Computer Vision-Based PV Module Fault Recognition Using a Transfer Learning Approach

With the continuous increase in the installed capacity of photovoltaic (PV), power generation around the world with its large demand the timely detection of faults and errors in PV modules has become a real challenge. Different methods based on deep neural networks are used for detecting and identifying PV system faults and anomalies. The electric characterization (I–V) and images processing (Infra-red thermal images, electroluminescent images, fluorescent images,…) of the PV modules are the commonly used methods. This study aims to develop a model for fault classification of PV modules using transfer learning approach, the well-known VGG-16 deep neural network was examined. The investigated faults are shadowing, dust deposit, cell cracks, and cell browning applied on the mono-crystal PV module. The dataset was prepared at the Solar Equipment Development Unit (UDES), located in the northern of Algeria, using an HD camera. The experimental results of the developed model for multiclass classification showed an accuracy of 98.90%.