Research On Methods Of Health Status Evaluation And Prediction Of Ball Screw

As the key equipment of mechanical processing,CNC machine tools are an important part of the modernization of high-end equipment manufacturing.As a key functional component of the CNC machine tool transmission system,the ball screw is long-time affected by the factors such as high-speed operation,load shock and high temperature.The ball screw will inevitably suffer from abrasion and performance degradation,which will lead to the decline of transmission accuracy and seriously affect the processing accuracy and workpiece quality of CNC machine tools.In order to monitor the running state of the ball screw in real time and perceive its performance,this paper studies the methods of health evaluation and state prediction methods of the ball screw,and a prediction model which combines convolutional neural network with long-short-term memory network based on the stage division of performance degradation state.Based on this evaluation model,a complete preventive maintenance architecture is constructed step by step from data acquisition to intelligent maintenance platform.(1)According to the structure and load characteristics of the ball screw,the contact characteristics of the ball and raceway,the friction characteristics of contact area and the force characteristics of the ball are analyzed.The effects of contact stress and friction moment on the wear and fatigue failure of the ball screw are studied,and the mechanism of performance degradation of the ball screw is revealed.The main factors which lead to the performance degradation and failure causes of the ball screw pairs,as well as the performance degradation law of the ball screw,are analyzed.The performance degradation evaluation method,evaluation model and intelligent prediction and maintenance system are constructed.(2)Aiming at the problem of how to obtain the degradation trend and state division of the ball screw,a method based on kernel principal component analysis,Mahalanobis distance transformation and DBSCAN density clustering is proposed.The performance indicators of the ball screw are reduced by using kernel principal component analysis(KPCA),and the features with strong representation ability are extracted.The dimension reduction features are mapped by Mahalanobis distance transformation to obtain a one-dimensional characterization space of the multi-dimensional feature,and the performance degradation trend of the ball screw is obtained.The degradation trend of the ball screw is clustered by DBSCAN density clustering,and several performance degradation stages based on time series segmentation are obtained.(3)To evaluate and predict the health status of the ball screw,an evaluation method based on the combination of convolution neural network and long-term memory network(CNN-LSTM)is proposed.This method does not need manual features selection,and directly learns the original signal data by own to realize automatic feature extraction.Combined with the experimental data and compared with other algorithms,the ability and effect of all kind of proposed algorithms on the health assessment and state prediction of the ball screw are verified.The results show that the evaluation method based on CNN-LSTM has better prediction performance and robustness.(4)An intelligent maintenance system for the ball screw is developed based on the above-described deep learning performance prediction method.The data acquisition system of CNC machine tool is developed by OPC technology protocol,the sensor signals on the ball screw of CNC machine tool are collected in real time,and the health evaluation and state prediction model of the ball screw is constructed by deep learning evaluation algorithm.The online evaluation of the running ball screw of CNC machine tool is realized which provides a theoretical basis and technical method for the health evaluation and stable operation of the ball screw.