About
A Windows desktop workbench for discrete element simulation: pick a template, adjust parameters, run the YADE solver, inspect results — no scripting required.
Why we build it
Discrete element work today lives at two extremes: script-driven open engines that put the whole workflow on the engineer's shoulders, and closed commercial packages that hide the middle of the pipeline. The repetitive part — pick a standard scenario, tune micro-parameters against a measured curve, record what was run — is the part software should carry, visibly.
Demulate takes the standard scenarios (angle of repose, direct shear, triaxial, box lift, slope) and makes them first-class templates with generated panels, pre-run estimates, result cards with provenance and an honest capability ledger. Everything else about the tool follows from that goal.
Engineering principles
The solver is never forked
The interface, templates and storage depend on the engine interface and the open .dres format — never on YADE internals. Physics stays exactly as published.
Open formats, versioned contracts
Cases (.demcase) and results (.dres, HDF5) are documented formats with version contracts: old files open in new versions, and a case folder travels by plain file copy.
The UI cannot outgrow the truth
Parameter panels are generated from the solver-facing schema; the manual is generated from the same source as the UI, with byte-exact sync guards. Docs and behaviour cannot quietly drift.
No fake affordances
If a button is disabled, its menu agrees. If a capability is not verified, the UI does not offer it. Claims live in the ledger, not in marketing.
Provenance or nothing
Every number on a result card traces to the frames it came from. The software never invents an estimate outside the solver.
Verified by regression
The desktop codebase carries 1767 automated tests; every change passes lint and the full suite before it lands. This site quotes that codebase with its own checks.
Solver traps we turned into guards
Each of these was measured on the 2022.01a YADE binary — not paraphrased from documentation. Every one is now a validation rule in the product.
| Trap | Why it is dangerous | What the product does |
|---|---|---|
| An unknown liquid-bridge model name degrades silently to none — zero bridge force, no error. | A wet material runs as dry, and nothing in the log reveals it. | Closed enum (V031) — bridge_model accepts only the names measured from the binary. |
| The documented nominal tensile strength sigmaT has no effect on the CPM response: identical pull trajectories for sigmaT = 100 and 10 000. | Tuning the parameter the docs point at changes nothing — the real knob is elsewhere. | Not exposed (V030) — the strength knob is eps_crack_onset (tensile strength = E·eps_crack_onset). |
| Damping typed for a model that never reads it is silently ignored — the scene runs undamped while damping is on screen. | Quasi-static results lean on damping; an undamped run pollutes the stress path without a warning. | Pairwise rules (V026 / V027) — kn/ks exist only under kn-ks; cn/cs only under viscoelastic or capillary. |
| A periodic-cell loading channel with a partial time window is silently inactive for its whole stretch. | The run finishes normally; the axis simply is never driven. | Whole-span channels (V033) — t_start = 0, t_end = None, at most one channel per axis, actuator refused. |
| A pure-Python Law2 contact law cannot be dispatched — overriding go is silently ignored and the base C++ law keeps running. | A “replaced” contact law is still the old one; the forces are not what the script promises. | Additive only (V035) — custom_hook is positioned as extra forces on top of the contact law, never a replacement. |
This is not a bug list for YADE — it is the boundary any integration has to handle. Software that claims to wrap the solver cannot dodge these; the difference is whether it tells you.
Why YADE in an isolated process
Scenario files describe physics in physical quantities, not solver API calls. That keeps the solver swappable: today YADE, tomorrow a vendored build, later an in-house kernel — the interface and every stored case survive the swap.
The staged engine plan
| Stage | What it means | Acceptance |
|---|---|---|
| 1 · Drive | YADE as a black-box solver behind a process boundary (current) | Real simulated results, reproducible |
| 2 · Vendor | YADE source in our own build, physics untouched | Bit-identical to stage 1 on the same version and seed |
| 3 · Build | Own GPU time-stepping loop, contact models carried over verified | Statistically equivalent to stage 2 (fixed seeds, repeated runs), with a performance win |
Stage 2 is not optional: it is the bridge that makes reading the solver source verifiable, and bit-identical output is the only hard evidence that the physics was not touched.
Independence note
Demulate is an independent workbench. It is not affiliated with YADE (OpenDEM), MatDEM or EDEM; template blurbs that reference their published cases do so to describe the physics scenario, and YADE is credited wherever it runs.