Prediction method for spiral bevel gear tooth surface roughness based on machine learning

LI Jiabin; CHEN Haifeng; LIU Guoliang; ZHOU Changjiang

doi:10.16578/j.issn.1004.2539.2026.08.011

Abstract：[Objective]To address the low prediction accuracy and parameter optimization difficulties of tooth surface roughness for spiral bevel gears,and overcome the limitations of traditional methods in handling complex nonlinear relationships and multivariate coupling effects,machine learning models were adopted to predict the tooth surface roughness of spiral bevel gears.[Methods]Firstly,based on the grinding test dataset of spiral bevel gears,three machine learning algorithms including decision tree(DT),support vector regression(SVR)and artificial neural network(ANN)were used to establish roughness prediction models for the convex and concave tooth surfaces,and the prediction performance of the three models was compared.Secondly,multiple linear regression was applied to derive a calculation formula for tooth surface roughness incorporating machining parameters of spiral bevel gears.Finally,Shapley additive explanations(SHAP)were employed to quantify the contribution of each input feature to the predicted roughness,providing references for the application of machine learning in high-performance gear manufacturing.[Results]The results show that the DT model suffers from underfitting and the SVR model suffers from overfitting,both yielding poor prediction performance.The ANN model achieves excellent data fitting and accurate roughness prediction at the cost of relatively slow computation speed.Its mean relative errors for predicting convex and concave surface roughness reach 3.5%and 6.09%,respectively.The influence degree of each machining input parameter on tooth surface roughness,sorted from highest to lowest,is grinding speed,grinding depth and generating speed.