Automatic roller bearings fault diagnosis using DSAE in deep learning and CFS algorithm

Abstract

A method based on denoising stacked auto-encoder in deep learning and clustering fast searching for roller bearings fault diagnosis automatically is presented in this paper. Unlike traditional classification methods, such as support vector machine, clustering methods can identify the faults without data label. However, most popular clustering methods like fuzzy-c-mean, Gustafson–Kessel, and Gath–Geva methods are needed to preset the number of the cluster. Different from these clustering methods, clustering fast searching model can select the cluster center points according to the local density and distance from any two points automatically. This paper presents a method based on denoising stacked auto-encoder in deep learning for feature extraction and clustering fast searching algorithm for fault diagnosis automatically without principal components analysis. Firstly, the denoising stacked auto-encoder is deployed to extract the useful fault feature from the different roller bearings vibration signals. Secondly, in order to visualize the data, the denoising stacked auto-encoder model with several hidden layers is used to reduce the dimension of the extracted features, then the extracted features are regarded as the input of the clustering fast searching model for fulfilling the roller bearings fault diagnosis. The experimental results show that the performance of the presented method is superior to the other different combination models include sparse auto-encoder, ensemble empirical mode decomposition, fuzzy entropy, fuzzy-c-means, Gustafson–Kessel, and Gath–Geva