This project is a collaboration of the BMW AG and the University Bamberg. We research novel AI-driven approaches to root-cause detection in the production chain of electric powertrains. Our aim is to combine machine learning methods and symbolic AI to incorporate interactively expert knowledge into data-driven anomaly detection methods.
A human-in-the-loop approach to machine learning shall enable the identification of more complex cause-effect relationships for product quality that go beyond current approaches of purely statistical anomaly detection in individual process sections (data silos). Through explicit use of expert knowledge as well as interactive corrections, adaptive quality models are developed that not only use local information but also map complex relational dependencies and thus contribute to the global optimization of process control. Through approaches of explainable AI (XAI), transparency and comprehensibility of the learned models is guaranteed. Thus, a contribution to acceptance and justified trust in the AI system is created. As an exemplary use case, the production chain of the electric powertrain is considered. The goal is to develop an interactive dashboard based on innovative methods of explanatory interactive learning. This enables the recognition of complex interdependencies in quality assessment and consequently increases production efficiency and minimizes rejects.
Focus of the University Bamberg:
- Formalizing expert knowledge in knowledge graphs
- Anomaly detection
- XAI for anomaly dection
- Interactively improving root-cause AI model by human feedback loops
Publications, Talks and Student Projects
We are always looking for students interested in contributing to KIProQua in the form of a thesis, project, or as a student assistant. If you are interested, feel free to contact Christoph Wehner.