Defect Detection Methods in Photovoltaic Systems

Defect detection in photovoltaic (PV) systems is critical to maintaining energy-conversion efficiency, safety and longevity of solar installations. Common defects include micro-cracks, hot spots, delamination and soiling, which can be localised through diverse imaging modalities such as electroluminescence (EL), infrared thermal and high-resolution visible-light capture. Traditional approaches rely on manual inspection or classical image-processing techniques that segment and threshold based on contrast and edge features. Over the past decade, deep learning and advanced computer-vision architectures have transformed the field, enabling end-to-end object detection and classification of multiple defect types with high throughput. Lightweight convolutional neural networks (CNNs) accelerate on-site analysis, while attention mechanisms and customised loss functions enhance sensitivity to small or visually subtle faults. Drone-mounted cameras and infrared scanners extend inspection range across utility-scale PV farms, providing rapid mapping of underperforming modules. Emerging hybrid methods fuse multispectral data to increase robustness under varying environmental conditions. By integrating these technologies, operators can detect faults earlier, schedule targeted maintenance and ultimately improve global photovoltaic yield and reliability.

Research from Nature Portfolio

Recent studies have introduced an enhanced object-detection framework for PV module inspection that simultaneously improves detection speed and accuracy. A modified VarifocalNet architecture replaces key bottleneck modules in the backbone with combinations of standard and dilated convolutions, expanding the receptive field without enlarging feature maps. A feature interactor in the detection head strengthens representation in the classification branch, while an improved intersection-over-union loss function refines bounding-box regression. Benchmarking on large EL imaging datasets demonstrates this method achieves the highest defect localisation precision among competing approaches, with real-time processing capability suitable for inline quality control and field diagnostics.

Defect Detection Methods in Photovoltaic Systems publication trend

The graph below shows the total number of articles in defect detection methods in photovoltaic systems across all publications each year (not limited to Nature Index journals).

Technical terms

Electroluminescence (EL) imaging: A technique where a reverse bias is applied to a PV module to produce luminescence, revealing internal micro-cracks and shunts.

Convolutional neural network (CNN): A class of deep learning model that uses convolutional layers to automatically extract spatial features from images.

Attention mechanism: A neural network component that weights feature maps or channels to focus on the most informative parts of an input.

Mean average precision (mAP): A standard metric in object detection that measures the precision–recall trade-off averaged across classes and thresholds.

Intersection over union (IoU): A metric for evaluating bounding-box overlap, defined as the area of intersection divided by the area of union between predicted and ground-truth boxes.

References

1. Defect detection of photovoltaic modules based on improved VarifocalNet. Scientific Reports (2024).
2. Anomaly Detection Algorithm for Photovoltaic Cells Based on Lightweight Multi-Channel Spatial Attention Mechanism. Energies (2023).
3. PV-YOLO: Lightweight YOLO for Photovoltaic Panel Fault Detection. IEEE Access (2023).
4. ST-YOLO: A defect detection method for photovoltaic modules based on infrared thermal imaging and machine vision technology. PLOS ONE (2024).