SparkNet–A Solar Panel Fault Detection Deep Learning Model
G. Rohith, Dikshithula Sai Manish, Rahul Narasimhan Arunagirinathan, Anurag Uttam Dhavale, Rohan Reji John · IEEE Access · 2025
Solar power is a clean, renewable energy source with minimal greenhouse gas emissions, combating climate change and increasing energy self-sufficiency. Early fault detection like Shading, cracking, or electrical malfunctions is crucial for maximum efficiency and system failure prevention. This work presents SparkNet, a deep learning model developed for the identification of faults due to contaminants on the surface of solar panels. SparkNet uses Fire Modules architecture, which comprises Squeeze and Expand layers. It is used to extract features and detect panel surface abnormalities, trained on a comprehensive dataset of clean and faulty solar panels under various environmental conditions, and improve computational efficiency by minimizing channel sizes while preserving rich feature representations. SparkNet, when implemented with a dataset of images of solar panels and different weather conditions, achieves an average of 95% in the quantitative performance metrics, including accuracy, precision, recall, and F1 score, better than the other state-of-the-art models. Through comprehensive ablation experiments, SparkNet features are added with best-in-class machine learning (ML) classifiers to test robustness in detecting faults, and results confirm the effectiveness of SparkNet features in detecting accurate fault detection without the use of ML classifiers.