Novel Preprocessing and Data Augmentation Method for Bearing Fault Diagnosis

To avoid financial and life risks, early diagnose of bearing faults is particularly important, and this has made intelligent fault diagnosis popular. Intelligent fault diagnosis faces challenges such as the need for sufficient data and suitable performance for noisy data. This paper presents a preprocessing and data augmentation method to improve the performance of a few-shot learning method based on these two challenges. In this method first vibration signal is converted to two-dimensional images by means of mel spectrogram, then two image processing methods, namely, histogram equalization and gamma correction are applied. Finally, a data augmentation method based on gamma correction method is used. By using the Case Western Reserve University dataset, it is shown that the proposed method can improve the fault diagnosis process.