Industrial data-driven machine learning framework for wafer quality-based decision making toward smart solar-cell manufacturing

초록

The manufacturing industry has evolved rapidly through mechanization, electrification, automation, and digitalization, and is now poised for another leap driven by artificial intelligence (AI). However, despite this progress, research on applying AI to photovoltaic manufacturing remains limited. To address this gap, this study proposes a machine-learning-based framework for smart automation in photovoltaic manufacturing, demonstrated using over 100,000 solar cell data points collected from an industrial manufacturing line. A machine-learning model was trained to predict solar cell efficiency solely from wafer quality. Among the evaluated models, the extra trees (ET) model achieves the highest predictive performance with R2 = 0.936, MAPE = 0.109 %, and RMSE = 0.0259 %, while maintaining high computational efficiency. This level of accuracy can enable reliable wafer screening prior to subsequent cell fabrication processes. Building on the trained ET model, tree-structured Parzen estimator (TPE) optimization was utilized to identify wafer-specific optimal process equipment paths, which are referred to as golden paths. This approach improves normalized efficiency by 0.0681 for samples in the bottom 0.25 % efficiency range. Feature importance and SHapley Additive exPlanations (SHAP) analyses are applied to enhance interpretability, providing intuitive insights into process behavior and feature–performance relationships. Overall, the proposed wafer screening and golden path-based process optimization methods enable machine-learning-level decision making without human intervention, positioning them as key enablers of intelligent automation, smart factory implementation, and Industry 4.0 in photovoltaic manufacturing. Integrating interpretable AI-driven process analysis enhances operator understanding and engagement, supporting the human-centric manufacturing paradigm envisioned in Industry 5.0. © 2026 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license. http://creativecommons.org/licenses/by/4.0/

키워드