Compilation Of Cutting Force Spectrum Of CNC Lathe Based On The Prediction Model Of Deep Confidence Neural Network

The inherent reliability of product is firstly determined by its design.Load spectrum is widely used in the reliability design and fatigue life analysis of mechanical and electrical products because it can reflect the changing law of product load.In order to solve the problem that it is difficult to select the typical working conditions and reappear the cutting process in the existing load spectrum compilation methods,this paper studies the cutting force spectrum compilation method of CNC lathe based on depth confidence neural network prediction model,including the determination of test conditions,the establishment of test system,the fitting of load signal distribution,the construction of average amplitude prediction model of cutting force,the extrapolation of load and the compilation of program loading spectrum.The main research contents are as follows:(1)In order to ensure the effectiveness and authenticity of the load spectrum and reduce the cost of load measurement,the variable speed cutting and constant speed cutting conditions are divided into three groups by empirical formula,that are light load25%,medium load 50% and heavy load 25%,according to the load data obtained from machine users.The determination method of typical test conditions is proposed based on hierarchical clustering.The pseudo-F statistics is introduced to determine the optimal number of clusters,and 29 groups of typical conditions of constant speed cutting and 26 groups of typical conditions of variable speed cutting are finally determined as the typical test conditions.Besides that,the dynamic cutting load test system of CNC lathe is established,and the cutting test is carried out based on the standardized test plan.Then the dynamic cutting force load signals of typical test conditions are obtained.(2)In order to reduce the influence of noise and external vibration on the cutting force signal,the pre-processing analysis is carried out for the measured dynamic turning force signal,including eliminating the trend term,removing singular points and filtering noise reduction.To keep the ratio of the obtained signal and the actual time length the same,the MCMC method is used to reconstruct the cutting load signal,and the rain flow counting method is used to count and analyze the reconstructed signal.To accurately describe the load distribution of each working condition,the mixed distribution and copula function is adopted to establish the joint distribution of load mean and amplitude,the Weibull distribution is adopted to fit the amplitude of constant speed and variable speed cutting,the Gaussian distribution is adopted to fit the mean of constant speed cutting,the double Weibull distribution is adopted to fit the mean of variable speed cutting,and the Gaussian copula function is adopted to establish the joint probability density function of mean value and amplitude under the constant speed and variable speed cutting condition.(3)To solve the problem that it is difficult to reappear the cutting process,the depth belief neural network(DBN-DNN)is used to predict the load distribution of turning force,and the feasibility of applying DBN-DNN prediction model to solve cutting force related problems of CNC lathe is analyzed.Then,the data is processed by zero mean and normalization to eliminate the error caused by different dimensions.Through the unsupervised training and the supervised reverse parameter adjustment,a prediction method of cutting force mean and amplitude of CNC lathe is proposed based on DBNDNN.The optimal parameters are determined by dividing the data set into training set,test set and prediction set,and comparing the accuracy of test set under different nodes and layers in the model structure.Finally,the parameters of cutting force mean value and amplitude are predicted respectively under constant speed cutting and variable speed cutting prediction sets.(4)To obtain the cutting force spectrum of CNC lathe in the whole life cycle,the frequency of extrapolation load is calculated by solving the total frequency of constant and variable speed cutting load cycles.The rain basin parameter method is used to extrapolate the load range and divide it into eight grades.To reduce the programming error of the program loading spectrum,the eight-level interval cycle frequency corresponding to the average amplitude under each typical test condition is obtained based on the fitting results and DBN-DNN prediction results,that is,three sets of working conditions are combined to obtain the two-dimensional program loading spectrum of CNC lathe under the constant and variable cutting conditions.The constant speed cutting and variable speed cutting process are carried out in the order of lowhigh-low alternating form,which will guide the reliability test of key functional parts of CNC lathe.