Machine Learning Application in Predictive Maintenance of Screwdrivers

With the growth of Industry 4.0, several advances in machine learning (ML) algorithms have emerged to predict the current and future states of equipment, particularly for preventive maintenance purposes. This paper presents the results of a practical application of ML algorithms to predict the maintenance of screwdrivers in a production line. Results were obtained by training and evaluating five regression models, primarily based on Decision Tree (Random Forest, Decision Tree, Extra Trees) and Gradient Boosting variants (LGBM and XGBoost). The experimental evaluation considered these five models, among which Random Forest achieved the best performance, reaching an R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> of 0.975 and the lowest error metrics (MSE = 27.308, MAE = 1.335, RMSE = 5.196). Decision Tree and Extra Trees also had competitive results (R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> =0.958 and R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> = 0.955, respectively), while LGBM showed moderate predictive capability. On the other hand, XGBoost presented the lowest performance (R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> = 0.32). These results demonstrate that ensemble methods, particularly Random Forest, are highly effective for the presented screwdriver predictive maintenance scenario. The best trained model was selected to integrate an application programming interface (API) used to predict the preventive maintenance period of screwdrivers.