Research On Optimization Strategy And Experimental Study Of State Identification Of Ball Screw Pair Of CNC Machine Tools

As the core component of modern industrial master machine,the state identification method of ball screw pair affects the accuracy and reliability of CNC machine tool to a certain extent.Based on the major national science and technology projects,this paper carries out theoretical and experimental research on the state optimization identification method of ball screw,focuses on the multi-domain feature extraction method of vibration signal and the optimization combination of state identification,further verifies the state identification strategy through test data and training models,so as to lay the foundation for the realization of on-line monitoring and health warning of ball screw.The fundamental to study the optimization strategy of ball screw state recognition lies in:firstly,clarify various characteristic states and corresponding extraction methods;secondly,carry out relevant vibration test experiments to obtain vibration signals under the target state;and then optimize the extraction by means of feature optimization feature set;finally,it is necessary to verify the effectiveness of the optimization strategy and establish an optimized recognition model.Therefore,this paper focuses on the multi-domain feature extraction methods for vibration signals,vibration signal acquisition experiments,feature vector optimization,and the establishment of state recognition models.The specific contents are as follows:(1)The characteristics of the vibration signals of the ball screw pitting,lubrication and preload are analyzed.Combined with the vibration feature extraction methods under three kinds of decomposition methods: time domain(with dimensional index and non-dimensional index),frequency domain and time frequency domain,the index parameters corresponding to three states are optimized for the first time: 12 feature sets and 259 feature vectors in total.(2)The vibration test requirements of three states of ball screw pair are analyzed,and the bench test for pitting corrosion state and the main engine test for lubrication and preload state are determined.Vibration test systems for the two types of tests of the bench and the host are set up respectively.Three types of vibration test schemes are designed and vibration signals are collected.(3)Furthermore,the feature sets of three kinds of state vibration signals are extracted,and the weight values of feature vectors in each feature set are calculated by using the Relief-F algorithm.The feature vectors of three states are optimized by the threshold screening method.(4)Taking the accuracy of recognition model as the optimization criterion,the eigenvectors of three states are optimized twice.By further comparing the recognition accuracy of the feature vector after the second optimization,the optimal recognition methods of pitting,lubrication and preload state are obtained in order: time domain index combined with Improved EMD's Hilbert envelope spectrum feature frequency extraction method,wavelet packet energy feature extraction method,time domain index combined with wavelet packet entropy feature extraction method.(5)Three-quarters of the optimal samples of each state are selected,and the improved PSO-SVM,GS-SVM and RBF neural network state recognition models are trained respectively,and the remaining 1 / 4 of the lubrication state samples are identified and predicted.The results show that the prediction accuracy of the model is 93.33%,84.44% and 80% respectively,which verifies the correctness of the optimization strategy of state recognition and the applicability of the improved PSO-SVM identification model.