Where no-load loss really comes from
The W/kg figure on the steel datasheet is measured on an Epstein frame. Your core is not an Epstein frame, and the gap between them has a name.
· 8 min read · Engineering
The datasheet number is a laboratory number
Grain-oriented steel is characterised on an Epstein frame or a single-sheet tester: flat strips, magnetised along the rolling direction, at a stated frequency and induction. The resulting specific loss in watts per kilogram is honest, repeatable and not what your core will do.
A real core has corner joints where the flux has to turn out of the rolling direction, burrs and stresses from slitting, clamping pressure, and a stacking pattern that leaves small air paths at every step. Each of those adds loss that the strip test never saw.
The building factor
The ratio between the loss your core actually produces and the loss the datasheet predicts for the same mass at the same induction is the building factor. For a well-built mitred core it sits modestly above one; for a poorly built one it can be substantially worse, and the difference is the whole margin on a tight guarantee.
The building factor is not a universal constant. It depends on the joint geometry, the number of steps, the core diameter, the induction you have chosen, and on the workshop itself. Two works buying the same coil from the same mill will not get the same building factor, and the honest way to find yours is to compare measured no-load loss on completed units against the calculated figure and keep the ratio.
Induction is the lever with the sharpest edge
Specific loss rises steeply with induction, faster than linearly across the working range. Pushing flux density up to shrink the core saves steel and tank, and it costs no-load loss at an accelerating rate. It also erodes your overexcitation headroom, which matters on a network where voltage runs high at light load.
This is why no-load loss guarantees and core cost trade off so sharply, and why the optimum is rarely at either end. It is also why a design that just meets the loss guarantee at nominal induction can fail it in service on a network that runs a few per cent high.
What to check in a calculation
When reviewing a no-load loss figure, ask four questions. Which grade and thickness were assumed, and is that what purchasing will actually buy? At what induction was the specific loss read, and was it interpolated correctly from the table? What building factor was applied, and where did that number come from? And does the core mass include the yokes and the limb steps as built, not as idealised?
If any of the four cannot be answered from the calculation itself, the number is an estimate wearing a decimal point.