Integrative deep learning architectures and convolutional neural networks for advanced fault classification in photovoltaic modules

The growing adoption of photovoltaic (PV) systems emphasizes the demand for effective fault detection approaches for maintaining the system’s performance. Conventional methods, like electroluminescence imaging and infrared thermography, usually need manual intervention and are less suitable for large-rated and real-time fault studies. Hence, deep learning techniques, especially convolutional neural networks (CNN), have been proposed and confirmed to efficiently automate fault detection by preprocessing images and determining patterns associated with defects like cracks, hotspots, and soiling. In this paper, we have reviewed around 125 research papers, the various fault detection and classification methods based on generalized CNNs, advanced CNN architectures, transfer learning, generative adversarial networks, support vector machine, YOLO-based, advanced image processing, feature extraction, lightweight CNN, multi-attention and ensembling to handle data imbalance, and real-time detection, navigating them suitable for large-rated PV farm monitoring. Some benchmark datasets and the proper deep learning model selection for optimized PV fault detection for a specific application context is discussed in detail. Despite advancements, practical drawbacks and challenges, such as unbalanced datasets, massive computations, and the necessity for lightweight architectures, have also been studied in detail. This study presents a practical feasibility of Deep learning-based hardware accelerator for VGG16 for real-time solar fault detection, optimizing throughput, memory, and scalability using drone-captured IR images. The paper concludes by providing future research directions on real-time deployment, combining IoT-based monitoring with cutting-edge lightweight CNN models to improve the expandability and efficiency of solar fault detection systems. • Comprehensive Survey of Deep Learning Models for PV Fault Diagnosis. • Focus on Real-Time and Scalable PV Monitoring. • Practical Implementation Using VGG16-Based Hardware Accelerator. • Addressing Data Imbalance and Model Robustness. • Future Vision: IoT-Enabled Smart PV Diagnostics.