ارزیابی ریسک بر اساس شبکه بیزی ایمنی برای پروژه های سازه های فولادی

There are four primary accident types at steel building construction (SC) projects: falls (tumbles), object falls, object collapse, and electrocution. Several systematic safety risk assessment approaches, such as fault tree analysis (FTA) and failure mode and effect criticality analysis (FMECA), have been used to evaluate safety risks at SC projects. However, these traditional methods ineffectively address dependencies among safety factors at various levels that fail to provide early warnings to prevent occupational accidents. To overcome the limitations of traditional approaches, this study addresses the development of a safety risk-assessment model for SC projects by establishing the Bayesian networks (BN) based on fault tree (FT) transformation. The BN-based safety risk-assessment model was validated against the safety inspection records of six SC building projects and nine projects in which site accidents occurred. The ranks of posterior probabilities from the BN model were highly consistent with the accidents that occurred at each project site. The model accurately provides site safety-management abilities by calculating the probabilities of safety risks and further analyzing the causes of accidents based on their relationships in BNs. In practice, based on the analysis of accident risks and significant safety factors, proper preventive safety management strategies can be established to reduce the occurrence of accidents on SC sites.

In Taiwan, steel structures have been the most common structure type for high-rise buildings. However, site accidents constantly occur because of work at heights in SC projects. The percentage of falls at SC projects in Taiwan has increased to 66% over the past decade (2000–2010). In addition to falls, object falls, object collapse, and electrocution comprise a high percentage of occupational accidents at SC sites. Fig. 1 shows a comprehensive occupational accident lists for steel construction projects in Taiwan between 2000 and 2010. Construction contractors have attempted to implement various safety measures to prevent occupational accidents, including safety training, site environment management, safety and health management, and appropriate health and safety plans. In addition, some systematic safety risk-assessment approaches such as fault tree analysis (FTA), failure mode and effect criticality analysis (FMECA), and decision trees are used to evaluate safety risks (Hartford and Baecher, 2004 and Kales, 2006). However, these methods ineffectively address dependencies among safety factors at various levels that fail to provide an early warning to prevent occupational accidents. To overcome the limitations of traditional safety risk-assessment approaches, several effective approaches have been developed to define the interplay between safety variables so that preventive safety measures can be proposed. Structural equation models (SEM) and Bayesian networks (BNs) are typical examples of these approaches (Kao et al., 2009, Martin et al., 2008 and Paul and Maiti, 2007). The safety of third parties during construction in multiple spaces has been assessed using BNs (Bedford and Gelder, 2003). BNs have been used to analyze workplace accidents caused by falls from heights (Martin et al., 2008). BNs, in addition to their good predictive capacity, possess satisfactory interpretative ability regarding workplace accidents (Matias et al., 2007). The critical causes of site accidents can be identified, and the relationships among these causes can be determined using BNs. Consequently, early and preventive safety measures can be defined through BN inference. Full-size image (19 K) Fig. 1. Occupational accidents of steel construction projects in Taiwan (2000–2010). Figure options Because of the constraint of data availability, expert knowledge is typically used to develop practical BNs that describe problems with causal relationships among nodes and their conditional probabilities. However, the method to develop a BN directly is more suitable for simple problems. It is difficult to directly develop complex BNs. Several systematic approaches to BN construction using FT transformation have been proposed (Franke et al., 2009, Marsh and Bearfield, 2007 and Xiao et al., 2008). The primary techniques of these approaches use “OR Gate” and “AND Gate” to change to a BN to perform probabilistic analyses of events. Most studies have regarded both events and logic gates in FT as nodes in BN. However, these have different definitions and purposes. Logic gates are primarily used to describe the relationship between events in a sequence. A BN node is used to represent a random variable in the problem domain. It is meaningless to convert logic gates to physical BN nodes. Therefore, this study combines FTA and BN to develop a more reasonable transformation process from FT to BN. A sub-BN, a fall risk-assessment model for SC building projects, was first validated against the safety inspection records of six SC building projects. The complete BN-based safety risk-assessment model was further validated against nine projects in which specific site accidents occurred. This shows that the ranks of posterior probabilities from the BN model are highly consistent with the accidents at each project site.

This study developed an effective method to construct a BN-based safety risk-assessment model for SC building project sites. The inference results of the BN were validated against nine SC building projects in which specific accidents occurred at each SC site. The analysis and comparison showed that the results of the BN inference were highly consistent with actual accidents at the sites. This indicates that the transformation process from an FT to a BN can create a realistic and accurate safety risk-assessment model. Consequently, based on the model assessment and sensitivity analysis, site project managers can prepare preventive safety measures and allocate resources to significantly reduce the risks of site accidents in SC projects. Although the transformation mechanism from an FT to BN has been thoroughly examined, the use of a BN relies on the inputs of expert experiences for the BN structures and CPTs in the BN. The information provided by experts directly affects the accuracy and assessment quality of the BN. More studies should explore expert elicitation. In addition, the BN can be learned from raw data combined with expert experiences. If reliable safety data are available, a sound BN framework and parameters can be explored and established. Finally, the proposed BN-based safety assessment model demonstrated its contribution to site accident prediction and assessment. It may be useful to extend the application scope of the BN to cover additional construction projects and to use the BN for overall safety diagnoses to enhance safety operations and management.