Chilled-water plants and heat rejection

In most large data centers, the heat collected by air handlers or coolant distribution units ends up in a water loop, and a central plant moves it out of the building. When the outdoor air is cool enough, the plant can reject the heat directly; when it is not, chillers use mechanical refrigeration to pump the heat "uphill" from the cool loop to a warmer condenser loop that can reject it to hot summer air. The plant is the largest single consumer of energy after the IT load, and its reliability determines whether the facility can keep running through equipment failures and power interruptions.

This lesson explains the refrigeration cycle and what sets a chiller's efficiency, the main chiller types, how heat leaves the site through cooling towers and dry coolers, how the water loops and pumps are arranged, how plants are made redundant and how they ride through a power failure, and the regulations affecting refrigerants.

The vapor-compression cycle

A chiller is a heat pump. Its refrigerant circulates through four components:

• Evaporator: a heat exchanger in which the refrigerant boils at low pressure and low temperature, absorbing heat from the chilled water.
• Compressor: raises the pressure (and therefore the condensing temperature) of the refrigerant vapor, using electrical energy.
• Condenser: a heat exchanger in which the hot, high-pressure vapor condenses, rejecting heat to condenser water or outdoor air.
• Expansion valve: drops the liquid refrigerant back to low pressure, completing the cycle.

Energy balance requires that the heat rejected at the condenser equals the heat absorbed at the evaporator plus the compressor work: \(\dot Q_{\text{cond}} = \dot Q_{\text{evap}} + W\). A chiller removing 5 MW from the data hall with 0.8 MW of compressor power rejects 5.8 MW outdoors.

What limits efficiency

The ideal (Carnot) coefficient of performance of a refrigerator depends only on the absolute temperatures at which it absorbs and rejects heat:

Real chillers achieve perhaps 50 to 65% of the Carnot value at full load. The key quantity is the lift, the difference between condensing and evaporating temperatures: the smaller the lift, the higher the COP. Lift falls when the chilled water is warmer (raising the evaporating temperature) and when the condenser water or outdoor air is cooler (lowering the condensing temperature). This is the thermodynamic reason behind the trend, discussed in Lesson 1, toward warmer chilled water.

Chiller types

Chillers are classified by compressor and by how they reject heat:

• Centrifugal chillers: a high-speed impeller compresses the refrigerant. They suit large capacities (about 1 to 10 MW per machine, 300 to 3000 tons) and achieve the best full-load efficiencies. Oil-free designs with magnetic bearings and variable-speed drives are very efficient at part load and low lift.
• Screw chillers: two meshing rotors compress the refrigerant; common from about 0.5 to 2 MW.
• Scroll chillers: orbiting scrolls, for smaller capacities, often packaged in modular units.

A water-cooled chiller rejects heat to a condenser water loop connected to cooling towers. An air-cooled chiller has its condenser coils and fans built in and rejects heat directly to outdoor air. Air-cooled chillers are simpler and use no water on site, but their condensing temperature follows the outdoor dry-bulb temperature, which is higher than the wet-bulb temperature a cooling tower can approach, so they are typically 30 to 50% less efficient at peak conditions. Many recent data centers use air-cooled chillers with integral free-cooling coils to avoid water use (Lesson 7).

Heat rejection

Cooling towers

A cooling tower rejects heat by evaporating a small fraction of the condenser water into the air passing through it. Because evaporation removes heat, the water can be cooled below the outdoor dry-bulb temperature, toward the wet-bulb temperature, which is the lowest temperature reachable by evaporation and on a humid summer day may be 8 to 10 K below the dry-bulb. Two quantities describe tower performance:

• Range: the temperature drop of the water through the tower (entering minus leaving), equal to the heat rejected divided by \(\dot m c_p\).
• Approach: how close the leaving water gets to the ambient wet-bulb temperature, typically 3 to 5 K for a well-sized tower. A smaller approach needs a larger tower.

Open towers expose the condenser water to the air, so water treatment and blowdown are needed to control scale, corrosion and biological growth, including Legionella bacteria, which can cause outbreaks of Legionnaires' disease from poorly maintained towers. Closed-circuit fluid coolers keep the process water in a coil, with a separate spray water circuit outside it, at some loss of efficiency.

Dry coolers and adiabatic coolers

A dry cooler is a large finned coil with fans, rejecting heat from a closed water or glycol loop directly to the air, without evaporation. It uses no water but can only cool the loop to some approach above the dry-bulb temperature, perhaps 5 to 10 K. Adiabatic coolers add pads or sprays that pre-cool the entering air by evaporation on hot days, using water only for the hottest hours. Where the facility water is warm, as with liquid cooling (Lesson 9), dry coolers alone can reject heat for most or all of the year in many climates.

Water loops and pumping

A typical chilled-water plant has two or three water loops: chilled water from the chillers to the air handlers or CDUs, condenser water from the chillers to the towers, and in many plants a separate loop through plate heat exchangers for economizing. Two pumping arrangements are common for the chilled-water loop:

• Primary-secondary: constant-speed primary pumps circulate water through each chiller; variable-speed secondary pumps distribute water to the loads. A decoupling bypass between them lets the two flows differ.
• Variable primary flow: variable-speed pumps send water through both chillers and loads, with a small bypass to maintain the minimum flow through each chiller. It uses fewer pumps and less energy, but needs careful control.

Pump power for a flow \(\dot V\) against a pressure rise \(\Delta p\) is

where \(\eta\) is the combined pump and motor efficiency. Because flow for a given load is inversely proportional to the water temperature difference, designing for a larger ΔT shrinks pipes and pumps dramatically: at a fixed system, pump power varies roughly with the cube of flow, as with fans.

Redundancy and ride-through

Chillers, towers and pumps are installed with redundancy, commonly N+1 (one spare of each type) or N+2 for large plants, and with dual piping paths or loops for higher reliability tiers. Concurrent maintainability, the ability to take any component out of service for maintenance without affecting cooling, usually requires valves arranged so that any chiller, pump or pipe section can be isolated, and sometimes a second, independent distribution path.

A power failure poses a special problem. The UPS keeps the IT equipment running, so the heat keeps coming, but chillers stop and take several minutes to restart once generators are online (often 5 to 15 minutes before full capacity returns). Without cooling, a dense hall can heat up beyond allowable limits in well under that time. Three measures bridge the gap:

• UPS-backed pumps and air handler fans, so that water and air keep circulating.
• Chillers with fast-restart capability.
• Thermal storage: an insulated tank of chilled water that the pumps draw from while chillers restart.

Refrigerants

Chillers use refrigerants with suitable boiling points and pressures. Older plants used CFCs and HCFCs, which deplete stratospheric ozone and have been phased out under the Montreal Protocol. Their replacements, hydrofluorocarbons (HFCs) such as R-134a and R-410A, do not deplete ozone but are powerful greenhouse gases: one kilogram of R-134a released has a global warming potential (GWP) of about 1430 kg of CO₂ over 100 years. The 2016 Kigali Amendment commits countries to phase down HFC production and consumption by about 85% by the mid-2030s. Newer chillers use low-GWP hydrofluoroolefins (HFOs) such as R-1234ze and R-1233zd, blends such as R-513A, or natural refrigerants such as ammonia, CO₂ and propane, each with trade-offs in flammability, toxicity, pressure and efficiency.

In the US, the American Innovation and Manufacturing (AIM) Act of 2020 directs the EPA to phase down HFC production and consumption by 85% by 2036, and the EPA's Technology Transitions rule sets GWP limits for refrigerants in many categories of new equipment, including chillers, with compliance dates phased in from 2025 (some data center cooling equipment has later dates). EPA rules under the Clean Air Act also govern refrigerant handling, leak repair and technician certification. Designers should check the limit and date that apply to each equipment category.

Many low-GWP refrigerants are mildly flammable (ASHRAE safety class A2L), which brings new requirements for machinery room ventilation, leak detection and equipment location under the mechanical refrigeration safety standards (ASHRAE 15 and, in Canada, CSA B52).

Lesson 6 follows the water from the plant to the coils: pump and system curves, pipe sizing, pumping arrangements, low ΔT and water quality.