Enhanced Classification of Faults of Photovoltaic Module Through Generative Adversarial Network

Abstract

The faults occurring in the photo voltaic system has to be detected to make it work efficiently .To detect and classify the faults occurring in the photo voltaic module infrared images, electro luminescent images, photo luminescent images of photo voltaic module is used .Using infrared images around 11 faults of photovoltaic module such as cell ,cell-multi, hot-spot-multi , hot-spot, cracking, diode, diode-multi, vegetation, shadowing, off-line module and soiling faults can be detected. In addition to the original infra-red images (IR) available in the IR dataset, the IR images are generated for each and every category of faults by using generative adversarial networks (GAN’s) to increase the dataset size. 45000 images are generated by GAN’s. Later the images are used to train and test the convolution neural network. The dataset visualization of original and that of GAN generated images are done in 2-dimensional space using uniform manifold approximation and projection. In this work 12 categories of IR dataset are considered for classification in which 11 belongs to fault category and the remaining one is the normal category of images. In earlier work only 11 category of faults or less than that is considered for classification. Compared the results with the existing work and it is found that by enhancing the dataset size by GAN’s accuracy of 91.7 % is obtained during the classification of 8 categories of faults.