Chiller Plant Analytics: kW/TR, Staging and Low Delta-T

Most chiller plants are judged on the chiller nameplate and nothing else. That is a mistake. The chiller is usually the most efficient thing in the room. The energy leaks are in the pumps, the towers, the staging logic and a building-side control problem called low delta-T. This is a narrow data study: how to compute plant kW/TR the honest way, what a healthy load curve looks like, and how to read six failure signatures straight out of the historian before anyone opens a panel.

Calculating plant kW/TR the honest way

kW/TR is the plant's fuel gauge. One ton of refrigeration (TR) is 12,000 BTU/hr of heat removed. Divide the electrical demand feeding the plant by the cooling it delivers and you get kW/TR, where lower is better. The trap is scope. Vendors quote the chiller compressor alone, where a good centrifugal machine sits near 0.55 to 0.62 kW/TR at design. The number that pays your bill is the whole plant.

Plant kW/TR needs three things trended together:

• Cooling delivered (TR): derived from chilled water flow and delta-T. TR = (GPM × delta-T in F) / 24, or in metric, kW of cooling = flow (L/s) × delta-T (C) × 4.19, then divide by 3.517 for TR.
• Total electrical demand (kW): compressors plus chilled water pumps plus condenser water pumps plus cooling tower fans. Every motor in the plant, not just the chiller.
• Both sampled at the same interval so they line up in time.

The gap between the two scopes is large. A compressor reading 0.60 kW/TR can sit inside a plant running 0.85 to 1.00 kW/TR once you add the pumps and towers. If you only ever meter the chiller, you will optimise the one component that was already fine and never see the pumps burning money at part load. Plant kW/TR is the only figure I trust for a whole-of-plant conversation.

Good and bad curves against load

A single kW/TR value tells you almost nothing, because efficiency depends heavily on how hard the plant is working. The right artefact is a scatter of plant kW/TR (y axis) against plant load as a percentage of installed capacity (x axis), built from months of data. The shape of that cloud is the diagnosis.

A healthy plant curve:

• Reaches its best kW/TR somewhere between 40 and 70 percent load, not at 100 percent. Water-cooled plants are usually most efficient at part load because condenser water is cooler and fans and pumps ride down on variable speed.
• Rises gently at very low load (below 25 to 30 percent), where fixed auxiliary draw dominates a small cooling output.
• Forms a tight band. Points at the same load percentage land close together.

A sick plant curve:

• Flat or falling toward zero load instead of rising, which means pumps and fans are not turning down. Classic fixed-speed auxiliary behaviour or a differential-pressure setpoint stuck too high.
• A wide vertical scatter at a given load, meaning the same demand produces different efficiency on different days. That is a controls or staging problem, not a mechanical one.
• A whole cloud that sits high, shifted up by half a point of kW/TR versus the design intent, which usually points at fouling or condenser water that is warmer than it should be.

The key move is to always read kW/TR relative to load percentage. A plant at 0.95 kW/TR at 30 percent load may be perfectly healthy; the same number at 80 percent load is a problem. Correct trend-log setup (covered at the end) is what makes this curve possible at all.

Spotting degradation over a season

Faults rarely announce themselves. They creep. The way to catch them is to bin the data by load percentage and compare the same bin across the season, so you are always comparing like with like. Take the 50 to 60 percent load bin in May and the same bin in August: if the median kW/TR for that bin has drifted up by 0.05 to 0.10 while ambient conditions are similar, something is degrading.

Two seasonal trends matter most. First, condenser approach temperature: the gap between the saturated condensing temperature and the leaving condenser water. It should be roughly flat within a load bin. A slow upward drift over weeks is fouling on the condenser tubes, the single most common efficiency loss in water-cooled plants. Second, evaporator approach, the mirror on the chilled water side, which drifts up with waterside fouling or low refrigerant charge.

Ambient is a confounder, so I normalise. Plot approach temperature and binned kW/TR against wet-bulb temperature, not calendar date, and the mechanical degradation separates cleanly from the weather. Without that step you will blame August heat for a dirty condenser.

Six signatures you can read from trends

These are the failure patterns I look for first. Each is visible in trend data alone, before anyone touches a valve. The table lists the specific points you need logged and the maintenance action each signature triggers.

Read them together, not in isolation. Low delta-T, bypass flow and pump overspeed usually travel as a set: they are three views of the same distribution problem on the building side. Condenser approach rise and evaporator fouling are genuine mechanical degradation. Excessive staging is a controls fault. The table tells you which trade to call.

Worked example: approach temperature to work order

Here is the full chain on one signature, using a week of 5-minute trends from a 500 TR water-cooled centrifugal machine. Two points are logged: leaving condenser water temperature and saturated condensing temperature (read from the chiller's refrigerant pressure). Condenser approach is the second minus the first.

Step 1: build the baseline.

A 5-minute interval gives 2,016 samples across the week. I filter to steady operation in the 60 to 70 percent load bin, discard start-up transients, and average. Leaving condenser water sits at 35.2 C, saturated condensing at 36.3 C. Baseline condenser approach is 1.1 C. That matches a clean condenser at this load.

Step 2: compare to the season.

The same 60 to 70 percent bin twelve weeks later, at a matched wet-bulb temperature, shows leaving condenser water at 35.1 C and saturated condensing at 37.7 C. Condenser approach is now 2.6 C. The rise is 1.5 C, and because I matched load and wet-bulb, weather is ruled out. This is fouling.

Step 3: translate to an energy penalty (indicative).

A widely used field heuristic is that each 1 C of condenser approach rise costs roughly 2 to 3 percent of chiller energy, because the compressor must lift against a higher condensing temperature. Treat this as an order-of-magnitude figure, not a guarantee; the true penalty depends on machine, lift and load. A 1.5 C rise therefore implies about a 3 to 4.5 percent chiller energy penalty. On this machine averaging 350 TR at 0.62 kW/TR, that is roughly 217 kW of compressor draw, so a 4 percent penalty is about 8.7 kW of avoidable demand. Across 4,000 run hours that is near 34,800 kWh a year; at AED 0.30 per kWh, close to AED 10,400 annually from one fouled condenser.

Step 4: raise the work order.

A condenser tube clean is a small planned job that costs a fraction of that annual penalty and pays back in weeks. So the trend triggers a specific work order: "Condenser approach on Chiller 1 has risen 1.5 C at matched load and wet-bulb over twelve weeks; brush-clean condenser tubes, inspect water treatment, and re-baseline approach after cleaning." That is a data-driven maintenance action, not a calendar-driven guess. The re-baseline after cleaning is what proves the job worked.

Low delta-T is a building problem, not a chiller problem

The most misdiagnosed signature in the sector is low delta-T syndrome. The plant is designed for, say, a 6.7 C (12 F) chilled water delta-T, but the trends show it running at 3 to 4 C. To hit the cooling demand at that thin delta-T the pumps push far more flow than design, auxiliary energy balloons, and chillers stage on early. Efficiency collapses even though every chiller is mechanically healthy.

Here is the practitioner truth: low delta-T is almost always a building-side control problem. The usual causes are stuck-open or oversized coil control valves, three-way valves left in place, dirty or undersized coils, and setpoints that keep valves wide open long after the space is satisfied. The chiller sees the symptom and never causes it. You confirm this from data by watching the return water temperature stay stubbornly low while coil valves sit near fully open, and by seeing bypass flow appear in the decoupler as excess primary water short-circuits back.

I have watched capital cases built to swap perfectly good chillers because the plant "could not keep up." The data said otherwise: the machines were fine, the valves were not. Fixing the building side recovered the delta-T and the apparent capacity problem vanished for a small fraction of the replacement cost. Always test the building side before you sign a chiller purchase order.

The trend-log configuration you need first

None of the above is possible without the right data, and this is where most plants fail before they start. If the BAS logs a handful of points every fifteen minutes, you cannot compute plant kW/TR, you cannot build a load curve, and you cannot separate fouling from weather. The analysis is only as good as the trend configuration underneath it. Getting these points specified in the controls contract is the real starting line; I cover how to write those clauses in BMS specification clauses.

Minimum points to trend, per plant:

• Chilled water supply and return temperature, and chilled water flow (for TR and delta-T)
• Condenser water supply and return temperature, and condenser water flow
• Saturated condensing and saturated suction temperature per chiller (for approach)
• Electrical demand (kW) for compressors, chilled water pumps, condenser water pumps and tower fans
• Per-chiller run status and load percentage, plus stage-up/down events
• Pump VFD speed and the differential-pressure setpoint and sensor value
• Outdoor dry-bulb and wet-bulb temperature (the normaliser)

Logging parameters:

• Interval: 1 to 5 minutes for plant analytics. Fifteen-minute logs hide staging and transients.
• Change-of-value plus periodic: capture events (a stage change) and a guaranteed periodic sample.
• Retention: at least 12 to 18 months, so you can compare seasons.
• Timestamps synchronised across controllers, so power and cooling line up for kW/TR.

Retaining months of dense trend data quickly outgrows the BAS itself, which is why plant analytics usually lives in a dedicated historian rather than the controller. For getting BAS trends into that historian cleanly, see SCADA historian integration; for the storage maths of keeping 5-minute points for years, see sizing a process historian. Set the trend configuration up on day one. Retrofitting a season of history you never logged is the one thing analytics cannot do for you.

Conclusion

A chiller plant tells you what is wrong if you log the right points densely enough and read them against load and weather. Compute kW/TR for the whole room, plot it against load percentage, bin by load to catch seasonal drift, and let the six signatures route work to the right trade. And when the delta-T collapses, look at the building before you look at the chiller. The cheapest fix is almost always upstream of the plant.

Independence note: I am not affiliated with any chiller manufacturer, BAS vendor or analytics platform, and nothing here is a product endorsement. The energy-penalty figures are indicative field rules of thumb, not guaranteed savings; always validate against your own metered data. For authoritative reference on cooling performance and efficiency method, see ASHRAE and the US DOE HVAC and water heating resources .