Blog

Working with a team at the University of Southampton researching photovoltaics, we collaborated to create our paper:

“Automatic fault detection in infrared thermographic images in photovoltaic arrays using deep convolutional neural networks.”

The Conference is an international forum that attracts the best articles and experts from all over the world in order to promote excellence.

The work is driven by Alois Klink who is a PhD student at the university. This project uses the learning ability of Artificial Intelligence to automatically identify problems on the cells using drones and software. Then the information is passed on to the manager that can speedily rectify problems. Ultimately, we hope to predict problems before they happen.

Abstract-

Solar photovoltaic (PV) arrays are now becoming more prevalent due to the expansion of renewable electricity supply systems globally. The size and footprint of such arrays will necessitate the use of automated systems to provide an overall solar module health status. The increasing power and decreasing cost of multi-rotor drones and camera technologies in recent years has allowed economical non-destructive status observation of individual modules in a solar farm. Infrared thermography (IRT) can be applied to take an image of the temperature of a PV module, therefore allowing the detection of the shape and efficiency loss of faults. These techniques are far superior to manually detecting and measuring faults in an image, which is labour intensive and requires a lot of time.

Currently, solar power generation supplies 3.4% of the UK’s electricity consumption, with growth rates of 7% in 2017. Such expansion of PV in the UK resulted in many solar farms being installed across all regions. Automating fault detection would allow for lower costs and inefficiencies, making solar power generation more economical. Although there has been some research on using simpler computer vision (CV) in PV fault detection, the recent “deep learning revolution” has created new advanced machine learning techniques, which often achieve higher accuracy than humans in many tasks. However, there is little work done on utilising such techniques for non-destructive detection for faults in PV arrays. This paper outlines the process of creating a suitable dataset for training a machine learning algorithm. Convolutional neural network (CNN), a deep learning technique most commonly used on images, was then employed. The paper discusses the results of applying such techniques to predict faults in PV modules in farms in the UK. Initial results indicate that such techniques have a high predictability of faults and can support operations and maintenance in solar farms. The paper also presents a review of the benefits and disadvantages of using these techniques in predicting PV modules faults.