Research On Fault Diagnosis Method Of CNC Machine Tool Spindle Bearing Based On Graph Neural Network

The spindle bearing is one of the core components of CNC machine tools.The machine tool spindle bearing support the machine tool to work for a long time.Faced with the harsh working environment of variable working conditions and high load,the spindle bearings are prone to failure,resulting in damage to the machine tool.This will seriously threaten the safety production of the enterprise and the personal safety of the workers.Traditional fault diagnosis methods based on signal processing and shallow learning often rely on the quality of manually extracted features,while rolling bearing fault diagnosis algorithms based on deep learning have poor generalization performance in the face of unbalanced fault samples,and rely heavily on the depth structure and require numerous parameters for training.Graph neural network,as a kind of connected neural network model,can fully combine graph theory with deep learning,extract the local feature information and global structure information of signals through graph data,make up for the shortage of fault label information,and greatly reduce the network parameters to be learned,which can effectively solve the problem of rolling bearing fault classification with unbalanced fault data.The research contents of this paper are summarized as follows:(1)Aiming at local time shifting problem of rolling bearing vibration signals,dynamic time warping(DTW)is introduced as the evaluation index.According to the similarity of DTW,independent time series samples are constructed as the nearest neighbor graph with topological structure.The geometric structure information between nodes in the graph provides supplementary information for the unbalanced fault samples,and then improves the diagnosis performance of the subsequent fault diagnosis model.(2)Aiming at the problem of unbalanced rolling bearing fault samples,a rolling bearing fault identification model based on topology adaptive graph convolution network(TAGCN)algorithm is proposed.Firstly,the model employs DTW to transform the time-domain vibration signals in Euclidean space into graph data in nonEuclidean space.Then the graph data is input into fault diagnosis model,TAGCN uses the nearest-neighbor relationship information between different healthy samples for training network parameters,so that the sample feature information of unlabeled samples can be propagated to their nearest-neighbor labeled samples,feature extraction and feature fusion of input data are carried out by graph convolution kernels of different sizes,which overcomes the problem that it is difficult to extract depth features by linear approximation of single convolution kernel.The effectiveness of the model is verified on the CWRU bearing data set and the CNC machine tool spindle bearing data set respectively.Compared with the classical deep learning algorithm,it is further verified that the TAGCN model has better performance in identifying rolling bearing faults under different types,different fault severity and different working conditions.(3)Aiming at the problems of poor generalization performance and high timeconsuming of rolling bearing fault diagnosis method based on deep learning under complex working conditions,a rolling bearing fault identification model based on multi head graph attention network(MHGAT)is proposed.Firstly,the bearing vibration signal is constructed into graph data with topological structure by using the DTW method,and then graph data is input to the MHGAT model.Relying on the application of multi head attention mechanism,distinguishing features are extracted on multiple scales for bearing fault diagnosis.The effectiveness of the model is verified by CWRU bearing dataset and CNC machine tool spindle bearing dataset.Compared with the traditional deep learning algorithms,the experimental results show that the rolling bearing fault diagnosis method based on MHGAT has the best diagnosis performance,and the required training time is greatly reduced.