CAREER: Generating Hierarchical Vector-Valued Data Summaries for Scalable Flow Data Processing, Analysis and Visualization

NSF IIS 1553329
PI: Guoning Chen

Key Outcome


This project has resulted in the following key publications:


(1) A Physics-based flow analysis and exploration framework

It has been demonstrated by the PI and his team in their previous work that accumulating certain physical properties of the flow along the integral curves of massless particles can be used to aid a number of flow visualization and exploration tasks, including integral curve selection, flow segmentation, and the discovery of the discontinuity in integral curve behavior. In this project, we further explored this framework and has applied it to support the generation of seeding curves for the construction of integral surfaces for 3D flow visualization, which was a challenging tasks previously. In addition, we discover that there exists more richful information in the profile of a certain attribute of interest associated with a given particle over time, which we refer to as the time activity curve (TAC). Specifically, the previously introduced accumulation framework can be considered as measuring the overall (or average) behavior of the corresponding attribute carried (or transported) by the particle over time, while the TAC and its fine analysis will provide a level-of-detail view on the behavior of the particle characterized by the corresponding attribute of interest over time. We exploited this observation and proposed a new TAC-based flow analysis and exploration framework. With the aid of this framework, the relations of different attributes, especially the relations between those physical attributes and the geometric attributes, are studied in detail.

(2) Summary and reduced representation of integral curves

We introduced new similarity metrics for the comparision of pairs of integral curves for their clustering in order to reduce the original dense representation of 3D flow into some sparse but informative representation.

(3) Correlation study of flow attributes

We proposed to study the co-variant behaviors of different flow attributes (including both geometric ahd physical attributes) to enhance our understanding of the flow behaviors, especially the relation between certain physical attributes and the geometric appearance flow patterns, aiming to address the question of "how much physics can the current geometric representation of the flow behaviors convey".

(4) Multi-scale coherent structure extraction and visualization for turbulent flow study

We proposed a number of framework for the extraction and separation of coherent structures in different scales for the study of different turbulent flow data in both steady and unsteady settings. Our current techniques allow the separation of large-scale coherent structures from the small ones, which is crucial to understand the global and dominant configuration of the flow behaviors.

(5) A machine learning framework for flow analysis

We developed a parametric modeling method that enables the generation of large sets of sample flow data with the known features to train a number of machine learning models (e.g., CNN, Resnet, Unet, etc.) to perform certain analysis tasks (e.g., vortex boundary extraction).

(6) Analysis and visualization of particle-based flow data

We developed a suit of techniques for the analysis and visualization of particle-based flow data generated from SPH or PBF simulations. One challenge of this kind of the data is the lack of the necessary neighborhood information among neighboring particles that is required for many analysis tasks. In addition, the trajectories of the individual particles are different from the mathematically well-defined integral curves, which makes most existing techniques developed for mesh-based vector fields unsuitable. To address these challenges, we adapted a number of techniques for mesh-based vector fields to the meshless setting, including FTLE, Jacobian-based computation, and the attribute-based processing (see the above sub-project). We have applied these adapted techniques to the analysis and visualization of various 2D and 3D particle-based flow data stemming from SPH or PBF simulations.

(7) Applications: wall shear stress study

This is a side sub-project enabled by the proposed research of the project. In this sub-project, PI Chen helped extend the streamline computation on curved surfaces to the pathline computation, which is applied to study the wall shear stress on surfaces.

(8) System: A Client-Server platform for disseminating the results.

(9) Optimization of Volumetric, Structured Meshes

Meshing is an indispensable step for many critical scientific computation. The quality of the meshes, especially structured-meshes often impacts the speed and accuracy of the computation. However, there currently lacks a robust and effective framework to optimize a structured mesh, i.e., a hexahedral mesh, to meet the requirements of a specific application, with certain quality guarantee (e.g., inversion-free). To address this, this project investigate a number of novel and effective optimization to improve the quality of the structure of the mesh and the quality of the individual elements.

(10) Microvascular network visualization and analysis