Description of field of research:

Accurately identifying defects in solar modules from luminescence images is currently a manual process and requires experienced domain experts. This approach is time-consuming, prone to error, and does not sufficiently scale to that required by the growing photovoltaic energy industry.

This project will use machine learning models (generative representation learning) to identify and classify visible defects in luminescence images of photovoltaic modules. this will enable a robust, scalable, and automated methodology to monitor and accurately identify defective modules, such that they can be replaced, and either re-used, repaired or recycled.

Through this research project you will gain an understanding of common defects in solar cells and modules, and how these present themselves in luminescence images, as well as develop valuable skills regarding image data analysis and the development and training of machine learning models

School

Photovoltaic and Renewable Energy Engineering

Research areas

Photovoltaic characterisation, Machine learning

Computer-based, high-performance compute server environment

  • Prepared training datasets (cell luminescence images)
  • Performance validation (statistical assessment) and visual examples of model output (generated solar cell luminescence images) for each configuration of learning model
  • Analysis of the influence of varying training dataset, model architecture, and learning methodology