BAYESIAN NETWORK AND RAILWAY TRACK DETERIORATION MODELING

Abstract

The safety and reliability of the railroad infrastructure are a priority for the railroad industry.The Federal Railroad Administration requires that inspections be carried out at least twice a week. This is particularly a challenge for the local railroad that relies on traditional inspection methods due to the high capital cost of the track inspection machine. The large amount of data generated by this machine is also a challenge for the local railroads to process since they lack the required expertise. This research proposes the use of a Bayesian Network to model the relationship between the components of the superstructure, substructure, and the geometry of the railroad. The proposed approach incorporates expert knowledge and a historical railway inspection dataset to construct a Bayesian Network model that captures the causal relationships among failure elements. A junction tree was constructed from the Bayesian Network model for exact inference. The study further evaluates the impact of individual component deterioration on the overall system performance. This approach is valuable to all classes of railroads, as it helps transform the large amount of railway data into a scalable, interpretable format for informed decision-making and risk minimization. Results from our model demonstrate the ability of the Bayesian Network to serve as a sensitivity analysis tool and as a mechanism to plan for scheduled maintenance operations on the railroad.