A Bayesian network model for prediction of weather-related failures in railway turnout systems. (1st March 2017)

• Record Type: Journal Article
• Authors: Wang, Guang; Xu, Tianhua; Tang, Tao; Yuan, Tangming; Wang, Haifeng
• Is Part Of: Expert systems with applications. Volume 69(2017)
• Page Start: 247
• Page End: 256
• Publication Date: 2017-03-01
• Subjects: Railway turnouts -- Failure prediction -- Bayesian network -- Causal noisy MAX -- Weather-related failure; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2016.10.011
• Languages: English
• ISSNs: 0957-4174

Abstract

Highlights: A failure prediction model based on Bayesian network is proposed. An Entropy Minimization-Based method is presented to discretize model variables. Learning parameters from small data sets by using a causal noisy MAX model. The causal relationship between the weather and failures of turnouts is captured. Experiments indicate that the proposed method outperforms other algorithms. Abstract: Railway turnout systems are one of the most critical elements in railway infrastructure. They are also one of the most vulnerable assets that are likely to be affected by the adverse weather. Therefore, effective methods for modeling weather-related failure of turnouts enable railway administrations to make optimal maintenance decisions. This paper presents a failure prediction model based on Bayesian network to evaluate the effect of weather on railway turnouts. Different failure causes related to weather are extracted as model variables. An Entropy Minimization based method is presented to discretize model variables for the purpose of reducing the input type and capturing better performance. By taking advantage of the independence of causal interactions, a causal noisy MAX model is applied to boost the efficiency of specifying the conditional probability table from small data sets. Prediction accuracy of the proposed method is compared to other advanced methods in order to evaluate the model's performance. Our experiments by using the data from a railway corporation demonstrate that the proposed method has high prediction accuracy.