A Hybrid Deep-Machine Learning Approach to Diagnose Partial Shading and Short-Circuit Faults in Solar Photovoltaic Systems using Infrared Thermography Images

Photovoltaic (PV) systems represent a key solution to global energy security, yet the need for reliable fault diagnosis is an essential operational challenge, preventing significant degradation in energy harvesting efficiency and system performance. This research presents a novel hybrid diagnostic system based on transfer learning-based deep feature extraction and ensemble machine learning classification for partial shading (PS) and short-circuit (S-C) detection in PV arrays. Features are extracted from infrared thermography (IRT) images with three state-of-the-art pre-trained CNN architectures (ResNet50, MobileNetV2, and VGG16); then, the feature vector dimensionality is decreased by principal component analysis (PCA) technique and classified by XGBoost and SVM classifiers. Extensive experimental evaluation on a database of 2,249 thermal images reveals that the MobileNetV2-SVM combination performs the best with 97.56% classification accuracy, surpassing other model combinations. The proposed method enables real-time, non-invasive, and very reliable fault detection that can aid predictive maintenance actions that minimize downtime, decrease operational costs, and enhance energy yield in commercial and utility-scale PV power plants.

• photovoltaic fault diagnosis
• partial shading detection
• short circuit detection
• solar photovoltaic array
• infrared thermography
• thermal images
• transfer learning
• MobileNetV2
• XGBoost
• support vector machine
• ensemble classification
• principal component analysis
• predictive maintenance

• 7 — طاقة نظيفة وبأسعار معقولة
• 9 — الصناعة والابتكار والبنية التحتية

• Photovoltaic System Optimization Techniques
• Thermography and Photoacoustic Techniques
• Electrical Fault Detection and Protection
• photovoltaic fault diagnosis
• partial shading detection
• short circuit detection
• solar photovoltaic array
• infrared thermography
• thermal images
• transfer learning
• MobileNetV2
• Renewable Energy, Sustainability and the Environment
• Mechanics of Materials
• Electrical and Electronic Engineering
• Energy
• Engineering