English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Fast and stable fluid simulations are an essential prerequisite forapplications ranging from computer-generated imagery to computer-aided designin research and development. However, solving the partial differentialequations of incompressible fluids is a challenging task and traditionalnumerical approximation schemes come at high computational costs. Recent deeplearning based approaches promise vast speed-ups but do not generalize to newfluid domains, require fluid simulation data for training, or rely on complexpipelines that outsource major parts of the fluid simulation to traditionalmethods.  In this work, we propose a novel physics-constrained training approach thatgeneralizes to new fluid domains, requires no fluid simulation data, and allowsconvolutional neural networks to map a fluid state from time-point t to asubsequent state at time t + dt in a single forward pass. This simplifies thepipeline to train and evaluate neural fluid models. After training, theframework yields models that are capable of fast fluid simulations and canhandle various fluid phenomena including the Magnus effect and Karman vortexstreets. We present an interactive real-time demo to show the speed andgeneralization capabilities of our trained models. Moreover, the trained neuralnetworks are efficient differentiable fluid solvers as they offer adifferentiable update step to advance the fluid simulation in time. We exploitthis fact in a proof-of-concept optimal control experiment. Our modelssignificantly outperform a recent differentiable fluid solver in terms ofcomputational speed and accuracy.

Learning Incompressible Fluid Dynamics from Scratch -- Towards Fast, Differentiable Fluid Models that Generalize