Akademik Veri Yönetim Sistemi
2025 Innovations in Intelligent Systems and Applications Conference, ASYU 2025, Bursa, Türkiye, 10 - 12 Eylül 2025, (Tam Metin Bildiri)
Predictive maintenance aims to forecast equipment failures before they occur, reducing unplanned downtime and maintenance costs. In this study, we evaluate the effectiveness of several automated machine learning (AutoML) frameworks-AutoKeras, H2O AutoML, rminer, AutoGluon, MLJAR, and TPOT-in modeling the time-to-next-failure of industrial manufacturing machines. Using a one-year failure log data set from five tire manufacturing machines, we engineered temporal features (month, weekday, hour), failure duration, and lagged interarrival intervals. Each AutoML tool performed model selection under comparable time budgets. Models were assessed via mean absolute error (MAE), total training runtime, and consistency of sequential failure-date predictions. When forecasting future failures, all frameworks maintained high date-ordering consistency compared to historical intervals. Our results demonstrate that AutoML can rapidly build highaccuracy, interpretable regression pipelines for predictive maintenance without manual feature-model tuning. We conclude by discussing the potential to incorporate richer sensor and operational data, extend toward classification tasks (e.g., categorizing failure types), and deploy these pipelines in a realworld production environment for continuous monitoring and maintenance scheduling.