Abstract

Predictive maintenance is considered a powerful practice for manufacturing assets health assessment, facilitating the identification of potential failure occurrences. By proactively addressing such failures, manufacturers can avoid unplanned downtime and allocate necessary resources for required maintenance activities. Machine Learning (ML) methods have emerged as a promising tool for preventing equipment failures in Predictive Maintenance applications. However, the effectiveness of Predictive Maintenance applications is largely determined by the Machine Learning techniques utilized and the quality of the data utilised. In this research, we adapted the cross-industry standard process for data mining to develop a predictive maintenance model for a unique, large, and complex manufacturing asset, utilizing various machine learning techniques. Specifically, the research incorporate Random Forest, Support Vector Machines, K-Nearest Neighbors, eXtreme Gradient Boost, and Logistic Regression algorithms to the asset failure records. Following the fitting of all models, Random Forest emerged as the best-performing model based on the recall parameter. However, the algorithm performance was not satisfactory due to the poor data quality. In addition, an exploratory data analysis process was conducted on the data to derive insights into the failure pattern of the machine.