A researcher from Jönköping University in Sweden has developed an innovative machine learning-based method to identify faulty solar panels using infrared thermography. This approach uses a hybrid local features method to overcome issues like scaling, noise, rotation, and haze, achieving impressive results with 98% training accuracy and 96.8% testing accuracy.
The method begins by capturing thermal images of the panels, followed by preprocessing to improve image quality using a dehazing algorithm and contrast adjustments. These images are then divided into 5×5 pixel sub-images, from which local features are extracted. A k-means algorithm reduces the data size, and the model is trained using a Support Vector Machine (SVM) classifier.
High Accuracy and Performance
Tested on a 44.24 kW crystalline silicon rooftop PV system in Lahore, Pakistan, the method achieved 98% training and 96.8% testing accuracy. The model’s precision and recall values were also strong, with 92% precision for faulty panels and 100% recall for healthy panels.
Comparison with Other AI Methods
The new method outperformed most other AI techniques, with only the RB scale-invariant feature transform achieving a slightly higher score (98.66%).
Conclusion
This machine learning approach provides a highly accurate and efficient way to monitor solar panel health, offering a robust solution for predictive maintenance and performance optimization in solar energy systems.”
