Agentic AI for Autonomous Fault Diagnosis and Root Cause Analysis in Industrial Maintenance

Abstract

Abstract Industrial maintenance is increasingly challenged by rare event failures and multi causal faults that traditional rule based or passive analytics fail to detect effectively. This paper presents an agentic AI framework for autonomous fault diagnosis and root cause analysis (RCA) in industrial maintenance environments. The Agentic AI module orchestrates multiple tasks involving multiple expert models and agents; (a) the perception agent performs continuous monitoring and detection of faults, (b) reasoning agent performs multi-causal reasoning and root-cause-analysis, and (c) the prescription agent provides actionable preventive and maintenance recommendations. The perception agent employs an Isolation Forest model to identify anomalies in highly imbalanced sensor telemetry data, while the reasoning and prescription agents maps failure signals to root causes and prescriptive actions, respectively. The system is developed using the Maintenance Dataset, which includes sensor telemetry, machine failure indicators, and type specific failure flags. Additionally, synthetic failure sequences generated using a neural generative model (LSTM-VAE) are used to stress test the system. Results demonstrate that the agent successfully identifies multiple concurrent root causes, produces actionable maintenance steps, and generalizes to synthetic failures, establishing a scalable, explainable framework for intelligent industrial maintenance.