Prediction of Thermally Induced Axial Displacement of Mechanical Components Using LightGBM

Abstract

The goal of this research is to create a machine learning model that can predict the thermally induced axial displacement of machine tool spindles. To achieve this goal, this study applied the Light Gradient Boosting Machine (LightGBM) learning framework to predict the thermally induced axial displacement of mechanical equipment by a heat source in a model that had an outer structure similar to that of a machine spindle. In the predictions using LightGBM, the time, temperature, and heat flux of equipment surfaces are measured and used to predict displacement. A similar trial study was conducted for a servomotor. A series of experiments clarified that the thermally induced axial displacement of the equipment can be predicted using a machine learning model created from the measured temperatures and heat fluxes of the target component and other parameters. Furthermore, the study focused on the feature importance in the prediction process. Through these considerations, the features that are most valuable for prediction among the features used for the trial measurement and subsequent prediction were extracted based on the feature importance. Using the feature importance, the top-ranked parameters were chosen to create a machine learning model for prediction. Consequently, equivalent prediction accuracy is possible, even if the number of features, namely sensors required for the acquisition of sufficient features for the prediction, can be reduced without significantly affecting the prediction accuracy. Specifically, it was confirmed that the number of sensors can be reduced from about 65 to about 4 for the spindle model and about 20 for the servomotor.