Implementation of a Photovoltaic Default Detection System using Deep Learning

In this paper, we propose a very compact implementation of real-time default detection in photovoltaic (PV) based on deep learning. The goal is to propose an innovative solution for PV fault detection to enhance the reliability and performance of photovoltaic installations. This involves early detection of faults and proactive maintenance of equipment. The dataset was created using Roboflow software, which allows to efficiently edit and manipulate the images. Furthermore, we used the YOLOv8 (You Only Look Once) model for real-time detection, which it offers good accuracy while being fast. In fact, we implemented our solution on the Raspberry PI board to create a real-time and portable system. Such implementation results demonstrated the effectiveness of our system to detect PV default using various evaluation metrics, and its ability to operate in real-time.