Recent advances in the application of infrared thermographic imaging and embedded artificial intelligence for fault diagnosis and predictive maintenance of photovoltaic plants: Challenges and future directions

Recently, fault localisation, detection and diagnosis of photovoltaic (PV) plants using infrared (IR) thermographic imaging combined with advanced deep learning (DL) methods have attracted significant interest from researchers and engineers. This paper presents a comprehensive assessment of recent advancements in fault detection, localisation and diagnosis of PV plants through IR thermal images. Available methods are compared with a particular focus on complexity, accuracy, hardware requirement, affordability, and deployment. Special attention is given to preparing the datasets and real-time deployment of DL methods (e.g., Deep Convolutional Neural Networks (DCNN), You Only Look Once (YOLO), and Vision Transformer (ViT)). In addition, deep discussions are provided on case studies involving the real-time implementation of embedded machine learning (TinyML). The paper offers new insights into the real-time inspection and diagnosis of large-scale solar PV plants using TinyML, Internet of Things (IoT) and IR thermal images. Furthermore, a modern monitoring and predictive maintenance method that integrates the concepts of Large Language Models (LLMs), Artificial Intelligence of Things (AIoT), and TinyML into Unmanned Aerial Vehicles is also presented. Proposed cutting-edge solutions through the design of end-to-end devices will help bridge the gap between academic research and industry. Finally, challenges, recommendations and future directions in this field are highlighted. • Conduct an in-depth evaluation of state-of-the-art methods for fault diagnosis in PV plants based on IR images. • Discuss the steps involved in building a high-quality database of IR thermal images. • Introduce the concept of embedded AI for real-time fault diagnosis in PV plants. • Present three real-world case studies of end-to-end prototypes for fault diagnosis in PV plants. • Propose a cutting-edge solution for predictive maintenance of PV plants based on LLMs and AIoT.