Research On Data-driven Adaptive Prediction Method For Remaining Useful Life Of Slotting Cutter

Rotor is the core component of turbine.The rotor slot connects the rotor and the blade root,and its machined surface quality directly affects the energy conversion efficiency and life of turbine in service.During the actual machining process of rotor slot,there are significant differences between tool wear and tool life due to the fluctuation of rotor material performance,tool performance change,machine tool difference and cutting parameter adjustment.How to accurately predict the remaining useful life of slotting cutter,change the tool reasonably and improve the machining efficiency under the premise of ensuring the machining quality of rotor slot's surface is an urgent problem to be solved.Traditional tool life prediction methods are generally for the same type of cutting tools,which need vast labeled training data,and require that testing data are independently and identically distributed with the training data.However,it is difficult for the actual machining data to meet the above conditions.Therefore,it is urgent to carry out accurate prediction of remaining useful life of slotting cutter in actual time-varying and nonlinear systems.The main contents of this thesis are as follows:(1)Studying on recognition and dynamic classification method of key factors causing wear of milling cutter under multistage machining process.In view of the problem that tool wear is influenced by multistage machining process factors,the factors are cross-linked,process wear data are missing,continuous and discrete data coexist,this thesis discretized the influencing factors on wear of milling cutter using dynamic hierarchical classification method,and built an incomplete information system of milling cutter wear in multistage machining process based on similarity relation theory.Furthermore,the key influence factors,i.e.cutting tool regrinding and tool supplier are found in this thesis and the dynamic classification of time-varying and nonlinear machining system is realized by calculating and analyzing the importance degree of each influencing factor.(2)Studying on hierarchical adaptive life prediction method for cutting tool in multiple degradation modes.As the cutter material,regrinding parameters are dynamically changing,it is difficult to precisely present the issues of degradation of cutter performance in overall system by using one single and fixed model.Therefore,the hierarchical and multi-model prediction method based on similarity is studied in this thesis.By using acoustic emission monitoring signal,the tool performance index is established by considering the distance similarity and spatial direction similarity,and it can make full use of the degradation information of samples to realize the adaptive and accurate prediction of tool remaining useful life.The experimental results show that the proposed algorithm can reduce the mean absolute error of the remaining useful life by about 2 slots.It can effectively improve the unreasonable tool changing in enterprise,and the comprehensive life utilization rate is increased by 9.68%.(3)Studying on transfer-learning adaptive prediction method for cutting tools in new degradation mode.In practical production,the adjustment of tool property and cutting parameters could cause large change in the performance of cutting tool,thus the historical data-based tool life prediction model is difficult to predict the above cutting tools effectively.Firstly,a LSTM-based model is designed,and the source domain prediction model is obtained by pre-training the historical data;Secondly,through adversarial training between historical data and the data of new degradation mode,the prediction model is updated;Finally,the pre-trained source model is transferred to the remaining useful life prediction of cutting tool in new degradation mode,and the tool life prediction under changing environment is realized.The experimental results show the effectiveness of our proposed method.