Deep Learning Based Fault Diagnosis of Rolling Bearing Using Spectrogram and CWT Images of Vibration Data

Rolling bearing plays an important role in fault detection of rotating machinery in the era of industrial 4.0. Very recently, machine learning based methods are gaining priority in automatic fault feature extraction and fault classification. Even though these are the promising field the presence of noise plays a vital role in fault diagnosis. To overcome this difficulty, this work is focused on a method based on Variational mode decomposition (VMD), Wavelet Threshold (WT) denoising and CNN based fault classification. Here VMD is used for decomposition, WT is used for denoising of each mode, and a well-known CNN named as AlexNet is used for Classification of Stacked STFT(S-STFT) and stacked CWT(S-CWT) images. Results show that S-CWT provides an accuracy value of 99.3% compared with S-STFT of accuracy 99.1%. Performance parameters like precision and recall also implicate the validation of the results.