An ensemble fault diagnosis method for rotating machinery based on wavelet packet transform and convolutional neural networks

Abstract

Deep learning has made fascinating achievements in fault diagnosis of rotating machinery. However, the individual deep learning models generally have poor performances under variable operating conditions or noise environment. Additionally, they are prone to overfitting when dealing with unbalanced fault data. Therefore, an ensemble method based on wavelet packet transform (WPT) and convolutional neural networks (CNNs) is presented for rotating machinery fault diagnosis. First, the raw signals are transformed into multiple wavelet packet coefficients with local information and a reconstructed signal with global information through WPT. Then, these signals are separately fed into the corresponding CNN models for diagnosis. Finally, the diagnosis results of multiple CNNs are combined into a more stable and accurate diagnosis result through the improved weighted voting strategy. The proposed method is applied to the fault diagnosis of the motor bearing and the CNC machine tool spindle bearing. Compared with the other traditional diagnosis methods, the results show that the proposed method achieves a better diagnostic performance.