Research On Load Spectrum Compilation Method For Typical Operating Conditions Of CNC Lathes

CNC machine tools are commonly used industrial equipment in the equipment manufacturing industry.Because of their low reliability,some malfunctions and failures often occur in the actual use.Therefore,it is necessary to improve the reliability of CNC machine tools.The load spectrum of CNC machine tools can be used in many aspects such as reliability design,reliability modeling and reliability testing,etc.The research on the load spectrum of CNC machine tools can improve the reliability of CNC machine tools and enhance the strength of the equipment manufacturing industry.With the continuous improvement of our country’s manufacturing level,the application range of CNC machine tools is becoming wider and wider,and the actual cutting conditions are becoming more and more complicated.The conventional typical working conditions are determined by the orthogonal test method and cutting force empirical formula combined with probability density function’s equal probability interval division method.The typical working conditions determined by them are of low representativeness and cannot reflect the real cutting conditions;in addition,after the rain-flow counting of the cutting force signals,when using the kernel density estimation method to fit the frequency distribution histogram of mean and amplitude,the bandwidth of the kernel function is not selected properly,and the fitting effect is not good,so the compiled load spectrum cannot truly and comprehensively reflect the cutting force data.Aiming at the above problems,this paper systematically studies the load spectrum compilation method of the typical working conditions of CNC lathes.It mainly includes:clustering method to determine typical working conditions,multi-layer wavelet noise reduction,Topsis comprehensive evaluation method to determine the optimal bandwidth of kernel function,building BP neural network kernel density estimation bandwidth model to obtain optimal bandwidth,and compiling two-dimensional load spectrum of CNC lathes.The main research contents of the paper are as follows:（1）In order to obtain the typical working conditions of CNC lathes,the cutting process data of CNC lathes collected on-site are sorted out,and the cluster analysis method is proposed to determine the typical working conditions.Selecting the feed amount,depth of cut,and the cutting speed as the three elements of cluster analysis,firstly,the 372 kinds of working conditions collected by the on-site tracking are clustered by the systematic clustering method,and the range of the number of clusters is obtained as[30,40],this paper selects the number of clusters as 36,and obtains the initial cluster center,and then uses the k-means clustering method to perform fast clustering,and iteratively updates to obtain the final cluster center,which is used as the CNC lathes’typical working conditions,36 groups of typical working conditions are finally determined,which is more in line with the actual cutting conditions.（2）A CNC lathe cutting force testing system is built,and the minimum sampling frequency is 1000 Hz.The collected cutting force signals are processed to remove singular values and eliminate trend items.For the noise signals existing in the cutting force signals,the multi-layer wavelet noise reduction method is used for noise reduction,and the traditional noise reduction effect evaluation indicators:root mean square error（RMSE）,signal-to-noise ratio（SNR）,smoothness（r）are not related,and the noise reduction effect cannot be comprehensively evaluated,proposes a multi-criteria decision-making wavelet denoising method based on grey relational analysis,carries out grey relational analysis and multi-index comprehensive evaluation on the denoising methods of one to six-layer wavelet,obtains the correlation between the noise reduction effect of various noise reduction methods and the optimal noise reduction effect.It is found that the four-layer wavelet noise reduction method has the highest correlation with the optimal noise reduction effect,so the four-layer wavelet noise reduction method is selected to denoise the cutting force signals.（3）The cutting force signals are counted by the rain-flow counting method,and the mean-amplitude rain-flow matrices of the cutting force signals are obtained,and the cutting force amplitude frequency histogram and mean frequency histogram are drawn.Amplitude frequency histogram and mean frequency histogram are fitted by using kernel density estimation,The traditional bandwidth determination methods in kernel density estimation are the empirical rule method and the cross-validation method,and the text chooses topsis comprehensive evaluation method to comprehensively evaluate the kernel density estimation effect of different bandwidths to determine the optimal bandwidth.In addition,this paper also introduces the BP neural network to find the optimal bandwidth,and builds a BP neural network kernel density estimation bandwidth model.A total of 90 sets of data for the first 30 typical working conditions are used as the training set,and a total of 18 sets of data for the last 6 typical working conditions are used as the test set.By continuously training the model,it is found that the coefficient of determination is greater than 0.75,and the BP neural network kernel density estimation bandwidth model has a good prediction effect,which greatly improves the efficiency of obtaining the optimal bandwidth.（4）The two-dimensional load spectra of the cutting force of the CNC lathes are compiled,and 36 typical working conditions are combined into a total working condition.Using BP neural network model to obtain the kernel density’s optimal bandwidths of main cutting force mean-amplitude distribution、backward force mean-amplitude distribution and feed force mean-amplitude distribution.A two-dimensional Gaussian kernel function is used to fit the mean-amplitude frequency histogram after rain-flow counting,the mean and amplitude were divided into ten grades at equal intervals,and the number of load cycles is extrapolated to10~6,and the mean-amplitude two-dimensional spectra of the main cutting force,back force,feed force are obtained.