On Application of Machine Learning Models for Prediction of Failures in Production Lines

Abstract

Fault detection of production lines is an important part of the manufacturing process. Production line monitoring focuses at detecting emerging faults at an early stage to make better maintenance. Therefore, at the present time the prediction of faults of production lines is a highly topical issue. The machine learning methods are an effective tool for solving this problem. The paper deals with the problem of predicting the failure of production lines using methods of extreme gradient boosting, random forest and light gradient boosting. Extreme Gradient Boosting presents an ensemble of weak models, which are transformed into a high performance model with the help combination algorithm. Light gradient boosting build the decision tree vertically while other boosting algorithms grow trees horizontally. Light gradient boosting applies the leaf-wise strategy while level-wise strategy is used by extreme gradient boosting. The main goal is to compare the prediction results obtained by different methods. The extreme gradient boosting and light gradient boosting demonstrate the results with the best performance.