A wide range of defects, failures, and degradation can develop at different stages in the lifetime of photovoltaic modules. To accurately assess their effect on the module performance, these failures need to be quantified. Electroluminescence (EL) imaging is a powerful diagnostic method, providing high spatial resolution images of solar cells and modules. EL images allow the identification and quantification of different types of failures, including those in high recombination regions, as well as series resistance-related problems. In this study, almost 46,000 EL cell images are extracted from photovoltaic modules with different defects. We present a method that extracts statistical parameters from the histogram of these images and utilizes them as a feature descriptor. Machine learning algorithms are then trained using this descriptor to classify the detected defects into three categories: (i) cracks (Mode B and C), (ii) micro-cracks (Mode A) and finger failures, and (iii) no failures. By comparing the developed methods with the commonly used one, this study demonstrates that the pre-processing of images into a feature vector of statistical parameters provides a higher classification accuracy than would be obtained by raw images alone. The proposed method can autonomously detect cracks and finger failures, enabling outdoor EL inspection using a drone-mounted system for quick assessments of photovoltaic fields.
The performance of high-efficiency silicon solar cells is limited by the presence of bulk defects. Identification of these defects has the potential to improve cell performance and reliability. The impact of bulk defects on minority carrier lifetime is commonly measured using temperature- and injection-dependent lifetime spectroscopy and the defect parameters, such as its energy level and capture cross-section ratio, are usually extracted by fitting the Shockley-Read-Hall equation. We propose an alternative extraction approach by using machine learning trained on more than a million simulated lifetime curves, achieving coefficient of determinations between the true and predicted values of the defect parameters above 99%. In particular, random forest regressors, show that defect energy levels can be predicted with a high precision of ±0.02 eV, 87% of the time. The traditional approach of fitting to the Shockley-Read-Hall equation usually yields two sets of defect parameters, one in each half bandgap. The machine learning model is trained to predict the half bandgap location of the energy level, and successfully overcome the traditional approach’s limitation. The proposed approach is validated using experimental measurements, where the machine learning predicts defect energy level and capture cross-section ratio within the uncertainty range of the traditional fitting method. The successful application of machine learning in the context of bulk defect parameter extraction paves the way to more complex data-driven physical models which have the potential to overcome the limitation of traditional approaches and can be applied to other materials such as perovskite and thin film.
End‐of‐line characterization of solar cells is necessary to filter out defective cells and bin cells to avoid power mismatch loss in photovoltaic modules. Current–voltage testers, used by almost any photovoltaic company and research laboratory, are costly to maintain and to adapt to recent and predicted morphological changes in solar cells: larger and thinner wafers, half or shingled cells, a wide range of busbar layouts, and more. In this study, we challenge this fundamental technique and propose to bin solar cells and detect defective cells based on a deep learning analysis of their electroluminescence images. The use of electroluminescence imaging addresses the above‐mentioned limitations of the current–voltage technique, as well as allowing faster measurements as it avoids any capacitance effects. By introducing LumiNet, a convolutional neural network end‐to‐end framework, solar cell efficiency bins can be accurately predicted from electroluminescence imaging with a mean error similar to that obtained by current–voltage measurements. The proposed framework is validated on several state‐of‐the‐art mono‐crystalline silicon solar cell structures. We show that photovoltaic modules fabricated using the proposed method would have similar mismatch loss as the traditional current–voltage binning. We then demonstrate the method on half‐cut silicon solar cells. Predicting the half‐cut cell efficiencies, from the deep learning framework, enables manufacturers to assess post‐cutting damages and reassess their binning strategy before module assembly. Furthermore, the deep learning framework is shown to work well even on datasets that have not been previously seen. The trained deep learning LumiNet models' structure and weight are shared with the community to accelerate the adaptation of deep learning for luminescence image analysis in the photovoltaic industry.
To keep improving the efficiency-to-cost ratio of photovoltaic solar cells, manufacturing lines must be continuously improved. Efficiency optimization is usually performed process-wise and can be slow and time-consuming. In this study, we propose a machine-learning-based method to perform simultaneous multiprocess optimization. Using the natural variation of a production line, we train machine learning models to investigate the relationship between process parameters and cell efficiency. We employ genetic algorithms to identify new process parameters in order to maximize cell efficiency. The proposed method is demonstrated on a simulated production line of monocrystalline aluminum-back surface field solar cells. Using neural networks, an accurate model is built to predict cell efficiencies from input process parameters with errors of <0.03% absolute efficiency. In five iterations, the mean cell efficiency increases from 18.07% to 19.45%. Provided strong process monitoring and accurate wafer tracking, the proposed method is directly applicable to production-type datasets, enabling the photovoltaic industry to build smart factories and join the fourth industrial revolution.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.