Description
The project develops an AI assistance system to support defect detection and evaluation in ultrasonic weld inspection. It focuses on human-centred development, high-quality data, DICONDE standards, and future education and training.
Challenge
The project develops an AI assistance system for ultrasonic weld testing. It supports the test personnel in detecting and evaluating defects. In addition to AI development, the project focuses on high-quality data, standardized data formats (DICONDE), human-centred human-AI interaction, and concepts for education and training.
Objective
The project aims to make ultrasonic weld testing more efficient, reliable, and user-friendly. The AI assistance system will support the test personnel in making well-informed decisions and enable the safe, trustworthy, and practical use of artificial intelligence in non-destructive testing. In doing so, it strengthens the competitiveness of industry.
Methods
The project combines ultrasonic testing, simulation, and artificial intelligence. AI models are developed using both real and simulated testing data and integrated through standardized data formats (DICONDE). Requirements for human-centred human-AI interaction are identified together with end users and translated into design recommendations. In addition, concepts for education and training are developed to support the future use of AI-assisted testing systems.
Ultrasonic testing plays an important role in ensuring the safety of welded joints. However, evaluating complex testing data is demanding and time-consuming and requires a high level of expertise and experience. At the same time, the NDT sector faces the challenge of experienced testing personnel leaving the workforce while too few qualified professionals are entering the field. AI-assisted systems offer the potential to support testing personnel in these demanding tasks. UltrasonAIc therefore develops an AI assistance system to support defect detection and evaluation while keeping human decision-making at the centre.
BAM contributes its expertise in three main areas: the generation and use of testing data, the standardization of data formats, and the human-centred development of the AI assistance system. Real ultrasonic testing data are complemented by simulated data to represent different defects and boundary conditions and provide a suitable data basis for AI development. At the same time, the use of the standardized DICONDE data format is being further developed to enable consistent storage and exchange of ultrasonic data and facilitate their use in AI applications.
Another focus of BAM's work is human-AI interaction. Together with testing personnel, the project investigates which tasks can meaningfully be supported by AI and which should remain with the human. It also examines what information the AI needs to provide so that testing personnel can understand and appropriately assess its outputs and make informed decisions. This includes transparency and explainability as well as information on uncertainty and confidence.
The methodology combines technical approaches to modelling and data processing with methods from Human Factors and human-machine interaction research. These include task analyses, interviews and workshops with experts, as well as user-centred evaluations. The resulting requirements and design recommendations are incorporated into the technical development of the assistance system.
In addition, new competency requirements for working with AI-assisted testing systems are identified, and recommendations for future education and training are developed together with the project partners. In this way, the project aims to establish not only the technical foundations but also the conditions for the safe, understandable and accepted use of AI-assisted ultrasonic testing in practice.
Partners
BAM
DGZfP Ausbildung und Training GmbH
Funding
Industrie.IN.NRW – Innovationswettbewerb Industrie im Rahmen des EFRE/JTF-Programms NRW 2021–2027
