Multi-burner gas-fired hot-oil heater
MultiBurnerFiredHeater is a native ProcessSystem unit with a fuel inlet, a common
air inlet, a flue-gas outlet and optional hot-oil inlet/outlet. It exposes individual
burner states while preserving the total supplied fuel. An optional Cantera backend
solves real chemical reaction rates and ideal-gas mechanism thermochemistry.
The intended application is a reduced engineering model of an onshore gas-fired hot-oil heater. It is useful for separating changes in burner loading, local air capture, finite-rate CO burnout and stack dilution. It does not certify industrial CO emissions or replace a resolved flame/CFD model, furnace vendor data or a plant combustion safety system. No plant measurements or proprietary geometry are used by the executable example.
Qualification status
The implementation is on master, but scientific qualification remains incomplete.
The native seven/five burner demonstration
now passes against workspace Java classes in hosted CI with the default 1e-6
projection tolerance unchanged. The earlier native hydrogen-element rejection
has not been reproduced in this hosted environment, including with the old
mappings. Its historical cause is therefore not established. Supported ammonia
and methanol are now retained in the EOS projection without claiming that they
caused that rejection.
For the seven/five burner cases, omitted hydrogen fractions are approximately
8.72e-7 / 8.62e-7; unmapped mass fractions are approximately 1.52e-7 / 1.50e-7.
The native integration job also reports the result with the old mappings,
allowing a direct comparison instead of assuming the cause of a prior failure. Exact species remain available after rejection.
Every backend result now includes elementProjectionDiagnostics: inlet, exact
outlet and omitted molar atom flows, exact-mechanism and EOS-projection relative
residuals, and omitted species ranked by their contribution to each element.
Relative values use the inlet elemental inventory, matching the Java gate. For
an absent inlet element the relative value is null; absolute outlet and omitted
flows remain available. unmappedMechanismMassFraction is a separate diagnostic.
No diagnostic changes the existing acceptance tolerances or renormalizes species.
A bounded comparison with supplied field observations did not establish plant calibration. Air capture, fuel assay, geometry and temperature remain insufficiently identified. The example-only staged-air curve is a conditional sensitivity; it has not demonstrated the cause or amplitude of the observed plant CO decrease. Measurements, site details and external mechanism files are excluded from the PR.
Physical network
Each lit burner has a fixed-volume perfectly stirred reactor (PSR). Fuel shares are normalized over lit burner weights. The local captured air is an independently configured fraction of the common air stream. Products mix by conserving species, mass and enthalpy, and explicitly uncaptured air joins that mixing point. A common constant-pressure reacting plug-flow region then integrates burnout and cooling until its accumulated swept volume reaches the configured chamber volume.
Switching from seven to five burners increases fuel per lit burner when the total fuel stream is unchanged. It does not, by itself, determine local air, temperature or CO. Two air-routing examples are available: redistribute all common air among the five lit burners, or retain caller-specified burner capture fractions and mix the explicitly uncaptured air downstream. Neither option asserts air leaks through closed burner ports.
Burner IDs are caller-defined labels. In the example, IDs 5 and 6 can represent the two burners nearest the air inlet under the caller’s ordering. The reduced network does not resolve burner distance or inlet position. Spatial effects must be expressed through justified air-capture fractions or resolved by an external flow model.
Inputs and units
| Input | Unit or meaning |
|---|---|
| Fuel and air streams | NeqSim streams; composition, flow, temperature and absolute pressure |
| Burner index/state | Zero-based installed index; on or off |
| Burner fuel weight | Positive relative share, normalized over lit burners |
| Burner air-capture fraction | Fraction of total air; active fractions must sum to at most one |
| Burner reactive volume | m3, fixed stirred-zone volume |
| Chamber diameter/axial length | m, distinct from stack height |
| Common reactive volume | Gross cylinder volume minus active burner-zone volumes |
| Effective tube area | m2 |
| Effective convection coefficient | W/(m2 K) |
| Effective gas/tube emissivity | Fraction from zero to one, including the assumed view-factor treatment |
| Primary heating-area fraction | Fraction of tube and refractory area assigned to burner PSRs |
| Effective oil-side tube sink temperature | K, prescribed boundary; not automatically set to inlet or outlet oil temperature |
| Brick thickness/conductivity | m and W/(m K), caller-sourced |
| Refractory emissivity/convection | Fraction and W/(m2 K), caller-sourced effective values |
| Ambient cold-side temperature | K |
| Ignition initialization temperature | K, initial burner inventory; cold reactant inflow is unchanged |
| EOS projection tolerance | Maximum omitted mass fraction and fraction of each inlet element, default 1e-6 |
The example’s 10–40 MW scaling changes reactive volumes and tube area explicitly. It does not extrapolate a laboratory flame solely by multiplying its mass flow. Firing duty is fuel LHV input. Useful oil duty, shell loss and stack energy are reported separately; a nominal 30 MW firing heater is not a promise of 30 MW oil duty.
Heat and energy balances
Tube exchange combines convection and grey radiation:
\[Q_{oil}=A_t[h_t(T_g-T_o)+\epsilon_t\sigma(T_g^4-T_o^4)]\]Here $A_t$ is effective tube area [m2], $h_t$ is the convection coefficient [W/(m2 K)], $T_g$ and $T_o$ are gas and prescribed tube-sink temperatures [K], $\epsilon_t$ is effective emissivity, and $\sigma$ is the Stefan–Boltzmann constant [W/(m2 K4)]. The common plug-flow region integrates the spatially distributed heat independently as extra CVODES state variables, alongside accumulated swept volume.
The refractory inner temperature $T_w$ solves a monotonic steady balance:
\[h_r(T_g-T_w)+\epsilon_r\sigma(T_g^4-T_w^4)=\frac{k_r}{d_r}(T_w-T_a)\]$h_r$, $\epsilon_r$, $k_r$, $d_r$ and $T_a$ are the specified refractory-side convection, emissivity, conductivity, thickness and ambient temperature. Shell loss is the conduction term multiplied by effective refractory area. The cold boundary is prescribed; external shell convection, wall thermal inertia and refractory density/heat capacity are not modeled in this steady version.
The reported complete energy split is:
\[Q_{fuel}+H_{in,sens}=Q_{oil}+Q_{shell}+H_{stack,sens}+Q_{chemical,out}\]Chemical power uses complete gaseous products at 298.15 K (LHV water-vapor basis).
Stack sensible heat uses the actual outlet composition relative to 298.15 K.
Residual chemical power includes incomplete combustion and fuel slip, rather than
calling all unabsorbed fuel energy a stack sensible loss. Inlet sensible heat is
included explicitly. The detailed result contains the relative residual, individual
PSR balances and independently integrated post-flame wall-heat balance.
Each burner now separately rejects mass-flow residuals above 1e-7, relative
pressure deviations above 1e-5 and energy residuals above 1e-6 of the local
enthalpy/heat scale (with a 1 MJ/kg mass-flow scale floor). This prevents local
errors from cancelling in the total furnace balance. postFlameVolumeM3 reports
the actual integrated swept volume; specifiedPostFlameVolumeM3 reports the
requested volume. A relative mismatch above 1e-6 rejects the solve.
airChemicalPowerW exposes any chemical energy carried by the air stream, so
that reactive contaminants can be included explicitly in the full energy split.
If a hot-oil stream is connected, HotOilHeatBalance applies only the useful tube
heat to a cloned oil fluid with a NeqSim PH flash, preserving pressure and flow.
The existing outlet stream object is updated, so downstream equipment stays
connected. Oil enthalpy gain is checked separately. Cantera NASA formation enthalpy
and NeqSim EOS enthalpy have separate reference contracts; equality across the two
packages is not assumed. The fixed effective tube-sink temperature is a reduced
boundary condition, not a solved oil-coil temperature profile.
Optional detailed-chemistry backend
Java requires no Cantera dependency and does not select a chemical mechanism
implicitly. CombustionKineticsBackend accepts a version-1 JSON SI-unit contract.
examples/combustion/cantera_backend.py implements that interface with JPype and
Cantera 3.2. The adapter has no fabricated global rate or CO factor.
The executable demonstration defaults to Cantera’s bundled GRI-Mech 3.0 only as a software/physics smoke test. GRI was developed principally for natural-gas chemistry; that choice is not a propane validation claim. For C2/C3-rich industrial studies, select and qualify a suitable mechanism against relevant published measurements. AramcoMech 3.0 can be supplied externally; no mechanism file is redistributed in this change. Verify its source, license and validity before use.
Default component mappings are one-to-one and independently checked against the
NeqSim element database. Methane, ethane, propane and common stable combustion
species are mapped when present in the selected mechanism. Methanol (CH3OH) and
ammonia (NH3) are also retained: both have component and elemental records in
NeqSim. This preserves real stable species; it does not recombine radicals or
replace missing species with surrogate molecules. C4 mappings are enabled
only when the mechanism contains C4H10/NC4H10 and/or IC4H10. Unsupported inlet
species fail explicitly. GRI’s C2H2 remains in the exact result because NeqSim does
not have independently supported acetylene component/element data for this mapping.
The exact mechanism species and atom counts are retained, including radicals and intermediates. The stable EOS projection must satisfy both the total missing mass fraction and every elemental omission limit. Missing atoms are never renormalized into available species. An unaccepted projection throws before downstream execution and leaves the exact result available for diagnosis. Significant negative numerical species fractions are rejected; sub-1e-12 negative trace fractions are clipped and their total is reported explicitly, without renormalizing the remaining species.
Mechanism provenance includes solver version, species/reaction counts and a SHA-256 fingerprint of canonical resolved phase/species/reaction input data. Generated timestamps and phase state are excluded from that fingerprint.
Benchmark and mechanism qualification gate
examples/combustion/benchmark_catalog.json is the durable public evidence index.
benchmark_qualification.py validates it without Cantera and fails closed on incomplete
provenance. A conditions-only record may define a reproducible apparatus, pressure,
temperature range, residence time, equivalence ratios and fuel composition, but it
cannot contain observations and never counts as quantitative input evidence.
Promotion to quantitative requires redistributable machine-readable observations,
an exact source URL and SHA-256, a verified SPDX license, attribution and modification
notices, a declared measurement basis and uncertainty coverage for every value.
Uncertainty provenance is explicit: an evaluated estimate is not mislabeled as a value
reported by the primary paper. A qualification mechanism separately requires the exact
mechanism-file SHA-256, a verified SPDX redistribution license and the
qualification-candidate role.
The catalog now includes ReSpecTh file x00014002.xml, one lean ethylene JSR series
from Cong, Bedjanian and Dagaut (2010): 1 atm, 950-1100 K, 0.12 s, 30 cm3,
C2H4/O2/N2 = 0.0017/0.0102/0.9881 and equivalence ratio 0.5. Its seven temperature
points report C2H4, CH4, CO and CO2 in molar ppm. ReSpecTh labels the species-specific
uncertainties as evaluated standard deviations estimated from scatter and estimated
uncertainty; they are not claimed as primary-paper reported uncertainties. The source
is the CC-BY-4.0 ReSpecTh OSF deposit (DOI 10.17605/OSF.IO/NBMZV), archive SHA-256
246876488dd39ce1830610191852a95b5e0b7950ed49a920d1e341ffac6605dc, and extracted
XML SHA-256 e4546690562c936314416762cee143ccb03d0b3cd832d477d962e53d962d0f6e.
CRECK-S 2.0.0 is acquired from Zenodo DOI 10.5281/zenodo.22982859 rather than
vendored. The acquisition command requires HTTPS and verifies 495804 bytes, the
source-reported MD5 and SHA-256 before atomically publishing the file:
tmp_dir="$(mktemp -d)"
python examples/combustion/acquire_qualification_mechanism.py \
creck-s-2.0.0-zenodo-22982859 "$tmp_dir/CRECK-S_mechanism.yaml"
python examples/combustion/run_jsr_qualification.py \
"$tmp_dir/CRECK-S_mechanism.yaml" --output "$tmp_dir/jsr-result.json"
python examples/combustion/run_ignition_delay_qualification.py \
"$tmp_dir/CRECK-S_mechanism.yaml" --output "$tmp_dir/ignition-delay-result.json"
With Cantera 3.2.0 and the exact CRECK-S bytes, the unadjusted isothermal constant-pressure JSR comparison converged in 10-16 residence cycles. Across seven temperatures, the largest absolute normalized residual was 1.686; species RMS normalized residuals were 0.819 (C2H4), 0.715 (CH4), 0.313 (CO) and 0.485 (CO2). No mechanism parameters, observations or direct emission factors were fitted.
The catalog also includes ReSpecTh file x10004030.xml, the 16-point methane
mixture-4 shock-tube series from Aul et al. (2013): 1686-2248 K, 1.13-1.44 atm,
CH4/O2/Ar = 0.006667/0.013333/0.980000 and nominal equivalence ratio 1.0.
The measured ignition marker is the maximum pressure-rise rate. ReSpecTh stores
point-specific relative evaluated standard deviations estimated from generic
uncertainty; the catalog preserves those values and their absolute microsecond
equivalents without claiming that the primary paper reported them. The CC-BY-4.0
archive SHA-256 is
c9e36ca63050d352344f5101dd8eec18c8a8578051c1a989d03762216c504383 and the
extracted XML SHA-256 is
c196194b2d34696e6f62aab9e4eacc064a557820ba859fdabf6c85609d4b7e45.
The source-matched comparison uses an adiabatic constant-volume ideal-gas reactor and locates the maximum pressure derivative on a uniform time grid. It repeats every point with 1000 and 2000 intervals and fails if their ignition delays differ by more than 1% of the observation. With exact CRECK-S and Cantera 3.2.0, all 16 points passed that refinement gate: the maximum difference was 4.285 microseconds, or 0.25% of an observation. The maximum absolute normalized residual was 1.950, the RMS normalized residual was 1.165 and the delay RMSE was 115.661 microseconds. The observed delay sets only a five-times-long numerical search window; it does not alter the simulated state, kinetic mechanism or reported residual.
This methane series is an explicit extrapolation check: all temperatures exceed the 800-1500 K range described by the CRECK-S deposit. Agreement therefore does not extend the mechanism’s stated validity range and does not qualify flame speed, heater mixing, heat transfer or emissions.
--readiness now means that licensed, fingerprinted inputs exist for an executable
quantitative comparison. It is not a claim of heater qualification. The lean C2 JSR
series and extrapolative C1 ignition-delay series do not qualify C3/C4 chemistry,
spatial mixing, heat transfer, dry or reference-O2 emissions conversion, plant
behavior, safety or design:
python examples/combustion/benchmark_qualification.py \
examples/combustion/benchmark_catalog.json --readiness
The LPG and propane records remain conditions-only. Licensed C3/C4 observations with uncertainty, followed by declared acceptance criteria and cross-condition validation, remain the next chemistry dependency. GRI-Mech 3.0 remains a software demonstration.
CO, fuel slip and other species
CO is obtained from the reacted mechanism state. Reports include kg/h, dry ppmv and dry mg/Nm3 at 273.15 K and 101325 Pa. Reference-O2 correction is separately identified; it is invalid and returns null near or above 20.9% dry O2. Compare plant values only after checking normal/actual volume, wet/dry basis and reference O2.
Low CO alone is not evidence of good combustion. Inspect useful oil heat, burner
temperature, carbon-to-CO2 conversion and total organic-carbon fraction, including
oxygenates. The BURNING diagnostic requires positive useful heat, residual chemical power
below 1% of the supplied chemical power, and (for carbon-containing fuel) more
than 99% carbon conversion to CO2 with less than 1% organic carbon. The energy
criterion prevents a carbon-only check from hiding H2 slip. Carbon-free fuel
uses the energy criterion without dividing by a nonexistent carbon inventory.
These thresholds are diagnostic conventions, not burner safety limits. Other points retain
their species and energy diagnostics. Extinguished cases cannot explain a plant
that continues to supply useful hot-oil heat.
The adapter also reports CH4 and organic-carbon/NMVOC concentration surrogates in
mgC/Nm3. The organic sum includes exact C/H-containing gas species and oxygenates;
NMVOC excludes methane. These are carbon-mass diagnostics, not a calibrated FID
instrument response. Pollutants absent from a selected mechanism are marked
NOT_COMPUTED, never reported as a modeled zero. AramcoMech 3.0, for example,
does not include NO, NO2, N2O, SO2, SO3 or HCl in the supplied research mechanism.
Executable example and validation
After building the branch and installing Cantera 3.2 plus the normal NeqSim Python
runtime, run examples/combustion/multi_burner_fired_heater.py with the repository
root and a selected mechanism. --sweep adds an explicitly assumed staged-air
entrainment hypothesis at constant total air. Its nominal capture coefficient is
not fitted to plant measurements and is not a default native-equipment law.
The optional sweep declares --sweep-projection-tolerance 1e-3 and prints the actual
mass and elemental omissions. Its largest omitted elemental fraction in the
synthetic run was 3.37e-4. The native default remains 1e-6; callers must assess any
explicitly relaxed downstream projection. Exact mechanism emissions remain available.
The prescribed 550 K effective sink makes the final stack temperature relatively
insensitive and does not validate the plant temperature response.
For a built checkout with the normal NeqSim Python runtime and Cantera 3.2:
python examples/combustion/multi_burner_fired_heater.py --project-root . --mechanism gri30.yaml --sweep --scale-study --output heater_results.json
python -m unittest discover -s examples/combustion -p "test_*.py"
The example constructs SRK fuel, common-air and hot-oil streams, attaches the
backend, adds the native unit to ProcessSystem, executes seven/five lit burners
at constant total fuel/air, and prints complete energy/CO/conversion/projection
diagnostics. The two off burner IDs can represent the nearest pair only under the
caller’s ordering; distance is not a resolved model input.
--scale-study executes explicitly scaled 10, 20, 30 and 40 MW cases. The optional
constant-air --sweep declares a separate 1e-3 projection bound for frozen trace
oxygenates/intermediates (maximum 0.1% of any element), adjustable with
--sweep-projection-tolerance. The native default stays 1e-6. Every sweep row
prints the selected bound, actual omitted elemental fractions and omitted mass;
exact species and CO reporting remain authoritative. A stricter bound can reject
a quenched EOS projection without invalidating the retained detailed chemistry.
The prescribed 550 K effective tube sink and large illustrative area keep the
example’s final stack temperature close to that sink over the sweep. This can
hide a stronger bulk-temperature trend seen in a real heater. Inspect the burner
temperature separately, and supply actual oil-side/geometry boundaries before
attempting a plant comparison. A CO peak is not numerically tuned to the plant.
The public synthetic 30 MW GRI smoke case produced 26.27 MW useful oil heat, 0.09 MW shell loss, 3.65 MW stack sensible energy, 574 K flue temperature and 3.05/2.86 ppm dry CO for seven/five lit burners with all air redistributed. The maximum unmapped mass fraction was 1.6e-7, below the default 1e-6 gate. Full mechanism energy residual was below 6e-8 relative. These numbers validate execution, conservation and a conditional trend; they are not the private plant’s CO curve.
Focused Java tests cover ports, total-supply preservation, switching, outlet identity, failure invalidation and utility PH closure. Optional Python tests cover finite-rate chemistry, carbon-free hydrogen fuel, independently integrated nonzero wall heat, species/element balances, radiation effects, reproducible provenance and seven/five burner cases.
The detailed heater result and executable example retain
zoneConservationDiagnostics for each burner inlet mixer and PSR, the common
mixer including bypass air, and the post-flame PFR. Each boundary checks mass and
every mechanism element at the native detailed-mechanism limit of 1e-7, using
mechanism molecular masses and a 1e-20 mol-atoms/s element scale floor. These
checks are independent of the EOS projection limit, which remains unchanged.
PSR checks use the actual controller outlet mass flow. Inventories, absolute and
relative residuals, and acceptance-scaled residuals identify the first failing
zone; a rejected solve raises an error naming that zone and element rather than
returning a converged result.
The existing outlet convention permits negative solver mass fractions down to -1e-12 to be zeroed, without renormalization. The zone records disclose the zeroed negative mass fraction and the signed raw element residuals as well as the checked inventories. Positive intermediates are never removed. An element absent from the supply still has an absolute acceptance bound of 1e-27 mol atoms/s; numerical creation of a trace element fails this check. The carbon-free hydrogen regression therefore selects a tighter integration absolute tolerance of 1e-28 rather than relaxing element acceptance. With Cantera 3.2.0 and Python 3.12.14, 1e-18 and 1e-22 produced rejected trace carbon, while 1e-28 and 1e-32 passed. This is a numerical qualification boundary, not evidence of physical carbon production. No automatic solver-tolerance adjustment is performed.
The additional staged-air test sends 15% of common air around the burner PSRs and transfers 2% of tube/refractory exchange area in the primary zones. It checks all 16 boundaries at unchanged acceptance limits. This synthetic case tests accounting and thermal coupling; its air-capture fractions remain assumptions. The example also retains the mechanism fingerprint and the validity/value of the reference-O2 concentration correction alongside raw CO mass and dry basis.
The Optional combustion physics workflow installs Cantera 3.2.0 and runs these
Python tests for combustion changes; Java-only CI is not chemistry validation.
A separate native job compiles workspace Java classes and runs the seven/five
burner demonstration through JPype with the default projection tolerance. The
example prints retained omitted-atom diagnostics before rethrowing a rejected
projection.
Analytical diagnostics tests can run without Cantera. This validates the public native seven/five burner demonstration. Broader native
operating-envelope validation, mechanism qualification and plant calibration
remain separate acceptance gates.