Fault Diagnosis for Pantograph Based on Time-frequency Decomposition and Approximate Entropy

Railway transport is one of the most critical mass transportation media in the worldwide. With the development of trains speed, safety and comfort levels of railways are gradually and progressively becoming much more significant. Besides the high level of the security requirement, detection of anomaly for rail and road shall be early identified for decreasing operation and maintenance expenditures. The pantograph-catenary system has vital role in collecting the current in electrical railways. The problem occurred in this system will affect the current collection performance of electrified trains. In this paper, a fault feature extraction model of pantograph vibration signal based on time-frequency decomposition and approximate entropy (ApEn) is constructed. Firstly, ensemble empirical mode decomposition (EEMD) is conducted for vibration signal, then calculate ApEn after optimizing parameters for the intrinsic modal function obtained by EEMD. The ApEn features are input to the support vector machine (SVM) based on particle swarm optimization (PSO) to identify the pantograph fault identification. And the results show that the EEMD-ApEn fault diagnosis method has an excellent accuracy result for the vibration data of the pantograph panhead top pipe, yet is unsatisfactory for the vibration data of the carbon contact strip. According to this, the second-generation wavelet decomposition (SGWD)-ApEn method is used to improve the fault diagnosis results for the vibration data of the carbon contact strip. It verified the effectiveness of the combination with time-frequency decomposition and information entropy for pantograph fault diagnosis.