Research On Health Status Evaluation And Prediction Of CNC Tool

CNC is the core equipment of modern industrial production.Monitoring the health status of CNC is an unavoidable problem in advanced manufacturing.Tool as a part of CNC directly contacts with the workpiece part,which will work in high temperature and high pressure environment for long-term.It's one of the most easily worn and problematic parts of CNC.Tool wear affects the size of the processing precision and surface quality.The tool works in failure state for long-term may even cause breakage and broken,which will cause serious economic losses.Therefore,the real-time online monitoring for tool is a very important topic at present.In order to implement an online assessment of tool health,the whole life cycle experiment of the tool is designed and carried out in this paper.The combination of traditional signal processing and advanced machine learning algorithm is used to analyze and model the signals.A method to extract abstract features of signals is proposed innovatively to research tool wear,which can implement tool health prediction effectively.Finally,a tool health prediction platform is developed based on an actual production line.The main contents are as follows:(1)Design and carry out the full factor and full life cycle experiment for tool wear for several major factors after researching the principles of tool wear.The temperature,current,vibration and other signals in the process are collected by means of multi-sensor fusion.Wavelet decomposition and differential calculation are used to realize the segmentation of different working state signals.The traditional signal processing method is used to extract the time domain,frequency domain,wavelet packet and other characteristics of the effective working signals.The VB value of tool is measured by a professional micrometer as the label of data.Finally,a complete tool wear data set is built.(2)T-SNE and K-means algorithms are used to reduce dimension and cluster features,which can realize the polymerization of similar working conditions and the separation of different working conditions successfully.The classification results of working conditions can provide more information for tool wear.XGBoost integrated learning algorithm is used to train and predict the features of small and unbalanced tool wear samples and compared with integrated learning algorithms such as random forest and GBDT by regression and classification.The results show that XGBoost algorithm has some advantages in both precision and time for tool wear prediction.(3)The traditional method of feature extraction needs a deeper understanding of signal processing,and the selection of signal features requires rich experience.To solve this problem,a signal analysis and prediction method based on abstract features is proposed in this paper.VMD algorithm is used to carry out mode decomposition for the signals of specific working procedure.Then,select the mode with the highest degree of reduction to analyze.Compressed sensing algorithm is used to compress the mode to obtain the abstract features while ensuring the reduction accuracy.Finally,random forest algorithm is used to train and predict abstract features,and compared with BP neural network,SVR,GBDT and other common algorithms.The results show that random forest algorithm has the best prediction accuracy.(4)A CNC health monitoring system is developed for an automobile engine production line.The system adopts the design architecture of cloud,edge and end separation,which reduces the data computing and transmission pressure and improves the system's expansibility.The edge end adopts the OPC protocol and edge computing technology to realize the fusion of signals from different channels and eliminate the transmission process of high-frequency signals to the cloud.The cloud adopts a design framework combining data dictionary,device management and pluggable algorithm model,which not only ensures the universality of data flow but also enables engineers at different levels to be completely decoupled.The terminal can directly interact with users.Through the unified data interface,the terminal can not only realize the display and early warning of various machine tool information,but also open the secondary development platform.Through the above four parts,the wear mechanism,the prediction method and the practical application of three aspects of tool wear has been studied in this paper.The existing prediction scheme has been greatly improved,and a relatively complete tool wear evaluation system has been established.It is of great help to improve the current level of prediction and the follow-up research.