In Situ Condition Monitoring of a Rail-based System

Train derailment can lead to loss of capital, damage to the environment, and loss of life. Machinery condition monitoring techniques can be used to develop a structural health monitoring system for railway track to detect faults at the incipient stage, before catastrophic failure occurs. This study uses vibration measurements from a novel material haulage unit to develop a decision support system for railway track maintenance. A variety of feature sets and classifiers were compared to determine the optimal configuration for fault detection. The results highlight the shortcomings and advantages of certain methods and describe the process in which the optimal result was obtained. The Parzen data descriptor paired with a refined statistical feature set performed the task of fault detection with less than five percent false positive and false negative classification error. Overall, the outcome of these experiments indicated that condition monitoring of railway systems is a promising approach for fault detection and has the potential to avoid catastrophic failure.