Feature extraction and tracking

Efficient data analysis for high-performance computing

Feature-based studies

The rapid growth in available computing power enables numerical modelling of ever more complex, scientifically and technologically important systems at higher spatial and temporal resolutions. Correspondingly, simulations begin to generate extreme amounts of data, limiting the ability to perform efficient post-processing. Therefore, enabling feature-based studies during simulation run-time is a critical aspect for accelerating research discovery in the era of exascale computing. Consequently, a new kind of meta-analysis is required that can efficiently extract, filter and compress the data without the need to interactively adjust input parameters.

Feature-based flow visualisation

The ability to detect and follow time-varying features in data obtained from numerical simulations enables characterisation and analysis of modelled physical phenomena. Feature-based flow visualisation shows only fragments of the results that are considered significant based on the application and the research problem; some examples of such patterns include vortices, shocks, eddies, critical points, etc. The saving in storage space with feature extraction can be significant and allows for data analysis and on-the-fly statistics at large scale. Additionally, as the scientific datasets often suffer from numerical noise, filtering techniques need to be employed before feature-based analysis can be performed.

Our work focuses on developing methodologies that allow important structures in numerical results to be extracted and studied with negligible impact on the overall simulation run-time. The images on the right show the result of applying our proposed level-dependent WienerChop filter to data corrupted with coloured noise. We obtain a substantial computational saving compared to classical approaches.

Colored noise


[1] M.J. Zimoń, J.M. Reese, D.R. Emerson,
A novel coupling of noise reduction algorithms for particle flow simulations,”
Journal of Computational Physics 321, 169-190, 2016.

[2] M.J. Zimoń, R. Prosser, D.R. Emerson, M.K. Borg, D.J. Bray, L. Grinberg, J.M. Reese,
An evaluation of noise reduction algorithms for particle-based fluid simulations in multi-scale applications,”
Journal of Computational Physics 325, 380-394, 2016.

Ask the experts

Robert Sawko

Robert Sawko

Małgorzata Zimoń

Małgorzata Zimoń

Chris Thompson

Chris Thompson

Our collaborator at STFC

David Emerson