Comparative analysis of optimization algorithm on DSAE model for bearing fault diagnosis

A rolling-element bearing is one of the most vital components in machinery and maintaining the bearing health condition is very important. Intelligent fault detection and diagnosis based on deep sparse autoencoder (DSAE) is presented to improve the current maintenance strategy. The conventional maintenance strategy suffers from manual feature extraction and feature selection. In this project, the DSAE model made up of multiple layers of neural networks that can perform automated feature extraction and feature dimensional reduction is proposed. The DSAE model is used to extract the important features from the Fast Fourier Transform (FFT) images by learning the high-level feature from the unlabeled images. However, the DSAE model requires hyperparameter selection of which manual hand-tuning is time-intensive. The DSAE model contains four hidden layers and requires 12 hyperparameters selection. The hyperparameter is automatically selected using an optimization algorithm. The comparative study is conducted on three optimization algorithms, namely particle swarm optimization (PSO), grey wolf optimizer (GWO) and genetic algorithm (GA). The overall analysis result shows that the proposed model achieved 100% diagnosis accuracy. Furthermore, the proposed model is tested with a completely new dataset and the result indicated that the DSAE model achieved 93.5% accuracy for the new dataset. The grey-wolf optimizer optimized quicker compared to PSO and GA.