PILOT: Physics-Informed Optimized LightGBM for Edge-Based Predictive Maintenance
In Industry 4.0 applications, predictive maintenance is one of the frequently used methods for optimizing equipment reliability and maintenance costs. This paper proposes a new framework for edge-based machine fault detection, Physics-Informed Optimized LightGBM for Edge-Based Predictive Maintenance (PILOT), using a public dataset, AI4I 2020. The approach derives physics-informed interaction features from sensor data and uses Optuna to optimize the LightGBM hyperparameters. To ensure a fair comparison, the proposed model is compared with a known reference study from the literature. Experimental results show that the proposed approach provides a relative improvement of 0.51% in accuracy, 7.52% in precision, 11.12% in recall, and 9.40% in F1 score compared to the basic LightGBM model. To test the suitability of deployment for edge environments, the compared models were deployed on a Raspberry Pi 4 device. The results confirm the suitability of the proposed model for edge environments with a 5ms prediction time, even though it requires additional training and prediction time compared to other models.
Authors
- Muhammed Golec (ORCID: https://orcid.org/0000-0003-0146-9735)
Institutions
- Boğaziçi University (TR)
Publication Details
- Journal
- Black Sea Journal of Engineering and Science
- Published
- 2026-11-15
- DOI
- https://doi.org/10.34248/bsengineering.2019354
- Primary Topic
- Machine Fault Diagnosis Techniques
- Type
- article
- Field-Weighted Citation Impact
- 0.00