Hybrid CNN-Vision Transformer Framework for Explainable Defect Detection in Solar Panel Manufacturing

Abstract

Solar panel manufacturing is becoming increasingly important due to rising global needs for efficient photovoltaic panels with high levels of performance. The quality of such products mostly relies on detecting potential manufacturing defects at an early stage of their development throughout cell manufacturing, connection, lamination, and final assembly processes. Currently existing manufacturing processes are faced with difficulties related to unreliable defect detection, manual inspections, variable production conditions, and inability to analyze complex features of defects in mass production. Such difficulties may lead to increased production losses, material waste, operational expenses, rejections, as well as decreased efficiency and reliability of manufacturing. A novel hybrid explainable deep learning model for automated detection of defects in solar panel manufacturing is proposed to combine the power of convolutional networks for feature extraction, transformers for contextual learning, multimodal information about manufacturing, and explainable artificial intelligence. A complete manufacturing dataset is created in the form of Solar Manufacturing Intelligent Defect Dataset (SMIDD), which consists of visual inspection images and synchronized manufacturing process parameters. Standardized preprocessing helps in improving the quality of data whereas multimodal feature fusion helps in discriminating the good from bad products. Explainability also aids in inspecting and understanding the regions of defects as well as the manufacturing environment that causes them, thus facilitating quality inspection and manufacturing decisions based on reliable quality assessments. Experiments show an improvement in detecting defects, better classification performance, decreased uncertainties, and adaptability in various manufacturing environments. Practical implementation aids in ensuring intelligent quality inspection, efficient manufacturing, less wastage of materials, and efficient production process.