Assessment of catenary condition monitoring by means of pantograph head acceleration and Artificial Neural Networks

Abstract

Measuring the pantograph–catenary interaction force is crucial for homologation and maintenance purposes. However, this is an intricate procedure that requires a specially-instrumented pantograph. As an alternative to the traditional method of measuring the interaction force, we propose in this paper a new procedure to predict the current collection quality (usually quantified by the standard deviation of the interaction force) from the vertical acceleration measurements of the pantograph collector head.The proposed approach consists of properly preprocessing the acceleration signals to train and validate Artificial Neural Networks (ANNs) to predict the standard deviation of the interaction force. First, the proposed method is defined with academic examples in which different ANNs identify dropper defects and severe contact wire wear from numerically generated pantograph accelerations. The normalization of the input data and proper resampling to make samples independent of the train speed are revealed as key aspects in the success of the proposed approach.Also from a numerical point of view, the method is then applied to predict the standard deviation of the contact force. The results obtained show that these predictions have an error of less than a 10%. The analyses of other potential scenarios such as different train operating speed ranges and the use of displacements instead of acceleration as input, show a poor generalization power of the ANNs used. They provide accurate results only when fed with input samples with the same features as those used in the training stage.Finally, the proposed approach is applied to the experimental data obtained from a pantograph lab test bench. In this case, the ANNs are able to predict the current collection quality with less than a 10% error in almost 95% of the cases. This work aimed to be an intermediate step between pure numerical analysis and real data measured on track.