Tuesday, February 18

Scientists from Russia have created a neural network to detect defects in solar panels

Google+ Pinterest LinkedIn Tumblr +

The first tests on open databases showed that the algorithm is able to detect imperfections of solar cells and their possible source with an accuracy of about 90-95%

Russian scientists have created a machine learning system capable of detecting defects in the structure of solar cells. The algorithm is also able to determine the most likely errors in the technological chain of production, leading to defects. The work of the software package was tested in real production, the press service of the AI Institute of Artificial Intelligence reported on Tuesday.

“The result of the joint work of the teams was a successful pilot test of the system at the existing production of the Hevel plant in Novocheboksarsk. When two or more images with defects of the same type with similar localization were detected at the sorting stage, the model informed in real time about the most likely nodes of the plant’s production line equipment associated with the defects found,” the report says.

The algorithm was developed by a research team led by Semyon Budyonny, head of the “Design of New Materials” group of the AI Institute of Artificial Intelligence, in cooperation with specialists of the IT company “Soltech” and the company “Hevel”, specializing in the production of photovoltaic modules.

As noted in the report, manufacturers of solar panels, as well as companies servicing them, often face defects in the manufacture of new solar cells, which occur at different links in the technological chain and significantly reduce the efficiency of manufactured panels. Defects can be detected by images from special cameras capable of tracking the glow emitted by solar cells when exposed to electric fields or current.

The researchers prepared a database of about 68 thousand images obtained during the operation of real industrial enterprises for the production of solar panels, and used it to train the neural network they developed. Her first tests on open databases showed that the algorithm is able to identify defects and their possible source with an accuracy of about 90-95%.

“At the moment, my colleagues and I are conducting additional large-scale data markup to improve the accuracy of localization of defects and, in case of successful completion of the second phase of test tests, we plan to put the software into pilot operation. I hope that our project will accelerate the transition to high-tech methods of optimizing production processes and will become a clear example of how science can help solve real business problems,” Budyonny summed up, whose words are quoted by the press service of the AI Institute of Artificial Intelligence.

In recent years, scientists have made significant progress in the development of artificial intelligence systems and created neural networks capable of performing non-trivial tasks, solving important problems of industrial production and even “thinking” creatively. In particular, now there are neural networks capable of drawing pictures, analyzing texts, conducting scientific experiments and making medical diagnoses.

 

A source: https://e-cis.info/news/569/112200/

Share.

Comments are closed.