Machine Learning-Based Predictive Maintenance and Risk-Aware Production Scheduling
5th International Ege Congress on Scientific Research, Muğla, Türkiye, 22 - 23 Temmuz 2026, ss.16-25, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Basıldığı Şehir: Muğla
- Basıldığı Ülke: Türkiye
- Sayfa Sayıları: ss.16-25
- İstanbul Üniversitesi-Cerrahpaşa Adresli: Evet
Özet
Unplanned machine failures in production systems constitute a significant operational problem, leading to delivery delays, capacity losses, and increased operating costs. This study proposes a machine learning-based hybrid approach that integrates predictive maintenance with production scheduling decisions. The main objective of the study is to generate production plans with lower failure risk by incorporating machine failure risks predicted through machine learning into the job shop scheduling problem. Accordingly, in the first stage, machine failure probabilities are predicted using the XGBoost model, and model performance is evaluated using metrics such as accuracy, ROC-AUC, F1 score, MAE, and RMSE. The resulting probabilities are transformed into machine-time-based risk scores and incorporated as inputs into the flexible job shop scheduling (FJSP) model; the model is tested under λ scenarios representing low, medium, and high risk penalties (λ = 0, 1, 5, 20). Accordingly, when transitioning from λ = 0 to λ = 1, risk exposure decreases by 83.6%, while the increase in makespan remains limited to only 12.1%. In the same transition, the utilization of the highest-risk machine decreases by 76.9%, and this reduction is not limited to a single machine but extends across the entire instance in proportion to the risk ranking. The findings demonstrate that the proposed hybrid decision model provides a measurable and balanced trade-off between production time and failure risk.