Application of machine learning in machine failure prediction

Abstract

Maximizing the life of machine is one of a manufacturer's key concerns. Machine failures might result in production line interruptions and expensive repairs. Therefore, manufacturing industries need solutions to save maintenance expenses, boost equipment uptime, and enhance availability. A more efficient and economical maintenance solution is offered by predictive maintenance. Predictive maintenance is not based on the equipment's real state, which might result in unnecessary and wasteful. Maintenance should be predictable to save time, costs as well as labor. To accomplish this, it is crucial to appropriately classify all machine failure cases in advance. This thesis proposes the Extreme Gradient Boost (XGBoost) model of machine learning combined with Synthetic Minority Oversampling Technique – Tomek data balancing technique to minimize error classification. Additionally, the outcomes are evaluated in comparison to a variety of machine learning approaches, including Gaussian Naive Bayes (GNB), Support Vector Machine (SVM), Logistics Regression (LR). The proposed method achieve a result of 96.6% accuracy and over 50.72 % F1 score, which demonstrating that the suggested strategies outperform other indicated techniques in terms of predictive ability.