Artificial Intelligence Backed Automation for Detecting Cracks in Solar Cells

The reliability and performance of photovoltaic (PV) systems are strongly influenced by manufacturing and operational defects that accelerate degradation and reduce energy yield. Conventional inspection methods are often time-consuming and ineffective for detecting early-stage or micro-scale defects. This research presents an automated PV defect detection framework that integrates electroluminescence (EL) imaging with deep learning techniques to improve inspection accuracy and efficiency.

The thesis is structured around five peer-reviewed publications covering key aspects of intelligent PV quality assessment. The early chapters review existing EL-based inspection methods and convolutional neural network (CNN) architectures, identifying limitations that motivate the need for automation. Subsequent chapters propose novel CNN-based models, including a Dual Spin Max-Pooling network and an ensemble architecture designed to enhance classification robustness and feature extraction.

Experimental results demonstrate validation accuracies exceeding 98% in distinguishing healthy and defective PV cells under varying imaging conditions. In addition to cell-level defect classification, the research develops a hybrid inspection framework combining cell-level and module-level analysis to support PV acceptance testing and quality assurance.

The proposed methods are validated using large-scale datasets obtained through in-house EL imaging of commercial PV modules, supported by photoluminescence and thermal measurements. Overall, the research contributes a scalable and data-driven methodology for automated PV defect detection and provides a foundation for integrating artificial intelligence-based inspection technologies into industrial PV manufacturing and maintenance processes.