تجزیه و تحلیل های پشتیبانی تصمیم گیری برای کنترل ایمنی در محیط های پروژه های پیچیده بر اساس شبکه های بیزی

This paper presents a novel and systemic decision support model based on Bayesian Networks (BN) for safety control in dynamic complex project environments, which should go through the following three sections. At first, priori expert knowledge is integrated with training data in model design, aiming to improve the adaptability and practicability of model outcome. Then two indicators, Model Bias and Model Accuracy, are proposed to assess the effectiveness of BN in model validation, ensuring the model predictions are not significantly different from the actual observations. Finally we extend the safety control process to the entire life cycle of risk-prone events in model application, rather than restricted to pre-accident control, but during-construction continuous and post-accident control are included. Adapting its reasoning features, including forward reasoning, importance analysis and background reasoning, decision makers are provided with systematic and effective support for safety control in the overall work process. A frequent safety problem, ground settlement during Wuhan Changjiang Metro Shield Tunnel Construction (WCMSTC), is taken as a case study. Results demonstrate the feasibility of BN model, as well as its application potential. The proposed model can be used by practitioners in the industry as a decision support tool to increase the likelihood of a successful project in complex environments.

In the past 10 years, metro construction has presented the powerful momentum for rapid development all over the world. Owing to various risk factors of uncertainty in complex environments, safety violations occur frequently in metro construction. On January 12, 2007, Pinheiros Station on Metro Line Four collapsed at Sao Paulo’s Aquarium in Brazil, causing death of seven people (Ti, 2007). On July 6, 2010, a tunnel boundary accident also took place in Prague, Czech Republic, causing a 15-meter-wide sunken pit (Ti, 2010). On August 23, 2012, metro line leak caused chaos in Warsaw. Water flooded into the tunnel at the planned Powisle station. Fortunately nobody hurt, but traffic had been redirected, causing considerable transport problems in the already gridlocked city (Zagranicy, 2012). Also in China, the number of construction accidents shows a big rise during large-scale metro construction projects. On July 1, 2003, great quantities of sand swarmed into the tunnel in Shanghai Track Traffic Line Four. The tunnel was severely damaged and discarded as useless, causing a total direct economic loss exceeding 1000 million Yuan (Huang Tj, 2003). On November 15, 2008, 21 people were killed as a result of a road cave-in above a metro tunnel under construction in Hangzhou (Yb, 2008). As for Shenzhen Metro Construction, four accidents happened with five deaths in October 2009 (Song, 2009). Also, three people were killed and five were injured during collapse accidents from April to May in 2011. As you can see, metro construction is a complicated project with high risks, which would bring enormous hidden dangers for the public safety. To avoid heavy casualties and property losses caused by the safety violations, innumerable studies have introduced risk-based analysis into safety control. Risk analysis can be divided into qualitative and quantitative risk analysis. The former includes fault tree analysis (FTA), comprehensive fuzzy evaluation method and check list; while the latter includes job risk analysis method (LEC) method, influence diagrams (de Klerk, 2001), neural network (Carr & Tah, 2001), support vector machines (Wang, Yuan, Chen, Yang, Ni, & Li, 2012) and decision trees (Vens, Struyf, Schietgat, Džzeroski, & Blockeel, 2008). These risk-based analysis methods make a significant contribution to the safety control in complex engineering projects (del Caño & de la Cruz, Piniella et al., 2009 and Vinod et al., 2003), but they are confined to static control management (Alaeddini & Dogan, 2011). Khakzad described FTA unsuitable for complex problems for its limitation in explicitly representing the dependencies of events, updating probabilities, and coping with uncertainties (Khakzad, Khan, & Amyotte, 2011). Yang et al. regarded LEC unsuitable for complex dynamic environment, resulting from its insufficiency in timely diagnosing and dealing with various problems (Yang, Liu, Huang, & Wu, 2011). When associated parameters, such as geological, design and construction parameters are changed, these methods cannot accurately illustrate the updated features of dynamic environments as the construction progress evolves. Nor can the professional supports or suggestions be provided in real time as these parameters are updated. Safety control is a dynamic management process in complex engineering environments, where the relative parameters would truly change along with the change in space and time. With the capacity of integrating prior knowledge and sample data, Bayesian Networks (BN) provide a powerful tool for knowledge representation and reasoning under the dynamic environments. Compared to neural networks, support vector machines, decision trees, and so forth, BN have shown superior for several high-level classification tasks such as datamining fault monitoring and bioinformatics (Hsu, 2004 and Xu, 2012). At present, BN have been widely used in quality evaluation (Correa, Bielza, & Pamies Teixeira, 2009), dynamic management (Barrientos & Vargas, 1998), knowledge discovery (Lee & Abbott, 2003), as well as decision support (Lauria & Duchessi, 2006). As a consequence, BN is introduced as the resolving method for safety control in dynamic and complex engineering environments. In recent years, relatively few researchers have adopted BN method to risk management in large engineering projects (Ordonez, 2007), mainly focusing on the application. Lee, Park, and Shin (2009) presented a large engineering project risk management procedure using BN and then identified the major risk items that affected project performance. Bayraktar and Hastak (2009) applied BN in the decision support system for safety maintenance in developing suitable contracting strategies among different project components in highway work zone projects. Sousa and Einstein (2012) presented a construction strategy decision model based on BN to systematically access and manage the risks associated with tunnel construction. However, there are primarily two problems in the current study related to BN application: (1) Excessive attention is given to the establishment of BN model, but the validation of an established BN model is rarely completed. As a matter of fact, the effectiveness of BN model is considered as an essential guarantee of the plausibility in its application; (2) BN models are mostly applied before an accident happens, also referred to as pre-accident control (Ledolter and Swersey, 2005 and Li et al., 2009). Few studies have explored during-construction continuous control and post-accident control, which are two indispensable links for safety control in the overall process. Applying an accurate BN model that has already been validated to decision support analysis for safety control in complex project environments and provide professional technical support for decision makers in the overall process constitutes an entirely new domain. We first propose a decision support model for safety control based on BN, then make detailed expatiation of the design and validation process of the BN model. At last, adapting its reasoning features, including forward reasoning, importance analysis and background reasoning, decision makers could be provided with scientifically documented and effective support in regard to safety control. A typical metro construction hazard ground settlement such as the one in Wuhan Changjiang Metro Shield Tunnel Construction (WCMSTC) is taken up as a case study. Results demonstrate the feasibility of proposed method, as well as its application potential. The structure of this paper is organized as follows. The fundamental theory and analyzing method of BN are introduced in Section 2, including model design, validation and application. Section 3 presents a BN model for safety control in shield tunnel construction, namely Ground Settlement Bayesian Network (GSBN). In Section 4, the proposed BN model, GSBN, is validated by two indicators: Model Bias and Model Accuracy. In Section 5, GSBN is applied for decision support analysis on safety control in the overall work process. Afterward, the conclusions are drawn in Section 6.

Owing to various risk factors in dynamic complex project environments, accidents occur frequently in many complicated construction projects. Traditional fault analysis methods, such as FTA, LEC, influence diagrams and others cannot accurately illustrate the dynamic and updated features of geological, design and construction parameters as the construction progress evolves. With the capacity of integrating prior knowledge and sample data, BN techniques provide a powerful tool for knowledge representation and reasoning in dynamic and complex engineering environments. Results of real projects demonstrate the feasibility of BN method, as well as its application potential. The effectiveness of BN model is an essential guarantee of the plausibility in application. Previously limited validation of BN model has been undertaken. Two indicators, Model Bias and Model Accuracy, are proposed to evaluate the rationality and validity of the established BN model. We extend the safety control process to the entire life cycle of risk-prone events, rather than restricted to pre-accident control. But during-construction continuous control and post-accident control are included. Adapting its reasoning skills, including forward reasoning, importance analysis and background reasoning, decision makers are provided with systematic and effective support for safety control in the overall work process. The decision support model put forward in this research, including BN model design, validation and its application, can not only be applied in the safety control of metro construction, but also other complex engineering projects, like highway construction, nuclear power plant maintenance, aircraft manufacture and others. The model can be used by practitioners in the industry as a decision support tool to provide some positive guidelines on safety control in the overall process under dynamic complex environments, as well as increase the likelihood of a successful project in complex environments. Large quantities of monitoring records served as training data were obtained from the web-based systems developed for this research. Numerous engineering technicians participated in the monitoring work, making an essential contribution to securing a regular scheduled input the daily monitoring records into the system from the project sites. This process was laborious and susceptible to human error. Our subsequent research goal will focus on automatic data acquisition regarding different geological conditions, as well as adopting the Internet of Things (IoT) to develop a real-time intelligent monitoring system.