基于红外热成像图像和轻量级深度CNN的光伏组件故障诊断端到端原型

With the expansion of solar energy in various sectors, effective and reliable fault detection and diagnosis techniques have been developed to ensure the safety and efficiency of solar photovoltaic (PV) installations. Recently, the fault diagnosis of PV plants using infrared thermography (IRT) images has gained significant attention from researchers. Designing an effective and affordable end-to-end prototype to assist operators and maintenance teams in fault diagnosis of PV plants, based on IRT images is extremely needed and remains a key concern within the photovoltaic community. In this study, first we compare a lightweight deep convolutional neural network (DCNN)-based model with five hybrid models to classify the severity of defects in PV modules using captured IRT images. The results show that DCNN, DCNN-SVM, and DCNN-XGBoost provide similar results in terms of classification accuracy (87%), while the other hybrid classifiers offer lower accuracy. Subsequently, we designed an endto-end prototype incorporating an IR camera and a low-cost edge device. Experimental tests were conducted to evaluate the effectiveness of the developed lightweight DCNN in real-time applications. To assess the generalizability of the developed model, the prototype was verified under different climatic conditions. The experimental results demonstrated the prototype's ability to classify defects with good accuracy. The designed prototype serves as a key bridge between laboratory-scale research and industrial PV applications for fault diagnosis, greatly assisting operation and maintenance teams in effectively diagnosing faults in their PV power plants.