Lab · Commissioning
The Green Light
Three faults. One reassuring screen. Find what it misses.
One coolant distribution unit. Three branches. A running pump does not tell the whole story. Follow the readings through three imagined commissioning tests.
Read the explanation
Three imagined commissioning tests, not real incidents or procedures. One coolant distribution unit (CDU) feeds three branches, A, B and C, each carrying 80 kW of rack heat into the water at the test load. A CDU's job is to supply the racks' water at the right temperature, pressure and chemistry (US DOE FEMP and NREL, 2024). At the design point each branch carries about 115 L/min, enough to take 80 kW with a 10 K rise: water leaves the CDU at 30 °C and comes back at 40 °C. Facility water reaches the CDU at 25 °C. In the third case the unit has a duty pump and a standby pump under one controller.
Case 1, the green light. The pump shows RUN. Branches A and C carry 115 L/min, 30 °C out and 40 °C back. Branch B carries 0 L/min; its return sensor is rising slowly in still water, and the rack's own inlet sensor is climbing. Check first: branch B's isolation valve. The pump is running and the other branches have flow, so the problem is local to branch B, and zero flow points to something closing that branch. The magazine's imagined test found a closed valve (pages 85 to 91).
Case 2, warm all over. The pump shows RUN and all three branches carry 115 L/min. The supply reads 33 °C and every return 43 °C instead of 30 °C and 40 °C. Facility water reaches the CDU at 28 °C instead of 25 °C. Check first: the facility water supply to the CDU, its temperature and flow. Flows and the 10 K rise are normal, so the technology loop is carrying the heat as designed; the supply is 3 K warm because the facility water arriving at the CDU is 3 K warm, and the exchanger cannot deliver water closer than 5 K to it. The problem is upstream, and a faster pump would not cool the water.
Case 3, the spare that did not start. Beforehand, P1 showed RUN on duty and P2 showed AVAILABLE on standby, with no fault on either. The test stops P1 to prove the changeover. P2 does not start, still reporting AVAILABLE with no run command, and every branch falls to 0 L/min. The controller shows FAULT. Check first: the controller both pumps depend on. A second pump removes one dependency but not the shared ones; this is the magazine's point about where the two lines meet (pages 97 and 100).
A green light is true about the part it watches. Where to look first comes from the pattern: one branch unlike the others points to something local, every temperature shifted by the same amount points upstream, and everything stopping at once points to something they all share.
Behind the model
Assumptions
- These are imagined teaching scenarios. They are not real incidents, test procedures or equipment.
- Each branch carries 80 kW of heat into the water, the liquid share of the magazine's 100 kW example rack (pages 18 and 23). Heat into room air, fans, other pumps and losses are left out.
- Water: specific heat 4.18 kJ/(kg·K) and density 1 kg/L. 80 kW at a 10 K rise needs 1.91 kg/s, about 115 L/min. Flows are shown to the nearest 5 L/min, as the magazine prints them.
- The CDU holds its supply at a 30 °C setpoint when it can. At the test load its exchanger cannot deliver technology water closer than 5 K to the facility water entering it, as in the magazine's worked example (page 28: facility water in at 25 °C, technology water out at 30 °C). With facility water at 28 °C, the best it can deliver is 33 °C.
- The pump holds the pressure the branches need, so closing one branch does not change the flow in the others.
- With no flow, a temperature sensor sits in still water, so it reports a trend (rising) rather than a steady value.
- The schematic shows functional connections, not a piping or control design.
Sources
- AI Data Center Energy Performance Framework: Commissioning & Performance Validation
ASHRAE, NEMA and PNNL · June 2026 - Best Practices Guide for Energy-Efficient Data Center Design
US DOE FEMP and NREL · July 2024 - NIST Chemistry WebBook (SRD 69), water: thermophysical properties of fluid systems
NIST · Undated web resource - Data Center Edition, Issue 01, pages 18, 23, 28 and 85 to 100
Data Center Edition · October 2026
Keep exploring
Cooling
Follow the Heat
Trace a rack’s 100 kW of heat to the air outside. Change the cooling and watch the consequences.
Planning
The Expansion Brief
Plan a 2 MW hall. Three decisions reveal the trade between space, water, energy and testing.
Capacity and load
Four Kinds of Number
Separate the size of a plan from the demand on a meter. Six worked examples make capacity, commitments and load easier to compare.