Deep learning for photovoltaic defect detection using variational autoencoders

Abstract

Faults arising in photovoltaic (PV) systems can result in major energy loss, system shutdowns, financial loss and safety breaches. It is thus crucial to detect and identify faults to improve the efficiency, reliability, and safety of PV systems. The detection of faults in large PV installations can be a tedious and time consuming undertaking, particularly in large-scale installations. This detection and classification of faults can be achieved using thermal images; use of computer vision can simplify and speed up the fault detection and classification process. However, a challenge often faced in computer vision tasks is the lack of sufficient data to train these models effectively. We propose the use of variational autoencoders (VAEs) as a method to artificially expand the data set in order to improve the classification task in this context. Three convolutional neural network (CNN) architectures – InceptionV3, ResNet50 and Xception – were used for the classification of the images. Our results provide evidence that CNN models can effectively detect and classify PV faults from thermal images and that VAEs provide a viable option in this application, to improve model accuracy when training data are limited.

Significance:

• Faults in PV systems can be labour and time consuming to detect and classify. This process can be automated by using computer vision and thermal images.
• CNN models (InceptionV3, ResNet50 and Xception) are effective in the detection and classification of PV faults from thermal images.
• Small data sets are a common barrier to entry for computer vision assessments. VAEs provide an effective method to artificially expand a limited data set to allow for the successful use of CNN models.
• The expansion of training data using VAEs for CNN models can improve the prediction accuracy in these models.