Agentic AI Framework for Automated Work Order Generation in Mining Drill-Rig Maintenance

High availability is essential for mining drill-rig operations, and unplanned downtime in critical subsystems directly disrupts production. Although predictive analytics can flag anomalies in streaming sensor data, converting those alerts into actionable, CMMS-ready maintenance work orders remains largely manual. Maintenance planners must interpret alerts, consult documentation, and encode decisions in CMMS with uneven consistency. This paper presents an agentic framework that automates work-order drafting through structured collaboration among four specialised agents: a Workflow Orchestrator, an Anomaly Detection Agent, a Diagnostic Agent, and a Work Order Generation Agent. Diagnostic reasoning is grounded in a domain Knowledge Graph and augmented with Retrieval-Augmented Generation over technical documentation, enabling evidence-linked recommendations. The prototype is evaluated on five representative failure scenarios spanning major drill-rig subsystems, and generated work orders are compared with procedure-derived ground truth using automated metrics (field completeness, diagnosis accuracy, action coverage, part recall). A hydraulic pump seal-wear walkthrough illustrates agent interactions and the resulting structured work order. The framework establishes a traceable and reproducible interface between predictive maintenance outputs and industrial work management systems.

Keywords: Agentic AI, Knowledge Graph, RAG, Predictive Maintenance.