This project will develop ways to identify faulty modules earlier, to reduce risk and lower the cost of PV power-plants. This report provides an update on what has been learnt at the mid-point of the project.
Due to the abundance of solar resources, photovoltaic has a key role to play in reducing global carbon emissions. In this project, we use machine learning to fasten the binning process of solar cells, automatize the fault classification to increase the throughput of production lines, and reduce the damp heat test time which otherwise takes 1,000 hours.
Luminescence images are used for the characterization of solar cells. They may also have information about the solar cell electrical parameters. However, there is no easy approach to extract this information. Here, we have used machine learning to extract the solar cell electrical parameters. We found that cell efficiency strongly correlates with image features and the efficiency can be predicted using only luminescence images with an R2 of 0.93 and a root mean square error (RMSE) of 0.10%. Moreover, these images can also be used for fault classification using machine learning. Our approach can predict faults with an F1 score of 98.3% and an accuracy of 99.6%.