Fraunhofer Institute for Casting, Composite and Processing Technology IGCV
Fraunhofer Institute for Casting, Composite and Processing Technology IGCV

TrustME – Certifiable AI applications in aviation production systems

Trustworthy AI for efficient and flexible aircraft production

Artificial intelligence offers great potential for making production processes in aviation safer, more efficient, and more flexible—for example, through predictive quality forecasts or automated image analysis. At the same time, the approval and use of AI in safety-critical environments is heavily regulated.

 

Challenges

To date, there are no standardized procedures for certifying AI applications in aviation production. Black box models are difficult to understand, data is often unstructured, and there are hardly any proven concepts for integrating AI into existing assembly and testing processes in a trustworthy manner.

 

Target

The project addresses aircraft manufacturers and suppliers, production planners, quality managers, automation and IT departments, as well as researchers and certification bodies who want to develop, evaluate, or use AI solutions in production.

 

Objectives of the TrustME project

TrustME develops architectures, methods, and tools for certifiable and trustworthy AI in aviation production. These include predictive quality models, AI-based image processing, ontologies, generative AI (LLM, RAG), and multi-agent systems. The approaches are being tested in realistic laboratory environments in Hamburg and Augsburg and aligned with regulatory requirements such as the EU AI Act and EASA Roadmap.

Certifiable AI as the key to tomorrow's aviation production

TrustME develops methods and architectures to make AI applications in aviation production certifiable and trustworthy. Based on guidelines such as the EU AI Act and EASA Roadmap, a framework is being created that defines technical, regulatory, and documentary requirements for AI systems in manufacturing.

Data-based quality forecasts and optical process monitoring
Predictive quality models are being developed for robot-based assembly processes that predict drilling quality from process data and specifically reduce testing effort. AI-based image processing detects faulty sealant applications; synthetic training data and GANs are intended to significantly increase detection accuracy while reducing data acquisition.

Semantic data models, ontologies, and generative AI in manufacturing
A PPR ontology (product–process–resource) maps production knowledge in a structured and machine-readable way and serves as a single source of truth for all partners. Retrieval-augmented generation links ontologies and knowledge graphs with large language models to answer domain-specific questions in a comprehensible manner and make AI decisions more explainable.

Multi-agent systems for planning, control, and assistance in production
TrustME is developing an LLM-based multi-agent system that supports production planning and execution. Using function calling, AI agents access planning tools, databases, and standardized interfaces to production resources. Voice-based assistance systems link quality forecasts, image processing, and ontologies, making it easier for non-experts to use AI.

Validation on demonstrators in Hamburg and Augsburg
All methods are tested in realistic laboratory environments: in Hamburg on an automated fuselage assembly line, and in Augsburg on a robot-based component assembly line. There, AI models for quality prediction, image processing, assistance, and robot accuracy are integrated, evaluated, and further developed with regard to certifiability and industrial scalability.