Machine and Deep Learning Methods for Fault Detection and Classification in Photovoltaic Modules

As the world looks for alternatives to high CO₂-emission electricity production systems, solar energy is seen by many as an attractive and sustainable method to generate electricity. As larger photovoltaic (PV) systems are installed, increased focus has been placed on the maintenance of these installations, which can be both demanding and time-consuming. One method to facilitate the maintenance of such PV installations is through the use of aerial thermal infrared images of the PV modules within the system. These images can be used to identify and classify faulty modules, which can lead to more specific and directed maintenance of the system. This study uses both machine learning and deep learning techniques for the identification and classification of defects and faults in PV systems using aerial thermal infrared imagery. These methods achieve good levels of accuracy in both the identification and classification of faults in these PV modules through the use of feature descriptors, random forests and support vector machines, as well as image augmentation and transfer learning methods. A comparison between the machine learning and deep learning methods in this application also provides valuable insight into the potential benefits and drawbacks of each, when implemented for large PV systems.