Gearbox Fault Diagnosis Classification with Empirical Mode Decomposition Based on Improved Long Short-Term Memory

Abstract

As the main part of the Rotary Machinery System, the health of the gearbox's internal gears and bearings are essential when the machine is running. It is necessary to analyze the vibration data of the gearbox healthy state in time, find the fault, exact fault locations and types. Some researchers are only taking advantage of Long Short-Term Memory (LSTM) to deal with fault diagnosis gearbox dataset, which has a poor precision actually. In this paper, a deep neural network which combines Empirical Mode Decomposition (EMD) method, Long Short-Term Memory (LSTM) model and Particle Swarm Optimization (PSO) algorithm to achieve a high precision rate of machine fault diagnosis. Compared with existing methods, the proposed method is faster on training and more accurate. Firstly, original sensor data is pre-trained with EMD, then EMD's output as input of the LSTM to identify the gearbox fault types. At the same time, the PSO algorithm optimizes the LSTM's hyper-parameter automatically to avoid the problem that random initialization makes the network fall into local optimum. In the meanwhile, the Root Mean Square Error (RMSE) is established for better performance the EMD-PSOLSTM model. Therefore, the proposed method can learn a robust and discriminative representation from the raw gearbox dataset. Compared with other machine learning methods, such as Back Propagation (BP) and Support Vector Machine (SVM), the proposed method shows state-of-the-art results on gearbox dataset, and it is effective and efficient for gearbox fault diagnosis.