Model-Agnostic Strategies to Align AI with Real-World Operational Goals in Predictive Maintenance

University Of Florida, Gainesville FL

Investigators

Abstract

This research project focuses on aligning artificial intelligence models with real-world operational goals in predictive maintenance for manufacturing systems. Predictive maintenance plays a crucial role in manufacturing by optimizing equipment reliability, minimizing downtime, and reducing costs, yet current artificial intelligence-driven solutions often underperform in practice due to a misalignment between how models are trained and how maintenance decisions are made. Many artificial intelligence models rely on standard statistical metrics that do not directly reflect maintenance objectives, leading to costly and inefficient decision-making. This research looks to introduce a systematic approach that ensures artificial intelligence models are designed to improve actual maintenance outcomes, not just predictive accuracy. By embedding operational objectives and constraints directly into artificial intelligence model training, the project seeks to enable manufacturers to deploy predictive maintenance solutions that are both effective and practical. The findings from this research look to support the broader adoption of artificial intelligence in predictive maintenance for manufacturing systems, contributing to national economic prosperity and the progress of science and engineering. Additionally, the project seeks to develop educational materials and outreach programs to integrate artificial intelligence alignment concepts into engineering curricula and prepare future professionals for artificial intelligence-driven industries. This research aims to establish model-agnostic strategies that improve the decision-making impact of predictive maintenance models in manufacturing. The objectives include: (i) designing machine learning models that explicitly incorporate maintenance cost and operational constraints into predictive modeling; (ii) extending the unit-level prognostics to fleet-level maintenance decision-making through Bayesian optimization to enhance scalability and adaptability; and (iii) assessing the framework’s effectiveness under various industrial settings using real-world and simulated manufacturing datasets. The project seeks to answer fundamental questions such as: (1) how can artificial intelligence models be trained to optimize for real-world operational objectives rather than simple proxy metrics, such as average prediction accuracy? and (2) how can unit-level predictions be integrated into fleet-level maintenance decision-making? By addressing these questions, the project strives to advance artificial intelligence alignment for decision-making in predictive maintenance and contribute to the broader field of artificial intelligence for manufacturing applications. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.