This release moves both Turbulence Realm SINDy and Tracker a long way forward. SINDy can now discover equations from video in six different ways, and Tracker has grown from a simple point-and-click tool into a full analysis console for bubbly and multiphase flows. Both install like any ordinary application on Windows and Linux.
It also introduces AeroFlow NS, a real-time Navier–Stokes wind tunnel that runs entirely in the browser, with two aircraft, a refined airliner and a stealth flying wing, to fly through it. Both simulations are embedded and fully interactive further down this page.
Why six methods, not one?
The promise of SINDy is disarmingly simple: watch a video, extract the motion, and find the simplest equation that explains it. The catch is that fluids do not behave in one way. A purely polynomial model loses the plot when the data is noisy, when the true equation has a fraction in it, or when the experiment is being pushed along by a pump or an actuator.
So rather than ask one algorithm to do everything, the new version offers six, chosen from a single menu. Classical SINDy is the natural starting point, returning clean, explicit equations from smooth data. SINDy-PI handles equations with fractions and unknown denominators. PDE-FIND discovers partial differential equations when the data covers a spatial grid rather than a single time series. WSINDy uses weak formulations to stay calm in rough, sparse or noisy data, while E-SINDy runs an ensemble and keeps only the terms it finds consistently. And SINDYc brings external forcing into the picture, for flows driven by pumps, valves or actuators. The interface adapts to whichever one is chosen.
In practice
A clean video of a vortex street will usually yield its equation through Classical SINDy in seconds. A pump-driven experiment calls for SINDYc; grainy footage for WSINDy or E-SINDy. The right method is now one click away.
Validation that does not give up
Earlier versions had a fragile streak. If a single cross-validation fold hit a numerical edge case, the whole run stopped and every other result was lost with it, and comparisons between model settings suffered the same fate. Now each fold and each comparison runs in isolation: one failure is reported, and the rest carry on. Settings can be compared in the confidence that the results are real rather than the side effect of one unlucky fold.
An installer that simply works
The Windows installer is now entirely self-contained. Previously the machine-learning features needed a separate PyTorch installation, which rather undermined the idea of a one-click installer. The new version brings PyTorch, the full computer-vision toolkit and its runtime along with it. The download is larger, at 231 MB, and that is a deliberate trade: open the file, accept the disclaimer, and start analysing.
Tracker: from clicks to a full physics report
While SINDy widened its repertoire, Tracker grew into a research instrument. Give it a video of bubbles, droplets or particles and it follows every one of them automatically, separating bubbles even when they touch and reporting their sizes, shapes and roundness. It computes the dimensionless numbers that fluid dynamicists live by, Reynolds, Weber and Eötvös, along with drag coefficients and terminal-velocity comparisons for common fluids. It builds full flow-field maps of speed, vorticity and divergence, spots events such as bubbles merging, breaking up or detaching, and measures the flux through any cross-section. Every reported speed carries its uncertainty, propagated from the calibration, and the whole analysis exports as a single PDF report or as a tidy table for further work.
Point it at a video of bubbles and it hands back a physics report.
AeroFlow NS: a wind tunnel in your browser
SINDy discovers equations from video, and Tracker measures what moves in it. AeroFlow NS works the other way round: it creates the flow from first principles, in real time, inside a web page. There is nothing to install. Open the page and orbit the camera.
Two aircraft are generated from mathematical shape descriptions: a refined airliner with winglets and engine nacelles, and a stealth flying wing with a swept lambda planform. Around them, the incompressible Navier–Stokes equations are solved afresh on every frame. The wingtip vortices, the wake and the stagnation point are not drawn by hand; they emerge from the physics. This release adds a slider for the number of tracer particles, a soft smoke mode that recalls the dye and smoke of a real tunnel, and a control for the number of streamlines, from a handful of diagnostic lines to a dense field.
Space freeze, G grid, O orbit.The story of how the solver itself works is told in the companion piece: AeroFlow NS: A Real-Time 3D Navier-Stokes Wind Tunnel in Your Browser.
Download and try
Both desktop applications are available as signed installers for Windows and Linux, with no Python required; each asks you to accept a no-liability disclaimer before installing. Download them from turbulencerealm.com/download, and find the tutorials at turbulencerealm.com/tutorials.
AeroFlow NS needs no download at all: the two simulations above are already running in your browser. You can also open the airliner full-screen or open the stealth flying wing full-screen in a new tab.