تجزیه و تحلیل شبکه های بیزی از حوادث محیط کار ناشی از سقوط از ارتفاع
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|28768||2009||9 صفحه PDF||سفارش دهید||6308 کلمه|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Safety Science, Volume 47, Issue 2, February 2009, Pages 206–214
This article analyses, using Bayesian networks, the circumstances surrounding workplace tasks performed using auxiliary equipment (ladders, scaffolding, etc.) that may result in falls. The information source was a survey of employees working at a height. We were able to determine the usefulness of this approach – innovative in the accident research field – in identifying the causes that have the greatest bearing on accidents involving auxiliary equipment: in these cases, the adoption of incorrect postures during work and a worker’s inadequate knowledge of safety regulations. Likewise, the duration of tasks was also associated with both these variables, and therefore, with the accident rate. Bayesian networks also enable dependency relationships to be established between the different causes of accidents. This information – which is not usually furnished by conventional statistical methods applied in the field of labour risk prevention – allow a causality model to be defined for workplace accidents in a more realistic way. With this statistic tool, the expert is also provided with useful information that can be input to a management model for labour risk prevention.
Techniques used to manage accident prevention in companies include accident analyses, accident investigations, safety inspections and incident recall, etc. (Bird and Germain, 1990; Ley 31/1995), which provide management with information on the causes of accidents among particular groups of employees. Knowing the circumstances and causes of accidents enables corrective and preventative measures to be implemented that exercise greater control over factors that may cause accidents. Different informations sources are used in order to apply these analysis tools. Fatality inspection records, which are completed after the accident, have been used by some authors for their research ( Janicak, 1998). These records reflect the circumstances of the accidents providing data, for example, on the job, type of activity, type of injury, direct cause of the injury, etc. In Spain, fatality records are the most widely used source of information for historical studies of workplace accidents ( Orden TAS/2926/2002; Begueria, 1988). Other information sources – including risk reports ( Bird and Germain, 1990) and worker surveys ( Gillen et al., 2002, Kines, 2003 and Paul and Maiti, 2007) – enrich the theories elaborated from the usual sources of information and provide additional, mostly subjective, information (largely on the behaviour of the worker during the risk activity). Irrespective of the information source, the data is usually analysed using conventional descriptive statistics (Kines, 2003), factorial analysis (Dedobbeleer and Beland, 1991), analysis of variance (Janicak, 1998), and multiple regression (Gillen et al., 2002). The conclusions obtained using these simple data processing techniques – which form the basis for many management models – enable the relationship between the accident and each causal variable to be analysed, but do not enable the interplay between causes to be determined. These techniques fail to reflect, therefore, the fact that an accident is usually the result of more than one factor – that is, the outcome may be greater than the sum of the parts (Bird and Germain, 1990). More effective approaches to defining the interplay between variables have been developed by other authors, for example, using structural equation models (Paul and Maiti, 2007). In the present work we use an approach based on Bayesian networks (BNs) to describe the circumstances (and relationship between circumstances) associated with tasks performed at a height that might result in personal injury or damage to property. BNs have been applied in several knowledge areas, such as medicine (Antal et al., 2007), ecology (Adriaenssens et al., 2004), environmental assessment impact (Baran and Jantunen, 2004, Marcot et al., 2001 and Matías et al., 2006), business risk and product life-cycle analysis (Zhu and Deshmukh, 2003), and more recently, to handling data obtained as a result of prospection for minerals and rocks (Rivas et al., in press). In the workplace risk area, Galán et al. (2007) applied a canonical probabilistic test (based on Bayesian models) to the analysis of nuclear system safety and Papazoglou and Ale, in press and Papazoglou et al., 2006 applied functional block diagrams and event trees to quantify the risk of falls. More specifically for construction and mining accidents, Matías et al. (2008) compared the predictive capacity of BNs with other expert systems, concluding that BNs, in addition to their good predictive capacity, possess a satisfactory interpretative capacity in regard to workplace accidents, given that: (1) they enable different circumstances to be simulated and their effects on each of the variables in play to be probabilistically analysed; (2) they enable the use of discrete qualitative variables (such as the many parameters that have a bearing on accidents); and (3) they enable the causal dependency relationship between variables to be mapped. Using BNs, we analysed workplace accidents caused by falls from a height in order to identify the most important causes of this kind of accidents and, most of all, to determine the relationships existing between these causes, which will allow the real circumstances of the unsafe work tasks performed at different heights to be defined. We focused on falls, from a height of more than 2 m above floor level, of employees working in a standing position. In Spain, workplace falls from a height are a major cause of workplace fatalities, third only to accidents involving vehicles and heart attacks (Ministerio de Trabajo y Asuntos Sociales, 2006). These accidents, which occur most frequently in the construction and mining sectors, have been studied by several authors (Gillen et al., 2002, Janicak, 1998, Kines, 2003 and Hale et al., 2007). A secondary aim of our analysis was to establish the need for reinforcing safety measures for this type of work. As an information source for the analysis, we implemented a survey of workers who were interviewed as they performed a range of work tasks at more than 2 m above floor level (on ladders, structures, scaffolding or platforms). The use of information obtained during or immediately after the risk activity in prompt/no-delay interviews have enabled some authors to draw interesting conclusions on the causes of falls from a height (Kines, 2003). In our opinion, the use of information collected in the course of a task allows the circumstances of immediate relevance to an accident to be better analysed. In general terms (Bird and Germain, 1990), the causes of accidents at work can be classified on immediate causes (both substandard practices and substandard conditions) and basic causes (both personal and job factors). In the case of worker accidents caused by falls, the adoption of worker unsafe behaviours contributes directly or indirectly to around 90% of incidents (Holnagel, 1993). This unsafe behaviour can be directly related to substandard practices (using protective equipment incorrectly or removing safety devices), but also to substandard conditions (the existence of inadequate protection or incorrect task location). Worker inexperience, lack of motivation and fatigue are basic causes which often underlie immediate causes. In the present work, the questions posed are specifically related to worker behaviour (how the task is performed, the reasons for the application of alternative safety measures…) but also to substandard conditions (difficulties in applying legislations) which can condition the worker decisions. Furthermore, issues in regard to immediate behaviour and decision-making – generally not recorded in accident reports – constitute non-measurable and non-quantifiable variables that are better modelled as categorical variables (or at least as ordinal variables). Although categorical variables are difficult to incorporate in conventional statistical techniques, they can be easily be analysed using BNs.
نتیجه گیری انگلیسی
In this research we used Bayesian networks to analyse the factors affecting the performance of tasks that involve a high risk of falls from ladders or from other auxiliary equipment. This enabled us to identify the circumstances that have the greatest bearing on workplace accidents during both activities. Task duration, as would be expected, seems to have a bearing on worker behaviour. In the case of ladder-based tasks, workers tend not to wear a harness if the task is of short-duration. In the case of scaffolding and similar equipment, haste is often the reason for adopting incorrect postures – a variable which, in this context, seems to be associated with a greater accident rate. In both situations, there appears to be a lower perception of potential danger in jobs of a short-duration. The level of hazard perception seems to be connected with the training workers receive from their companies and with worker experience in the job. The higher the level of training and experience, the higher the level of hazard perception. It would seem, therefore, that experience has a bearing on accident rates in these tasks, although the relationship appears to be an indirect one (through its relationship with the level of training). This would indicate that – as found in other studies (Paul and Maiti, 2007) – although experience can help avoid risks, it is the immediate cause related to behaviour that ultimately causes an accident. In the case of work performed on scaffolding and similar equipment, minimal or no training seems to lead to a higher frequency of incorrect postures – which are directly linked to higher accident rates. This fact is reflected in the majority of the companies within the group analysed; new or recently employed workers received inadequate or no training, despite the fact that Spanish legislation (Ley 31/1995) both requires workers to be trained before taking up their posts and indicates that this training must be theoretical and practical, as well as sufficient and suitable for the post. In the groups analysed, immediate causes (following Bird and Germain, 1990) such as unsafe behaviour (like incorrect postures) and basic causes (e.g. lack of knowledge of regulations), respectively, have the greatest bearing on accident rates in the case of work carried out on scaffolding and similar equipment and in the case of work performed using ladders. In the case of ladders, corrective measures should be aimed at providing better training for workers; in the case of work carried out on scaffolding and similar equipment, unsafe behaviour could be avoided through task-specific training. The importance of training is undeniable, furthermore, in a sector in which employment is temporary and short-term and the physical location varies. Both the Bayesian networks presented in this work have shown that the level of training has an important bearing on accident rates. From the results of other works centred in the analysis of the causes of accidents by falls from ladders and scaffolds, it can be also deduced the importance of training in minimising the risks: in Papazoglou et al., 2006 and Papazoglou and Ale, in press, the variables introduced in the functional block diagrams are related to job factors and substandard conditions (like the type of ladder and its placement), contrary to the variables used in the present work, which are mainly related to the user behaviour. Nevertheless, is undoubted that the correct choice of the material needed for developing the work and also its correct use are conditions which are only achieved following a task-specific training. Furthermore, Hale et al. (2007) obtain a high percentage of falls from ladders related to substandard practices (i.e. the use of a ladder in situations in which this is not the appropriate equipment, the incorrect placement and use of the ladder, etc.) and these practices can also be corrected through training. Training is the most effective way to minimise the risks that arise from incorrect posture and other unsafe behaviours, but it is only effective if long-term and if given in circumstances of perseverance and control – above all in sectors such as construction and mining, which have high levels of temporary employment. Independently of whether long-term training is provided, our results lead us to believe that a short-term solution that could reduce accident rates would be to lower the working height at which it is mandatory to adopt protective measures (2 m in Spain). Other authors have found that accidents happen at lower heights and in situations perceived as non-hazardous (Kines, 2003). Although adopting measures to correct inappropriate behaviour is crucial to attaining greater integration of risk prevention in the management of these kinds of tasks, the adoption of collective protection measures would represent less effort and would ultimately be more effective. The results of this research indicate that Bayesian networks are very useful in explaining the causes of falls. By identifying the dependency relationships between different variables (and expressing these relationships in probabilistic terms), Bayesian networks offer a broad-based perspective on the circumstances surrounding work performed at a height that will enable us to define a preventative strategy that reflects a particular reality. Bayesian networks represent a statistical tool of huge potential in investigating the causes of accidents in the workplace. As an expert system, moreover, Bayesian networks allow us to build a knowledge base that progressively and incrementally grows with the inclusion of new data. In this research we have chosen to structure the networks a priori, but in earlier research – in this field of knowledge (Matías et al., 2008) as well as in others (Matías et al., 2006 and Rivas et al., in press) – we established that fully or partially automated structuring (using algorithms that deduce the structure directly from the data) can throw up causal relationships that escape the attention of experts. In the case of data on the circumstances surrounding incidents in the workplace, automated models would reflect the effectiveness of a company’s safety management and internal risk prevention procedures (this will be the subject of future research). However, using BNs has its limitations, the most obvious being the use of discrete variables. If a large number of categories is selected for each variable – which is desirable so as not to lose information – then a large quantity of data that is representative of all possible combinations is required. As far as discretization is concerned, therefore, a compromise solution is required that does not endanger computational capacity. As regards the sources of information used, surveys of workers have proved to be very useful in identifying causes other than those which can be quantitatively measured and evaluated. The inclusion of variables with high levels of subjectivity in problems like that analysed in this article – in which the protagonists are human beings whose behaviour is the cause of workplace accidents – offers a more realistic vision of a problem. In research into accidents or reports on workplace accident rates, the combined use of this source of information and Bayesian networks enables us to establish hypotheses more in keeping with reality compared to conventional techniques.