In situ adaptive timestep control and visualization based on the spatio-temporal variations of the simulation results
Yoshiaki Yamaoka, Kengo Hayashi, Naohisa Sakamoto, Jorji Nonaka · 2019
Effective visualization of time-varying volumetric data is considered challenging, since in addition to the traditional visualization parameter selections, there is also the need to take care about the dynamic changings between the timesteps. In situ adaptive sampling based on the amount of change between the simulation timepsteps might be helpful for the I/O and visualization cost savings. In this paper, we propose the use of Kernel Density Estimation (KDE) and Kullback-Leibler (KL) divergence, in order to estimate the amount of internal variations between the timesteps, and use this as a parameter for controlling the sampling rate. Needless to say that these selected timesteps should be sufficient to enable the understanding of the underlying phenomena without missing important data features. We confirmed the effectiveness on reducing the number of timepsteps, by using an OpenFOAM CFD simulation with irregular volume data. However, we could also observe that the user-defined parameter selection can highly influence the amount of data size reduction as well as the smoothness of the visualization results, and an optimal parameter selection method remains as a future work.