Research On Tool Wear State Identification Method Based On Force And Vibration Signals

As the continuous advancement of "made in China 2025" strategy,the position of manufacturing industry has become more and more important.Intelligent manufacturing is a new strategic highland of international competition in the new round of industrial revolution.In the mechanical processing field,with the continuous improvement of the intelligence and automation in the production process,how to accurately monitor the tool wear status in real time has become a key issue to guarantee the production tempo,improve the production efficiency and product quality.In the mechanical processing,tool wear is inevitable.When the tool wear is more serious,it will affect the quality of the workpiece or even cause machine tool accident.However,changing the tool too early also will reduce its service economy,which improves the production cost of the enterprise.Therefore,it is necessary to provide a suitable strategy for tool changing based on accurately identifying tool wear state.Nowadays,with the intelligence and automation is carried out in factory,the real-time monitoring of the tool wear state through related physical signals can be realized.In this article,the monitoring methods for tool wear state based on the vibration and force signals is proposed and verified.The main research contents are as follows.(1)A test bench is set up,based on which the cutting tests are carried out and the wear state data of tools are collected.Firstly,all kinds of signal characteristics for cutting are analyzed,then the force signals and vibration signals are chosen for monitoring.After that the milling test bench is set up and the experimental scheme is carried out.Finally,the vibration signals,force signals and wear state of tools in their whole life cycle are collected.(2)A method for dividing wear stages of tool life cycle is proposed.In this paper,the mechanism of tool wear is studied,and a standard for dividing tool wear stage is established by analyzing the wear band width VB of the tool and the surface roughness Ra of the workpiece.Based on the standard,the tool is divided into four wear stages: initial wear stage,stable wear stage,sharp wear stage and tool failure stage.(3)A tool wear stage identification method based on multidimensional feature fusion is proposed.Based on the experimental data and the divided tool wear stage,the tool wear characteristics of time domain,frequency domain and time-frequency domain are extracted and fused from the vibration and force signals.Then the principal component analysis(PCA)method is used to optimized the features.After that the identification model of tool wear stages based on least squares support vector machine(LS-SVM)is constructed,and its sensitive parameters of c and g are optimized by using the particle swarm optimization(PSO)method to improve the identification accuracy.(4)A tool wear stage identification method based on deep learning network LSTM is proposed.In the traditional machine learning method,the features need to be extracted by user' experience,which will sharply affect the identification accuracy.While the proposed deep learning method based on LSTM network overcomes the mentioned weakness.Using the vibration and force signals as input,the model adaptively extracts the features of different wear stage and realizes the accurate identification.The test results show that the LSTM model improves the identification accuracy than the LS-SVM model.