In-situ Condition Monitoring Of Thread Tool Based On Unsupervised Learning

In-situ tool condition monitoring can improve the efficiency of tools used and prevent accidents due to tool chipping and broken.At present,there are still some problems existing in the research area that are remaining unsolved: insufficient training sample size,high acquisition cost of life cycle data,and the model being extremely sensitive to changes of processing parameters.In order to solve the above problems,the idea and concept of the real-time training of an unsupervised model for monitoring by processing data are proposed.The sensor signals of real-time tool cutting are used as training samples,and One Class support vector machine(One Class SVM)is used to establish a monitoring model to detect abnormal signals.In order to improve the accuracy of the monitoring model,an idea of improving One Class support vector machine(One Class SVM)is proposed,which provides a new idea for the online monitoring of the tool status.The main research content is as follow:1.Multi-sensor fusion technology.The power and vibration of the spindle during the cutting process were collected experimentally.The signal analysis methods of time domain,frequency domain,and time-frequency domain were used to extract the features of multiple sensor signals and establish a multidimensional feature vector matrix,and using the correlation analysis technology to screen.The results show that the use of multi-sensor fusion technology can make up for the tool condition monitoring defects between various information and better monitor the tool condition.2.An effective signal-cleaning scheme.The test uses the same PLC signal trigger to set the synchronization of the power signals and the vibration signals to start acquisition.Based on the characteristics of the power signal derivative and the time domain consistency of the vibration signals,a real-time and effective signal-cleaning algorithm has been proposed and tested effectively.3.Establishing an in-situ monitoring model of tool conditions based on the One Class SVM.A support vector machine model suitable for small samples was selected to establish the One Class SVM network model based on the real-time processing data,which solves the generality of the model problem,and the loop modeling,which transforms the monitoring samples into training samples,was used to improves the model accuracy.