A bearing detection method based on convolutional neural network

Abstract

The invention discloses a bearing detection method based on a convolutional neural network. The method includes the following steps: generating a one-dimensional vibration signal based on a convolutional neural network based on the vibration acceleration signal of the bearing, and converting the one-dimensional vibration signal according to the time sequence and ratio It is divided into training set, verification set and test set in turn, and the convolutional neural network visualization structure based on Grad-CAM is established, and the Grad-CAM graph corresponding to the input sample of the convolutional neural network is obtained through ReLU function activation. In the Grad-CAM Dimensions of the vibration signal sampled on the graph, different activation thresholds are set in the value range of Grad-CAM, and the coordinates corresponding to the activation area of ​​Grad-CAM are used as the index to mark the activation of the original signal after passing through the convolutional neural network, and the neural network is established. Connections between network regions and target categories.

Description

technical field [0001] The invention belongs to the field of bearing fault detection, in particular to a bearing detection method based on a convolutional neural network. Background technique [0002] Nowadays, under the background of industrial big data, the rapid progress of artificial intelligence and machine learning has made fault diagnosis gradually become intelligent, and the use of data-driven intelligent algorithms for fault diagnosis has attracted more and more attention, becoming a new research hotspot in the field of fault diagnosis. Especially in recent years, the deep learning method represented by convolutional neural network has achieved great success in pattern recognition. This type of method can automatically mine the deep features of the input information, directly input the original information at the input end, and directly input the original information at the output end. The output result can be obtained, so it is also called "end-to-end" learning met...

AI Technical Summary

Problems solved by technology

[0003] Although the convolutional neural network has a remarkable effect on the results, it has been controversial because of the incomprehensible "black box model" and potential safety hazards. The root cause is that the neural network itself lacks the support of basic mathematical theory. Aspect neural network results lack engineering interpretability

In particular, convolutional neural networks and other common overfitting problems make it difficult for people to be convincing in the results

Especially in the field of mechanical fault diagnosis, many convolutional neural network methods simply apply existing deep learning models directly to the analysis of mechanical signals, which still lacks basis.

Images