Research on fault diagnosis method of aviation bearing based on improved DRSN

Abstract

Aiming at the traditional bearing diagnostic methods with complex arithmetic and low accuracy. In this paper, an improved deep residual shrinkage network model is designed by integrating the advantages of long short-term memory network (LSTM) and deep residual shrinkage network (DRSN). Firstly, the original one-dimensional vibration signal is imported into the LSTM module to fully extract the timing features, and then the extracted feature information is convolved and imported into the residual shrinkage network module for deep feature mining, and finally the classification of faults is accomplished based on the fully connected layer. The model is validated on the aviation bearing dataset, and the experimental results show that compared with the traditional DRSN network model, the improved model proposed in this paper not only saves 93.7% of the running time, but also achieves 97.4% of the fault diagnosis accuracy; at the same time, in the presence of noise interference, the model proposed in this paper still has a higher accuracy compared with other methods. Therefore, the model proposed in this paper not only saves a lot of time, but also has better robustness and accuracy.

Unsupervised Image Classifier based on Manifold Learning

“…In recent years, manifold learning has received much attention and has been applied to many different fields, such as data visualization, information retrieval, pattern classification, fault diagnosis, etc. [15] The most widely used of mainfold learning, on the other hand, are isomap algorithm and LLE algorithm.…”