Vibration Signal Prediction of Gearbox in High-Speed Train Based on Monitoring Data

Abstract

Vibration signals contain abundant information which can reflect the running state of high-speed trains. Accurate vibration signal prediction can provide references for anomaly detection of the gearbox in high-speed trains. This paper develops a hybrid model combining ensemble empirical mode decomposition (EEMD) with auto regression (AR) and support vector regression (SVR) models. First, the EEMD method is applied to decompose the vibration acceleration signal of gearbox. Second, AR models are employed to predict the intrinsic mode functions and the outputs are aggregated as the final result of AR. Third, reconstruct phase space and establish SVR models to predict the components; The predictions are aggregated as the final result of SVR. Finally, the results predicted using the AR and SVR models are weighted and summed together, with the weights being optimized by the chaotic particle swarm optimization algorithm. The actual operation monitoring data are used to validate the hybrid model. Data analysis demonstrates that the proposed method has better approximation compared with the AR model, the SVR model and the RBF neural network model.