Implementation of Convolutional Neural Network in Mobile Applications for Solar Panel Crack and Efficiency Prediction

Solar panels, as a renewable energy source, are susceptible to efficiency degradation due to cracks in solar cells. Manual crack detection has many limitations, while the use of specialized tools like electroluminescence imaging is not economical for small-scale users. Therefore, this research aims to develop an image-based automatic detection system using the Convolutional Neural Network (CNN) method, specifically the YOLOv8 architecture, integrated into a web-based mobile application using the Flask framework. Solar panel image datasets were collected and annotated using Roboflow, then trained in Google Colab with the help of a GPU. The trained model is integrated into a web-based mobile application, allowing users to upload panel images, detect cracked areas, and estimate panel efficiency based on linear regression of the crack area. Testing results show that the system can function in real-time, although the accuracy of efficiency estimation can still be improved due to limitations in data quantity and variation. This research is expected to be an economical and practical solution for solar panel monitoring.