Advanced Backbone for Service-Oriented Orchestration of Linked Universal Twins
ABSOLUT
The ABSOLUT project (Advanced Backbone for Service-Oriented Orchestration of Linked Universal Twins) aims to advance research data management in aviation research and improve the reusability and long-term availability of large datasets in accordance with the FAIR principles (Findable, Accessible, Interoperable, Reusable). Building on the results of the previous DigTwin and DigECAT projects, ABSOLUT is developing the twinstash platform from a functional prototype into an operational system that can be used across institutes. The platform serves as the heart of data management and pays particular attention to the citation of data in order to make scientific reuse clearly traceable and to link publications directly to the underlying data. In addition, ABSOLUT forms the technical and infrastructural basis for the application of AI methods as an end-to-end process, giving users easy access to modern software tools such as pattern recognition, anomaly detection or predictive maintenance.
As part of this project, the Institute of Data Science contributes to the detection and explanation of anomalies such as sensor malfunctions or other problems in research data. Such anomalies can both complicate the use of data in research - especially if researchers encounter them without warning - and provide indications of necessary maintenance work and thus save resources. Here, the high-dimensional and diverse nature of research data poses a particular challenge: research aircraft can record hundreds of different measured values at high temporal sampling rates using a wide variety of sensors - from spatial position and orientation to environmental information such as air pressure and temperature to technical displays. The sheer size of the available research data makes manual inspection impractical. Therefore, ABSOLUT develops automated causal anomaly detection algorithms that can efficiently flag anomalies in data and trace them back to their origin by combining machine learning and causal inference.
This combination is intended, in particular, to avoid the "black-box" nature of many AI methods. This term refers to the problem that, especially with powerful machine learning methods, users often have no way of knowing how the various methods arrive at their decisions or predictions. Causal inference addresses this problem by shifting the focus from “What happens?” to “Why does it happen?” This shift provides explanations that are more stable and fairer across different environments (e.g., different research aircraft) than pure AI methods, and from which concrete action guidelines can be derived. Together, machine learning and causal inference allow a transparent and efficient analysis of research data that users can rely on. Additionally, by combining data-driven analysis with existing expert knowledge, the goal is to expand and make process understanding actionable in order to support researchers and engineers.
The core objectives of the Institute for Data Science in the ABSOLUT project are:
• Development of domain-specific models for knowledge representation that effectively enable further use by specialist users.
• Development of condition monitoring for all research aircraft sensors of the DLR Flight Experiment facility in the form of condition indices.
• Implementation of a software tool for the effective detection of anomalies in time series of research aircraft sensors and identification of their root causes.
These goals increase the ability of researchers and engineers to analyze and solve complex problems in aeronautical research.
