Large Language Model-based defect detection in context of industry 5.0

The transition from manual labor to automation, driven by Industry 4.0 and the emerging Industry 5.0, has significantly transformed manufacturing. However, defects are still unavoidable in manufacturing, especially in the metal domain. To avoid substantial economic and reputational losses, defect detection is a must to guarantee long-term quality and durability of metal products. Despite high accuracy in defect detection, advanced Artificial Intelligence (AI) models like Large Language Models (LLMs) and Multimodal Large Language Models (MLLMs) can optimize production processes, improve efficiency, and lower operational costs. This drives the emergence of AI agents, which are autonomous systems powered by advanced AI models to efficiently perform complex tasks. Thus, this paper is aimed to address the question of how LLMs and MLLMs can enhance defect detection efficiency and interpretability in Industry 5.0 manufacturing systems. To achieve this, a comparative and conceptual analysis is conducted among the studies applying LLMs and MLLMs in a workflow, in terms of data collection, models, and evaluation metrics. In addition, this paper also further studies the practical applications and limitations encountered among the studies.