Perancangan Model Predictive Maintenance untuk Mesin Filling pada PT. Es Siantar dengan Pendekatan Data Mining

Abstract

High machine downtime can disrupt operational continuity and affect production efficiency. This study was conducted at PT. Es Siantar, which experienced 45 failures with a total downtime of 360 hours on two filling machines during the period from May 2024 to April 2025. The aim of this study is to design a data mining-based predictive maintenance model that can predict failure times and Remaining Useful Life (RUL) of the machines, enabling maintenance to be carried out in a more planned and efficient manner. The methods used include calculating MTTR, MTBF, and applying Random Forest and XGBoost algorithms. The results show that the MTTR for machine FL_1 is 4.25 hours, and the MTBF is 8.11 hours. The Random Forest model was selected as the best model with an R² = 0.9930 and MAE = 0.4156, compared to XGBoost, which resulted in R² = 0.9875 and MAE = 0.5264. The predicted RUL for machine FL_1 is 19.58 days, with the next failure expected to occur on May 14, 2025, while machine FL_2 has an RUL of 65.94 days with the next failure predicted for July 3, 2025. This model provides the company with a stronger foundation for more planned and efficient maintenance scheduling.