Gast ::
| Hauptseite > Publikationsdatenbank > Scalable Anomaly Detection in High-Speed Combustion Imaging |
| Hauptseite > Publikationsdatenbank > Scalable Anomaly Detection in High-Speed Combustion Imaging |
| Contribution to a conference proceedings | FZJ-2026-00850 |
; ; ; ; ;
2025
Astronautical Federation (IAF)
Paris, France
This record in other databases:
Please use a persistent id in citations: doi:10.52202/083090-0032
Abstract: In aerospace-combustion research, advanced high-speed imaging enables the observation of dynamic processes, suchas turbulence–flame interactions and instability mechanisms, that may have a significant influence on propulsionsystems. To analyze such complex and data-intensive recordings, we leverage high-performance computing (HPC) forscalable anomaly detection in high-speed combustion video data using Python. Specifically, we developed a parallelizedimplementation of the Local Outlier Factor (LOF), a well-established density-based algorithm for detecting local (i.e.,small-scale) anomalies, to support massively parallel execution across multiple GPUs. This allows for the identificationof critical instabilities and localized extinction events that are otherwise difficult to detect, due to the immense volumeand complexity of data, in the investigated high-speed video containing 116,487 images. Our results illustrate thatmemory and runtime constraints—which in the standard implementation of the LOF would require to evaluate and storethe result of more than 10 billion image comparisons on a single computation node—can be significantly mitigated. Byenabling scalability to substantially larger datasets, our approach shows the potential of HPC-driven anomaly detectionto overcome existing computational bottlenecks in order to advance aerospace research
|
The record appears in these collections: |