journal article·2023-09

Automated classification of electroluminescence images using artificial neural networks in correlation to solar cell performance parameters

Marko Turek, Manuel Meusel

Publisher page

Abstract

The quality control in solar cell production relies on the acquisition of quantitative performance data (I–V-data) together with imaging diagnostics such as electroluminescence (EL) imaging. The classification of the solar cells into different quality groups based on the performance parameters or the identification of specific defects in the EL-images, such as cracks, is rather straightforward and well established. These techniques are extended in our work as we analyze the quantitative correlation between the two approaches. As a specific example, a quality issue related to the contacting of the solar cells during the solar simulator measurement is considered. We demonstrate that a reliable and fully automated classification of the images with respect to the considered quality issue can be achieved by our artificial intelligence (AI) approach. This approach is different from the majority of state-of-the-art approaches as it does not consider specific image objects or features, such as cracks or spots, but rather predicts an image category for the entire image. The performance of the developed supervised image classification approach based on a neural network is analyzed quantitatively. We show that a clear quantitative correlation between cell I–V-data and AI-predicted EL-image classes can be established. In this way, an early-warning procedure can be implemented to detect this performance related issue using the AI-analysis of the EL-images even when its appearance is rather weakly pronounced and before it can be discovered in the I–V-data. We also present results on the impact of the training data set on the performance of the neural network and key parameters describing the quality of the network.
Source: expertdata · Quality: candidate unverified · Source record ↗

Research context

OpenAlex topics: Spectroscopy and Chemometric Analyses · Spectroscopy Techniques in Biomedical and Chemical Research · Industrial Vision Systems and Defect Detection

SciVX discovery tags are terminology-based. OpenAlex topics are source-provided classifications.

References (9)

Show source-provided references
1. Machine-learning techniques and their applications in manufacturing
2. Data mining in manufacturing: a review10.1115/1.2194554 ↗
3. A review of machine learning for the optimization of production processes10.1007/s00170-019-03988-5 ↗
4. End-of-Line binning of full and half-cut cells using deep learning on electroluminescence images
5. Machine learning for optimization of mass-produced industrial silicon solar cells
6. A combined numerical modeling and machine learning approach for optimization of mass-produced industrial solar cells10.1109/JPHOTOV.2020.3004930 ↗
7. Electroluminescence imaging and automatic cell classification in mass production of silicon solar cells
8. Convolutional neural network: a review of models, methodologies and applications to object detection10.1007/s13748-019-00203-0 ↗
9. Improved lab-to-fab solar cell performance assessment by statistical data analysis in an automated, high-throughput metrology line10.1063/5.0141831 ↗