Egyptian researchers have developed an artificial intelligence (AI) model to detect damage to solar panels used in space. In this study, we analyzed images of solar arrays affected by arcs, a phenomenon in which high voltage solar panels generate electrical discharges when they interact with plasmas in space.
According to the authors, “Arcs can cause serious damage to cell interconnects because peak currents were generated, and thus have a significant impact on the performance and reliability of spacecraft systems.” In this study, arcs were found to occur most frequently at intermediate cells, interconnects, and solar cells where electric fields are the strongest.
Deep learning to analyze arc damage
This study applied deep learning to understand ARC behavior by scientists from the National Institute of Astronomical and Physics and Venice F University. Defective cells were classified and detected from image data using convolutional neural networks (CNNS) and transfer learning.
The team analyzed 2,624 black and white images of solar cells taken from 44 individual modules. Some cells were operating normally, while others showed distinct faults such as cracks and surface contamination from the arc.
They tried two different AI models to see which one is best. Initially, the network built from scratch was almost perfect when looking at the images we had already learned, with 95.98% accuracy. However, when tested with new images, its accuracy fell to 83.24%, suggesting it is not so reliable outside the lab.
The second approach used transfer learning using a pre-trained EfficientV2L model. This version achieved 89.05% verification accuracy and proved better with the handling of new images.
Both models were able to find ARC-related damage in the solar cell parts where discharge is most likely, especially those in interconnects, edges, and mid-cell regions.
Next Steps
Researchers concluded that deep learning is an effective approach to identifying arc damage in solar panels. They stated: “This work provides valuable insight into image processing and analysis, and provides further application recommendations for artificial intelligence (AI) in the engineering and aerospace industries. This study could enhance understanding of arc processes, improve predictive capabilities of AI models, and support the design of more robust solar array systems for space application.”
Future research includes simulations that predict the behavior of ARC events using machine learning techniques applied to scenarios that include “arc currents, potentials, and flashovers of solar arrays.”
