756,000
cells, six-storey residential block, 0.2 m grid
Fire-safety CFD · installed, not hosted
The same solver everyone uses. Everything around it, done properly.
Ashbeck runs NIST’s Fire Dynamics Simulator — the tool your engineers already trust — inside a complete, checked, auditable assessment workflow on a machine that sits in your office. We are not claiming better physics. We are claiming that the fortnight of setup, meshing, post-processing and report writing around the physics should not be done by hand.
68 min
wall clock for that model, 24 cores
6 of 6
workflow stages independently checked
FDS 6.11.1
NIST solver, unmodified
Figures measured on the reference specification, not estimated. The model above is the one in the specimen report.
Scope
Worth being exact about what this is
Fire engineering is a field where overstatement is expensive. So, plainly:
What it does
- Takes a building from floor plan to issued assessment report in one workflow
- Builds the FDS model, meshes it, and checks the grid against D*/δx practice
- Evaluates tenability to PD 7974-6 — visibility, temperature and toxic FED
- Derives ASET and RSET, and states the margin between them
- Runs the case with and without the proposed measures, so the difference is measured rather than asserted
- Records who confirmed what, and when, against the exact model that was run
What it does not do
- It does not improve on FDS. The solver is NIST’s, unmodified, and its validation is theirs
- It does not decide compliance. It reports what the model shows
- It does not sign anything. A competent chartered engineer reviews and signs off, exactly as now
- It does not replace judgement about whether the model represents the building — that is the one thing no software can do for you
- It does not read DWG directly. Drawings come in as PDF or image
Workflow
Six stages, each one checked before it can carry the next
Every stage runs two independent checks: deterministic rules that can stop the job, and a separate AI review that raises queries for the engineer. The rules block; the AI never does, and never marks its own work as correct.
Project and drawings
Plans are uploaded as PDF or image and read for dimensions, storey count, corridor and stair positions.
Checked: the drawing is read a second time, independently, and the two readings are compared. A dimension the reader cannot reproduce is flagged for you to measure by hand.Building model
Geometry, design fire, grid and duration. A 3D view and an overlay on your own drawing show exactly what will be simulated.
Checked: grid resolution against D*/δx, slab thickness against cell size, burner area against the declared fire, run duration against the expected fire growth — plus an AI comparison of the plan against the model you entered.Confirmation
Nothing is simulated until a named person records that the model fairly represents the drawings. Blocking errors cannot be signed away.
Recorded: who, when, their note, and a hash of the exact model specification — so the report can prove which model the confirmation applied to.Works and measures
Describe the scheme and select the fire-safety measures the mitigated case should include.
Checked: measures selected that the model cannot represent, measures described but not selected, and interactions that make things worse — an enclosed stair whose doors are not self-closing becomes a chimney.Simulation
Baseline and mitigated cases run on the machine, one at a time, with live progress and a notification when they finish.
Checked before committing the machine: whether the run is long enough for anyone to have escaped within it. A run that stops early reports “no breach” because it stopped, not because conditions held.Results and report
Tenability per location and storey, ASET against RSET, findings, and an issued report — with the before-and-after comparison.
Checked: an AI reading of the results is drafted for the engineer to accept or reject. No report is issued until both the model and the interpretation have been confirmed by a person.Evidence
Two things you should ask for, before anything else
Any supplier can describe a workflow. These are the artefacts that let you check the claim yourself.
A specimen assessment, unedited
A real six-storey residential block. Client, site and personnel details are replaced; every number, finding and audit entry is exactly as the software produced it. It is a failing result, and we have not softened it — a specimen that only ever shows a pass tells you nothing about the tool.
| Quantity | Result | Basis |
|---|---|---|
| Peak heat release rate | 1,018 kW | Upholstered furniture, fast t², PD 7974-1 |
| ASET, fire floor | 88 s | Visibility below 10 m in the corridor |
| RSET, fire floor | 101.5 s | PD 7974-6 hand calculation from the model |
| Life-safety margin | −13.5 s | Inadequate — reported as a critical finding |
| Governing criterion | Visibility | Temperature reached 60 °C at 139 s; toxic FED never breached |
| Grid resolution | D*/δx = 4.8 | Coarse; disclosed in the report, not buried |
A validation dossier for the pipeline
FDS is validated by NIST and we do not re-run their work. What needs independent checking is our layer — the mesh we generate, the tenability criteria we apply, the ASET we extract and the RSET we derive. The dossier compares the pipeline against established correlations and quantifies the numerical uncertainty rather than asserting it away.
| Check | Reference |
|---|---|
| Ceiling-jet temperature and velocity | Alpert (1972); SFPE Handbook ch. 14 |
| Plume centreline temperature | Heskestad; SFPE Handbook ch. 13 |
| Asphyxiant FED | PD 7974-6:2019 Annex D (Purser) |
| Visibility from obscuration | PD 7974-6:2019 Annex C (Jin) |
| Grid convergence and numerical uncertainty | ASME V&V 20-2009; Richardson / GCI |
| Energy conservation | Delivered HRR against specified HRR |
Available on request, with the input decks, so your own engineers can re-run and check every case.
Deployment
One machine, in your building
Ashbeck is installed on dedicated hardware at your premises. It is not a shared service and there is no multi-tenant cloud.
Your drawings never leave the building
Plans, models, results and reports stay on the machine. Access is over your own network.
Runs as an appliance
Services start with the machine and restart themselves. Daily backups, weekly housekeeping, and a health check that reports honestly.
Works from a phone
An installable web app for reviewing progress and results, with a notification when a run finishes.
Capacity you can plan against
The six-storey model in the specimen — 756,000 cells, 180 s of fire — took 68 minutes on the reference machine. The software estimates the cost of a run before you commit to it, and refuses jobs that would be too short to answer the question being asked.
Enquiries
There is no sign-up, deliberately
Every installation is specified around a particular practice, its drawings and its review process. That starts with a conversation between engineers, not a credit card.
Useful to know before you write
- Roughly how many CFD assessments your practice runs in a year
- What your drawings usually arrive as, and from whom
- Who signs off the fire strategy, and what they need to see to do it
- Whether you have somewhere to put a workstation-class machine