Research On Fault Diagnosis System Of Spindle Rotor Of CNC Machine Tool Based On Hybrid Intelligence

Spindle rotor system is the core functional component of CNC machine tools,and it is also one of the parts with the highest failure rate in machine tool operation.Therefore,it is particularly necessary to carry out research on fault diagnosis of machine tool spindle rotor system.However,in the monitoring of machine tool operation status,there are many samples of normal status and few samples of fault status,resulting in an imbalance in the number of samples of various categories,affecting the identification of a few types of fault status.In addition,due to the variety of monitoring sensors and different sources of signals,the obtained information may have conflicts or redundancy,affecting diagnostic accuracy.In this paper,a hybrid intelligent fault diagnosis system for machine tool spindle rotor is developed based on the research on these two issues.The specific content is as follows:Firstly,three representative fault mechanisms and manifestations of the spindle rotor system of CNC machine tools are analyzed,and a fault simulation experimental scheme for the rotor system is designed.On the mechanical fault simulation experimental platform,four fault states of the rotor system,including imbalance,misalignment,rubbing of moving and stationary parts,misalignment,and imbalance coupling,as well as one normal rotation state,are simulated.Vibration acceleration data,vibration displacement data,and infrared temperature images are collected,providing data support for subsequent research on multiple types of imbalance and multi-source information fusion.Secondly,an improved multi class unbalanced oversampling algorithm based on MCSMOTE is proposed,that is,by improving the MCSMOTE two class unbalanced oversampling algorithm,it is suitable for multi class unbalanced data.The algorithm in this paper and the MDO,CCR multi class unbalanced oversampling algorithm are used in the UCI dataset and the measured rotor system fault dataset.The balanced samples are visualized by tsne,and three different types of classifiers KNN,SVM,BP are used to identify the balanced samples.By comparing the classification effect indicators m GM and m AUC,the results show that the data balanced by the improved MCSMOTE algorithm can obtain relatively good classification effect.Afterwards,a multi classifier fusion diagnosis method based on weighted D-S evidence theory was proposed to address the potential conflict or redundancy in the multi-source information collected by the rotor system fault simulation experimental platform,which may affect the diagnostic accuracy.Firstly,the vibration data collected under various faults in the rotor system is feature extracted and input into three neural network classifiers: BP,RBF,and GRNN to obtain preliminary diagnostic results.At the same time,the collected infrared temperature image is dimensionally reduced using SNMF to extract feature vectors,and the preliminary diagnostic results are also obtained by inputting them into the SVM classifier.These preliminary diagnostic results are then assigned as the original basic probability values,Calculate the weights and fuse them using D-S evidence theory to obtain the diagnostic results after weighted fusion.Compared with direct D-S evidence theory fusion diagnosis,the effectiveness of the proposed method can improve the accuracy of diagnosis.Finally,a rotor fault diagnosis software with multi-class imbalance and information fusion as the core is developed by using the MATLAB App Designer tool.This software,combined with a machine fault simulation bench,enables a hybrid intelligent fault diagnosis system integrating multiple intelligent machine learning methods in data import,class imbalance processing and fusion diagnosis of rotor system fault samples.