Bearing fault diagnosis method based on hierarchical extreme learning machine

Abstract

The invention provides a bearing fault diagnosis method based on a hierarchical extreme learning machine, and belongs to the technical field of mechanical part fault diagnosis, and the method comprises the steps: decomposing a vibration acceleration signal into a plurality of modal components through a VMD algorithm; selecting the first four modals sorted according to the size of the center frequency, carrying out feature extraction through an SVD algorithm, mapping input feature data into a random sparse hidden layer space, obtaining hidden information between training samples, and carrying out random mapping again on the feature data processed by the previous layer by each hidden layer through a sparse automatic encoder; and obtaining an optimal neural network weight through a fast iterative shrinkage algorithm (FISTA), so that the actual output is close to the specified label data. According to the method, noise reduction and accurate classification are realized at the same time, and the recognition accuracy and the feature information utilization rate can be improved under the condition of a hierarchical extreme learning machine; compared with an original extreme learning machine, higher recognition precision and higher training speed can be achieved in fault diagnosis of rolling bearing signals.

Description

technical field [0001] The invention relates to the technical field of fault diagnosis of mechanical parts, in particular to a fault diagnosis method based on a layered extreme learning machine. Background technique [0002] Bearing fault diagnosis is an effective technology to ensure the stable operation of rotating machinery. Therefore, rapid identification of fault states plays an important role in social production and life. However, in the existing bearing fault diagnosis algorithms, accuracy and speed have always been two research issues that are difficult to balance. How to find an algorithm that can ensure both recognition accuracy and recognition speed is of great significance in real life. However, traditional methods such as BP neural network and legacy algorithm only guarantee the accuracy in actual use, but sacrifice a lot of computing time. The extreme learning machine (ELM) is an algorithm that has an extremely obvious advantage in speed compared with other ...

AI Technical Summary

Problems solved by technology

However, it also has disadvantages such as unstable learning effects. For its improved composite extreme learning (ML-ELM), deep extreme learning machine (AE_ELM), etc., each sacrificed in accuracy and speed.

Images