Condition Monitoring of Rail Vehicle Suspension Elements: A Machine Learning Approach

Abstract

With an increasing demand for safe and efficient rail transportations with high availability, there is an interest to apply condition based maintenance on railway systems to increase the total system reliability. A condition based maintenance system utilizes data collecting, data processing and decision making to schedule maintenance based on the actual condition of components. In this paper a rail vehicle is simulated at varying operational conditions, and with degraded dampers in the primary and secondary suspension. A large database of simulations is generated and is used to train and test classification algorithms to detect upcoming damper faults, introduced as a fault factor multiplied with the damper coefficients. Frequency response functions between accelerometer signals in the carbody, bogieframes and axles are used as fault indicators, predictors, fed to the classification algorithms. The algorithms are evaluated for a varying number of included frequency response functions, as well as varying operational conditions in the training datasets. The linear Support Vector Machine and 1-Nearest-Neighbour classifier both indicate high capability of correctly classifying damper degradations.