Advanced Technologies for High Energetic Atmospheric Flight of Launcher Stages
ATHEAt
The DLR project AHTEAt is researching technologies for reusable space transport systems that are highly cost-efficient. The reusability of launch vehicles is to be made more reliable and safety margins reduced with a view to improving economic efficiency. As part of the ATHEAt project, the technologies developed are being tested in practice in two space flights and in ground tests.
In the ATHEAt project, the Institute of Software Technology is responsible for analysing selected ground test and rocket flight data using algorithms from the field of artificial intelligence. The analysis is divided into several detailed tasks.
On the one hand, it involves determining the mass loss of the material of rocket fins in ground facilities and determining the regression rate during fuel burn-up using optical methods. To do this, we apply neural networks based on a U-Net architecture to high-speed image data in order to automatically detect structures such as the test models of the rocket fin or the rocket propellant. We determine the uncertainties of our detection results using statistical methods (Uncertainty Quantification).
We also analyse ground and flight test data for anomalies in order to detect faults and problems at an early stage. For anomaly detection, we use both density-based algorithms and customised neural networks based on autoencoders or vision transformers.
Image of the rocket fin burn-off experiment (left) and comparison of two approaches to segmenting the combustion flame (centre and right).
The centre of the image shows the result of a classic filter from image processing (computer vision). Similar results, but with greater robustness, can be achieved with neural networks based on a U-Net architecture (right). Source: DLR Institute of Software Technology, Department of High-Performance Computing and DLR Institute of Aerodynamics and Flow Technology, Department of Supersonic and Hypersonic Technologies
Automatic detection of anomalies in the ground test data of a rocket motor
Each test image is assigned an "outlier score", which indicates the probability of an anomaly from the algorithm's point of view. For comparison, the red vertical lines show anomalies that an expert found manually in the data set. Source: DLR Institute of Software Technology, High-Performance Computing Department and DLR Space Operations and Astronaut Training, Mobile Rocket Base Department