Research On CNN-Wiener Based Power Servo Tool Holder Power Head System Residual Life Prediction Method

CNC machine tools as the manufacturing industry "industrial mother machine",is an important strategic equipment that affects the national economic lifeline and national defense construction.Power servo tool holder as a high-end CNC machine tools-turning center key functional components,in addition to turning,drilling and boring processing,because of the addition of power head system can also realize milling processing and non-center hole drilling and boring processing,etc.,greatly enhance the processing capacity and efficiency of the machine tool.However,at present,the domestic power servo tool holder is only functionally available,but has not yet achieved stable and reliable performance,with frequent failures in service and serious reliability problems.How to reasonably predict the remaining life according to the actual operating condition in the process of service and provide the basis for predictive maintenance,so as to reduce the maintenance cost and improve the reliability of use is the focus of current research.Therefore,it is of great theoretical significance and engineering application value to carry out the study on the remaining life prediction of power servo tool holder power head system.Relying on the key laboratory of CNC equipment reliability,this paper carries out the research on the remaining life prediction of phased degradation,the remaining life prediction considering complex working conditions and the remaining life prediction interval estimation based on nonlinear Wiener process for a certain type of domestic power servo tool holder power head system with complex and variable service conditions and performance degradation characteristics,and also carries out the comparative analysis of the three methods using the bearing public data.The main work of this paper is as follows.1.Introduce the functional structure of power servo tool holder power head system,carry out the reliability test of power head system and test data processing.The mechanical structure and transmission form of the power head system of the tool holder are analyzed to find out the weak parts of the power head system of the tool holder and determine the key monitoring positions and parameters in the reliability test;the reliability test and state data acquisition plan of the power head system are formulated according to the actual service conditions;the state performance data obtained in the reliability test are subjected to noise reduction and time-frequency domain feature image extraction to provide basic data for the subsequent development of the remaining life prediction.2.For the problem that the performance degradation of the power head system of the power servo tool holder presents multiple stages during the service process,a twodimensional convolutional neural network(Convolutional Neural Networks,CNN)-based power head system remaining life prediction method is proposed in a phased degradation.Based on the extracted time-domain features and frequency-domain features,the health indicators of the powerhead system are constructed by fusing the time-frequency-domain features,and the performance degradation process of the powerhead system is divided into four stages according to the characteristics of the degradation trend of the health indicators,and each stage is linearized as the remaining life label value of the powerhead system of the tool holder.The extracted timefrequency domain feature images are used as the data set and divided for training,and a two-dimensional convolutional neural network residual life prediction method for the power head system degradation in stages is proposed.3.In view of the complex and variable service conditions of the power head system,a CNN-SVM-based two-dimensional convolutional remaining life prediction method for the identification of service conditions of the power head system is proposed for complex service conditions.Based on the time-frequency domain feature images,the last pooling layer features of the two-dimensional convolutional neural network condition classification model are extracted and input to the support vector machine classifier to build a CNN-SVM classification model to identify the condition types based on the time-frequency domain feature images,and to identify the condition types of the power head system at each acquisition moment.Based on the results of the condition recognition,the remaining life label values are constructed to degrade by condition.A 2D convolutional neural network residual life prediction method is proposed for complex working conditions of the power head system.4.To address the problem that the remaining life prediction point estimates based on deep learning are not conducive to the optimization of system maintenance strategies,a CNN-Wiener-based method for predicting the remaining life interval of the power head system for complex operating conditions is proposed.The remaining life prediction results of the two-dimensional convolutional neural network are used as degradation data,and a nonlinear Wiener process remaining life prediction model considering random effects is established to find out the probability density function of the remaining life of the power head system at the observed moment of degradation stage,and the remaining life interval estimation of the power head system is carried out by the fold-and-half search method.