A Convolutional Neural Network for Rotating Machinery Fault Diagnosis Based on Fast Independent Component Analysis

Efficient and accurate fault diagnosis is essential to ensure the safe operation of rotating machinery. An intelligent fault diagnosis based convolutional neural networks (CNN) and fast independent component analysis (FICA), is proposed to improve the classification and recognition ability of rolling bearings and. Firstly, the intrinsic mode function (IMF) components of the raw vibration signals are obtained by empirical mode decomposition (EMD) preprocessing method. Secondly, FICA method is used to extract additional feature components of IMFs and ICA components are obtained. Finally, a shallow CNN model is constructed to learn feature and diagnosis from different signal-to-noise ratio (SNR) and different working load of rolling bearings. The proposed method can conduct high accuracy of fault recognition and classification, which is more efficient than the IMFs feature. To verify this method, Back propagation neural networks (BPNN), stacked autoencoder (SAE), multilayer perceptron (MLP) are used as comparative models. The results demonstrate that the proposed method can achieve higher accuracy than other comparative methods.