Towards Supporting Non-Destructive Testing Inspectors with a Multi-Modal Generative AI Copilot

Non-destructive testing (NDT) involves complex processes that are subject to numerous standards and expert knowledge. In addition, inspectors must handle heterogeneous inspection data, while existing digital tools typically support only isolated tasks, which motivates the need for an integrated and context-aware support system. To support inspectors through planning, analysis, and documentation in NDT, this work presents a concept of a multi-modal generative AI copilot tailored to computed tomography (CT) NDT inspections. Through workshops with related-industrial stakeholders and analyzing state-of-the-art literature, requirements and potentials of the generative AI copilot were derived and assessed. Based on the requirements an agent-based system architecture combining language processing and computer vision components was conceptualized and demonstrated in a prototype. Initial feedback from industry partners confirmed the relevance of the concept, verified that their requirements had been met, and provided information on further improvements and suggestions.