network. It is found from the autocorrelation signal functions that the autocorrelation coefficient is greater than 0.8 in the time interval 0.1 sec. for speed ( )tv  of 0.5 sec. for digging depth ( )th  and 0.2 sec for the resistance force ( )tP . Length of delay lines TDL taking into account the sampling frequency of 10 Hz are up to 1, 5 and 2 accordingly (Figure 4). G S 3 - V ot  148-4 [14] ( ) ( ) ( ) ( ) ( )tetFttttt ×∇×∆−−∆−= PXX ; Covariance matrix of the vector Х(t) of neural network parameters used in the algorithm: [15] ( ) ( ) ( ) ( )( )[ ] ( ) ( ){ } ( )[ ] ( ) λλ tttFtFtttFtFtttttTT ∆−×∇×∇×∆−×∇+∇×∆−−∆−=PPPPP; 4 ADAPTIVE CONTROL OF BULLDOZER’S WORKFLOWS Applying a hybrid neural network consisting of a combination of traditional neural networks and neural networks of higher order (Figure 5.). Thus, the neural network has the ability to switch between linear connections and connections of high order that can be described by the following dependencies. Linear coupling: [16] 𝑦𝑦𝑖𝑖 = ∑�𝑤𝑤𝑖𝑖𝑖𝑖𝑥𝑥𝑖𝑖 + 𝑏𝑏𝑖𝑖0�; High order coupling: [17] 𝑦𝑦𝑖𝑖 = 𝑓𝑓 �∏𝑥𝑥𝑖𝑖𝑝𝑝𝑖𝑖𝑖𝑖𝑖𝑖∗ 1𝑏𝑏𝑖𝑖0�; Activation function: [18] f(x) =11+e−ax; where wij – coupling weight coefficients; yi – output neuron signal; xi – input neuron signal;   Figure 5: Hybrid Neural Network Structure This implies that each layer depending on the operating mode, may change the type of connection between neurons. For example, for a neural network consisting of 3 layers, the following options are possible (linear - L, higher order - HO): L-L; L-HO; HO-L and HO-HO. To optimize the created neural network is possible with the help of the genetic algorithm adaptation (Figure 6).  . . . . . . Fuzzification procedure Hidden layers Defuzzification procedure . . . . . . . . . . . . Output Input  148-6  Figure 7: Architecture of a management and control systems 5 CONCLUSIONS AND RESULTS Automatic control function of blade positions precisely adjusts the cutting edge. Depending on the content of correction signals, the regulating dual hydraulic valve automatically lifts or drops the cutting edge of the blade, constantly keeps it in position that ensures the accuracy of work and ensures an optimum level of productivity. Identification technique of the dozer’s working processes and models obtained on its base, are intended to be used in the development of adaptive systems of automatic control of the dozer’s working process. Methods of development of adaptive systems of control of the dozer’s working process, is based on neural network technology. For the formation of control actions on a dozer, and of the electrical switch signals of the hydraulic directional valves of the lifting and dropping hydraulic cylinders of the working body, in particular, the structure and functioning algorithms of the adaptive neural network controller have been designed. References Krapivin D.M., Nefedov V.V., Tokmakov G.E. Mathematical model for the movement of mechatronischen devices for the intelligent building site, Mechatronik, Lik, Nowotscherkassk, 2010.- S. 50-54. Min-Yuan Cheng, Hsing-Chih Tsai, Erick Sudjono. Evolutionary fuzzy hybrid neural network for construction industry. Automation in Construction 21 (2012) S. 46-51. Mecheryakov V.A. Recurrent algorithm neural identification working process earthmoving machinery. Siberian State Automobile and Highway Academy, Omsk, 2007. - S. 63-66 Kureychik V.M. Genetic algorithms. State problems, prospects. Izvestiya RAN, Theory and control system. 1999. № 1. S.144-160. Rutkovskij L., Pilinskij M. Neural networks, genetic algorithms and fuzzy systems. Gorzachaza Liniza TELEKOM, 2004. Input Data; Desired results Fuzzification Procedure Fuzzy logic block The hybrid neural network Defuzzification Procedure Predicted Results Optimization Procedure Membership Function Neuron  Network  Paramete  Defuzzification Parameters Data flow    Control flows   Functional objects  Databases  148-8 M.E. Georgy, L.M. Chang, L. Zhang, Prediction of engineering performance: a neurofuzzy approach, Journal of Construction Engineering and Management 131 (5) (2005). S.548–557 T.M. Cheng, C.W. Feng, M.Y. Hsu, An integrated modeling mechanism foroptimizing the simulation model of construction operation, Autom. Constr. 15 (2006). S.327–340.  148-9 


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



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"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items