Research On Fault Diagnosis Of CNC Machine Tool Spindle Bearing Based On Deep Learning And Ensemble Learning

The spindle is the core component of the CNC machine tool.The failure of the spindle often leads to the abnormal operation of the CNC machine tool,performance degradation,the decline of the product quality,and even the failure of the product.The spindle failure is usually caused by the failure of the bearings supporting the spindle.Therefore,studying the fault diagnosis technology of CNC machine tool spindle bearings has important practical significance and research value to ensure the safe operation of CNC machine tools,improve production efficiency,and avoid losses and catastrophic accidents.Spindle bearings of CNC machine tools usually operate under heavy load,high speed,and strong impact conditions,and their fault signals often have the characteristics of strong noise,nonlinearity,non-stationarity,and non-Gaussian distribution.In addition,the labeled fault samples that can be obtained on the production site are usually very limited,but the number of normal samples is sufficient,thus resulting in the problem of sample imbalance.In these cases,the traditional fault diagnosis methods are difficult to effectively mine the potential fault information contained in the samples and extract the representative fault features,which leads to the decline of the model diagnosis performance.In response to these problems,this dissertation proposes a fault diagnosis method for CNC machine tool spindle bearings based on deep learning and ensemble learning.The main research of the dissertation is as follows:(1)The improper parameters selection of the deep learning model will result in the decline of model diagnostic performance.Therefore,the dissertation studies the influence of important parameters in the rolling bearing diagnosis model based on convolutional neural network on the diagnosis performance,provides guidance for the subsequent integration model to provide individual classifiers with good performance,and then improves the generalization performance of the integration model.(2)Aiming at the problem that the original vibration signal of rolling bearing carries noise and fault sample is unbalanced,a fault diagnosis method based on wavelet packet transform and convolutional neural network for rolling bearing integration(EWCNN)is proposed.The method decomposes the original signal into orthogonal sub-signals through wavelet packet transform,which ensures the independence of the classifiers and is beneficial to improving the performance of the ensemble model.The effectiveness of the EWCNN method is verified by the data of the rolling bearing test bench and the spindle bearing of the machine tool.Compared with the other machine learning methods and ensemble learning methods,the experimental results show that the EWCNN has the best diagnostic accuracy and stability.(3)Aiming at the problems of insufficient feature extraction ability and poor generalization ability of single model of shallow machine learning rolling bearing fault diagnosis method,an ensemble learning method based on convolutional neural network and ensemble support vector machine(CNN-ESVM)is proposed.In the feature extraction stage,the method uses the convolutional neural network to extract multi-level features layer by layer,then sends the multi-level features to the integrated support machine,and finally combines multiple classifiers into a more accurate and stable classifier through the designed combination strategy.The effectiveness of the CNN-ESVM method is verified by the machine tool spindle bearing data,and the results show that the CNN-ESVM has better diagnostic performance than the single-model diagnostic methods,and the other intelligent diagnostic methods.