Abstract
Defect detection systems that are automated are increasingly mixing machine learning (ML) with big language models (LLMs) to enhance accuracy and interpretability in industrial quality control. Albeit ML methods such as convolutional neural networks (CNN) and image transformers are powerful in abnormality recognition for visual data, the black-box nature of their models often limits transparency and user trust. Combining the LLMs with explainable AI (XAI) techniques such as post-hoc methods like Grad-CAM and SHAP makes it possible for these systems to produce human-understandable stories about reports on defects as well as offer reasons why they happened. LLMs can also be used to not only classify defects but also provide real-time correction advice while avoiding slowing down operation during transmission of data. By this combined method introduction, not only does defect detection performance increase but it also supports regulatory compliance, and boosts user confidence.