There is no disputing the potential that AI has for many fields of application and, in particular, for quality control in the context of production. However, it is equally undisputed that there are also some unknowns in the introduction of AI technologies. These can include issues such as regulatory and normative requirements such as the much-discussed "EU AI Act", potential risks in terms of personnel or financial damage, or even the underlying data and its meaningful processing. A particularly important factor is also the people who will have to deal with AI in the future. They should trust it and also be able to help shape it with the help of their comprehensive specialist and domain knowledge.
In order to address precisely these aspects in the introduction of AI in automotive production, Audi commissioned the Fraunhofer Institute for Industrial Engineering IAO with its Research and Innovation Center Cognitive Service Systems KODIS in Heilbronn and the Fraunhofer Institute for Manufacturing Engineering and Automation IPA. During the several months of collaboration at Audi's Neckarsulm site, there were three project stages in which possible hurdles in the introduction of AI and possible solutions for these were developed on the basis of the use case "Welding of car bodies". In essence, the aim was to be able to adequately safeguard and document future AI applications and to involve all parties involved at Audi in the development process.
Promoting a positive attitude toward the new AI technology was particularly important to the project partners on the Audi side. That's why the Fraunhofer experts started by interviewing the staff. They asked specialists in production, quality assurance, IT and management about the potential and wishes they associate with the use of AI. Discussions with external certification bodies supplemented the internal Audi interviews.
In the second project step, resistance spot welding (WPS) was selected as an exemplary application and evaluated with regard to its suitability for AI use. Up to now, quality control has been carried out here under the guidance of a specialist. This means that the specialist inspects spot welds using ultrasound and classifies them as "okay" or "conspicuous. This is a time-consuming process. In the future, this process will be supported by AI by evaluating the quality of the spot welds on the basis of certain process parameters and thus proposing a preselection of conspicuous spots for inspection by ultrasound.
The third and final stage of the project resulted in the development of a company-specific guideline for the successful and secure use of AI at Audi. Its structure is based on the common development process of an AI application, i.e. it goes from the development of the use case to data use and model development to the deployment of the model. "With the guide, we combine current research findings with best practices from AI application development and upcoming legislative and regulatory efforts. In this way, we can provide companies with guidance for the development of reliable AI solutions," explains Prof. Marco Huber, in whose Cyber Cognitive Intelligence department at Fraunhofer IPA the project was worked on. To this end, the authors refer to the EU AI Act, which is available in draft form, the "Guide to the Design of Trusted Artificial Intelligence" of the Fraunhofer Institute for Intelligent Analysis and Information Systems IAIS, and the "Whitepaper Trusted Artificial Intelligence" of TÜV Austria.
Finally, the guide also incorporates feedback from interviews with staff. "For the professionals at Audi, it is crucial that the functionality and security of an AI system are made transparent," says Janika Kutz, head of the "Public Service Innovation" team at Fraunhofer IAO's KODIS and an expert in technology acceptance, summarizing the interview results. Audi thus has a comprehensive recommendation on hand for successfully introducing AI in quality control or even in other production processes. The company is thus ideally positioned to meet the legal requirements for the use of AI applications as soon as they come into force.
The Fraunhofer experts use the general, company-independent findings from the project on the one hand as a basis for further consulting projects around AI for production. In this way, they can create similar guides for companies of all sizes for all AI applications around the production context. In addition, in coordination with Audi, they have also published the project results in a scientific context as a