Year: 2026
Type: article
Abstract: Industrial enterprises generate large volumes of operational data through equipment connectivity and production digitalization. However, a clear gap remains between analytical outputs and maintenance decisions. Failure prediction often stops at risk identification, which limits its direct contribution to resource allocation and scheduling. To address this issue, a data driven decision support framework was developed using the AI4I 2020 Predictive Maintenance Dataset. LightGBM was applied to estimate equipment failure probabilities, while SHAP was used to identify the key operational factors behind the predictions. The predicted risks were then combined with downtime losses, equipment importance, and maintenance hours in a resource constrained scheduling model. The results showed that LightGBM achieved an F1 score of 0.814 ± 0.012 and an ROC AUC of 0.973 ± 0.003, outperforming logistic regression and random forest. Sensitivity analysis further indicated that the total operating cost decreased to 266.6 thousand standardized units when the risk threshold was 0.30 and the daily maintenance capacity was approximately 70 hours. The framework establishes a continuous link between risk prediction and maintenance action. It can support industrial enterprises in improving maintenance priorities, reducing downtime losses, and using limited resources more efficiently.
Author: Yuwei Li
Institution: The University of Queensland
Language: English
Subfield: Safety, Risk, Reliability and Quality
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7225637892
FWCI:
Cites:
Cited by:
Related to: