Exploring insights on deep learning-based photovoltaic fault detection for monofacial and bifacial modules using thermography

Abstract

Routine maintenance of photovoltaic (PV) power plants is critical to mitigate module faults, which can result from environmental factors, reducing power output and accelerating module degradation. To effectively detect faults across the entire PV module array, aerial infrared thermography (AIRT) is employed, using unmanned aerial vehicles (UAVs) to capture thermal images via predetermined waypoints. Afterward, these images are analyzed by a deep learning (DL) model known for its objec detection accuracy, identifying modules requiring further inspection. While prior research has focused on monofacial modules, limited studies have examined bifacial modules, which are rapidly gaining market share due to their albedo characteristics that increase energy yields in high-albedo areas. Thus, research on bifacial performance and faults is essential to support the development of PV maintenance systems across diverse environments. This study tests bifacial modules under PV fault conditions using thermography, adhering to established inspection standards, which enables comparative analysis with monofacial modules. Furthermore, our PV fault detection model uses a novel dataset from thermal images of both module types to train a mask region-based convolutional neural network (Mask R-CNN). The experiment demonstrated that, under similar irradiation conditions, bifacial faults exhibit higher temperatures and show distinct surface patterns in their thermal images. Despite these variations, our model detected PV faults in both module types, achieving a mean average precision (mAP) of 84.27 %. The model's performance could be further enhanced by expanding the bifacial dataset to address challenges in detecting soiling defects, which vary in shape, size, and location.