Traceable AI for the aerospace industry of the future
TrustME
Certifiable AI applications in aerospace production systems
Symbolic picture
Credit:
AI generated (llava)
Holistic deployment of trustworthy AI in aviation Modern AI algorithms have the potential to significantly improve efficiency in aerospace manufacturing. AI-powered prediction and automated inspection can reduce energy, time and material costs in manufacturing processes considerably. However, the transition from research to industrial practice requires AI models, training data and development processes to fully meet the aviation industry’s stringent verification, validation and certifiability requirements – requirements which have so far hindered the widespread adoption of AI in this sector. This is where the TrustME research project comes in. The project aims to make AI in aerospace production powerful and demonstrably safe, making it eligible for regulatory approval. At the heart of the project is a central AI platform that consistently ensures quality assurance, data provenance, and traceability of AI workflows throughout the entire development process, ensuring compliance with the EU AI Act and the certification requirements of the European Aviation Safety Agency (EASA).
TrustME's central AI platform with integrated quality assurance To overcome the issue of fragmented, individual solutions, TrustME is developing a reusable platform that covers the entire lifecycle of AI applications, from data acquisition and model development to validation and certification. The platform incorporates mechanisms for monitoring data quality and provenance, and for consistently versioning models and datasets. Modular components for automated verification and validation enable the systematic checking of AI models against legal and regulatory requirements. Automated report generation provides comprehensive documentation of the entire development process in a format suitable for auditing.
Contribution from the Institute for AI Safety and Security The Institute for AI Safety and Security is contributing its expertise in secure, standards-compliant AI methods to the project consortium. The Institute's contributions include helping to design the overall system architecture and defining the requirements for verification and validation (V&V) processes. Specifically, the Institute is developing modules for evaluating training results, generating automated reports and automating processes in the V&V domain. Additionally, new metrics for evaluating AI models are being developed, particularly with regard to learning guarantees, robustness, and stability in transfer learning and domain adaptation scenarios.
External partners involved The Institute collaborates within a consortium that brings together leading industrial and academic partners: Airbus Operations GmbH and Airbus Aerostructures GmbH as key representatives of the aviation industry, Fraunhofer-Gesellschaft e.V., as well as Helmut Schmidt University Hamburg and Hamburg University of Technology. The close integration of industry, universities and research institutions ensures that the project results are practical and industrially scalable. The developed platform is not exclusively geared towards aviation; it is intended to serve as a transferable basis for the deployment of certifiable AI in other safety-critical sectors.
