Scholarlink

Smart cities are dedicated to mitigating the effects of climate change and reducing their carbon footprint. Solar Photovoltaic (PV) panels generate electricity in an environmentally friendly and sustainable manner, devoid of any emissions of greenhouse gases or contaminants, making them a key component of sustainable energy strategies. Unfortunately, solar panels experience a range of defects over their lifespan, resulting in their reduced performance and overall system output. To identify these defects, it is vital to have human professionals who can examine electroluminescence (EL) images manually, but this method is both time-consuming and expensive. This paper introduces an advanced fault diagnostic technique for solar panels using YOLOv8 and Mobilenet v2 deep learning algorithms. These models are trained on improved and processed EL image datasets over four critical faults. The performance of models is compared and evaluated with two other popular deep neural networks, i.e., InceptionV3 and ResNet50. Results show that MobileNet-v2 provides the highest accuracy of 80.40% and outperforms other algorithms. The results are validated by deploying Mobilenet v2 on edge device NVIDIA Jetson TX2.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deep Learning-Based Fault Diagnosis System for Solar Photovoltaic Panels Using EL Images

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Pramod R. Sonawane,
  • Sheetal U. Bhandari

摘要

Smart cities are dedicated to mitigating the effects of climate change and reducing their carbon footprint. Solar Photovoltaic (PV) panels generate electricity in an environmentally friendly and sustainable manner, devoid of any emissions of greenhouse gases or contaminants, making them a key component of sustainable energy strategies. Unfortunately, solar panels experience a range of defects over their lifespan, resulting in their reduced performance and overall system output. To identify these defects, it is vital to have human professionals who can examine electroluminescence (EL) images manually, but this method is both time-consuming and expensive. This paper introduces an advanced fault diagnostic technique for solar panels using YOLOv8 and Mobilenet v2 deep learning algorithms. These models are trained on improved and processed EL image datasets over four critical faults. The performance of models is compared and evaluated with two other popular deep neural networks, i.e., InceptionV3 and ResNet50. Results show that MobileNet-v2 provides the highest accuracy of 80.40% and outperforms other algorithms. The results are validated by deploying Mobilenet v2 on edge device NVIDIA Jetson TX2.