To main content

Automatic solar cell diagnosis and treatment

Abstract

Solar cells represent one of the most important sources of clean energy in modern societies. Solar cell manufacturing is a delicate process that often introduces defects that reduce cell efficiency or compromise durability. Current inspection systems detect and discard faulty cells, wasting a significant percentage of resources. We introduce Cell Doctor, a new inspection system that uses state of the art techniques to locate and classify defects in solar cells and performs a diagnostic and treatment process to isolate or eliminate the defects. Cell Doctor uses a fully automatic process that can be included in a manufacturing line. Incoming solar cells are first moved with a robotic arm to an Electroluminescence diagnostic station, where they are imaged and analysed with a set of Gabor filters, a Principal Component Analysis technique, a Random Forest classifier and different image processing techniques to detect possible defects in the surface of the cell. After the diagnosis, a laser station performs an isolation or cutting process depending on the detected defects. In a final stage, the solar cells are characterised in terms of their I–V Curve and I–V Parameters, in a Solar Simulator station. We validated and tested Cell Doctor with a labelled dataset of images of monocrystalline silicon cells, obtaining an accuracy and recall above 90% for Cracks, Area Defects and Finger interruptions; and precision values of 77% for Finger Interruptions and above 90% for Cracks and Area Defects. Which allows Cell Doctor to diagnose and repair solar cells in an industrial environment in a fully automatic way.
Read the publication

Category

Academic article

Language

English

Author(s)

  • Alvaro Rodriguez
  • Carlos Gonzalez
  • Andres Fernandez
  • Francisco Rodriguez
  • Tamara Delgado
  • Martin Bellmann

Affiliation

  • SINTEF Industry / Sustainable Energy Technology
  • Spain
  • Universidade da Coruña

Year

2020

Published in

Journal of Intelligent Manufacturing

ISSN

0956-5515

Volume

32

Page(s)

1163 - 1172

View this publication at Norwegian Research Information Repository