Image-based Fault Detection of Photovoltaic Modules using Machine Learning

Student: Alois Klink

Photovoltaic (PV) solar energy production is the fastest growing form of renewable energy, with forecasts of generation doubling by 2024. This growth is led by a decrease in capital cost, causing the relative cost of PV maintenance to increase, from ~50% of the total cost of UK’s large-scale PV farms in 2019, to ~67.5% in 2030. Because of this, research is being undertaken into optimising solar PV maintenance using automated fault detection using AI. In this work, infrared thermography is used to capture an image of the temperature of a solar PV module. State-of-the-art deep-learning-based image-classification algorithms are then used to detect if there is a fault and the type of the fault providing warning to farm operator.

Our current work includes benchmarking and evaluating different forms of image-classification artificial intelligence on the field of fault detection in photovoltaics. The methods include both more traditional machine learning, such as support vector machines (SVMs) and neural networks, as well as modern deep computer vision algorithms, such as convolutional neural networks (CNNs).

Our latest results show that on our dataset of infrared images, our algorithm performs with similar accuracy to the state-of-the-art, although performing twice as quickly and using half the computer memory.

Publications and Conferences

  • Alois Klink, Abubakr Bahaj, and Patrick James. aloisklink/flirextractor v1.0.0: An efficient GPLv3 Python package for extracting temperature data from FLIR IRT images. Nov. 2019. url: https://eprints.soton.ac.uk/435549/.
  • Alois Klink et al. “Automatic fault detection in infrared thermographic images in photovoltaic arrays using deep convolutional neural networks”. Presented at ICREN 2019. ICREN 2019, Paris, France, Apr. 25, 2019. url: https://icren2019.exordo.com/programme/presentation/154.
  • Alois Klink, Abubakr S. Bahaj, and Patrick A.B. James. “Infrared Thermographic Fault Detection using Deep Convolutional Neural Networks on Building-Mounted Photovoltaics”. Presented at ICEC 2019. ICEC 2019, Southampton, United Kingdom, Jul. 10, 2019.

A VGG-19 Convolutional Neural Network applied on a IRT PV module from Klink’s PhD