Ignimetheus is a browser-based, reduced-order 1D spacecraft microgravity combustion screening and teaching tool. It pairs an interactive HTML5 Canvas flow-duct simulation with 0g vs. 1g flammability envelope comparisons, modeled active and passive suppression levers, NASA Saffire / BASS-II / FLEX / NASA-STD-6001 reference scenario matching, and a client-side local WebGPU Telemetry Assistant backed by deterministic replies and unit-aware numeric checking.
-
HTML5 Canvas Duct & Flame Simulation (
src/components/SimulationStage.tsx,src/components/CombustionDuctCanvas.tsx):- Interactive canvas rendering of velocity streamlines, vector arrows, heat-flux shading, a 6-point thermocouple array (
TC-1throughTC-6), interactive CAD spatial probes (P-n) and calipers (D-n), and a toggleable1gbuoyant plume silhouette overlay. The flow field visualization is illustrative (analytic), not a computed flow field. - Supports
msg-12cm(12 cmMicrogravity Science Glovebox duct,Dh = 120 mm) andsaffire-40cm(40 cmCygnus Saffire flow duct,Dh = 400 mm,0.88xwall-loss spread scaling) confinement scales acrossrod,sheet,channel, anddropletspecimen geometries.
- Interactive canvas rendering of velocity streamlines, vector arrows, heat-flux shading, a 6-point thermocouple array (
-
Reduced-Order 1D Thermo-Fluids Engine (
src/engine/thermoCore.ts):- Evaluates effective velocity
$U_{\mathrm{eff}} = \sqrt{U_{\mathrm{flow}}^2 + U_{\mathrm{buoy}}^2}$ , oxygen-enriched exploration atmospheres (e.g.,34% O₂at56.5 kPa), quiescent radiative quenching (0 cm/sin0g), high-flow convective blow-off (Da* < 1.00), and a two-sided thin-sheet radiative-loss shift for charring/ash-forming fabrics (nomex,silicone).
- Evaluates effective velocity
-
0gvs.1gFlammability Envelopes:- Tracks pressure- and flow-adjusted effective Limiting Oxygen Index (
LOI_eff),LOImargin, Extinction-Proximity Index (Da*, an algebraic index derived from solver LOI margin and flow strain rather than independent Arrhenius kinetics), andNASA-STD-6001152 mmburn-length comparisons acrossSTANDBY,IGNITING(80 Whot-wire pulse over3.5 s),SUSTAINED_BURN, andSELF_EXTINGUISHEDstates. Simulation runs in this app represent0gor duct test conditions, not theNASA-STD-6001Test 1 upward-burn configuration.
- Tracks pressure- and flow-adjusted effective Limiting Oxygen Index (
-
Modeled Suppression Levers (model assumptions, not NASA data):
- Models upstream Wake Baffle (
0.72xroot flow factor at95 mm, raising the nominal blow-off velocity limit from38 cm/sto48 cm/s), Flame Arrestor Mesh (280 mmburn clamp up to32.0% O₂, overwhelmed above32.0% O₂), and Pulse N₂ Suppression (-8.0% O₂transient for5.0 sat160 mm).
- Models upstream Wake Baffle (
Six Cygnus Saffire microgravity concurrent-flow records (SAFFIRE-I-01, SAFFIRE-II-05, SAFFIRE-II-06, SAFFIRE-III-01, SAFFIRE-II-07, SAFFIRE-II-01) use published summary conditions and measurements from Urban et al., ASGSR 2017 (NASA NTRS 20180005168, Table 1, pp. 30–31) alongside adapted BASS-II, FLEX, and NASA-STD-6001 reference scenarios. Within the Saffire records, CONCURRENT_THIN_SHEET_SPREAD_FACTOR (0.07756; 0.07756 * 0.88 = 0.06825 in the 40 cm Saffire duct) is fitted to the two 20 cm/s cotton-fiberglass (SIBAL) runs (SAFFIRE-I-01 and SAFFIRE-II-05) and evaluated on the two 25 cm/s SIBAL runs (SAFFIRE-II-06 and SAFFIRE-III-01):
| Record ID | Role | Material (specimenGeometry, ductScale) |
Conditions (0g) |
Published / Ref V_f (mm/s) |
Simulated V_f (mm/s) |
Relative Error / Match |
|---|---|---|---|---|---|---|
SAFFIRE-I-01 |
Fitted | cellulose (sheet, saffire-40cm) |
21.6% O₂, 101.3 kPa, 20 cm/s concurrent |
1.80 |
1.888 |
+4.9% (+4.89%) |
SAFFIRE-II-05 |
Fitted | cellulose (sheet, saffire-40cm) |
22.1% O₂, 101.3 kPa, 20 cm/s concurrent |
2.10 |
2.017 |
-3.9% (-3.94%) |
SAFFIRE-II-06 |
Held-out | cellulose (sheet, saffire-40cm) |
22.1% O₂, 101.3 kPa, 25 cm/s concurrent |
2.60 |
2.296 |
-11.7% (-11.69%) |
SAFFIRE-III-01 |
Held-out | cellulose (sheet, saffire-40cm) |
21.1% O₂, 101.3 kPa, 25 cm/s concurrent |
2.30 |
2.005 |
-12.8% (-12.83%) |
SAFFIRE-II-07 |
Extinction check (calibration-informed) | nomex (sheet, saffire-40cm) |
22.1% O₂, 101.3 kPa, 20 cm/s concurrent |
0.00 |
0.000 |
Outcome matches (no sustained spread; LOI margin -1.4%) |
SAFFIRE-II-01 |
Extinction check (calibration-informed) | silicone (sheet, saffire-40cm) |
22.1% O₂, 101.3 kPa, 20 cm/s concurrent |
0.00 |
0.000 |
Outcome matches (no sustained spread; LOI margin -0.8%) |
BASS-II-14 |
Adapted Reference (unverified) | pmma (rod, msg-12cm) |
21.0% O₂, 101.3 kPa, 15 cm/s opposed |
1.22 |
1.824 |
Excluded from headline error metrics (+49.5% / +49.48% vs. unverified adapted value) |
- Radiative-loss term note:
src/engine/thermoCore.tsappliesthinSheetCharRadiativePenalty(+2.2%to+4.2%LOI shift on charring thin sheets:+2.2%fornomex,+4.2%forsilicone). Because git history is unavailable in this workspace (.gitabsent), it cannot be determined from history whether that term was introduced before or afterSAFFIRE-II-07andSAFFIRE-II-01were first checked; those two rows are therefore labeledExtinction check (calibration-informed). - Duct scale and specimen width handling:
sampleWidthMminsrc/data/nasaPsiExperiments.json(406 mmforSAFFIRE-I-01andSAFFIRE-III-01;50 mmforSAFFIRE-II-05,SAFFIRE-II-06,SAFFIRE-II-07, andSAFFIRE-II-01) is metadata only and is not read bycomputeThermoTelemetry; all sixSAFFIRE-*rows useductScale: 'saffire-40cm'(Dh = 0.40 m,0.88xwall-loss spread scaling), whileBASS-II-14usesductScale: 'msg-12cm'(Dh = 0.12 m,1.0xscaling) withAdapted Reference (unverified)values (excluded from headline error metrics). - 4-Fold Leave-One-Out Cross-Validation (LOOCV,
n = 4): Across the four cotton-fiberglass (SIBAL) Saffire runs (SAFFIRE-I-01,SAFFIRE-II-05,SAFFIRE-II-06,SAFFIRE-III-01), least-squares refitting of the scalar thin-sheet spread factor on each 3-run fold produces held-out absolute errors from4.46%(SAFFIRE-II-05) to16.33%(SAFFIRE-I-01), withSAFFIRE-II-06at7.20%andSAFFIRE-III-01at8.11%(MAPE =9.02%). Unlike the2-fit /2-held-out split (which calibrates once on the20 cm/sruns and extrapolates to25 cm/s), 4-fold LOOCV refits on3runs at a time across both flow speeds, so each fold evaluates interpolation or extrapolation from a different training subset.
The onboard Telemetry Assistant (src/hooks/useSafetyOfficerLLM.ts, src/engine/safetyAdvisor.ts, src/utils/hardwareProfiler.ts, src/workers/webllm.worker.ts) executes 100% client-side in the browser with zero cloud inference calls:
- Device-Adaptive Hardware Tiering & URL Overrides:
Append
?tier=constrained,?tier=balanced, or?tier=performanceto override device probing:?tier=constrained(default when WebGPU is unavailable, on mobile user agents,deviceMemory <= 2 GB, ormaxStorageBufferBindingSize < 128 MB): Disables WebLLM loading (allowWebLlm: false,contextWindowSize: 1024,maxTokens: 140,systemPromptBudget: 'compact') and runs strictly in deterministic mode.?tier=balanced(default whendeviceMemory <= 8 GB,maxStorageBufferBindingSize < 1 GB, orhardwareConcurrency <= 4): Defaults toQwen3-0.6B-q4f16_1-MLC(falling back toq4f32_1withoutshader-f16, orQwen2.5-0.5B-Instruct/Qwen2.5-1.5B-Instructif missing),context_window_size: 1536,max_tokens: 320,systemPromptBudget: 'standard', with 1-token shader warmup.?tier=performance(higher-headroom desktops): Defaults toQwen3-0.6B(q4f16_1/q4f32_1),context_window_size: 2048,max_tokens: 320,systemPromptBudget: 'full', with 1-token shader warmup. Desktop users can also manually chooseQwen2.5-1.5B-Instruct(q4f16_1/q4f32_1) from the composer backend menu.- Sampling parameters (
getGenerationParamsForModel):Qwen3-0.6Bruns on bothbalancedandperformancewithtemperature: 0.6,top_p: 0.95,top_k: 20,max_tokens: 320, andenableThinking: true;Qwen2.5-1.5B-Instructusestemperature: 0.15,top_p: 0.9,max_tokens: 320, andenableThinking: false.
- Pre-LLM Intent Router (
classifyIntent): Every prompt is classified before any model call intoINJECTION,AUTHORITY,OUT_OF_SCOPE,SMALLTALK,ABOUT,GLOSSARY,COMPARE_0G_1G,WHAT_IF,VALUES,EXPLAIN, orUNKNOWN. OnlyEXPLAINprompts (excluding pre-ignitionSTANDBYquench queries) invoke the local WebGPU model; all other intents are resolved deterministically or refused without calling the LLM. - Output Guardrails, Numeric Audit, and Badges:
- Before displaying an LLM reply (
evaluateAuditedReasoningResponse), the assistant isolates<think>...</think>blocks and checks the final answer for prompt-framing leakage, forbidden authority/safety/certification terms, truncation, bulleted telemetry dumps, and numeric values matched by physical-unit family against the live telemetry snapshot (bare numbers without units are rejected). Failures fall back to a deterministic reply with a visible audit note. - The audit checks that figures match the live snapshot; it cannot check every qualitative statement.
<think>traces fromQwen3-0.6Bare parsed into the collapsibleThoughtLineaccordion shown to the user above the response; the numeric and authority audits cover only the final answer and do not audit numbers inside<think>traces.- Response badges indicate provenance:
Deterministic,Local model (figures checked),Not modeled, andInfo.
- Before displaying an LLM reply (
- Not validated for spacecraft design or certification: Ignimetheus is a reduced-order screening and teaching tool, not validated for flight hardware qualification, crew safety decisions, or
NASA-STD-6001compliance. - Reduced-order steady physics: Uses steady-state 1D algebraic correlations with first-order thermocouple lag and linear burn-length integration rather than transient CFD or multi-step chemical kinetics;
Da*is an algebraic index derived from LOI margin and flow strain. - Illustrative flow rendering: Streamlines, velocity arrows, and flame plume shapes on the canvas are analytic illustrations, not a computed flow field.
- Modeled hardware assumptions: Wake Baffle (
0.72xroot flow factor at95 mm), Flame Arrestor Mesh (280 mmclamp,32.0% O₂limit), Pulse N₂ (-8.0% O₂for5.0 s), and the80 W/3.5 signiter pulse are model assumptions, not NASA test data. - Unmodeled sample width & radiative history:
sampleWidthMm(50 mmvs.406 mm) is not an input tocomputeThermoTelemetry.thinSheetCharRadiativePenalty(+2.2%to+4.2%LOI shift) is calibration-informed onSAFFIRE-II-07/SAFFIRE-II-01, and git history is absent here to verify when it was added. - Unmodeled hazards: Toxic species (
CO,HCN,HCl,HF), soot/smoke transport and obscuration, cabin depressurization timelines, electrical arc ignition energy (mJ), bulk cabin air heating, and future multi-minute flame spread are not modeled. - Dataset & assistant limits: For
BASS-II(PSI-25), the DOI (https://doi.org/10.60555/4qc4-de67->PSI-25Version 4) andCC0-1.0license are verified, whereasBASS-II-14's observed PMMA rod values (1.22 mm/sand112 mmat21% O₂, 15 cm/sopposed flow) were NOT FOUND acrossNASA/TM-20210011385(Section 4.1 Figs. 4.1–4.3 pp. 49–51 & Table A.3 pp. 106–107),PSI-25_Experimental table_BASS-II.csv, andBASS-II Test Matrix by PI Final.xlsx(neither tabular dataset contains spread-rate or burn-length columns; opposed-flow PMMA rods had57.15 mminitial length, were tested at<= 5 cm/snear21% O₂inP13a/P13b/P22a/P22b/P25, and spread at~0.04–0.08 mm/srather than1.22 mm/s/112 mm) and are kept asAdapted Reference (unverified)(excluded from headline error metrics). Other non-Saffire dataset records remain unverified (pending_verification) with assumed or synthetic fields, andPSI-98,PSI-99,PSI-100, andPSI-69identifiers remain(pending verification);FLEXdroplet surrogates are excluded from solid-fuel matching. Numeric auditing does not inspect<think>traces or every qualitative statement.
| Record ID | Source | Status | PSI ID / DOI Status |
|---|---|---|---|
SAFFIRE-I-01 |
Urban et al. 2017 (NASA NTRS 20180005168, Table 1) | in_scope_solid; summary verified in NTRS (O₂ midpoint, 101.3 kPa & igniter W assumed) |
PSI-98 / 10.60555/0t15-1z43 (pending verification) |
SAFFIRE-II-05 |
Urban et al. 2017 (NASA NTRS 20180005168, Table 1) | in_scope_solid; summary verified in NTRS (O₂ midpoint, 101.3 kPa & igniter W assumed) |
PSI-99 / 10.60555/2chp-m469 (pending verification) |
SAFFIRE-II-06 |
Urban et al. 2017 (NASA NTRS 20180005168, Table 1) | in_scope_solid; summary verified in NTRS (O₂ midpoint, 101.3 kPa & igniter W assumed) |
PSI-99 / 10.60555/2chp-m469 (pending verification) |
SAFFIRE-III-01 |
Urban et al. 2017 (NASA NTRS 20180005168, Table 1) | in_scope_solid; summary verified in NTRS (O₂ midpoint, 101.3 kPa & igniter W assumed) |
PSI-100 / 10.60555/2chp-m469 (pending verification) |
SAFFIRE-II-07 |
Urban et al. 2017 (NASA NTRS 20180005168, Table 1) | in_scope_solid; summary verified in NTRS (O₂ midpoint, 101.3 kPa, thickness & igniter W assumed) |
PSI-99 / 10.60555/2chp-m469 (pending verification) |
SAFFIRE-II-01 |
Urban et al. 2017 (NASA NTRS 20180005168, Table 1) | in_scope_solid; summary verified in NTRS (O₂ midpoint, 101.3 kPa & igniter W assumed) |
PSI-99 / 10.60555/2chp-m469 (pending verification) |
BASS-II-14 |
Adapted Reference (unverified) (PSI-25 Version 4; checked vs. NASA/TM-20210011385, PSI-25_Experimental table_BASS-II.csv, and BASS-II Test Matrix by PI Final.xlsx) |
in_scope_solid; Adapted Reference (unverified) (identifier/license verified; observed 1.22 mm/s & 112 mm at 21% O₂, 15 cm/s opposed flow adapted/unverified — not found in NASA/TM-20210011385 or PSI-25 tables, where 57.15 mm PMMA rods at <= 5 cm/s in P13a/P13b/P22a/P22b/P25 spread at ~0.04–0.08 mm/s) |
PSI-25 / 10.60555/4qc4-de67 (PSI-25 Version 4, CC0-1.0) |
BASS-II-22 |
Adapted reference (psi.nasa.gov, BASS-II) |
in_scope_solid; pending_verification (conditions & spread assumed) |
PSI-25 / 10.60555/4qc4-de67 (PSI-25 Version 4, CC0-1.0) |
BASS-II-31 |
Adapted reference (psi.nasa.gov, BASS-II) |
in_scope_solid; pending_verification (conditions & spread assumed) |
PSI-25 / 10.60555/4qc4-de67 (PSI-25 Version 4, CC0-1.0) |
BASS-II-52 |
Adapted reference (psi.nasa.gov, BASS-II) |
in_scope_solid; pending_verification (conditions assumed, spread derived) |
PSI-25 / 10.60555/4qc4-de67 (PSI-25 Version 4, CC0-1.0) |
BASS-II-45 |
Adapted exploration atmosphere (psi.nasa.gov, BASS-II) |
out_of_scope; pending_verification (synthetic 34% O₂ / 56.5 kPa scenario) |
PSI-25 / 10.60555/4qc4-de67 (PSI-25 Version 4, CC0-1.0) |
FLEX-09 |
ISS CIR FLEX droplet surrogate (psi.nasa.gov) |
out_of_scope for solid-fuel validation (liquid droplet mapped to cellulose surrogate; synthetic fields) |
PSI-69 / 10.60555/mbq8-0451 (pending verification) |
FLEX-18 |
ISS CIR FLEX droplet surrogate (psi.nasa.gov) |
out_of_scope for solid-fuel validation (synthetic elevated-O₂ silicone surrogate) |
PSI-69 / 10.60555/mbq8-0451 (pending verification) |
STD-6001-1G-NOMEX |
NASA-STD-6001B Test 1 criterion |
out_of_scope; pending_verification (<= 152 mm 1g screening criterion, illustrative 31 mm char, no spread rate) |
NASA-STD-6001B · Ref PSI-99 / 10.60555/2chp-m469 (pending verification) |
STD-6001-1G-PMMA |
NASA-STD-6001B Test 1 criterion |
out_of_scope; pending_verification (> 152 mm 1g failure example, illustrative 240 mm char, no spread rate) |
NASA-STD-6001B · Ref PSI-25 / 10.60555/4qc4-de67 (PSI-25 Version 4, CC0-1.0) |
- Live Demo:
https://ignimetheus.pages.dev/demo - Offline Usage: Local model weights download once; load online first, then it works offline.
# Install dependencies
npm install
# Start the local development server on http://localhost:3000
npm run dev
# Override hardware tier in the browser if desired
# http://localhost:3000/?tier=constrained
# http://localhost:3000/?tier=balanced
# http://localhost:3000/?tier=performance
# Run the unit test suite (thermoCore, units, ductProbes, safetyAdvisor, hardwareProfiler, toastQueue, inferenceBackend)
npm test
# Type-check with TypeScript
npm run lint
# Build offline-ready production bundle (dist/ + PWA service worker)
npm run buildTBD: add npm run validate(currentlynpm testrunsthermoCore.test.tsassertions against the Saffire benchmarks without printing the comparison or LOOCV tables).
AI Assistance Disclosure Code was written with AI coding assistants (Gemini and Claude) under the author's direction; the author ran the tests and checked results against NASA sources. Submission text and documentation were drafted with AI and edited by the author. Early reference-data entries drafted with AI were later replaced or relabelled against NASA's published table. The app embeds open-weight local models (Qwen) that phrase explanations; every figure comes from the solver.
- Source Code: Licensed under the Apache License, Version 2.0 (
Apache-2.0). - Dataset Metadata License:
CC0-1.0(src/data/nasaPsiExperiments.json). - NASA NTRS & PSI Attribution:
- Urban et al., ASGSR 2017 (
NASA NTRS 20180005168, Table 1, pp. 30–31). - NASA Physical Sciences Informatics (
psi.nasa.gov) references included insrc/data/nasaPsiExperiments.json:PSI-98(DOI10.60555/0t15-1z43, pending verification)PSI-99(DOI10.60555/2chp-m469, pending verification)PSI-100(DOI10.60555/2chp-m469, pending verification)PSI-25(DOI10.60555/4qc4-de67,PSI-25Version 4, licenseCC0-1.0;BASS-II-14observed values areAdapted Reference (unverified))PSI-69(DOI10.60555/mbq8-0451, pending verification)
- Urban et al., ASGSR 2017 (