Machine learning framework for photovoltaic module defect detection with infrared images

This paper develops an automatic defect detection mechanism using texture feature analysis and supervised machine learning method to classify the failures in photovoltaic (PV) modules. The proposed technique adopts infrared thermography for identifying the anomalies on PV modules, and a fuzzy-based edge detection technique for detecting the orientation of PV modules with anomalies. Further, the gray level co-occurrence matrix is used for extracting texture features of the image. These extracted features are labelled and trained with the support vector machine classifier to classify the failure type in the PV modules. The classifier is trained with 99.9% accuracy and tested with multiple samples for three different scenarios to monitor the defects in modules. The average testing accuracy is 94.4% for all the samples in the testing scenario. The results show the advantage of the developed algorithm with early failure detection to prevent the catastrophes that would happen in the future.