Micro-Fracture Detection in Photovoltaic Cells with Hardware-Constrained Devices and Computer Vision

TL;DR

This work tackles the challenge of detecting micro-fractures in crystalline silicon photovoltaic cells using computer vision under hardware constraints. It proposes a hardware-aware ML pipeline evaluated across three deployment scenarios (unconstrained, edge, and drone-mounted microcontroller) and tests three architectural families: InceptionV3, a quantized EfficientNet-B0, and VGG16-block-based small CNNs. Through a 2,624-image EL-based dataset with mono- and poly-Si modules, the study demonstrates that larger models achieve >90% accuracy in unconstrained settings, while compressed models reach ~85% on edge devices and ~82% on microcontrollers, highlighting the trade-offs between accuracy, memory, and transmission needs. The findings underscore the feasibility of autonomous UAV- and edge-assisted PV inspection, while emphasizing the need for tailored tiny architectures and hardware-aware training to enable practical deployment at scale.

Abstract

Solar energy is rapidly becoming a robust renewable energy source to conventional finite resources such as fossil fuels. It is harvested using interconnected photovoltaic panels, typically built with crystalline silicon cells, i.e. semiconducting materials that convert effectively the solar radiation into electricity. However, crystalline silicon is fragile and vulnerable to cracking over time or in predictive maintenance tasks, which can lead to electric isolation of parts of the solar cell and even failure, thus affecting the panel performance and reducing electricity generation. This work aims to developing a system for detecting cell cracks in solar panels to anticipate and alaert of a potential failure of the photovoltaic system by using computer vision techniques. Three scenarios are defined where these techniques will bring value. In scenario A, images are taken manually and the system detecting failures in the solar cells is not subject to any computationa constraints. In scenario B, an Edge device is placed near the solar farm, able to make inferences. Finally, in scenario C, a small microcontroller is placed in a drone flying over the solar farm and making inferences about the solar cells' states. Three different architectures are found the most suitable solutions, one for each scenario, namely the InceptionV3 model, an EfficientNetB0 model shrunk into full integer quantization, and a customized CNN architechture built with VGG16 blocks.

Figures (8)

• Figure 1: Electroluminiscence image of a single mono-Si solar cell containing type A, B, and C cracks.
• Figure 2: Novel ML pipeline to detect micro-fractures in solar cells.
• Figure 3: Example images of each label. 0.0 means a fully functional solar cell, 1.0 is a wholly damaged solar cell, and 0.3 and 0.6 are not confident because the expert had doubts about annotating data.
• Figure 4: Environments' configurations to detect cracks in solar cells in a real-world setting.
• Figure 5: Classification accuracy of DL models set in environment A with second model configuration.