Embedded Hybrid Model (CNN–ML) for Fault Diagnosis of Photovoltaic Modules Using Thermographic Images
M. Benghanem, A. Mellit, C. Moussaoui
Abstract
In this paper, a novel hybrid model for the fault diagnosis of photovoltaic (PV) modules was developed. The model combines a convolutional neural network (CNN) with a machine learning (ML) algorithm. A total of seven defects were considered in this study: sand accumulated on PV modules, covered PV modules, cracked PV modules, degradation, dirty PV modules, short-circuited PV modules, and overheated bypass diodes. First, the hybrid CNN–ML has been developed to classify the seven common defects that occur in PV modules. Second, the developed model has been then optimized. Third, the optimized model has been implemented into a microprocessor (Raspberry Pi 4) for real-time application. Finally, a friendly graphical user interface (GUI) has been designed to help users analyze their PV modules. The proposed hybrid model was extensively evaluated by a comprehensive database collected from three regions with different climatic conditions (Mediterranean, arid, and semi-arid climates). Experimental tests showed the feasibility of such an embedded solution in the diagnosis of PV modules. A comparative study with the state-of-the-art models and our model has been also presented in this paper.