Hybrid AI Agentic Framework and Machine Vision for Automated Preliminary Root Cause Analysis in Wind Turbine Alarm Systems Using Multi Modal Data Fusion

Wind turbine systems are frequently subjected to complex operational conditions, leading to frequent quality issues that trigger warnings and alarms, and this demands rapid and accurate quality check with Root Cause Analysis (RCA). Conventional RCA processes rely fully on manual inspection and expert interpretation that involves human, resulting in significant time delays, increased operational costs, and high turbine downtime. To address these challenges, this paper presents a Hybrid AI Agent Framework that integrates Machine Vision and Multi Modal Data Fusion for automated preliminary RCA in wind turbine alarm systems. The proposed system consists of YOLO based machine vision for physical defect detection, advanced thermal image analysis for temperature anomaly identification, and historical SCADA data integration to generate a comprehensive diagnostic context. A novel AI agent-based RCA engine is developed, which synthesizes multi source inputs and produces structured root cause insights using a DMAIC (Define-Measure-Analyse-Improve-Control) framework and that also includes Fishbone and 5-Why quality analytical methodologies from six sigma quality principle. The architecture is implemented using a scalable Fast API based backend, integrated with a user-friendly web interface that has turbine data retrieval, and real time analysis. The system supports both single instance and batch processing, enabling efficient handling of large datasets. Automated alerting mechanisms are incorporated through email notifications, ensuring timely communication of detected anomalies and quality issues. Validation of the system is carried out using synthetic and open-source datasets, demonstrating its capability to accurately identify defects, detect thermal hotspots, and generate meaningful preliminary RCA reports. The proposed framework significantly reduces dependency on manual inspection, minimizes turbine downtime, reduces loss production factor (LPF) and enhances operational efficiency. Overall, the system establishes a scalable and intelligent foundation for future development of fully autonomous wind turbine diagnostics and predictive maintenance solutions.