Fault diagnosis for train plug door using weighted fractional wavelet packet decomposition energy entropy

•Signal reconstruction method based on hybrid criteria for IMFs selection is proposed.•Novel efficient feature named WFWPDE is developed.•Synchronous optimization strategy is proposed to optimize weights and SVM. As the passage for passengers to get on and off, train plug doors directly affect the operation efficiency of the train and the personal safety of passengers. This paper proposes a non-contact fault diagnosis method for train plug doors based on sound signals. First, empirical mode decomposition (EMD) is utilized to process the raw sound signals. A signal reconstruction method by selecting intrinsic mode functions (IMFs) using hybrid selection criteria is then proposed. Second, novel feature named weighted fractional wavelet packet decomposition energy entropy (WFWPDE) is developed by introducing the idea of fractional calculus and weight to wavelet packet decomposition energy entropy (WDPE). Third, a synchronous optimization strategy is proposed to optimize the weights and hyperparameters of support vector machine (SVM) synchronously. Finally, the superiority and feasibility of the proposed method are verified on field-collected data. By comparing with different fault diagnosis methods, the proposed method performs best on fault diagnosis of train plug doors, with accuracy of 97.87%.