Rolling bearing fault diagnosis method fusing attention mechanism and twin network structure

Abstract

The invention discloses a rolling bearing fault diagnosis method fusing an attention mechanism and a twin network structure. The method comprises the following steps: 1) obtaining an original vibration signal and carrying out standardization processing on data to construct a training and test sample; 2) constructing a fusion attention module and twin network fault diagnosis model, taking a twin network as a framework, taking a feature extraction part as a composite dynamic convolutional network fused with a long-short term memory structure, and performing training by using a training sample; and 3) inputting to-be-diagnosed test data into the trained fault diagnosis model to obtain a fault type result. According to the method, a composite dynamic convolutional network fused with a long-short-term memory structure is arranged in a twin network framework, the dynamic convolutional network preliminarily extracts related features of a time-frequency diagram, and the long-short-term memory network further extracts bearing fault degree features in a complex scene; therefore, a rolling bearing fault diagnosis function under a limited data set condition can be realized.

Description

technical field [0001] The invention belongs to the technical field of mechanical state monitoring and fault diagnosis, and in particular relates to a rolling bearing fault diagnosis method integrating an attention mechanism and a twin network structure. Background technique [0002] In the field of artificial intelligence, the attention module has become an important part of the neural network structure. It can leverage the mechanisms of human vision for intuitive interpretation, incorporating these notions of relevance to improve model performance by allowing the model to dynamically focus on certain parts of the input that help perform the task at hand. The attention module is introduced into the traditional convolutional network by introducing the dynamic convolutional network. The attention dynamically adjusts the weight of each convolution kernel according to the input, thereby dynamically aggregating multiple parallel convolution kernels. Stacked convolution kernels ...

AI Technical Summary

Problems solved by technology

There are two major problems in the existing deep learning fault classification methods: a large amount of fault sample data is required, and the data must meet the requirements of independent and identical distribution, but the acquisition of fault samples usually requires high experimental costs. In terms of equipment, it is obviously difficult to obtain a large number of samples that meet the requirements

Images