Cooling Tower Module Validation

Stage 1 — Isolated single cooling tower

Single-unit cooling tower across single-cell, multi-cell and free-cooling regimes, plus a Hospital DOE Reference Building 8760-h matrix covering all three fan-control types (Single-Speed, Two-Speed, Variable-Speed) and both polynomial curves (CoolTools 35-coefficient, YorkCalc 27-coefficient). 90 cases: 18 scored hour-by-hour against EnergyPlus (12 EnergyPlus-coupled + 6 Hospital DOE 8760-h), and 72 analytical part-load / curve cases scored against CogenS self-consistency gates with NO reference engine.

Stage 1 is the comprehensive single-tower validation envelope. It covers what one cooling tower does on its own, across the conditions a commercial chiller plant actually sees: warm summer operation, cold winter operation, single-cell and multi-cell configurations. Two layers of evidence support the result.

Layer one: 72 analytical part-load and curve scenarios

Seventy-two scenarios exercise the cooling tower module across two performance-curve formulations (CoolTools 35-coefficient and YorkCalc 27-coefficient), three commercial capacity classes (500 / 2000 / 5000 kW heat rejection), three part-load levels (30%, 50%, 75%), two wet-bulb conditions (cool 15 °C and warm 22 °C), and steady plus variable load profiles over one-week and one-month simulation periods. Every case passes five analytical engineering gates that together prove the simulator's outputs are physically consistent: heat rejection balances, the approach temperature is in the physically valid range, fan energy stays inside its installed nameplate, makeup water is non-negative, and supply temperature equals wet-bulb plus approach exactly.

The curve math underlying every one of these cases is EnergyPlus-canonical. The 35-coefficient CoolTools and 27-coefficient YorkCalc coefficient tables CogenS uses are byte-identical to the tables in the EnergyPlus 26.1 example file CoolingTower_VariableSpeed.idf. This means the approach-temperature correlation that drives every downstream output is intrinsically aligned with the EnergyPlus reference engine.

Layer two: 12 EnergyPlus-coupled hourly cross-validation cases

Twelve additional cases run the cooling tower side by side with EnergyPlus on a fully coupled chiller + cooling tower plant simulation. The topology is the same EnergyPlus uses in its own documentation (PlantApplicationsGuide_Example1.idf): a chilled-water loop with a synthetic cooling demand drives a Chiller:Electric:EIR; the chiller's condenser rejects heat to a condenser-water loop served by a CoolingTower:VariableSpeed. EnergyPlus runs the whole plant. The hourly chiller condenser heat-rejection profile is then fed to the CogenS cooling tower engine so both engines face the same load. The cooling tower's fan power, supply temperature, and makeup water are scored against ASHRAE Guideline 14-2023 tolerance bands.

The 12 cases span three operating regimes a commercial cooling tower will actually see in service:

Layer three: 6-test Hospital DOE 8760-h matrix at Tier 1 strict

The flagship Stage 1 result is the 6-test Hospital DOE Reference Building matrix. The DOE Reference Hospital model is run against Chicago O'Hare TMY3 weather in EnergyPlus to extract the 8760-hour cooling-demand profile (24/7 operation, ~4.79 GWh annual, ~2.2 MW peak). That identical profile is then fed to BOTH engines (CogenS and EnergyPlus) so they face the same chiller condenser heat-rejection load hour by hour. CogenS uses its production capacity_control + curve-type code path with the new ep_4step engine mode; EnergyPlus uses the matching native cooling-tower object — CoolingTower:VariableSpeed (polynomial), CoolingTower:TwoSpeed (Merkel UA-LMTD), CoolingTower:SingleSpeed (Merkel).

All six fan-control × curve combinations land at Tier 1 strict ASHRAE Guideline 14-2023 on annual fan kWh: Variable-Speed × CoolTools +2.65%, Variable-Speed × YorkCalc +5.43%, Two-Speed × {CoolTools, YorkCalc} +2.21%, Single-Speed × {CoolTools, YorkCalc} +1.84%. Supply temperature holds at the 30 °C condenser-loop setpoint exactly on both engines. The Variable-Speed VFD reduces annual fan kWh by roughly an order of magnitude compared with Single-Speed cycling on the same building (~5,000 kWh vs ~78,000 kWh) — a real engineering finding that both engines capture consistently.

Modeled on the same physics EnergyPlus uses

The CogenS CoolTools 35-coefficient and YorkCalc 27-coefficient approach correlations use the byte-identical coefficient tables shipped in the EnergyPlus example library, in the EnergyPlus-canonical form (air-flow ratio is cubed before being fed to the polynomial). CogenS' Single-Speed and Two-Speed paths use a Merkel UA-LMTD ε-NTU heat exchanger with saturated-air-enthalpy slope as the air-side capacity rate; UA is autosized so the model matches the cooling tower's rated nominal capacity at CTI standard design conditions (35 °C inlet water, 29.4 °C outlet water, 25.6 °C inlet air wet-bulb), with a separately autosized low-speed UA for Two-Speed.

The Variable-Speed fan-control logic replicates the four-step algorithm described in the EnergyPlus Engineering Reference: try the full-speed fan, then check whether free convection alone would meet setpoint, then check whether cycling the fan at the minimum air-flow ratio would meet setpoint, then modulate fan speed continuously between min and max if the setpoint can be held mid-band. Two-Speed cooling towers extend this with a HIGH / LOW cycling step for the regime where low speed alone is insufficient. The range temperature fed into the polynomial is solved iteratively from the coupled Twb + Tapproach + Trange = T_water,inlet energy balance — the same self-consistent operating point EnergyPlus computes.

Where the two engines model the same equipment differently

The dominant remaining differences after the same physics is on both sides are noise-floor: per-hour cycling distributions at very low load (heat rejection below 100 kW where the absolute fan power is a few hundred watts) where EnergyPlus' sub-hourly fan duty integration produces slightly different counts than CogenS' hourly resolution; and EnergyPlus' temperature-dependent water Cp (via its glycol object) versus CogenS' constant 4186 J/(kg·K) — about 0.3% effect on the energy balance at typical loop conditions. Both are documented in the audit-trace side-by-side CSVs; neither moves any annual-aggregate gate.

Stage 2 — Multi-unit / multi-cell staging

Multi-unit / multi-cell staging. 59 cases: 11 scored against EnergyPlus (8 summer unit/cell configurations at two PLR levels + 3 Hospital DOE 8760-h multi-cell scenarios), and 48 analytical staging cases with no reference engine. Two of the 11 EnergyPlus cases FAIL — the cause is a published design divergence, not an unexplained gap; see the limitations section.

Stage 2's analytical layer is an independent re-implementation of the CogenS CT staging band-fit rule that predicts unit and cell counts for each hour from the same load + capacity inputs the production simulator sees. The simulator's emitted units_operating and cells_operating arrays must match the predictor exactly (tolerance 0.01) — a stronger gate than any aggregate energy check.

Observed staging patterns confirm the band rule. At system PLR=0.75 in a 2-unit configuration the simulator stages 2 units (each at 75% utilization, in the [0.7, 1.0] band); at PLR=0.50 it consolidates to 1 unit at 50% (the band-fit fallback). The multi-cell variants (1u/2c, 1u/3c, 2u/2c) confirm all cells fire when a unit is on (no per-cell staging by design, fan power scales with cells_operating).

DOE Hospital 8760-h cross-engine layer

Three additional multi-cell scenarios run the cooling tower side by side with EnergyPlus on the full DOE Reference Building Hospital 8760-hour cooling-demand profile against Chicago O'Hare TMY3 weather. Both engines see the same hourly chiller condenser heat-rejection load; CogenS' multi-cell aggregate fan power, supply temperature, and makeup water are scored against ASHRAE Guideline 14-2023 Tier 1 strict tolerance bands. All three configurations (1-unit × 2-cells × CoolTools, 1-unit × 2-cells × YorkCalc, 1-unit × 3-cells × CoolTools) PASS with annual fan kWh deviation between +2.65% and +4.84%.

Stage 3 — Free cooling / waterside economizer

Free cooling / waterside economizer. 24 analytical cases (economizer enabled vs disabled, three wet-bulb conditions, two load levels, two load shapes) plus 1 EnergyPlus cross-engine case — Chicago January, a full month. The EnergyPlus case FAILS and is reported as a failure; its cause is measured and published in the limitations section.

Stage 3 validates the simulator's three-mode classifier: Mode 0 (no free cooling), Mode 1 (full free cooling: load <= qecomax), Mode 2 (partial free cooling: load > qecomax). At every hour the emitted fc_cap must match the mode's expected value, and any Mode 1 or 2 hour must have an actual supply temp at or below the 10 C SI activation threshold.

Empirically the simulator's WB-Reset-Min floor (21.11 C SI / 70 F IP) clamps supply-setpoint up so that achieved supply stays above 10 C even at WB=0 C (supply lands at 10.39 C — just barely above the threshold). Free cooling correctly does NOT activate at these conditions. This is the intended behavior: real cooling towers don't operate at chilled-water-range supply temps under default WB Reset control. Stage 3 confirms the code paths are exercised and produce strict zeros where expected.

Stage 4 — Chiller + CT coupling (chillers_ct mode)

Centrifugal water-cooled chiller + cooling tower in the production multi-pass feedback loop, mirroring the orchestrator Pass 1 -> Pass 4 sequence. 16 cases: 4 scored hour-by-hour against EnergyPlus over a full Chicago July, and 12 analytical coupling cases with no reference engine.

Stage 4 validates the production chillers_ct dispatch mode where chiller and cooling tower interact via a 4-pass feedback loop: chiller @ initial constant CT supply (30 C design) -> CT @ chiller heat rejection -> chiller @ actual CT supply feedback -> CT @ updated heat rejection. The test mirrors the orchestrator's exact pass sequence.

All eight gates pass with deviations at exactly 0.000% across every scenario. The chiller's average COP varies cleanly with PLR (4.41 at PLR=0.30 -> 5.25 at PLR=0.75) reflecting the part-load efficiency curve. The CT supply temp varies with wet-bulb (20.9-23.6 C at WB=15 C, 25.9-28.4 C at WB=22 C) reflecting the CoolTools approach correlation. Multi-pass convergence at exactly 0% across all 12 scenarios — under steady load + constant WB the chiller's operating point stays inside its calibrated range so feedback corrections vanish.

Stage 5 — CT Total Cost of Ownership + VFD-upgrade NPV / IRR / payback / MACRS

Eight cost bases — two synthetic capex/opex bases (SMALL 500 kW and LARGE 5 MW) plus six PHYSICS-DRIVEN bases derived from the Stage 1 6-test Hospital DOE 8760-h matrix (CogenS' validated annual fan kWh + makeup water at $0.12/kWh + $7/m³ commercial rates) — across six financial profiles (discount rate, inflation, study period, with/without 25% tax + 10-yr MACRS depreciation).

Stage 5 validates the financial layer against two independent reference families. The operating-stream present value is validated against EnergyPlus LifeCycleCost (NIST Handbook 135 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 — a polynomial-root IRR solver (vs CogenS's bisection method), a direct NPV recompute, and the published IRS GDS percentage tables.

All gates pass with deviations at exactly 0.0000% across all twelve scenarios. The financial engines are technology-agnostic (shared with boiler and chiller Stage 6 validations), so Stage 5 CT primarily confirms the cost-basis sweep — different absolute numbers and a different mix of capex vs energy vs O&M — doesn't expose any numerical edge case unique to the CT financial profile. It doesn't.

Stage 6 — Dry fluid cooler (air-cooled, no water)

A dry fluid cooler rejects heat sensibly to the outdoor DRY BULB with no evaporation and no makeup water — a different machine from the evaporative wet tower in Stages 1-5, and a separate equipment class in the platform. 92 scenarios: a 90-case parameter sweep (2 fan-control strategies x 3 capacities of 250 / 500 / 1500 kW x 5 part-load ratios from 0.25 to 1.00 x 3 contrasting weeks of the Chicago O'Hare TMY3 year) plus 2 full 8760-hour annual canonicals on a weather-driven load that traverses the whole operating envelope (part-load 0.10 to 1.00, dry bulb -22.8 to 35.0 degC, a 16x spread between the lightest and heaviest month).

All 92 scenarios pass at STRICT ASHRAE Guideline 14-2023 bands; no relaxed regime is claimed anywhere in this stage.

Two-Speed scores wider than Single-Speed (3.27% vs 0.99% worst-case annual fan energy) because its blended low-to-high fan state is a two-point interpolation of a curve EnergyPlus solves continuously — still comfortably inside strict.

Documented exclusions (not scored): part-load ratios at or below 0.10 against a climate-limited leaving setpoint, where both engines sit on the operating-envelope edge; and sub-freezing ambient, where a real dry cooler runs a glycol loop and the water-property assumptions of both models stop applying.