| An essential part of the steam generator of the AP1000 nuclear power plant is the water chamber head.It has the characteristics of large volume,complex shape,and unique materials.It is made of 508Ⅲ steel forging rough which is difficult to process on heavy-duty machine tools.Up to 70% of the raw material has to be removed from the forging billet weighing more than 200 tons in order to process it into the finished product.Extreme manufacturing techniques,such as heavy milling and turning,dominate its processing.In the cutting process,the front and rear tool sides are prone to severe wear due to strong extrusion and friction,which will drastically impair the tool’s service life and the quality of the workpiece.However,at present,the tool wear status of heavy machining almost still needs manual judgment,and the production efficiency is low,therefore,for the above problems to predict the wear of milling 508Ⅲ steel tools and reasonable cutting parameters selection has become an urgent problem to be solved.First,Deform software was used to establish a finite element simulation cutting508Ⅲ steel model,analyze the cutting heat distribution and tool wear form;design milling 508Ⅲ steel carbide tool wear test,explore the tool wear mechanism and determine the tool wear stage division,provide the theoretical basis for tool wear prediction.Analyze the cutting force signal with the change of tool wear and use the cutting force signal as the source of tool wear prediction signal.Second,to reduce the contamination of outliers and noise in the signal.Therefore,the statistical theory and the improved wavelet threshold denoising method are used in the preprocessing of the force signal to improve the signal quality.Then,the processed force signal’s features were extracted in the time domain,frequency domain,and time-frequency domain,respectively,and it was found that the force signal was mainly distributed in the low frequency part.15 features closely related to tool wear were screened by Pearson correlation coefficients and used as input for traditional machine learning to provide the basis for the traditional machine learning model.Furthermore,Firefly Algorithm(FA)is used to optimize the traditional machine learning BP neural network.The above 15 features are used as input components and the corresponding tool wear as output to build the FA-BP tool wear prediction model and test and validate it.To avoid the tedious and time-consuming process of manual feature extraction,a multi-scale Dense Net-Res Net-GRU tool wear prediction model with deep learning is established.The multi-scale convolution kernel is used to convolve the force signal matrix as input,and the signal space features are extracted according to the respective characteristics of Dense Net and Res Net,and the timing features are extracted using GRU.Testing and validating the prediction accuracy of the model.The interface of milling tool wear monitoring system is built by Lab VIEW,and the signal can be processed and analyzed based on MATLAB scripts to obtain the signal time and frequency waveforms and the predicted value of tool wear and wear status.Finally,the three cutting elements are selected as optimization variables,and tool life,material removal rate,and cutting force are set as optimization objectives to achieve multi-objective optimization of carbide tools for milling 508 III steel.Analyzing the link between the three cutting factors and the change in tool life using orthogonal experiments,and the three cutting elements are used as inputs to establish an approximative model of tool life and cutting force using BP neural networks;Input the approximate model and material removal rate formula into the NSGA-II algorithm to produce 50 sets of Pareto optimum solutions.To increase the efficiency of process parameter decision-making,the entropy-based TOPSIS method is used to sort and analyze 50 groups of plans.This determines the best decision solution and provides a basis for the selection of cutting parameters in actual machining. |
PDF Full Text Request