One rating, three consequences: transformer sizing and selection

Sizing a distribution transformer takes one line of arithmetic, and that line is not the hard part. The hard part is that the same choice sets the short-circuit current at every board below it, decides whether the switchgear you already specified is still adequate, and locks in a loss bill for twenty-five years — and those three consequences pull in opposite directions. A bigger unit runs cooler and costs more in no-load loss; a lower impedance costs less in load loss and raises the fault current your breakers have to interrupt. This guide runs the transformer sizing and selection tool through all of it. The worked example is a real run — every figure below was taken from the calculator, not typed in by hand.

What the tool is for

A preliminary two-winding transformer selection, extended to the winding configurations a real substation uses: two separate windings, three windings, split LV, and auto-connected. It sizes the unit, checks the impedance against the recognised minimum, and computes what that impedance does to the fault level — with the corrections IEC 60909-0 requires and hand calculations routinely skip.

S_required = S_demand × (1 + future/100) / target loading
In         = S_r × 1000 / (√3 × U_LV)
I″k        = c_max × In / (u_k/100)

For a 720 kVA demand with 25 % allowed for future load and a target loading of 80 %:

The rating ladder is the R10 preferred-number series every manufacturer's catalogue follows. It is practice, not a clause — IEC 60076-1 does not tabulate preferred ratings — so the tool says so and lets a project rating outside the series be entered directly.

The impedance floor is a clause. IEC 60076-5:2006, Table 1 — recognised minimum values of short-circuit impedance for transformers with two separate windings — reads: 4,0 % from 25 to 630 kVA · 5,0 % from 631 to 1250 · 6,0 % from 1251 to 2500 · 7,0 % from 2501 to 6300 · 8,0 % from 6301 to 25 000 · then 10,0 · 11,0 · 12,5 % and above. It exists to stop exactly the specification that looks attractive on a loss calculation: a low-impedance unit that quietly doubles the fault current on the board.

The correction that raises the fault current

I″k = c_max·In/u_k is the number in every handbook, and it is not the number IEC 60909-0 asks for. Clause 6.3.3 applies an impedance correction factor to the transformer:

u_Rr = P_krT / S_rT × 100 = 12 / 1250 × 100 = 0,96 %
u_Xr = √(u_kr² − u_Rr²) = √(5² − 0,96²) = 4,9070 %          (10d)

K_T  = 0,95 · c_max / (1 + 0,6·x_T)
     = 0,95 × 1,05 / (1 + 0,6 × 0,049070) = 0,9690           (12a)

K_T is below 1, so it reduces the transformer impedance and raises the fault current: 39,10 kA at the LV terminals instead of 37,89 kA, +3,2 %. The tool computes both and its validation panel pins the corrected figure. A hand calculation that omits K_T is not conservative — it is 3 % optimistic, in the direction that matters when a 40 kA board is being justified against a 37,9 kA calculation.

The one place the correction is properly left out is an OLTC auto-transformer whose u_k+ and u_k− both sit well above u_kr, per the NOTE to 6.3.3, and that is a switch in the form rather than an assumption in the code.

The tolerance nobody puts in the fault study

u_k = 5 % on the nameplate is a guaranteed value with a tolerance, and IEC 60076-1:2011 Table 1 — Tolerances, item 3, states it: for a two-winding transformer on the principal tapping, ±7,5 % of the specified value when the impedance is ≥ 10 %, and ±10 % when it is below 10 %. On the same 1250 kVA unit that turns one fault current into a band:

Switchgear has to be rated for 42,10 kA, and protection has to grade with 34,44 kA. Reporting only 37,89 kA gets both wrong.

And there is a second row in that table that catches people out: for any other tapping of the pair, the tolerance widens to ±10 % at impedances ≥ 10 % and ±15 % below 10 %. A transformer on a tap other than the principal one — which is where it will actually sit for most of its life — can legitimately come in at 4,25 %, and 1,05 × 1804,22 / 0,0425 is 44,6 kA. The tolerance is an input field for that reason. Enter what your case demands rather than accepting a default.

The loss guarantees carry tolerances of their own, in item 1 of the same table: +10 % on total losses, and +15 % on each component loss provided the total is not exceeded. A loss capitalisation is a commercial figure with a documented 10 % upside.

Losses, and where the curve actually sits

With P0 = 1,8 kW no-load and Pk = 12 kW load loss:

P_total(S) = P0 + Pk · (S/S_r)²

At the 72,0 % operating point that is 8,02 kW. The tool plots the curve and marks the loading at which the specific loss — loss per kVA delivered — is lowest, which is at √(P0/Pk) = 38,7 %, not at full load and not at the 80 % target either. Over ten years at 8760 h and 0,12 per kWh, the capitalised loss for this unit comes to 84 315, and that number moves more with the loading you actually operate at than with anything on the nameplate.

Which is the real tension in the sizing line at the top. A 25 % future allowance divided by an 80 % target loading is a 1,56 multiplier on the present demand, and it puts the unit at 72 % today — comfortably above the efficiency optimum and comfortably below the thermal limit. Halve the future allowance and you save no-load loss for twenty years and lose the headroom. That is a judgement, not a calculation, and the tool's job is to price both sides of it rather than to hide the choice inside a rule of thumb.

Three windings, and the negative reactance that is not a bug

Select three windings and the model changes shape. The three measured pair impedances are referred to side A, corrected individually, then converted to the star equivalent of IEC 60909-0 Formulas (11a) to (11c). An 11 / 3,3 / 0,4 kV unit with pair impedances of 10 %, 17 % and 6 %:

Z_B has a negative reactance, and that is arithmetically correct rather than broken. Formula (11b) is Z_B = ½(Z_AB + Z_BC − Z_AC), and on the corrected pair reactances above that is ½(9069,7 + 5553,3 − 14867,0) = −122 mΩ: whenever the HV–LV pair impedance exceeds the sum of the other two, the middle branch of the equivalent comes out negative. Real three-winding transformers do this routinely. It is a property of the star equivalent, not of the iron — no physical element is negative, and the terminal behaviour the model reproduces is entirely physical. With a 25 kA network feeder this unit gives 2,37 kA at the MV winding and 12,08 kA at the LV winding.

Each K_T is applied to its pair before the star conversion, which is what 6.3.3 requires and the order that a spreadsheet gets backwards.

Earthing, the vector group, and I″k1

The line-to-earth current depends on things the impedance alone cannot tell you, so the tool asks for them:

• the vector group decides whether a neutral exists on the faulted side at all — a YNd11 secondary has none, and no line-to-earth current is defined there;
• the neutral earthing enters the zero-sequence circuit as 3·R_N and 3·X_N, so a 10 Ω earthing resistor adds 30 Ω to Z(0), and an isolated or high-impedance neutral gives no low-impedance zero-sequence path at all;
• the transformer's Z(0)/Z(1) is a manufacturer figure. 0,95 to 1,0 is usual for a Dyn unit, which is why the earth-fault current on a Dyn-fed board can exceed the three-phase one — a result the LV short-circuit tool works through in detail.

Where IEC 60076-5 has no system short-circuit power to work with, the standard itself provides a documented fallback in its Table 2: 500 MVA for U_m of 7,2 to 24 kV, 1000 MVA at 36 kV European practice, and so on, with a NOTE that a system Z(0)/Z(1) between 1 and 3 should be assumed when it is not specified. A stated default from the standard beats a number nobody can trace.

How the implementation is checked

Eighteen cases run on every load. Five pin the sizing chain and the impedance band; the rest pin the IEC 60909-0 machinery the fault estimate rests on:

• Z_Q = 1,10 × 11 000 / (√3 × 10 kA) = 0,698594 Ω, Formula (4) · X_Q = Z_Q/√1,01 = 0,695127 Ω with R_Q/X_Q = 0,1, Formula (5)
• κ = 1,02 + 0,98·e^(−0,3) = 1,7460 for R/X = 0,1, IEC TR 60909-1:2002 Formula (68)
• u_Xr = √(5² − 0,96²) = 4,9070 %, Formula (10d) · K_T = 0,9690, Formula (12a) · corrected I″k = 39,10 kA
• the star conversion of Formulas (11a) to (11c): three equal pair impedances give each branch exactly half, and Z_A + Z_B reproduces the corrected A–B pair reactance
• the neutral earthing entering Z(0) as 3·R_N; Z(0) = Z(1) giving I″k1 = I″k3 by Formula (54); a delta secondary and an isolated neutral each giving no line-to-earth current; a solidly earthed Dyn11 giving one
• IEC 60076-5:2006 Table 1 at two different bands, and the sizing line itself

All eighteen pass.

What it does not do

This is a screening tool, and the specification it feeds is longer than what it computes. It does not settle:

• temperature rise and cooling class — IEC 60076-2, and the site ambient, which in the Gulf is the whole question
• insulation level and dielectric tests — IEC 60076-3
• overload capability — the tool takes an overload class as a note; the loading guide is IEC 60076-7, and a cyclic rating is a thermal calculation on the actual load curve
• inrush — the multiple is an input used for a screening current (10 × In here, 18 042 A) and a coordination note; the real curve is manufacturer data, and the HV protection has to clear it while still grading with the damage curve
• the short-circuit withstand of the transformer itself — the thermal and mechanical ability to survive the fault, which is the actual subject of IEC 60076-5 beyond its Table 1
• losses as guaranteed values, efficiency indices, noise, tank strength, or the utility's own requirements

P0 and Pk here are the values you entered, and they are the values a tender will guarantee to +10 %.

The short version

Two decisions come out of this exercise, and they are coupled. The rating decides your loss bill and your headroom; the impedance decides your fault level and your voltage regulation. IEC 60076-5 Table 1 puts a floor under the second so the first cannot be optimised into a switchgear problem — and IEC 60909-0's K_T, plus the ±10 % (or ±15 %) impedance tolerance of IEC 60076-1 Table 1, mean the fault current you hand to the switchgear supplier should be a band with a stated basis, not a single number carried over from a nameplate.

Size your own unit →