Damage to the pantograph or sliding strip may cause the blocking of the railway line. This is the main reason for which the prediction of pantographs' failure is important for railway carriers and researchers. This article presents a sliding strips failure prediction method as a main means of preventing disruptions to the transport chain. To develop the best predictive model based on the decision tree, the complex tree, medium tree and simple tree machine learning methods were tested. Using a decision tree, the categorization of the given technical conditions can be properly realized. The obtained results showed that the presented model can reduce sliding strip failure by up to 50%. Special attention was paid to the current collector (AKP-4E, 5ZL type), measured during periodic reviews of locomotives EU07 and EU09. To assess the reliability of the selected pantograph strips, a non-destructive degradation analysis was carried out. On the basis of the wear measurements of the strips and the critical value of wear, a failure distribution model was developed. Operational data, collected during periodic technical reviews, were provided by one of the biggest railway carriers in Poland. The results of the performed analyses may be used to build a preventive maintenance strategy to protect pantographs. The applied reliability models of wear propagation can be extended by the parameters of the cost and repair time becoming the basis for estimating the costs of operation and maintenance.
Keywords: AI methods; artificial neural network; damage prevention; failure distribution model; machine learning; pantograph strip; reliability assessment.