Data centers / Cooling control

Cooling loop optimization.

Cooling is the largest controllable load on your site. When the loops beneath it hold their targets, the same plant cools more racks on less energy — no chiller replaced, no new hardware in the critical path, nothing changed until you approve it.

  1. PUE pointsEfficiency the control layer was giving away — recovered, then held
  2. kW · $/MW-yrEnergy back from loops hunting, saturated or left in manual
  3. CapacityCooling headroom you can fill with racks, without a capital project
  4. ExcursionsStable supply temperature through AI load steps, not just at steady state
01 / What it costs you

A cooling plant rarely fails loudly. It degrades quietly.

Each of these is defensible on its own. Together they are a standing kilowatt charge, billed every hour, that never raises an alarm.

 What the site seesWhat it is quietly costing
01
Symptom
Chilled water supply temperature hunting around setpoint
Cost
Staging thrash and compressor cycling — wear on the most expensive asset in the plant.
02
Symptom
Pumps and tower fans never settling; valves hunting near closed
Cost
Continuous excess head and fan power, plus travel that shortens every VFD and valve in the loop.
03
Symptom
Operators running colder and wetter than design "for margin"
Cost
Every degree of unneeded margin is compressor work you paid for and threw away.
04
Symptom
Loops quietly left in manual after a commissioning issue
Cost
Efficiency now depends on who is on shift — precisely what continuous monitoring removes.
05
Symptom
An optimization platform on top, underdelivering its pilot number
Cost
It writes setpoints to a control layer that can't hold them. The brain is fine; the hands are shaking.
02 / What we optimize

The loops we take responsibility for.

Scoped from your points list and historian on day one. Select a control point to see what we do there.

 
Point 5 · Central plant
Differential-pressure reset

Typically the single largest recoverable pump-energy loop on a site. Often the first fix we rank.

Central plant

Largest energy impact per loop. Usually where the assessment pays for itself.

  • Chiller staging and sequencing — add/shed logic, minimum-run hysteresis, load balancing across unequal machines
  • Chilled water supply temperature and reset schedules; condenser water temperature, tower fan speed and approach control
  • Differential-pressure setpoint reset — typically the single largest recoverable pump-energy loop on a site
  • Primary/secondary balance, decoupler bypass, and the low-ΔT syndrome that quietly caps plant capacity
03 / The method

Why more tuning stops working — and what replaces it.

Past a certain deadtime, the problem is no longer the tuning. It is the model.

In a large chilled-water loop, or a CDU secondary, the sensor sits well downstream of the valve. The controller is acting on a measurement of the past. Conventional PID is built for deadtimes of a few seconds; a plant loop can run minutes.

Past that point, no set of PID constants both tracks load and stays stable. Tune it tight and it oscillates. Detune it and it stops controlling. Operators widen the margins, run cold — and the cost disappears into the baseline.

ControlSoft's MANTRA Internal Model Control acts on where the variable will be when the current move lands — not on error that has already happened. Same instruments, same valves, same BMS.

Long-deadtime loop · response
 

Model-based control predicts where the loop will be when the move lands — and settles on the first pass.

04 / Proof

The same physics, already solved.

From plants where an unstable loop does not merely waste energy. Client names withheld; full case studies on request.

±0.2ppm
held · 25%+ chemical cut
Municipal water utility
Long-deadtime dosing

A loop nobody could tune, because tuning was never the problem

Transport deadtime meant the sensor reported a dose injected minutes earlier. PID overshot and oscillated, so operators permanently overdosed to stay compliant. Model-based control with flow feedforward replaced reaction with prediction.

→ coolingAny loop whose sensor sits minutes downstream of the valve: chilled water supply, CDU secondary, large-loop return temperature.
~38hp/MGD
from ~58 · cycling ended plant-wide
4.2 MGD treatment plant
U.S. Army installation

Two minutes of deadtime had every pump and blower short-cycling

350 feet between the pumps and the flow meter meant a two-minute lag, so flow never stabilized and every downstream blower, UV system and pump cycled continuously. Model-based control settled the flow and the cascade unwound.

→ coolingPump and fan cycling — the least visible energy leak in a cooling plant and the fastest route to reclaimed capacity.
±2°F
from ±10°F · live in under two weeks
Continuous process line
4°F margin to product loss

Holding a tight band while flow and inlet conditions swung underneath it

One controller could not absorb steam pressure, inlet temperature and flow changes at once, so the outlet cycled through its limit and fouled the exchangers. Coordinated control with feedforward on both disturbances pre-empted them instead of chasing them.

→ coolingSupply-air and CDU supply temperature under AI load steps, where the safe band is narrow and the disturbance always arrives first.
05 / How it works

Start with one hall. Standardize across the fleet.

The assessment works from historian and BMS trend exports — no downtime, no controller changes, no site visit to begin.

01read-only

Assess

Worked from your historian and BMS trends. You receive a ranked loop inventory, quantified instability, recoverable kW, and a costed fix plan.

From $7,500 per site
02your approval

Optimize

Tuning validated across the full operating range, not just today's load — plus model-based control where the physics puts a loop beyond PID's reach.

Scoped from the assessment
03ongoing

Hold, then replicate

Continuous loop-performance monitoring keeps the savings. The proven hall becomes the template, so a new site inherits a tuned plant instead of rediscovering one.

Site or fleet
06 / Request a consultation

Show us your most difficult hall.

Two weeks tells you whether the kilowatts are there. You end up with a ranked list of the loops worth fixing and what each one is worth. If the numbers are not there, we will say so.

Generation closer to your critical path? Power generation control →
Cooling Loop Performance Assessment
Duration
Two weeks
Access
Historian / BMS trend exports. Read-only.
Systems
Any DCS, BMS or PLC — not tied to a manufacturer
Deliverable
Ranked loop inventory, quantified instability, recoverable kW, costed remediation plan
Suggested price per site
Trend reviewRemote, using exported BMS trend data
$7,500
Core offerRemote, live data, 2 weeks, read-only
$15,000
On-site, coolingOn site + live data, cooling plant
$20,000–25,000+ travel and lodging