GPU-Native · Cloud-Based · AMR-CFD · AI-Accelerated

AgniKawach

GPU-Native, Cloud-Based Process Safety Consequence Modelling

Gas dispersion · Jet fire · Pool fire · Flash fire · VCE · Detonation & DDT · BLEVE · Cryogenic · QRA

🧪 Try Demo → 📊 View V&V Results →
40–73×
Faster Than CPU
GPU-native consequence solvers
14×
Fewer Mesh Cells
Adaptive octree refinement
8
Consequence Types
Dispersion · Fire · Explosion · Cryogenic
1
Geometry Import
Feeds every consequence solver
Five Technical Pillars

A modern, scalable, regulator-ready architecture

Each pillar individually exists elsewhere. Combining all five — with no compromises — is what makes the difference.

⚡
GPU-Native Compute
Every physics solver runs natively on modern GPUs. Order-of-magnitude faster than CPU-bound legacy tools. Same physics, same accuracy, a different compute era.
40–73× faster
🧬
Adaptive Mesh Refinement
Octree mesh automatically refines near hazard sources — flame fronts, jets, leak points — and coarsens far away. Full accuracy where it matters, dramatically reduced cell count overall.
14× fewer cells
☁️
Cloud-First Delivery
Browser-based. No installation, no HPC queue, no file transfers. Pay-per-scenario cloud GPUs auto-scale from one run to thousands. Collaborative workspace, always-current physics.
Zero install · pay-per-scenario
🪜
Two-Tier Architecture
Tier-1 integral models (Gaussian, HEGADAS, MEM/BST, solid-flame) give screening results in seconds. Tier-2 GPU AMR-CFD delivers high-fidelity accuracy in minutes — auto-escalates when geometry, congestion, or risk demands it.
Tier-1 seconds · Tier-2 minutes
🧩
Shared Physics Backbone
Six engineered modules — thermodynamics, combustion kinetics, spectral radiation, soot, geometry, atmospheric boundary layer — power all eight consequence types. Build once, validate once, reuse everywhere.
6 engines · 8 consequences
📄
Standards-Ready Outputs
Purpose-built outputs — ERPG / AEGL / IDLH limits, probit dose-response, F-N curves, risk contours, regulator-ready PDF reports. Traceable and auditable by design — not bolted on.
ERDMP · OISD · COMAH · Seveso III
How It Works

From geometry to risk report in four steps

No CFD expertise required. Safety engineers drive the workflow; the solver tier runs unattended.

🗺️ 1
Upload Site Geometry
Import a CAD file or draw plant layout in the 3D editor. Tag buildings, pipe racks, vessels, and dyke walls. The mesh generator handles the rest.
💥 2
Define Release Scenario
Choose substance, leak diameter, inventory pressure and temperature, meteorological class, and wind speed. Tier-1 pre-check runs instantly to flag boundary conditions.
⚙️ 3
Run Analysis
Submit to the GPU cluster. Live progress shows solver step count, plume extent, and current CFL. Tier-2 AMR refines automatically around the hazard zone.
📥 4
Download Report
Receive a regulatory-formatted PDF with isopleths, σy/σz tables, Hanna metrics, and risk contours. Upload directly to PNGRB portal or include in COMAH Safety Report.
Interactive Demos

Try the platform now

All four modules run in the browser. No account required. Sample Prairie Grass and LNG dispersion cases pre-loaded.

🧪
Scenario Builder →
Define release source, meteorology, and Pasquill stability class. Parametric sweeps across wind speed, leak size, and substance. Tier-1 Gaussian results in real time.
Tier-1 · Gaussian · HEGADAS
🌐
3D Plume Viewer →
Interactive WebGL visualization of CFD concentration fields. Toggle between horizontal and vertical cross-sections. Time-scrub through transient simulation results.
Interactive 3D · CFD field data
📊
Results Dashboard →
Centerline concentration vs. distance, σy/σz dispersion widths, Hanna FB/NMSE/FAC2 metrics. Overlay Prairie Grass Barad (1958) field trial data for Tier-2 validation.
Prairie Grass · V&V metrics
⚠️
QRA Dashboard →
F-N curves, individual risk contours, and event trees. Aggregate multiple consequence types into overall site risk. Export to PNGRB ERDMP Table 5 format automatically.
F-N curves · ERDMP · LOPA
Workflow

Gas-dispersion pipeline, end to end

Seven concrete pages on this site, one per stage. Click a tile to jump in. The Tier-1 screening result auto-sizes the Tier-2 CFD setup, so the pipeline can run unattended once the scenario is defined.

01 · DEFINE Scenario Builder Substance, release rate, leak geometry, atmosphere class. scenario_builder.html 02 · GEOMETRY Geometry Viewer STL/STEP/IGES → EB · Brinkman · PDR. The mesh the solver actually uses. geometry_viewer.html 03 · VERIFY Grid Convergence 3 grids, Richardson ratio, per-arc PASS/FAIL. Required before production. grid_convergence.html 04 · SCREEN Tier-1 Dispersion Gaussian / Britter-McQuaid / HEGADAS-S. Seconds. Auto-sizes Tier-2. tier1.html#dispersion 05 · SOLVE Tier-2 CFD GPU AMR-CFD low-Mach LES — adaptive refinement at the plume. Minutes on a single GPU. tier2.html#sc-dispersion 06 · POST Results Dashboard Centerline, σ widths, Hanna FB · NMSE · FAC2 versus Prairie Grass. dashboard.html 07 · RISK QRA Probit dose, F-N curves, ERDMP-formatted output. qra_dashboard.html

Tier-1 ↔ Tier-2 — same physics, two fidelities

Each Tier-1 dispersion model maps to a specific Tier-2 solver entry. Use Tier-1 for screening & QRA loops; pivot to Tier-2 when the result approaches a regulatory threshold or the geometry stops being point-source-like.

Tier-1 · screening Gaussian Plume (Briggs-Pasquill) Closed-form algebraic; rural & urban Briggs coefficients; Pasquill A–F. tier1.html#dispersion-gauss
→
Tier-2 · CFD GPU AMR-CFD · low-Mach LES + Smagorinsky Validated config: PG-turb-8m, FAC2 = 0.84 at x = 500 m. Same wind / Pasquill input as Tier-1. tier2.html#sc-dispersion
Tier-1 · screening Britter-McQuaid (dense gas) HSE CRR 17/1988 nomogram; Cl₂, NH₃, LPG vapour ground-level continuous. tier1.html#densegas-bm
→
Tier-2 · CFD Boussinesq + multi-component (MW) Explicit ρ-perturbation from concentration + MW. Validated within |Δρ/ρ| ≤ 0.03 budget. tier2.html#buoy
Tier-1 · screening HEGADAS-S (Witlox / Holt) Steady 1-D ODE with crosswind similarity profile; Burro / Coyote LNG calibration. tier1.html#densegas-hegadas
→
Tier-2 · CFD Synthetic HIT turbulent inflow + log-law Friction-velocity log-law inlet, optional homogeneous isotropic turbulence box for fluctuations. Stable velocity clamp on inflow extremes. tier2.html#turb-inflow
Tier-1 · screening Geometry preprocessor + PDR helper SiteDefinition (Building / PipeRack / EquipmentCluster) → solver-ready blocks for Tier-2 PDR / Brinkman. tier1.html#geom
→
Tier-2 · CFD EB · Brinkman+IBM · PDR (composable) Three obstacle representations side-by-side. EB for sharp interface, Brinkman+IBM for runtime, PDR for sub-grid. tier2.html#geom
Tier-1 · cross-check AERMOD shim EPA reference comparator. Reports AgniKawach, AERMOD, and percent deviation side by side. tier1.html#dispersion-aermod
→
Tier-2 · validation Validation table — 19 / 20 cases pass Hanna FB · NMSE · FAC2 against PG-23/24/38, CODASC, Maplin Sands, Burro/Coyote LNG. coverage.html#val-dispersion
Compliance & Validation

Designed for regulators. Validated against field data.

Every solver claim is backed by published V&V cases against internationally recognised experimental datasets.

Regulatory Frameworks
PNGRB ERDMP OISD-116 OISD-118 COMAH 2015 Seveso III API RP 752
▸ PNGRB mandates consequence modelling for all petroleum, natural gas, and LPG pipelines >25 mm under ERDMP 2010 (amended 2019).
▸ Reports include probit dose, consequence distance, and land-use compatibility tables as required by PNGRB Technical Standard TS-ERDMP-001.
Validation Datasets
Prairie Grass (Barad 1958) CODASC Wind Tunnel NIST firemodels/exp Pasquill-D Analytical Hanna Metrics Smagorinsky LES
▸ All V&V cases target Hanna criteria: |FB| < 0.3, NMSE < 0.5, FAC2 > 0.5.
▸ Current status: 19/20 cases pass. Raw postprocessing scripts and plotfiles available under open V&V policy.
Get Started

Ready to run your first consequence scenario?

Try the interactive demo for free. Full API access and regulator-ready report generation on the cloud platform — cost-effective, pay-per-scenario pricing. No installation. No HPC queue. No waiting.

🧪 Try Demo → ✉️ Contact Sales