Part-Load DLC Startup Runs Hotter Than Full Load If Valves Throttle

The first racks are on. Information technology load is 10% to 20% [SENSITIVITY]. A branch is running below 35% of rated flow [TARGET]. Do you pinch the valve to hold temperature difference, or do you close the branch and send the flow to bypass? Recommendation R-09 answers that question in the direction operators find least intuitive: close the branch. Throttling to hold ΔT is the move that makes the plate laminar.

Two documents reach the same conclusion from opposite ends, and both put the danger at the start of commissioning rather than at design load. Conclusion 17 of the hydraulic basis reports that at full duty every pipe section runs far above a Reynolds number of 10⁴ and is fully turbulent, then says where the risk actually sits: at part load. A rack branch at 40 °C falls to a Reynolds number of 20,000 at 33% of rated flow, about 10,000 at 17%, and leaves the turbulent region entirely near 7%, at roughly 4,000 [MODELED]. The 35% line corresponds to a Reynolds number of 20,000, so that the heat-transfer correlations stay applicable and no laminar dead zone forms inside the cold plate [TARGET]. Below that line the instruction is explicit: do not throttle valves to hold temperature difference — close the branch and let the bypass or minimum-flow circuit carry it.

Cold Fill Moves the Reynolds Floor From 17% to 42%

On a cold start — 10 °C after filling, first circulation — the branch reaches a Reynolds number of 10,000 at 42% of rated flow rather than 17% [MODELED]. The order of events on a real project puts that case before any of the others: the loop is filled cold, circulated, and only then does information technology load arrive in stages.

If the minimum-flow line is not written into the control sequence and the rack energisation order, the low-load commissioning phase produces worse local overheating than full load. The failure has nothing to do with whether system capacity is sufficient.

The heat-rejection basis records the same error on the primary under a different name. At low load the control system tries to drop pumps to their minimum speed. If a branch then falls below its own minimum, approach temperature on the unit heat exchanger worsens. The symptom presents as short unit capacity, not short flow, and is easily misdiagnosed as a failed or undersized unit. The prescribed match is to take branches offline, not to lower flow everywhere.

A 30% Load Ride-Through Pass Is Not Full-Load Evidence

Recommendation R-B27 of the heat-rejection basis arrives at the same window from the evidence side. Ride-through time is inversely proportional to heat load, so a loss-of-power test run at 30% load returns a ride-through 3.3 times the full-load value — one divided by 0.3 — and will therefore systematically pass a system that would fail [SENSITIVITY]. The requirement attached is that the integrated test be run at the highest real thermal load obtainable, that dummy load make up any shortfall, and that the acceptance report record the actual heat load fraction and give the result scaled back to full duty by the same inverse law.

Low load is thus dangerous twice over. It is the operating condition in which the flow regime is worst, and it is the test condition in which every timed result looks best.

Stage deadbands have the same trap. Deadband tuned at full load almost necessarily oscillates on the ramp, and the ramp is the longest period of the plant's life. That check has to be repeated at actual low load. A dummy-load pass during commissioning does not substitute.

Minute Trends Cannot See a Ten-Second Plate Event

Conclusion B-69 of the same document sets a measurement floor that most of the argument depends on. The cold-plate-level event is on the order of 5 to 15 seconds, and the primary-loop temperature rise after rejection stops is about 2.4 K per minute, or 0.04 K/s [MODELED]. A building management system trending at one to five minute intervals records zero or one point across a ten-second event, so the instrumentation normally installed cannot observe the transient the acceptance criterion is written about.

The heat-rejection basis therefore requires sampling and storage at ≥ 1 Hz [TARGET] on the temperature, flow, pressure and electrical-state points that ride-through is read from, delivered with the model and the point list at Gate 2. A transmitter with a 30 s [MODELED] response time, used on a 10 s [MODELED] transient, returns the instrument's curve, not the system's.

That is an equipment specification finding hiding inside a hydraulics discussion. Whoever writes the point list decides whether the low-load overheating case can be demonstrated at all.

Close-the-Branch Is Settled; 35% Is Still Assumed

The controls basis carries the minimum-flow line as interface item TI-14, in the same terms: no rack branch flow below 35% of rated, and below it the branch closes while the bypass or minimum-flow circuit takes over, written into the control logic. It grades the criterion as settled and the 35% value itself as still an engineering assumption, which is the right split — the rule is sound, the number is a placeholder pending the cold-plate data that would confirm it.

The same basis fixes what the distribution unit does when its supervisor is unreachable: hold the current setpoint and keep running, never stop the pump, never close the main valve — and marks that behaviour as still requiring the supplier's written confirmation. Commissioning gate G3-07 closes the loop by requiring the minimum-flow line to be written into the management system logic and demonstrated under a genuine low-load condition, not inferred from a full-load run.

Limits and open items

Confirmed on the current record as sourced items: below the minimum-flow line the branch closes and bypass carries the flow; throttling to hold ΔT is forbidden; a low-load ride-through pass is not full-load evidence.

Target, not confirmed: 35% of rated flow. [TARGET]

Sensitivity, not evidence: a ride-through measured at 30% load is 3.3 times the full-load value. [SENSITIVITY]

Modelled and not frozen: Reynolds numbers of 20,000 / 10,000 / 4,000 at 33% / 17% / 7% of rated flow, 42% on a 10 °C first fill, 5 to 15 seconds, and 2.4 K per minute. Fluid properties and branch bore are assumed. No cold-plate internal geometry is included. [MODELED]

Held open, beside the value it governs: 35% of rated flow remains [TARGET], derived from a Reynolds screen using assumed fluid properties. The cold-plate manufacturer's own minimum flow, allowable temperature difference and maximum return temperature are not held [HOLD]. If those arrive with a stricter internal minimum, the branch line rises and the staged energisation plan has to be rebuilt around it; if they arrive with a looser one, the line is conservative and the bypass circuit is carrying flow it need not carry.

One condition inverts the recommendation's effect. If the minimum-flow line is implemented by throttling branch valves to hold temperature difference rather than by closing branches — which is the intuitive operator response and the one R-09 forbids — the Reynolds screen is defeated at exactly the load where it was written to apply, and the control system will report a correct temperature difference while a cold plate runs laminar. Nothing above reflects Engineer of Record review, supplier confirmation, commissioning acceptance or observed operation.

© 2026 K&K Data Service Inc. All rights reserved. Reproduction or republication is permitted only with clear attribution to K&K Data Service Inc. and a working hyperlink to the canonical URL of this article. Excerpts must preserve the technical context, maturity labels, assumptions, and limitations. No excerpt may imply project approval, field validation, certification, or endorsement that the original article does not state.

Source: K&K Data Service Inc., “Part-Load DLC Startup Runs Hotter Than Full Load If Valves Throttle,” https://www.kkdatasvc.com/lab/hydraulics-and-transients/low-load-startup-is-hotter-than-full-load/.