Fault monitoring and diagnosis of motorized spindle in five-axis Machining Center based on CNN-SVM-PSO

Abstract

A spindle fault diagnosis method based on CNN-SVM optimized by particle swarm algorithm (PSO) is proposed to address the problems of high failure rate of electric spindles of high precision CNC machine tools, while manual fault diagnosis is a tedious task and low efficiency. The model uses a convolutional neural network (CNN) model as a deep feature miner and a support vector machine (SVM) as a fault state classifier. Taking the electric spindle of a five-axis machining centre as the experimental research object, the model classifies and predicts four labelled states: normal state of the electric spindle, loose state of the rotating shaft and coupling, eccentric state of the motor air gap and damaged state of the bearing and rolling body, while introducing a particle swarm algorithm ( PSO) is introduced to optimize the hyperparameters in the model to improve the prediction effect. The results show that the proposed hybrid PSO-CNN-SVM model is able to monitor and diagnose the electric spindle failure of a 5-axis machining centre with an accuracy of 99.33%. In comparison with the BP model, SVM model, CNN model and CNN-SVM model, the accuracy of the model increased by 10%, 6%, 4% and 2% respectively, which shows that the fault diagnosis model proposed in the paper can monitor the operation status of the electric spindle more effectively and diagnose the type of electric spindle fault, so as to improve the maintenance strategy.