Time-dependent particle traces in a real-time visualization environment

Real-time visualization (RTV) is a process where visualization is performed on a local graphics workstation while the computationally intensive part of the simulation is performed either locally or remotely on a high performance workstation or supercomputer. This paper describes issues involved in generating time-dependent particle traces in the RTV environment. The technique proposed in this paper uses visual programming environment across heterogeneous computer architectures to create a flexible environment for visualizing particle trajectories released in a flow field. A numerically simulated non-circular jet is used to evaluate the use of the technique. Particle velocities at each time step are calculated based on local tri-linear interpolations of the instantaneous flow field data; particles are displayed as spheres that can be optionally colored using other scalar variables in the flow field. Only the particle positions are sent to the workstation, thus overcoming the need to send and store vast amounts of data. At any time of the simulation, the interactive implementation of the method allows users to clear trajectories or start new ones at chosen seed points. 1. Current address: Concurrent Technologies Corp. 1450 Scalp Ave, Johnstown, PA 15904 2. Associate Fellow, AIAA 3. Senior Member, AIAA This paper is a work of the US Government and is not subject to copyright protection in the United States. Introduction Flow visualization is an important step in understanding and interpreting results from laboratory and numerical experiments. Based on the databases generated in computational fluid dynamics (CFD) simulations, scientists can obtain detailed insights on the unsteady dynamics and topological features of the flow. Instantaneous flow visualization is based on a "snapshot" of the data at a selected timestep, involving scalar or vector fields. Many visualization techniques such as cut-planes, isosurfaces, and direct volume rendering, are available for visualizing scalar fields. For vector fields, several visualization techniques such as particle tracing and streamlines are described in [1,2,3]. Most of these papers describe techniques for generating streamlines for instantaneous flow fields. These techniques provide valuable information for understanding steady flows but are generally less useful for unsteady flow systems, where passive particle traces, are more appropriate. For time-dependent and spatially advancing flow fields, dynamical visualization techniques can provide valuable insight into understanding dynamics and topology of the flow field. For example in [4], animating volume rendered images allows to understand the formation of coherent structures in the flow field. In [5], a technique is reported where the entire time-dependent database is stored in disk in order to load the data for consecutive timesteps to the physical memory of the computer for generating time-dependent particle traces. Another commonly used technique for visualizing time-dependent data is to load the data as a fourdimensional field (time as the fourth dimension) and to generate animations of graphical objects (isosurfaces, 3D arrows, etc.) based on their variation in the time-axis. In all of these methods, the amount of data that has to be stored in disk or loaded into physical memory can be very large. The storage and memory requirements increase rapidly as the number of grid points in the computational space or as the number of timesteps in the simulation is increased. One of the valuable dynamic visualization techniques available for CFD research is generating particle traces. A particle trace is a line joining the positions, at an instant in time, of all particles that have been released previously from a specified location [6]. Particle trace visualization is especially useful for flow systems that move in space and time. For example, one can use this method to track fuel particles inside a jet; in addition to tracking the physical locations of the particles, coloring the spheres that represent them with scalar values such as mixedness or temperature can be very useful to determine jet control mechanisms (e.g., by which the efficiency of the jet combustion can be improved by optimal selection of unsteadiness and/or fuel injection location). One of the major difficulties in implementing the use of particle traces is the amount of data that needs to be stored for calculating particle positions for each timestep. This paper proposes a method to overcome this difficulty by using RTV environment for generating particle traces. First a description of how the real-time visualization is implemented under a heterogeneous computer environment is given. Then a parallel implementation for overcoming network delays is explained. Next, the modulardistributed method for generating particle trajectories is described. Finally, results from implementing the technique for our test case, based on 3D non-circular jet simulation are presented. Real-Time Visualization With the current methods used for generating particle traces and visualization in general, a considerable amount of time is spent on data storage, retrieval, and data transfer from supercomputers to visualization platforms. To avoid this intermediate step, we propose the RTV technique where minimal data transfer to the visualization platform is performed in real time during the simulation process. In this framework, specific details of the data transfer are handled by the software packages designed with portable transfer schemes which are transparent to the user. The only basic requirement of RTV is for the availability of a network connection between the supercomputer (or high performance workstation) and the visualization workstation. With the availability of visual programming software for scientific visualization such as AVS [7], IRIS Explorer [8], Data Explorer [9], and KHOROS [10], implementation of RTV has become practical. In these software environments that are based on data flow architecture, independent modules can be combined very flexibly to form a "network" or an application. Simulation programs can be converted into independent modules that pass data to other standard modules that perform visualization and other functions. These modules can run on either local or remote computers. Once the simulation code is converted into a module, a visualization network is constructed by mixing visualization modules in the local workstation with the simulation module on the remote computers. There is no special programming requirement necessary to develop modules that function in a heterogeneous computer environment. In the present work, the AVS software was selected as the basis for implementing particle traces with in the RTV environment since AVS offers source code compatibility, and it is currently available on UNIX platforms and most supercomputers (including Cray Research's Cray computers and Thinking Machine