关键词: 表面粗糙度, 振动信号, 小波包变换, 残差网络, 钛合金

Abstract: To improve the prediction accuracy of the model for surface roughness and avoid the dependence of feature extraction and selection on prior theoretical knowledge in traditional machine learning prediction method,a surface roughness prediction method based on Wavelet Packet Transform (WPT)and Residual Network (ResNet)was proposed.In this method,the vibration signal was decomposed into wavelet packet coefficients of different frequency bands by using wavelet packet transform,and the coefficient matrix was formed by fusing the wavelet packet coefficients of each frequency band for capturing the relationship between adjacent frequency bands.The input of ResNet was obtained by superposing the coefficient matrices of different directions of centerless lathe,and the features with strong ability to characterize surface roughness were adaptively extracted by ResNet.The prediction of surface roughness was realized.Compared with other prediction methods,the prediction results of the proposed method were close to the actual measurement results,and the accuracy was improved,which proved that the proposed method was more effective.

Key words: surface roughness, vibration signal, wavelet packet transform, residual network, titanium alloy