Agentic AI for autonomous preventive maintenance policy governance: a multi-agent framework for dynamic industrial environments

Agentic Artificial Intelligence (Agentic AI) is emerging as a practical paradigm for coordinating autonomous decision workflows in industrial asset management. This paper proposes an event-driven multi-agent architecture for preventive maintenance (PM) policy governance, implementing a closed-loop cycle that ingests maintenance data, re-estimates reliability under right-censoring, optimizes preventive replacement intervals through deterministic cost–time efficiency evaluation, and produces stakeholder-oriented explanations of the selected policy. The framework has been implemented both as an academic prototype in Python, enabling controlled experimentation under censored and non-stationary conditions, and as a realistic industrial architecture, where the agentic layer operates as a supervisory system integrated with enterprise maintenance platforms. Specialized agents (scenario/data, Weibull fitting, numerical optimization, orchestration, and explanation) interact through asynchronous JSON messaging, ensuring traceability and auditability. Importantly, the policy decision is computed exclusively through statistical estimation and numerical optimization, while the large language model is strictly confined to explainability and governance, translating quantitative evidence into human-readable justifications. The results show that, under changing failure regimes and cost structures, preventive maintenance intervals can be autonomously revised while preserving transparency and reproducibility, supporting the feasibility of Agentic AI for continuous PM policy management in Maintenance 4.0 environments. The contribution of this work is architectural and procedural: it demonstrates how preventive maintenance policy governance can be operationalized as an autonomous, and explainable decision process, without altering existing reliability estimation or optimization methods