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

A statistical safety control model for construction sites using location systems Edrei, Tsah; Isaac, Shabtai Jun 30, 2015

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

Item Metadata


52660-Edrei_T_et_al_ICSC15_040_Statistical_Safety_Control.pdf [ 377.44kB ]
JSON: 52660-1.0076383.json
JSON-LD: 52660-1.0076383-ld.json
RDF/XML (Pretty): 52660-1.0076383-rdf.xml
RDF/JSON: 52660-1.0076383-rdf.json
Turtle: 52660-1.0076383-turtle.txt
N-Triples: 52660-1.0076383-rdf-ntriples.txt
Original Record: 52660-1.0076383-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   A STATISTICAL SAFETY CONTROL MODEL FOR CONSTRUCTION SITES USING LOCATION SYSTEMS Tsah Edrei1, Shabtai Isaac1,2 1 Ben Gurion University of the Negev, Israel 2 Abstract: A statistical safety control model is presented that utilizes data from a Real Time Location System of relatively low accuracy, to alert of unsafe situations at construction sites. Based on the Statistical Process Control methodology, predefined statistical rules are used to detect trends of increasing exposure to hazards, and provide proactive alerts before a critical exposure takes place. An initial verification of the performance of the model was carried out through extensive laboratory tests. 1 INTRODUCTION Various studies indicate that the incidence rate of fatal workplace accidents in the building industry, with an estimated 60,000 fatal casualties a year around the world, is higher than in any other industrial sector (Aires et al. 2010).  The most common cause for fatal accidents on construction sites is usually falling from heights (Aneziris et al. 2012; Wu et al. 2010). In recent years, a number of studies have been dedicated to the improvement of safety on construction sites through the application of a Real Time Location System (RTLS) in order to track the movement of workers and prevent accidents from occurring (e.g. Nattichia et al. 2013; Maalek and Sadeghpour 2013). The main purpose of the RTLS is to facilitate an alert when a worker enters an area that has been defined as being of “high risk”.  Such an approach can be considered deterministic, and requires a highly accurate tracking system in order to be able to detect in real-time when a worker moves from a low risk area to a high risk area. Highly accurate indoor tracking systems such as Ultra Wide Band (UWB) (i.e. systems that, unlike standard GPS, can be used inside buildings that are under construction) are relatively expensive technologies that require significant time and effort to deploy (Khoury and Kamat 2009). Other, less expensive technologies such as WLAN-based tracking systems are economical but provide a much lower accuracy (i.e. 1.5 to 2 m, as opposed to centimeter level positioning accuracy for UWB). Such a low accuracy is not compatible with a deterministic approach, in which any penetration into a high risk area needs to be immediately identified. An additional factor that needs to be taken into account, in the application of a RTLS to improve worker safety on construction sites, is the high uncertainty that typically exists regarding expected worker locations.  This uncertainty stems from the fact that the project plan often does not reflect in real time decisions on changes that are made by the project management team (Isaac and Navon 2013). In addition to that, even an accurate and up-to-date project plan cannot fully reflect the complexity and unpredictability of worker movements, which often deviate from expected work envelopes (Isaac and Navon 2013).  040-1 The use of a RTLS on construction sites can therefore enhance worker safety. However, the economic rationale for using a cheap but inaccurate RTLS, and the uncertainty regarding actual worker behavior onsite, means that one might not be able to rely solely on the definition of safe work areas with clear-cut boundaries, and deterministic alerts in case these boundaries are crossed. In fact, the uncertainty regarding worker behavior may imply that improving the accuracy of real-time location measurements is likely to be less beneficial than is often assumed. Transgressions on the site are likely to be frequent, due to both workers entering dangerous areas, and the location of those areas changing in ways that differ from the project plan.  One solution could be to compensate for the expected inaccuracy and uncertainty by significantly enlarging the areas defined as being of high risk, and therefore off-limits for most workers on site. However, this would reduce the efficiency of the construction processes.  An additional limitation of the current deterministic approach for applying RTLS systems to enhance construction worker safety is the fact that for some safety risks, the temporal dimension is of importance. In other words, the duration of risk exposure has to be taken into account. For example, in case of a hazardous noise exposure, the duration of exposure is an essential factor. The current deterministic approach for using a RTLS for safety control on construction sites therefore has three main limitations:   1. It requires highly accurate and expensive tracking systems 2. It is not fully compatible with the highly uncertain nature of construction projects 3. It does not take into account the duration of risk exposure, which is of importance for certain hazards. In order to provide an answer to these limitations, the present research is based on a statistical approach that complements the existing deterministic methods. Specifically, this research deals with the development of a statistical safety control model that utilizes data from a RTLS, with the objective to alert of unsafe situations at construction sites of multistory buildings. An unsafe situation is defined here as one that causes workers to be exposed to hazards which were initially created by other teams of workers.  While there are many methods and models available to assess the risks that the workers’ own activities pose to themselves, few studies have dealt with the hazards derived from the concurrent activities of other workers on site to which workers are also frequently exposed (Hallowell et al. 2011). 2 THE PROPOSED STATISTICAL SAFETY CONTROL MODEL The proposed model is implemented in four modules, as follows:  1. Identification of hazards which are related to the execution of an activity;  2. Risk assessment according to a structured method (though sufficiently flexible to accommodate different types of hazards);  3. Division of the site into risk areas according to predefined parameters of injury level and distance from hazard;  4. Definition of safety zones using a statistical methodology. In the first module, an identification of potential hazards, based on the construction site layout and the project's schedule, is required before carrying out an activity. The related hazards are those to which other teams might be exposed. This stage contains several sub-stages. Preliminary Hazards Analysis breaks down the activity into a chain of actions. This list of actions makes it easier to detect the hazards. Following this, the required work area is analyzed in terms of the type of space, and the expected worker movement patterns. This is done in order to identify the potential teams that might be exposed to the related hazards.  In the second module, a risk assessment of the identified hazards is carried out. The assessment is defined by 2 parameters: 1) the level of the potential injury when an accident occurs, based on the classification of the Israel Institute for Occupational Safety and Hygiene; 2) The distance from the 040-2 hazards, taking into account different factors such as the RTLS accuracy, the movement velocity of the workers, data processing by the RTLS, worker reaction time to alerts, etc.  In the third module, the construction site is divided into risk areas. This division is a product of the risk assessment. The combination of assessments regarding injury level and distance from hazards creates a matrix which defines the risk level at a certain location. There are several risk area categories, as follows: very high, high, medium and low risk. The main applicable difference between those areas is the level of hazard exposure to which non-authorized teams are allowed. In fact, the risk area classification defines the level of tolerance towards the presence of non-authorized workers in a certain risk area. Therefore, each worker must be familiar with the risk area classification.  In the fourth module, statistical safety zones are defined. This stage is related to the statistical operation. Therefore, statistical alerts rely on this division. The main impact of the division is the definition of a “medium risk” area, where the proposed statistical model can provide benefits not provided by deterministic models. The medium risk area may appear to be “free” of hazards, since the risk is not defined there as critical or immediate. However, a continuous exposure to a certain hazard may harm a worker, or might increase the probability that that worker will penetrate into the high risk area. Therefore, the model contains several statistical rules that detect an increasing exposure to a hazard, and that provide alerts before a critical exposure occurs.  The remainder of this paper will focus on the fourth module, which is based on an existing methodology of process management, called Statistical Process Control (SPC). 3 STATISTICAL PROCESS CONTROL (SPC) SPC is a statistical methodology for process management, which has been mainly used for quality control in manufacturing. SPC uses statistical tools to observe specific characteristics of the manufactured product, and identify significant variations in those characteristics. Instead of defining deterministic rules for rejecting a product that does not meet certain specifications, SPC assumes that some variation in the process is to be expected, and that only a statistically significant variation needs to be addressed, and the factors causing it identified. The assumption in SPC is that the measured characteristic of a specific object has a normal distribution, which distributes symmetrically around the Mean (µ), with three zones defined according to the Standard Deviation (𝜎𝜎): 1. Zone1: ~68% of population, which defined as: µ±𝜎𝜎; 2. Zone2: ~95% of population, which defined as: µ±2𝜎𝜎; 3. Zone3: ~99% of population, which defined as: µ±3𝜎𝜎;  An application of the SPC methodology for safety control on construction sites, instead of for quality control in manufacturing, obviously requires significant adjustments. Firstly, it is assumed, based on the Central Limit Theorem, that the distribution of the movements of workers will be normal relative to the main “work” location. Secondly, the division into zones is based on safety aspects. Thus, Zone1 is related to a low risk area, Zone2 and Zone3 are related to a medium risk area, and all locations above Zone3 are related to a high risk area. The hypothesis of the present research is that a statistical model can deal with issues such as the relatively low-level accuracy (2m) of a cost effective RTLS and the high uncertainty regarding workers' movements, by providing alerts based on a statistical analysis of locations within medium risk areas, in addition to the existing deterministic alert when a worker enters a high risk area.  In practice, such medium risk areas would constitute buffers between planned work areas and areas identified as being of high risk to the workers. The application of SPC for quality control is based on certain statistical rules. The rules that are nowadays commonly used in the manufacturing industry were defined in the middle of the 20th century by an American company called Western Electric Company (Western Electric Co. 1958). These rules can be used to detect statistical trends relative to the desired mean, based on the expected normal distribution. 040-3 For example: measurements that fall within Zone3 have a low probability (under 1%), and are therefore considered abnormal. Other rules are based on sequential measurements related to the predefined zones and their probabilities. For example: the probability of four out of five consecutive measurements being in Zone2 or Zone3 is about 3%, and therefore considered to be statistically significant. An advantage of the statistical approach is thus the ability to detect trends of increasing exposure to hazards by calculating sampling rates, either for a single worker or an entire crew. This supports proactive actions in the form of alerts, which are received before a critical exposure takes place. As will be demonstrated, the model can ensure that alerts will not be ignored, by using such statistical rules to avoid an excessive number of alerts, and by discerning who should be the client of an alert. We can thus increase both safety and efficiency (in terms of multiple teams working simultaneously on site). This, on contrast to the present situation, in which areas with moderate risk exposure levels are either ignored (therefore increasing safety risks), or included in deterministic no-entry zones (therefore reducing the efficiency of the construction processes). 4 IMPLEMENTATION OF THE MODEL The implementation and initial verification of the model was based on laboratory tests that were carried out at the Structural Engineering Department at Ben-Gurion University of the Negev, Israel. Simulated movements of workers were monitored in the department using a Wi-Fi-based RTLS that was installed for this purpose. The RTLS included Access Points, Low Frequency tags and a software platform. The accuracy of the RTLS was 2 meters, and it was chosen due its low cost, which is expected to increase the likelihood that it will be used in actual construction projects. In fact, this type of RTLS is based on a regular Wi-Fi network that is installed in any case in most buildings. The only extra costs would therefore be those of the tags, which can be reused from one project to another. It was assumed that the implementation of a statistical model would make it possible to overcome the relatively low accuracy of the RTLS. For the tests, one floor of the department was divided into three different zones and a high risk area. The location of those zones changed in the tests according to different scenarios of construction activities that were defined. The definition of the zones according to one such scenario is shown in Figure 1. Zone1 (related to a low risk area) is shown in green, Zone2 and Zone3 (related to a medium risk area) are shown in yellow, and the high risk area is shown in red. During the tests, teams of participants carrying tags moved around the department and their locations were tracked. These locations were then converted into their distance from a simulated hazard location, according to the scenario.   Figure 1: Definition of zones in a laboratory test Using the model, different types of alerts were provided. These included alerts related to a penetration into a high risk zone, as well as statistical alerts according to the predefined rules. The model was used to track movements and provide statistical alerts at 3 different levels: • At the level of a single worker (Figure 2);  • At the level of a team of workers (Figure 3);  • At the project level (Figure 4).  Hazard 040-4 The trend charts in Figures 2-4 present the locations of the tags that were tracked in tests, relative to the statistical zones. The horizontal axis is the time axis, whereas the vertical axis presents the location of a tag relative to the predefined statistical zones. Black marks indicate locations for which statistical alerts were provided. These alerts depend on the definition of the statistical zones, which depend, in turn, on the structured risk assessment that was carried out. The alerts can be related to a single worker (when the location of that worker deviates from the safe work area), to an entire team (for example, when the team starts working in a medium risk area), or to all the teams (when a certain event or management decision puts them in danger). Accordingly, the clients of such alerts can be the worker, a site manager, or the company administration.   Figure 2: Trend chart of employee #4 in one of the tests  Figure 3: Trend chart of Team #2 in one of the tests  Figure 4: Trend chart of all the Teams The verification criteria for the testing of the statistical model were based on a comparison of the alerts provided by the model, with a manual measurement of the duration for which participants were located in 040-5 a certain zone. The results of this comparison reveal that about 7% of the samples in the medium risk area were mistakenly identified as being in a high risk area (i.e. a false alert) (Table 1). In addition, about 12% of the samples in the high risk area were mistakenly identified as being in a medium risk area. One might assume that in those cases in which the second type of error occurred, the risk potential was high, since there wasn’t any deterministic alert of the participant entering a high risk area. However, all of those cases were in fact detected through the statistical rules in the model, and statistical alerts were accordingly provided. Therefore, all of the events in which a worker would have been exposed to a high risk, were either warned of through a deterministic alert of a penetration into a high risk area (~88% of the events), or through a statistical alert that was provided according to the predefined rules in the model (the remaining ~12% of the events). Nevertheless, it should be noted that an implementation of the model in a real project, would of course include an adjustment of the model's parameters in order to improve its accuracy.  Table 1: Comparison of the model's output with manually collected data Color of manually identified zone  Identification by model -type of mistake  % of samples Green No mistake 90 Green Yellow Yellow instead of Green No mistake 10 87 Yellow Green instead of Yellow 6 Yellow Red instead of Yellow 7 Red No mistake 88 Red Yellow instead of Red 12 5 CONCLUSIONS A statistical model was developed in this research, with the objective to alert of unsafe situations on construction sites. The model utilizes data from a cost-efficient, yet relatively low-accuracy RTLS, to prevent workers from being exposed to hazards which were initially created by other teams of workers. The statistical approach complements existing deterministic models by detecting trends of increasing exposure over time to hazards, and providing proactive alerts that are received before a critical exposure takes place. This enables it to deal with issues such as a high uncertainty regarding workers' movements on construction sites, which is usually not observed in other types of industry. The results of the laboratory tests show that the model has the ability to overcome the medium accuracy of the RTLS, based on the division of the site into statistical zones. In addition, it was found that the percentage of errors in zone identification is about 10%. Nevertheless, the statistical rules were found to be efficient, since all the critical zone detection errors that were related to a penetration into a high risk area would have been prevented by alerts provided according to the predefined statistical rules, and this in turn would have prevented the entrance of workers into to a high risk area. Obviously, these results still have to be confirmed through an implementation in a real construction project.    References Aires, M. D. M., Gámez, M. C. R., & Gibb, A. 2010. Prevention through design: The effect of European Directives on construction workplace accidents. Safety science, 48(2): 248-258. Aneziris, O. N., Topali, E., & Papazoglou, I. A. 2012. Occupational risk of building construction. Reliability Engineering & System Safety, 105: 36-46. Hallowell, M., Esmaeili, B., & Chinowsky, P. 2011. Safety risk interactions among highway construction work tasks. Construction Management and Economics, 29(4), 417-429. 040-6 Isaac, S., and Navon, R. 2014. Can project monitoring and control be fully automated? Construction Management and Economics, 32(6), 495-505. Khoury, H. M., & Kamat, V. R. (2009). Evaluation of position tracking technologies for user localization in indoor construction environments. Automation in Construction, 18(4), 444-457.  Maalek, R., and Sadeghpour, F. 2013. Accuracy assessment of Ultra-Wide Band technology in tracking static resources in indoor construction scenarios. Automation in Construction, 30, 170-183. Naticchia, B., Vaccarini, M., and Carbonari, A. 2013. A monitoring system for real-time interference control on large construction sites. Automation in Construction, 29, 148-160. Western Electric Co. 1958. Statistical Quality Control Handbook, 2nd edition Wu, W., Gibb, A. G., & Li, Q. 2010. Accident precursors and near misses on construction sites: An investigative tool to derive information from accident databases. Safety science, 48(7), 845-858.   040-7 


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