Evaluation of Deep Learning Neural Networks with Input Processing for Bearing Fault Diagonosis

Deep learning networks have been widely used as effective methods for bearing fault diagnosis. Deep learning neural networks such as convolutional neural network (CNN) use the images as inputs while the others such as long-short term memory (LSTM) may apply data sequences as inputs. This paper focuses on performance evaluations of deep learning networks by utilizing various signal transforms to form the network inputs. The CNN and LSTM are adopted as our deep learning network structures. Besides raw data, the algorithms for processing input signals include short-time Fourier transform (STFT), Cepstrum, wavelet packet transform (WPT), and empirical mode decomposition (EMD). Our simulations validate the effectiveness for each network input formulation based on the Case Western Reserve University's (CWRU) bearing dataset.