Research On Methods Of Tool Condition Monitoring And Remaining Useful Life Prediction In End Milling Machine

Automation and intelligence of manufacturing processes are the development trend of industry.CNC milling machines have become important parts of production by its advantages of high automation,stability,precision and flexibility.As the end part in milling process,tool is one of the key factors for the success of milling,and is also the most vulnerable and wasteful part.It is particularly important to identify and monitor its status timely and effectively.However,traditional condition monitoring methods are greatly restricted in milling tool condition monitoring(TCM)due to its characteristics of small sample size,low signal-to-noise ratio and strong time-varying.Therefore,how to improve the accuracy of tool condition and its remaining useful life(RUL)prediction,and monitor tool condition in time and effectively,has become an urgent problem to be solved in intelligent milling and is also one of the main interests in intelligent processing technology.This dissertation focuses on the problem of TCM and RUL prediction in end milling process,combining machine learning,intelligent calculation and stochastic process theory,studies the monitoring model,feature selection method and RUL prediction in order to develop novel and effective methods for milling TCM.In order to improve the accuracy of TCM in milling,using the advantages of fast learning speed of kernel extreme learning machine and the ability of hierarchical angle kernel function to simulate large-scale neural network to calculate vector similarity,a two-layer kernel extreme Learning Machine(TAKELM)algorithm is proposed,it overcomes the shortcomings of the extreme learning machine in feature learning of complex non-linear high-dimensional data,and avoids the selection of kernel function and the pre-set of its parameters.The TAKELM algorithm is showed that it can improve the learning performance without significantly affecting the learning speed through several benchmark data sets of classification and regression problems.Then,a TCM model for milling process based on TAKELM is proposed,and its feasibility and effectiveness is demonstrated by testing on the condition recognition of two TCM benchmark data set and one single-sensor TCM experiment for milling processes.For improving the accuracy of TCM,a TCM method based on blind source separation and TAKELM is proposed,in which the stationary subspace analysis is introduced to separate the original signal firstly,and then the TAKELM model is used to identify the tool conditions.Experimental analysis verifies the effectiveness of the method that greatly improves the accuracy of TCM.In order to overcome the local dependence of feature parameter selection methods currently,a TCM feature selection method based on global diagnosis error and improved differential evolution(DE)algorithm is proposed.A candidate feature parameter set of multi-domain is constructed based on time-,frequency-,and time-frequency domains.A two-objective optimization model is established for minimizing the global diagnosis error and the number of feature parameters.Then,the optimization model is transformed into a single-objective unconstrained optimization combination problem,which is optimized by DE algorithm.Aim to accelerate the search speed of DE,an improved DE is provided,the solution information of individual population in past generations are used to update the value characteristics of feature parameters,which can find the optimal solution more quickly by optimizing the chromosomes of individual population.The applications in TCM benchmark data and multi-sensor end mill monitoring experiment show that this method can achieve high precision in TCM while reducing the number of sensors.Aiming at the insufficient accuracy of RUL prediction of milling tool under small sample size,an improved inverse Gaussian(IG)-based RUL prediction method is provided with stochastic process theory.According to the rule of experimental data,the tool wear degradation process is assumed that the unit increment of tool wear obeys the IG process based on tool wear value,the parameters are estimated by maximum likelihood estimation,and the RUL of tool is predicted based on slice sampling technology.Through the analysis of TCM benchmark data set and end mill TCM experiment,it is demonstrated that this method can obtain effective prediction value and 95% confidence interval of tool RUL.A portable end mill TCM system based on DSP is developed,which can accurately monitor the tool condition in end mill,and effectively predict the RUL of the tool.This system could provide effective information support for improving the effective utilization of end mill and reducing the cost of process.