Abstract
Health monitoring of photovoltaic (PV) systems is crucial to maximizing performance and ensuring operational efficiency. Traditionally, this process relies on software as a service (SaaS) solutions such as digital twins, performance monitoring platforms, and asset management software to detect and categorize failures in PV plants. These platforms are essential for managing large-scale PV installations, where manual oversight is impractical. However, current solutions are often constrained by rigid assumptions and threshold-based rules, limiting their ability to accurately detect failures. Additionally, the O&M (Operations & Maintenance) pipeline is fragmented across multiple platforms that have very limited interoperability, requiring different software solutions for various tasks. Many of these tasks still necessitate human intervention, reducing overall efficiency. This study introduces a novel architecture for a fully automated PV health monitoring pipeline using a multi-AI agent system. The proposed framework consists of two main layers: a multi-AI agent system and an autonomous robotics layer. The AI system is composed of different agents working collaboratively to automate the entire O&M process. The failure detection agent identifies failure symptoms, maps them to potential causes, and performs final diagnostics. The inspection agent coordinates field inspections using robotic systems equipped with multi-spectral sensors, ensuring that all necessary data is collected for accurate diagnostics. The monitoring orchestration agent manages workflow and communication between AI agents, optimizing operations and minimizing errors. Finally, the documentation agent records inspection results, diagnostic decisions, and system reports, automating processes such as issuing maintenance tickets, generating performance reports, and placing purchase orders. The second layer of the system, the autonomous robotics layer, enhances real-time, on-demand inspections through field robots equipped with multi-spectral sensors such as visual and infrared cameras. These robots operate based on AI-driven commands, improving the efficiency and accuracy of inspections while reducing the need for human intervention. By integrating AI and robotics, this framework advances the digitalization of PV O&M, automating decision-making processes and streamlining workflows. Preliminary results show that the proposed system significantly enhances the accuracy of failure detection, inspection, classification, and documentation within a unified platform. By drastically reducing the time from failure detection to ticket issuance from days to mere minutes, this approach improves operational efficiency while minimizing human involvement in routine maintenance tasks, paving the way for a fully autonomous PV O&M pipeline.