Thermal storage: ice banks, chilled-water tanks & shifting the load

1 · Why store cold (or heat)

A conventional plant makes cooling at the exact instant it is used. That sounds efficient until you look at when the use happens: cooling load tracks the sun and the working day, so it climbs through the morning and peaks in the mid-to-late afternoon. Unfortunately that is also when electricity is most expensive (peak time-of-use tariffs and demand charges) and when the outdoor air — the sink the condenser rejects into — is at its hottest, so the machine's COP is at its worst. You are asked to make the most cooling at the moment it is hardest and dearest to make.

Storage rearranges the schedule. If you can make cold ahead of time and bank it, three good things follow at once:

• Shift the run-time to night. Off-peak power is cheaper, and the night air (or the water off a cooling tower) is cooler, so the same chiller runs at a higher COP than it ever could at 3 p.m. You pay less per unit and you make each unit more efficiently.
• Shave the afternoon peak. The store discharges through the hours of highest demand, so the site never draws its full instantaneous load from the grid. That directly attacks the demand charge — the part of a commercial bill set by your single worst 15-minute kW spike, which can be a third to a half of the whole bill.
• Downsize the chiller. This is the structural win. A plant without storage must be sized for the peak load. A plant with storage only has to be sized for the average load, because the store absorbs the difference. A smaller chiller is cheaper to buy, cheaper to run near its sweet spot, and cheaper to maintain.

The same logic runs in reverse for heat — a stratified hot-water buffer, a phase-change "thermal battery", or the sensible mass of a district-heating network all let a heat pump charge when conditions favour it and discharge when the building calls. Cold simply happens to be where the money and the equipment are most concentrated, so this guide leans on the cooling case.

2 · Sensible vs latent storage

There are only two physical ways to park thermal energy in a material: change its temperature (sensible heat) or change its phase (latent heat). Every cold store is one or the other.

Chilled-water storage is sensible. You cool a big volume of water down and let it warm back up on discharge. The catch is the usable temperature difference: chilled-water systems typically supply around 4–7 °C and return around 12–15 °C, so you only have a swing of roughly 6–10 K to work with. A modest ΔT means you need a large mass of water — thousands of cubic metres for a serious store — but the water is cheap, benign, and the chiller runs at its normal, efficient evaporator temperature.

Ice storage is latent. Freeze water to ice and you bank 334 kJ for every kilogram, all of it released back at a steady 0 °C as the ice melts. For the same stored energy an ice store is roughly a quarter to a fifth of the volume of a chilled-water tank — decisive when floor space or tank height is tight. The price of that compactness is thermodynamic: to make ice the evaporator must run several degrees below freezing (refrigerant around −5 to −10 °C instead of the +1 to +3 °C a chilled-water plant enjoys), and as the Carnot ceiling guarantees, a colder evaporator means a bigger lift and a lower COP. Ice-making typically costs on the order of 10–20 % more energy per ton-hour than chilled-water making — a penalty you accept in exchange for density, tariff arbitrage and a much smaller chiller.

3 · Ice storage: static and dynamic

Because ice is the dense option, it is where most of the interesting engineering sits. There are two families, split by how the ice is formed and released.

Static ice — ice-on-coil / ice bank

The classic and most common form. Submerged coils sit in a tank of water; a cold secondary fluid (usually a glycol/water brine at around −5 °C) circulates through the coils during charging, and ice grows outward on the coil surface until much of the tank is frozen into a solid block around the tubes. To discharge, you reverse the heat flow — warm return fluid runs through the same coils (or the building's chilled water passes through a heat exchanger) and melts the ice from the inside out, delivering near-0 °C cooling. Nothing physically moves except the fluid; the ice stays put. It is robust, simple, and forgiving, which is why "ice bank" is almost a generic term.

The one subtlety is that ice is a poor conductor. As the ice layer thickens on the coil, it insulates the fresh water beyond it, so the charging rate slows and the brine has to run colder to keep freezing — the COP penalty grows toward the end of the charge. Good designs stop short of a total freeze-up to keep that penalty in check.

Dynamic ice — harvesting and slurry

Dynamic systems keep the ice thin or fluid so the insulating-layer problem never bites. Ice harvesting freezes a thin sheet on a plate evaporator, then briefly reverses to hot gas to shear the sheet off; the flakes drop into a storage tank of water below. Because the ice is always thin at the moment of formation, the evaporator can stay comparatively warm and the charging COP holds up better than a fully-frozen ice bank. Ice slurry systems go further and generate a pumpable suspension of micro-crystals in water or brine — effectively a "cold fluid" you can move through pipes and meter precisely, with a very high heat-transfer rate on discharge. The trade is mechanical complexity: harvesters cycle a defrost, and slurry generators need scraped-surface or supercooling hardware, so there is simply more to maintain than a tank of coils.

4 · Chilled-water & stratified tanks

Where floor space allows and the COP penalty of ice is unwelcome, a plain tank of chilled water is the elegant answer — and its central trick is stratification. Cold water is denser than warm water, so if you feed and draw carefully it will layer itself: cold at the bottom, warm at the top, with a thin transition zone between them called the thermocline. Charge by pushing cold supply in at the bottom and drawing warm off the top; discharge by doing the reverse. A single tank thus behaves like two — a cold reservoir and a warm reservoir separated only by that mobile boundary — with no membrane or moving part.

The whole art is keeping the thermocline thin. A sharp boundary means almost all the stored water is delivered at full useful coldness; a thick, mixed thermocline blurs cold and warm together and wastes capacity. That is why stratified tanks use carefully engineered low-velocity diffusers at top and bottom to spread the flow gently and avoid stirring. A tall, slim tank stratifies better than a short, wide one, because the buoyant boundary has less area to smear across.

Buffer tanks — a reliability win, not just storage

Not every tank exists to shift load by hours. A much smaller buffer tank (or "volumiser") solves a different and very common problem: on a light cooling load, a chiller with too little water in its loop satisfies the setpoint almost instantly, shuts off, then restarts moments later — short-cycling. Every start is hard on a compressor: inrush current, poor oil return, and thermal cycling all accelerate wear, and the machine never runs long enough to reach steady efficiency. A buffer tank adds thermal mass and flywheels the loop, so the chiller runs in longer, smoother cycles and starts far less often.

That makes the humble buffer tank a genuine reliability component, not merely an energy one. Short-cycle count is one of the cleanest early-warning signals a condition-monitoring platform can trend: a rising start frequency on a compressor is often the first sign of an undersized loop, a failing control, or a load that has drifted away from the design point — a theme the final guide in this series picks up in full.

5 · "Ice chests" in the real world

The idea of a big insulated box of cold that you fill and draw down is everywhere once you look for it. Some run for hours, some for months:

• Cold-storage warehouses. The refrigerated mass of product and structure is itself a store: operators pre-cool hard overnight and coast through peak-tariff afternoons with the compressors throttled back, riding on the building's own thermal inertia.
• Ice rinks. The most literal ice chest of all — a slab kept frozen by kilometres of pipe, buffered by the enormous latent mass of the ice sheet itself. Rinks are natural candidates for ice or brine storage because the whole facility is already built around making and holding ice.
• District cooling. Central plants serving campuses, business districts or whole cities lean heavily on large ice or chilled-water stores to make cooling at night and distribute it by day, smoothing the load across thousands of connected tons and cutting the peak draw on the grid.
• Process & batch cooling. Dairies, breweries, food plants and chemical processes with short, intense cooling demands (a batch quench, a cleaning cycle) use ice banks so a small chiller charged over hours can dump its cold in minutes — sizing to the average, not the spike.
• Thermal batteries for buildings. Packaged ice-storage and phase-change modules now ship as "behind-the-meter" batteries that store cold instead of electrons — charged off-peak or off surplus solar, discharged to cut the afternoon demand peak.

All of them are sized in the same currency: ton-hours (one ton of refrigeration sustained for one hour, ≈ 3.517 kWh-thermal) or directly in kWh-thermal. A store rated at 2,000 ton-hours can deliver 500 tons for four hours, or 250 tons for eight — capacity and discharge rate are separate specifications, and both matter. A store that holds plenty of energy but cannot discharge it fast enough to cover the peak is as useless as one that discharges fast but empties in twenty minutes.

6 · The economics — where the trade tips

Strip away the hardware and thermal storage is a financial instrument: you are arbitraging the price of electricity across the day and buying down your demand charge, and paying for it with a modest efficiency penalty and some capital. Three numbers decide whether it pays:

• The tariff spread — how much cheaper night power is than peak power. A wide day/night spread makes shifting run-time immediately profitable.
• The demand charge — the $/kW you pay for your worst spike. Shaving the peak with storage attacks this directly, and it is often the single largest line on a commercial cooling bill.
• The COP penalty — the extra energy to make ice (or to run a colder evaporator). This is the cost side of the ledger; a chilled-water store pays almost none of it, an ice store pays 10–20 %.

There is a fourth, harder to price but real: resilience. A charged store is a few hours of cooling you already own — a ride-through if the grid dips or a chiller trips, and a hedge against demand-response events. Put the first three on a 24-hour profile and watch the peak flatten:

Two patterns show up as you play. Add storage and the chiller line drops toward the average load while the blue night-charging block grows and the red afternoon-discharge block does the peak-shaving work. Widen the night-tariff window and more of that charging lands on cheap power, so the saving climbs even though the chiller size does not change. There is a knee, though: past the point where the store can fully level the day, extra ton-hours stop shrinking the chiller and only buy a little more arbitrage — the classic partial-storage sweet spot, where a store sized to shave the peak (not to carry the entire load) gives most of the benefit for a fraction of the tank.

7 · Maintenance & monitoring

A thermal store is mostly a tank, a heat exchanger, and a control scheme — which means its failure modes are quiet and slow, and easy to miss until capacity has silently bled away. The things to watch:

• Tank integrity & insulation. A store only works if the cold stays in. Degraded insulation, a breached vapour barrier, or condensation tracking into the lagging turns your store into a slow leak of everything you paid to make. Rising standby losses — the store warming faster than it should between charge and discharge — is the tell.
• Coil & heat-exchanger fouling. Ice-bank coils and plate exchangers foul like any other surface, and a fouled charging coil forces the brine colder to hit the same ice fraction, deepening the COP penalty. The approach temperature drifting apart is the direct indicator.
• Glycol & inhibitor. The secondary brine is a chemistry to be managed, not filled and forgotten. Concentration governs the freeze point (get it wrong and you either can't make ice or you gel the loop), and the corrosion inhibitor depletes over time — annual sampling for concentration, pH and inhibitor level is basic care that protects the whole loop.
• Controls & state-of-charge sensing. The store is only as smart as its charge/discharge logic. Ice-inventory probes, tank temperature strings and thermocline sensors must stay calibrated, or the plant either wastes cheap night hours undercharging or runs the chiller into the peak it was built to avoid.
• Short-cycling as a monitored fault. Whether from an undersized buffer or a mis-tuned control, a climbing compressor start-count is a maintenance signal in its own right — cheap to trend, and a leading indicator of both wasted energy and accelerated compressor wear.

Key takeaways

• Storage decouples supply from demand — make cold when it is cheap and efficient (night, cool condenser), use it when it is dear (the afternoon peak).
• Downsizing is the structural prize — a plant with storage is sized for the average load, not the peak, and shaves the demand charge on top.
• Sensible vs latent — chilled water is simple but bulky (small ΔT); ice is ~5× denser via 334 kJ/kg latent heat, at the cost of a colder evaporator and 10–20 % less charging COP.
• Know your two families of ice — static ice-on-coil is robust and common; dynamic harvesting/slurry holds COP better but adds mechanical complexity.
• Buffer tanks are reliability, not just storage — thermal mass that stops compressors short-cycling, and short-cycle count is a clean monitored fault.
• Size in ton-hours, weigh the trade — arbitrage and demand savings against the ice-making penalty; partial storage usually wins the economics.

Continue the series