DOE Big Hill atmospheric fractionation qualification
Issue #3305 separates refinery validation into explicit quality gates. The first DOE Big Hill increment qualified refinery-assay bookkeeping. These benchmarks advance the next gate by sending public assay representations through NeqSim’s rigorous DistillationColumn and checking process-level conservation, boiling-order separation, and repeatability.
Public source and evidence boundary
The source is the U.S. Department of Energy Strategic Petroleum Reserve Big Hill Sweet, sample MLI 009, assay date 1998-05-04:
- GovInfo SPR Crude Oil Comprehensive Analysis: https://www.govinfo.gov/content/pkg/CFR-2000-title10-vol4/pdf/CFR-2000-title10-vol4-part625-appA.pdf
- DOE SPR Crude Oil Assay Manual, 5th edition, 1 August 2024: https://www.spr.doe.gov/reports/docs/CrudeOilAssayManual.pdf
- DOE SPR assay landing page: https://www.spr.doe.gov/reports/Crude_Oil_Assays.html
The DOE manual documents ASTM D2892 atmospheric/light-vacuum distillation followed by D5236 for the residuum. It states that the distillation fractions are measured on a mass-percent basis and that volume-percent values are calculated using each fraction’s specific gravity.
This benchmark uses the measured mass-yield basis for the five intervals that have finite lower and upper boiling boundaries in the public table:
| Cut range (degF) | Weight % of whole assay | SG 60/60 F |
|---|---|---|
| 175-250 | 8.6 | 0.7815 |
| 250-375 | 15.2 | 0.8305 |
| 375-530 | 15.2 | 0.8623 |
| 530-650 | 11.1 | 0.9226 |
| 650-1050 | 30.3 | 0.9477 |
The five bounded cuts are normalized on their own basis. The C5-/175 degF light tail and 1050 degF+ vacuum residuum are not included because assigning finite boiling limits to those open-ended fractions would invent data that the published table does not provide.
Accordingly, this first tier is an integration and numerical-robustness qualification, not an independent validation of full crude-column product yields. It does not claim that the normalized five-cut slate represents the complete Big Hill crude.
Column benchmark
DoeBigHillAtmosphericFractionationTest performs the following workflow using public NeqSim APIs:
- Build an SRK system from the normalized measured DOE mass yields, cut specific gravities, and finite boiling ranges using
OilAssayCharacterisation. - Generate the five real NeqSim TBP pseudo-components.
- Feed the resulting broad-boiling slate to an eight-tray
DistillationColumnwith a reboiler, partial condenser, and one liquid side draw. - Solve at near-atmospheric pressure with the residual-monitored
MESH_RESIDUALcolumn solver; the solved state must satisfy the active full-MESH residual gate and guarded fallback products are not accepted. - Re-run the same initialized column to qualify repeatability rather than accepting a one-off solution.
The operating point is deliberately a reproducible screening case, not a reconstruction of a proprietary or historical refinery design:
| Quantity | Benchmark value |
|---|---|
| Feed flow | 5000 kg/h |
| Feed temperature | 550 K |
| Feed pressure | 1.5 bara |
| Internal trays | 8 |
| Feed tray | 4, bottom-up |
| Top pressure | 1.2 bara |
| Bottom pressure | 1.5 bara |
| Condenser mode | Partial |
| Condenser temperature | Solved; no fixed set point |
| Reboiler temperature | 600 K |
| Condenser reflux ratio | 1.0 |
| Liquid side draw | 10% of tray-4 liquid traffic |
The condenser remains partial and no fixed condenser-temperature specification is imposed. Top pressure and reflux ratio define the terminal controls while the top temperature remains a solved equilibrium/energy-balance result. Fixing both condenser temperature and reflux ratio overconstrains this broad-boiling screening case and can leave a thermally converged-looking profile with open tray material balances. The 600 K equilibrium-reboiler set point lies inside the bounded assay range and below the representative normal-boiling temperature of the heaviest cut, preserving both vapor traffic and a positive heavy bottoms product.
MESH_RESIDUAL uses inside-out initialization followed by rigorous residual monitoring. Interior tray temperatures remain solver variables governed by the stage energy balances. No tuning to a commercial process simulator is used.
Acceptance contract
The regression requires all of the following on each accepted solve:
- the column reports a solved state from
MESH_RESIDUAL, not guarded fallback products; - the full MESH infinity norm satisfies the solver’s active residual tolerance;
- the maximum normalized per-tray component material imbalance satisfies the column’s dedicated tolerance;
- overhead, liquid side draw, and bottoms are finite and strictly positive;
- external mass closure is within 5%;
DistillationColumn.getMassBalanceError()is finite and no greater than 5%;DistillationColumn.getEnergyBalanceError()is finite and no greater than 5%, with the energy-balance convergence gate enabled;- every pseudo-component closes across overhead + side draw + bottoms within 5% on a molar-flow basis;
- the lightest bounded DOE cut is enriched in overhead relative to bottoms;
- the heaviest bounded DOE cut is enriched in bottoms relative to overhead;
- composition-weighted representative boiling points order as overhead < side draw < bottoms;
- repeated-solve product mass flows and representative boiling points agree within 1%;
- the JUnit benchmark has a 120 s upper execution bound to fail closed on pathological solver stalls without treating shared-runner wall time as a performance claim.
The 5% process-balance gates are screening tolerances for this first broad-boiling integration case. They are intentionally much looser than the 1e-10 pseudo-component creation mass-closure gate because the column itself is an iterative process solver. Tighter refinery-specific balance gates should be introduced only after this public heavy-slate case establishes a stable baseline.
Complete modeled slate benchmark
DoeBigHillCompleteAtmosphericFractionationTest adds a second integration tier using the reusable DoeBigHillSweetAssay factory. Its primary sources are the official DOE SPR Big Hill Sweet comprehensive assay, reported 24 September 2021, and companion PIANO workbook.
The complete modeled feed contains all 12 components qualified by the characterization campaign:
- ethane, propane, i-butane, and n-butane allocated across the 1.70 mass% gas slice by normalizing DOE’s reported C2-C4 PIANO subset;
- the C5-175 degF cut with its PIANO-derived number-average molar mass and published upper boundary;
- six bounded 175-1050 degF petroleum cuts;
- the 1050 degF+ residue with its published lower boundary, specific gravity, and Watson factor.
The light-end allocation and zero sulfur or nitrogen values used where DOE leaves a cell blank remain explicit modeling assumptions. They are not additional measurements.
The complete-slate screening point deliberately reuses the proven bounded-column topology:
| Quantity | Complete-slate benchmark value |
|---|---|
| Feed flow | 5000 kg/h |
| Feed temperature | 550 K |
| Feed pressure | 1.5 bara |
| Internal trays | 8 |
| Feed tray | 4, bottom-up |
| Top pressure | 1.2 bara |
| Bottom pressure | 1.5 bara |
| Condenser mode | Partial |
| Reboiler temperature | 650 K |
| Condenser reflux ratio | 1.0 |
| Liquid side draw | 10% of tray-4 liquid traffic |
The test requires all 12 components to reach the column, a non-fallback MESH_RESIDUAL solution, positive overhead/side-draw/bottoms products, external mass closure, enforced energy closure, internal tray material closure, and per-component molar conservation. It also requires C2-C4 enrichment toward the overhead, 1050 degF+ enrichment toward the bottoms, composition-weighted normal-boiling-point ordering of overhead < side draw < bottoms, and 1% repeated-solve agreement for product flows and boiling descriptors. A 120 s timeout fails closed on a solver stall.
The complete-slate case is still an integration qualification. The DOE workbooks do not publish a matching atmospheric-column tray count, feed condition, pressure profile, furnace duty, stripping steam, reflux, pump-around duties, or product specifications. Therefore agreement with the DOE assay cut table would not by itself validate simulated plant product yields.
What this benchmark advances
This increment exercises, in one regression:
DOE assay facts -> reusable complete modeled slate -> standard and TBP components -> Stream -> rigorous DistillationColumn -> three refinery-style product draws
That closes an important integration gap between the characterization foundation and the refinery fractionation workstream. It also protects against future changes that would make heavy pseudo-components impossible to use in a near-atmospheric column even when assay bookkeeping still passes.
Remaining scientific gaps
These benchmarks do not validate:
- the normalized C2-C4 allocation as a complete measured gas analysis;
- generated pseudo-component molecular weights, critical properties, or acentric factors against independent laboratory property data;
- full-crude atmospheric product yields or cut-point recovery;
- pump-around heat duties, side strippers, or a refinery preheat/furnace train;
- vacuum fractionation;
- ASTM D86/D1160-to-TBP conversions;
- refinery conversion units.
The next scientific gate should use a public atmospheric operating case that includes enough column design and operating information to compare product yields and boiling ranges without tuning to an under-specified target. Only after that gate should #3305 advance to vacuum fractionation or conversion-unit models.