An Analysis of Construction Accident Factors Based on Bayesian Network

In this study, we have an analysis of construction accident factors based on bayesian network. Firstly, accidents cases are analyzed to build Fault Tree method, which is available to find all the factors causing the accidents, then qualitatively and quantitatively analyzes the factors with Bayesian network method, finally determines the safety management program to guide the safety operations. The results of this study show that bad condition of geological environment has the largest posterior probability; therefore, it is the sensitive factor that might cause the objects striking accidents, so we should pay more attention to the geological environment when preventing accidents.


INTRODUCTION
Nowadays, the control of the project construction accidents has become more stringent and how to reduce the project construction accident has become more and more urgent, so we need a viable solution for a detailed analysis of the failures to develop appropriate measures to reduce the occurrence of similar accidents.Xie et al. (2004) study the Bayesian networks to improve the fault tree method.Liu and Qin (2004) analyze the network safety assessment based on Bayesian networks.Zhang et al. (2005) have a research of the quantitative analysis of fault tree based on bayesian network.Li (2006) study the quantitative risk assessment of long-distance pipeline based on fault tree analysis.Lou (2004) has a research of the Bayesian network in mechanical fault diagnosis.Liu and Zeng (2007) study the applications of Bayesian networks in coal mine production safety evaluation system.Zhou (2006) study the probabilistic safety assessment and application based on bayesian networks.
This study focuses on statistical analysis of the accident and the development of safety management solutions.Use Bayesian network to do the statistical analysis of the occurred objects striking accidents in Xiluodu project from 2004 to 2006.

ANALYSIS OF THE FACTORS CAUSING ACCIDENTS
From 2004 to 2006 there had been 50 cases of construction accidents in Xiluodu project, in which included 22 death cases causing 27 deaths and 20 serious injury cases causing 29 seriously injured persons.In this 50 cases of accidents, the objects striking accidents accounted for 17 cases, in which there are 13 cases of serious injury and 4 cases of death.It can be seen the objects striking accidents is a major part of the accidents in Xiluodu project.Due to limited space, I only analyzed the objects striking accidents in this study.
According to the analysis of the cause of the objects striking accidents carefully, we can know all of the reasons can be divided into two cases: direct causes and indirect causes.Fig. 1 The Bayesian network of objects striking accidents

BAYESIAN NETWORK ANALYSIS
Bayesian network, also known as Bayesian reliability networks, which is combination of graph theory and probability theory (Liu and Qin, 2004).It can be intuitively expressed as an assignment causal relationship graph and can get joint probability distribution which contains all nodes according to the prior probability distribution of the root node and the conditional probability distribution of the non-root nodes (Zhou, 2006).Exactly this study did qualitative and quantitative research on the factors in the accidents by this method.
In order to build a Bayesian network of the objects striking accidents in Xiluodu project, we set the safety events in Table 2 according to Table 1 (Liu and Zeng 2007).
According to the relevant principles of Bayesian networks and Table 2 and 1, we can get the Bayesian networks of the objects striking accidents shown in Fig. 1 (Xie et al., 2004).
Counting all of the objects striking accidents in Xiluodu project, we can get Table 3 and Table 4, in which the table column refers to the every objects striking accident, two types of importance, 1 and 0.8, which is used to distinguish between fatalities and serious injury accidents, 1 is expressed as deaths and 0.8 is for the serious injuries, to emphasize the seriousness of the accidents.The probability of each basic cause represents its contribution to the objects striking accidents on the basis of that the objects striking accident has happened, For example, in the 1.05 accident, The probability of X3 is relatively large, indicating X3 plays a leading role in the occurrence of the accident, at the same time we also consider that an accident cause serious injury or death, accumulating the degree of these impacts we can get a Initial priori probability of each basic cause.
As we know, posterior probability can reflect the basic cause influence on the top event, As it can be seen from Table 6 , the Posterior probability of basic cause X3 is 0.00156, which is the largest of all the posterior probability of basic causes, Therefore, X3 is the most sensitive factor in the objects striking accidents, so we can analyze the possible factors of un-safety according to prospecting the geological details of construction sites and then develop appropriate programs and operating procedures of construction, so that we can reduce the probability of occurrence of X3 and then reduce the probability of the objects striking accidents.

CONCLUSION
For the Bayesian networks, the current use of accident analysis is a static Bayesian networks, can not be reasoning over time.With the development of Dynamic Bayesian Network, the dynamic Bayesian model and its prediction algorithm has more advantages when it is used to analyze the unexpected incidents and predict consequences.At the same time, considering adding information based on expert experience and the state transfer function of subjective judgments to calculate the probability distribution of each variable of the next time slice will be focus in future research.
This analysis method was validated reasonable through a variety of examples.However, in practical engineering applications, there may be a lot of problems, so must be handled accordingly with the actual situation.Since there may be a deviation between Bayesian network model built and the actual system, therefore need to do some necessary improvements to the model to ensure the consistency of models and systems.The structure and parameter learning function of Bayesian network just provides a good idea for this problem, there are many problems to be solved in this direction.Apply the model to test its feasibility in practical engineering and then revise, develop and perfect constantly.Use the Probabilistic Safety Assessment based on Bayesian network to guide safe design, safe growth, safe diagnosis and other engineering work.At present, China has not yet matured software for building the Bayesian Network Model, In addition existing tools can not be extended to develop graphical modeling software of Bayesian network analysis.

Table 3 :
The statistical probability of objects striking accidents' basic causes

Table 4 :
The probability of objects striking accidents' basic causes

Table 6 :
The prior probability and posterior probability of basic cause