Certified Manufacturer Data vs Curve Fits in AHU Selection Software

Two AHU selection tools can quote the same fan, the same coil and the same wheel and disagree by 15 % on absorbed power. The difference is rarely a bug. It is that one of them asked the component manufacturer's own selection engine, and the other evaluated a polynomial fitted to a handful of published catalogue points. Both produce a number. Only one of them is the number the supplier will stand behind at the factory acceptance test.

What a curve fit actually is

A component maker publishes a catalogue: a fan's pressure/flow curve at a few speeds, a coil's capacity at a few entering conditions, a wheel's efficiency at a few face velocities. A curve-fitted tool digitises those points, fits a surface through them, and evaluates that surface at your duty point. Inside the fitted grid, and away from its edges, the result can be very close. The failure modes are predictable:

Extrapolation beyond the fitted range. The polynomial keeps returning smooth, plausible values well past the point where the physical component stalls, chokes or surges. Nothing in the arithmetic announces that you have left the map.
Collapsed dimensions. A catalogue curve is a slice at fixed geometry. Fin pitch, tube pattern, circuiting, motor winding and drive type are usually held constant in the fit and varied only by lookup table — so the tool cannot tell you what happens when the real selection would have changed the circuiting.
Efficiency treated as a constant. Affinity laws scale a fan's flow and pressure correctly. They do not scale motor and drive efficiency, which is where part-load absorbed power actually lives.

Where the divergence shows up

Component	What the supplier's engine resolves	What a fit typically flattens
Fan	Impeller + motor + drive at the exact operating speed, with the real efficiency map	Constant η, affinity-scaled from one reference point
Water coil	Circuiting sized to tube velocity, per-row air and water dP, condensate/bypass behaviour	Fixed circuit count, capacity interpolated from a table
Rotary HRS	Rotor depth, matrix geometry and face velocity, with the dP that follows	Efficiency vs. flow only; dP under-reported at overload
Filter	Media-specific initial and design dP for that exact cassette	One dP per ISO 16890 class — which the standard does not define

The rotary row is the one that catches people out. Push a wheel above its nominal duty and pressure drop rises steeply — a rotor sized for roughly half the airflow you are actually pushing through it can land in the several-hundred-pascal range rather than the 150–200 Pa a fitted curve suggests. That error goes straight into fan total pressure, straight into SFP, and straight into the energy class.

The boundary where real data still goes wrong

Calling the manufacturer's engine is necessary, not sufficient. Vendor libraries return values in the units their own application expects, and those are not always the units their field names suggest — a field called pressure_drop_kPa returning pascals, a capacity_kW returning watts, a length_m returning millimetres. Each of these is a factor-of-1000 error that produces output which is internally consistent and completely wrong.

The discipline is to convert exactly once, at the read boundary, and to keep every downstream calculation in one unit system. A tool that converts opportunistically, at each point of use, will eventually convert twice somewhere — and a coil quoted at 33.5 W instead of 33.5 kW is the kind of thing that survives all the way to a data sheet if nobody runs the reconciliation arithmetic.

Three questions for a demo

"Which supplier engines are called, and what version?" A concrete answer names products — fans, coils, filters, heat recovery — and can show a selection reference on the printout. "Manufacturer data" as an unqualified phrase means catalogue-derived.
"Change the airflow by 8 % and re-run." Real engines re-select: circuit count, rotor size or fan speed can change discontinuously, because the component range is discrete. A fitted surface moves smoothly and never re-selects anything.
"Show me the operating point against the fan curve." If the tool can plot where you sit on the actual curve, it has the actual curve.

Why this decides compliance, not just accuracy

ECP-05-2026 classes are boundaries. A unit whose fan power is 6 % optimistic does not become 6 % wrong — it becomes a class B unit reported as class A, on a document a consulting engineer will treat as a commitment. ErP (EU) 1253/2014 pass/fail behaves the same way at its own thresholds. When the margin between the quoted result and the limit is smaller than the uncertainty in the method, the class letter is not information.

AirSelect3D calls the suppliers' own selection libraries directly — Camfil, Ziehl-Abegg, Friterm and the rest of the wired-in makers — converts units once at the read boundary, and derives the data sheet, the drawing and the DXF from the same model, so the number you quote is the number the component maker computed.

Run a selection against real supplier data →