Estimating Types of Faults on Plastic Injection Molding Machines from Sensor Data for Predictive Maintenance

Abstract

Fault type detection for the plastic injection molding machines is an important problem in order to take failure-specific actions to prevent any problem in production, hence providing continuity in procurement. In this study, we treat this problem as a multi-class classification task and proposed a novel machine learning model to achieve reliable and accurate results. We applied the Random Forest (RF) and Extreme Gradient Boosting (XGBoost) algorithms with and without SMOTE (Synthetic Minority Over-sampling Technique) to a real-world dataset for predictive maintenance. According to the results, XGBoost performed better than RF. With the combination of SMOTE method, the performances of both methods increased in terms of accuracy. XGBoost with SMOTE outperformed other techniques by achieving about 98% accuracy on average.