Chiller Module Validation

Stage 1 — Isolated single chiller

Single-unit isolated chiller under variable chilled-water demand. Three commercial air-cooled equipment classes (screw, scroll, reciprocating) swept across part-load ratios, capacities, evaporator leaving temperatures, condenser entering temperatures, and simulation periods. Each scenario runs the equipment in CogenS and EnergyPlus with identical CAPFT, EIRFT, and EIRFPLR performance curves drawn from the production coefficient database.

Stage 1 is the foundational chiller validation. It answers the question every independent engineer asks first: does the chiller module predict the electric input that EnergyPlus would predict, across the equipment classes a project will actually use, at the chilled-water leaving temperatures and condenser conditions a project will actually see? On 189 of 189 scenarios, the answer is yes — within the strict ASHRAE Guideline 14-2023 tolerance band on annual electric energy, NMBE, CV(RMSE), and peak demand.

What the matrix covers

The matrix sweeps three air-cooled equipment classes (screw, scroll, reciprocating compressor families) across part-load fractions from 25% to 100%, three nominal capacities (350 / 700 / 1,400 kW cooling), three evaporator leaving temperatures (5.5 / 6.7 / 7.2 °C), and two outdoor dry-bulb regimes (mild 25 °C and hot 35 °C). Each combination runs through both engines with identical bi-quadratic CAPFT and EIRFT curves and quadratic EIRFPLR curves drawn from the production coefficient database — same equipment, same load, same condensing conditions. Then every output gets scored against the ASHRAE Guideline 14-2023 tolerance bands.

Headline result

Annual electric input, NMBE, CV(RMSE), annual cooling delivered, annual COP, and peak hour electric input all land inside the strict ASHRAE Guideline 14-2023 band across every one of the 189 scenarios. The CogenS chiller engine matches EnergyPlus at the level a lender's independent engineer expects.

Stage 1B - Annual 8,760-hour realistic-load head-to-head

Five chiller types -- Air-Cooled Electric Screw, Air-Cooled Electric Scroll, Water-Cooled Electric Centrifugal, Indirect-Fired Absorption (Hot Water / Steam), Direct-Fired Absorption (Natural Gas) -- each run for 8,760 hours against a DOE Reference Building Large Hospital cooling demand at Chicago O'Hare TMY3 weather. Plant capacity sized to 1.15x peak load. Real Neon DB curves loaded for every chiller variant; absorption Indirect / Direct backfilled with Single-Effect and Direct-Fired BiQuadratic CAPFT + EIRFT + (Cubic) SteamUseFPLR coefficient sets.

Stage 1B extends Stage 1 from constant + sinusoidal short-horizon scenarios to the full 8,760-hour annual operation a real chiller plant sees. The exercise surfaced and closed two production-engine bugs: (i) the BiQuadratic EIRFPLR coefficients shipped on Path B Centrifugal were truncated when the IDF builder serialised the curve to EnergyPlus' 1-D Quadratic field -- closed by collapsing the BiQuadratic at the constant condenser-temperature setpoint to a 1-D Quadratic that both engines consume identically. (ii) the chiller_simulator's cycling floor was using the curve-validity ep_min instead of the mechanical min_part_load_ratio -- closed in production; Path B Centrifugal where ep_min=0.12 < min_plr=0.25 was the exposed case.

Absorption Indirect runs at -0.57% / -0.57% (electric / cooling) using Single-Effect Cubic SteamUseFPLR coefficients backfilled into the Neon DB. Absorption Direct runs at -0.67% / -0.57% using Direct-Fired Quadratic SteamUseFPLR. All 5 PASS Tier 1 strict ASHRAE Guideline 14-2023 on annual sum, peak, and hourly NMBE / CV(RMSE).

Stage 2 — Multi-unit identical staging

Multi-chiller plant where two identical units share a single chilled-water load. Six exclusions documented in the result envelope reflect a known control-law difference between CogenS and EnergyPlus at part-load-low conditions where the two engines apply different staging philosophies. The remaining 39 in-scope scenarios pass under ASHRAE Guideline 14-2023.

Stage 2 tests how two identical chillers share a single chilled-water load. The relevant engineering question is the staging philosophy: should the controller load one chiller heavily and bring the second one in only when the first hits maximum, or should both chillers operate in parallel at roughly equal part-load?

EnergyPlus's Optimal scheme picks the first approach: load chiller 1 to its optimum part-load ratio and stage chiller 2 in at minimum capacity. CogenS picks the second approach: load both chillers near-equally. Both are physically valid. CogenS's equal-loading philosophy reflects the actual sequence-of-operation logic shipped by the major chilled-water-plant controls vendors (TRANE Tracer, Johnson Controls Metasys, Carrier i-Vu) in commercial buildings.

On screw and scroll chillers, the quadratic shape of the part-load efficiency curve means the two staging philosophies produce annual electric energy within ASHRAE Guideline 14-2023 strict tolerance (5-12% deviation across the in-scope envelope). On reciprocating chillers at deep part-load (PLR <= 0.25), the curve nonlinearity is steep enough that the two staging philosophies diverge by 16-28%. Those six reciprocating low-PLR scenarios are documented as an out-of-scope regime — a control-law philosophy difference between two physically valid engines, not a model error.

Stage 2 - Annual 8,760-hour multi-unit identical plant

Five chiller types x {N=2, N=3} identical-unit plants running 8,760 hours of DOE Reference Building Large Hospital cooling demand. Per-unit capacity = (1.15 x peak_load) / N units. EnergyPlus plant load distribution scheme set to SequentialUniformPLR to mirror the CogenS minimum-units-within-optimum staging + equal-PLR sharing convention.

Stage 2 multi-unit extends to 8,760-hour Hospital demand. Two more production-engine bugs surfaced and closed: (i) the chiller_simulator _calculate_staging fallback opened all N units at sub-minimum PLR when load fell below the smallest unit's mechanical floor; production fix keeps a single unit cycling on/off matching EnergyPlus' behaviour. (ii) the default EnergyPlus 'Optimal' load distribution behaves as sequential-fill on parallel-chiller plants -- switched to SequentialUniformPLR for cross-engine consistency.

The single Direct-Fired Absorption N=2 documented edge sits at -5.54% on the annual electric gate (0.54 percentage points over the strict +/-5%). All other six gates PASS strict on that case. Cause: per-unit capacity 1,275 kW is large relative to Hospital winter low-load hours, so most of the year is in deep sub-min cycling where absorption COP=1.0 amplifies any cycling-fraction divergence. Direct-Fired N=3 (per-unit 850 kW) PASS strict at -4.42%.

Stage 3 — Multi-size (2-type) dispatch

Two chillers of different sizes serving a single load — the standard topology where a smaller chiller handles base load and a larger chiller stages in for peaks. Ten scenarios span the dispatch regimes where the size-aware staging rule matters. CogenS uses a system-COP-optimized dispatch ordering; EnergyPlus uses Optimal load distribution.

Stage 3 tests the two-size dispatch rule. A small chiller and a large chiller share a chilled-water load. CogenS picks the dispatch order that maximizes system COP at each hour; EnergyPlus's Optimal scheme prefers the fewest-units-near-optimum-PLR approach. The same Jensen's inequality on the EIRFPLR curve that drives the Stage 2 staging difference appears here in a different form.

System-wide annual cooling delivered matches inside ASHRAE Guideline 14-2023 strict tolerance on every case. Per-chiller annual electric energy matches strict at full-load; at sub-full-load with two chillers operating simultaneously the per-type split lands inside the FEMP relaxed band (+/-15% annual), the documented Tier 2 regime where two correct engines apply different sequencing philosophies.

Stage 3 - Annual 8,760-hour 2-type / 2-size plant

Same-family 2-size plants on a shared chilled-water loop with SequentialUniformPLR distribution, 8,760-hour DOE Reference Building Large Hospital cooling demand. Air-Cooled Screw 1x1500kW + 1x1100kW (= 2600kW aggregate) and Air-Cooled Scroll 1x1500kW + 1x1100kW.

Stage 3 surfaced and closed a real CogenS production bug in the 2-type COP-optimization dispatch (cop_optimization.py find_best_combination): the prior _evaluate_fill_scenario rejected ALL combinations whose per-unit PLR fell below the chiller's mechanical min_plr -- producing zero cooling at all winter-load hours (140 zero hours / 8,760, 15.5% annual cooling deficit). Fix: clamp the per-unit PLR to min_plr for the COP-ranking calculation only and pass the actual sub-min PLR through to the chiller_simulator so its cycling_fraction logic produces the right heat input.

Three EP-test fixes also landed: (i) chiller flow mode NotModulated -> LeavingSetpointModulated with per-outlet SetpointManagers, (ii) CHW pump topology aligned with Stage 1B (98 kPa head, 0.72 wire-to-water efficiency, affinity-law cubic '0,0,0,1' eliminating phantom pump heat), (iii) LoadProfile:Plant flow_fraction Schedule:File scaling with load (Schedule:File Column Number=1; a previous draft used 0 which silently kept loop fraction at 1.0). Combined: -7.5% -> -2.7% electric gap, -3.7% -> +0.05% cooling gap. The Scroll +10.83% peak failure is a small CogenS-side curve-evaluation edge at h4789-h4792; all other gates PASS strict.

Stage 4 — Chilled thermal energy storage (standby physics)

Insulated stratified chilled-water tank holding cooling capacity for later delivery. Stage 4 tests the standby physics — how fast a charged tank loses its cooling capacity through ambient gains over a quiet period. Two tank volumes (10 / 50 m³) × three insulation levels (0.3 / 0.5 / 1.0 W/m²·K) = 6 cases, one week at 5 axial nodes, tank at 5 °C in a 25 °C plant room, validated against EnergyPlus WaterHeater:Stratified.

Stage 4 validates the chilled-water-storage standby physics. A tank charged to its low-temperature setpoint (5 °C) is held with no further charge or discharge in a 25 °C plant room, so ambient heat slowly degrades the stored cooling capacity over a week. Two tank volumes (10 / 50 m³) × three insulation levels (0.3 / 0.5 / 1.0 W/m²·K) are scored against EnergyPlus 26.1 WaterHeater:Stratified, which is validated for both hot and chilled stratified tanks.

These figures were re-scored against the current stratified-tank engine in July 2026, and chilled service is where the re-score moved most. Trajectory CV(RMSE) improves by roughly a factor of twenty (2.61-7.90% to 0.05-0.36%) and the week-long temperature rise now matches EnergyPlus to the reported precision on every case. Both engines PASS the Guideline 14 bands, so the verdict is unchanged at 6 / 6; what changed is how closely the trajectory tracks.

Standby capacity loss previously sat inside a wider band that was attributed to a surface-area convention difference with EnergyPlus. It was not a convention difference but an under-count: the engine integrated the cylindrical side wall only, which for this tank geometry is 8/10 of the surface, so the two end caps were missing. The current engine integrates the full surface and the diagnostic is now within 0.1%.

Stage 4 - Annual 8,760-hour chiller + chilled TES coupled (in progress)

1,000 m3 chilled-water tank + scaled DOE Reference Building Hospital 8,760-hour discharge profile + source-loop chiller charging during 22:00 - 06:00 off-peak window. EP topology: ThermalStorage:ChilledWater:Stratified with Source Side wired to a PlantLoop carrying Pump + DistrictCooling; Use Side wired to a PlantLoop carrying Pump + LoadProfile:Plant with negative-signed cooling demand.

Stage 4 Annual Phase 1 + Phase 2 close the architectural gap documented in the 2026-06-01 'runtime instability' note (the tank's Nominal Cooling Capacity field is SIZING info for an external source-loop chiller; EnergyPlus does NOT use it as an internal magic chiller when Source Side ports are blank). Phase 2 adds the source-loop wiring (Pump + DistrictCooling + ChargeAvailabilitySchedule + per-outlet SetpointManager + Sizing:Plant), three real fixes (LoadProfile cooling sign convention -> negative on CHW loop, EP 26.1 ThermalStorage variable name changes 'Heat Loss Rate' -> 'Tank Heat Gain Rate', source-side heat-transfer-rate sign negative during charging in EP).

Phase 3 (next session) replaces the CogenS-side hand-built daily charge / discharge schedule with the proper coupled dispatch from modules/technologies/chiller/profile_calculation/tes_dispatch.py::compute_tes_dispatch_pre_pass. EP topology is correct as-shipped; only the CogenS-side schedule generation needs the final wiring.

Stage 4b — Chilled storage charge/discharge cycling

Deep daily charge/discharge cycle on the chilled-storage tank: charge during off-peak hours, discharge to serve a daytime cooling load. Four scenarios across two cycling depths validate the cycling energy balance using an analytical engineering approach (cumulative energy in = cumulative energy out plus shell losses to within +/-1%).

This stage has NOT been re-scored against the current stratified-tank engine — every figure below describes the engine that shipped before July 2026. Re-running it would also prove little on its own: an energy-balance closure check verifies that the numbers are self-consistent, not that they match an independent simulation, so it closes on either engine. The regime itself — chilled charge and discharge cycling — is now cross-validated against EnergyPlus by the Thermal Energy Storage module validation, which covers 36 chilled-water scenarios spanning both duties against the 4 here, and against a real EnergyPlus reference rather than a self-check.

Stage 4b validates the chilled-storage tank under deep daily cycling. The tank charges to its setpoint during off-peak hours and discharges to serve the daytime cooling load. Energy balance must close (energy charged in = energy discharged out plus shell-loss gains) to within engineering tolerance over the full cycle.

All four scenarios pass the cumulative-energy-balance gate within +/-1%. The tank trajectory is monotone during each phase (strictly cooling during charge, strictly warming during discharge with no spurious reversals). The Stage 4 standby physics validated above already proves the underlying shell-loss model against EnergyPlus.

Stage 5 — Chiller + storage setpoint-buffer coupling

Coupled chiller-plus-storage system operating as a single plant. Twelve scenarios span the setpoint-buffer composition (chiller maintains the tank within a temperature band, tank serves the cooling load) under varied load profiles and storage sizing.

Stage 5 validates the chiller-plus-storage coupling using the setpoint-buffer composition. The chiller's job is to keep the tank inside a temperature band; the tank's job is to serve the actual cooling load. Twelve scenarios span two compressor types × two tank volumes × three part-load ratios, with the chiller's electric draw scored against the combined (load plus shell gain) demand.

These figures were re-scored against the current stratified-tank engine in July 2026, with only the storage engine swapped so the movement is attributable to the tank alone. All twelve still pass. The change is small by construction: the tank's shell gain is only one to two tenths of a percent of the cooling load, so correcting it by about a fifth cannot move the energy-balance or COP gates much. Tank drift roughly doubles, from 0.63-0.69 K to 1.04-1.16 K, and remains well inside the 2 K band.

This stage is still scored analytically rather than against EnergyPlus, and the re-score does not change that. An energy-balance check closes on either engine — it verifies that the numbers CogenS reports are self-consistent, not that they match an independent simulation. The chilled-water tank physics themselves ARE cross-validated against EnergyPlus, in Stage 4 above and across the complete charge and discharge duty cycle in the Thermal Energy Storage module validation.

Stage 6 — Total Cost of Ownership and financial KPIs

The investment-grade financial outputs for chiller-plus-storage projects: total cost of ownership, net present value, internal rate of return, simple payback, and MACRS-200%-DB depreciation. Twelve scenarios cover two cost bases and six financial profiles.

Stage 6 validates the financial layer to the standard a lender's independent engineer requires. Operating present value runs against EnergyPlus LifeCycleCost (NIST Handbook 135 reference implementation, EndOfYear discounting with per-cost-stream escalation). The investment KPIs (NPV, IRR, simple payback) and the MACRS-200%-DB depreciation schedule are validated against independent analytical references.

All gates pass at exactly 0.0000% across all twelve scenarios. The financial engines are technology-agnostic so the chiller Stage 6 results reuse the same financial-engine validation that supports boiler and cooling-tower projects.