Intelligent Machine Fault Diagnosis Using Accurate CNN and Transfer Learning

Abstract

The efficient fault diagnostic techniques are needed to assure stability and dependability of mechanically operated equipment in the evolution integrated large scale industrial applications. The Deep Learning (DL) based approaches have a broader range of potential applications because of their end-end encrypted qualities, which are in contrast to the time commitment and poorly maintained performance of standard Machine Learning (ML) based approaches. Nevertheless, the DL methods has some issues including more number of activation functions, challenging control parameter tuning and restrict the system performance etc. Therefore, this paper suggests using accurate Convolutional Neural Network (CNN) and Transfer Learning (TL) to determine problems in intelligent machines. With help of TL algorithm, the high accuracy is attained. Furthermore, Continuous Wavelet Transformation (CWT) is used in data processing to transform vibration signals into 2-D images, and CNNs is used in place of fully connected layers to improve classification. The obtained results of the confusion matrices and convergence curve is evaluated in Python Jupiter platform. These findings shows that the proposed technique is accomplished the highest accuracy under a wide range of conditions.