Development of a Prediction Model for the Gear Whine Noise of Transmission Using Machine Learning

This study aims to develop a prediction model using machine learning to predict gear whine noise using the inspection data of gear tooth and noise bench test data in semi-anechoic chambers. To secure the reliability of the collected dataset, we select features that affect the prediction results by the lasso regression method. Based on the selected features, the prediction model was developed according to the characteristics of gear teeth under each vehicle's driving load condition. Random forest, decision tree, gradient boosting, extreme gradient boosting (XGB), light gradient boosting model algorithm, and the statistical regression method called ordinary least squares (OLS) are used to construct the prediction models. In addition, the performance of the machine-learning-based models and models using conventional statistical methods are compared. Prediction error reduction rate and prediction performance improvement rate are used to assess the performance of each model. The prediction model using the XGB algorithm exhibits the best performance. The obtained results demonstrate that the machine-learning-based prediction model can be used to predict gear whine noise with higher accuracy than OLS.