Aims & Objectives

Problem Formulation – Mathematical Model – Simulation – Big Data: This scientific process of a virtual experiment is well-established in many disciplines. However, there is a challenge on the way from the experiment to findings – the storage, analysis and visualization of huge amounts of data. In practice, data reduction strategies are used which are often afflicted with loss of information, e.g. storing only every n-th time step of a simulation.

Our approach for data reduction is to use Proper Orthogonal Decomposition (POD). This mathematical method extracts the characteristic features from simulation data and stores them as POD basis functions. Only few such basis functions are enough to build reduced mathematical models with high accuracy and minimal loss of information. These reduced models can be used for memory-saving storage, accurate reconstruction and mobile visualization of simulation data.

Research Topics

Conceptual studies for creating, storing and reconstructing reduced models

• Explore data formats for input and output
• Develop appropriate data structures and functionalities for usage with the POD method

Development of a data compression module

• Design and development of interfaces to simulation data
• Development of flexible controlling for the compression module

Implementation of the POD method in HiFlow³

• Extension of the finite element software package HiFlow³ with a POD module

Development of a Visualization module

• Design and development of a visualization module for reconstructing and visualizing the information from reduced models using VTK

Funding

This project was funded as a Feasibility Study of Young Scientists (FYS) by the Council for Research and Promotion of Young Scientists (CRYS) at Karlsruhe Institute of Technology (KIT).

The funding period was 2012-06 to 2013-05.

Contact 

Martin Wlotzka

Download

The tarball contains the python scripts for the compression and the reconstruction module and a user guide: Big-Data-for-Take-Away.tar