Aluminium Smelter Transformer: How to Size Six Rectiformers and a 220 kV Supply for a 420 kA Potline, and Why the DC Bus Decides the Design

The answer first: a 420 kA potline is a 514 MW DC load at 1,224 V, so the rectiformers and the DC bus — not the AC demand — decide the substation

A 288-pot aluminium potline running at 420 kA and 4.25 V per cell is a 514 MW DC load at 1,224 V. Add 45 MW of auxiliaries and the plant draws 567 MW, 590 MVA at 0.96 power factor — which is two 400 MVA 220/33 kV mains at 74% each. But nothing in that AC arithmetic tells you how to buy the transformers that actually matter. Six 87 MVA rectiformers, each delivering 70 kA at 1,224 V, decide the design, and three numbers on their nameplates decide whether the potline runs: 8.5% impedance (which sets current sharing to ±2% and DC bus fault current to about 500 kA per group), a vacuum on-load tap changer with at least 16 steps (which sets power factor and harmonic injection), and an eddy-loss-tested winding built to IEC 61378-1 rather than IEC 60076 alone. Get those right and the transformer package costs USD 30–40 million, roughly 4–5% of an 840 million smelter; you can reproduce the DC and MVA steps with the power system calculators. Get them wrong and the failure mode is not a failed transformer — it is a frozen potline, four to six weeks of lost production and USD 20–40 million of scrap bath and re-bricked cells.

Key takeaways

• Size the rectifier bank on DC, not AC: 288 pots × 4.25 V = 1,224 V; 420 kA = 514 MW DC; 45 MW of auxiliaries brings the plant to 567 MW / 590 MVA.
• Six 87 MVA rectiformers, not one large unit: 70 kA each, 12-pulse per group staggered to 72-pulse overall, and N-1 operation at 5/6 current instead of a potline trip.
• 8.5% impedance is a three-way compromise: current sharing between parallel groups, DC bus fault current, and reactive power consumption. Specify impedance tolerance at ±2.5%, not ±7.5%.
• Control voltage with the on-load tap changer first and the thyristor second — every 1% of phase-angle control below rated voltage costs about 0.6% power factor and adds harmonics.
• A smelter is a 590 MVA load that must survive the loss of one 400 MVA main, so N-1 compliance comes from a contractual current-reduction protocol (to 285 kA within 15 minutes), not from oversizing the mains.

What electrical load does a 420 kA potline actually present?

The potline is a single series circuit. Every pot carries the same 420 kA, and the voltages add: 288 cells at 4.25 V average give 1,224 V DC from end to end. That is the number the rectifier must produce, and it is a low voltage as far as insulation is concerned — 1.2 kV DC — and an enormous current as far as busbars, rectifiers and current sharing are concerned.

Two things follow immediately. First, 590 MVA is the largest single load most grids will ever connect at 220 kV, and it is remarkably steady — a potline is a constant-current load, not a motor load. Second, 14.8 kWh/kg means electricity is roughly 40% of the cost of the metal, so every kilowatt of transformer loss is a line item in the smelter's cash cost, which is why total cost of ownership rather than purchase price decides these tenders. That is why the loss evaluation in this application is done at USD 0.05–0.07 per kWh and 8,000–8,500 hours — the same arithmetic as the transformer sizing and budget formula, and why a transformer that is 0.1% less efficient can lose a tender worth USD 1.2 million per unit.

Why 4.25 V per cell and not 4.0 V? The energy trade-off behind the DC voltage

Cell voltage is a balance between energy and current efficiency. Roughly 1.2 V is the reversible decomposition potential and cannot be removed; the rest is ohmic drop in the bath, the anodes, the cathodes and the busbars. Reducing anode–cathode distance lowers the bath drop and saves energy, but it also raises the chance of anode effects and reduces current efficiency. In practice a smelter settles at 4.15–4.35 V, and each 100 mV is worth about 2.4% of the DC energy. For this potline, moving from 4.25 V to 4.15 V saves 4.5 MW — about 39 GWh a year, or USD 2.3 million at USD 0.06/kWh — without changing a single transformer. That is why the electrical design conversation with a smelter always starts with cell voltage and current efficiency, not with kVA.

Why does a 420 kA potline use six rectifier groups instead of one large rectifier?

Four reasons, and only the first is about manufacturing. One 514 MW rectifier does not exist; even if it did, its DC bus fault current and its single point of failure would be unacceptable.

Manufacturing and transport. A 90 MVA, 33 kV converter transformer weighs 110–130 t with oil. Six of them fit the same transport envelope as one 540 MVA unit that could not be built, shipped or replaced. Spare-unit strategy also becomes practical: a smelter keeps one 90 MVA spare rather than a 540 MVA one.

Harmonics. Each group is a 12-pulse rectifier (two secondaries, one delta and one star). Six groups staggered in phase by 5° give an effective 72-pulse characteristic, pushing the first characteristic harmonic from the 11th to the 71st — the mechanism behind why harmonic currents overheat transformers — and cutting the filter bank from 200 MVAr to 120 MVAr. The staggering is done in the transformer windings — extended-delta or zig-zag secondaries — so it is a specification item, not a rectifier setting.

Fault current on the DC bus. The potline busbar is a 420 kA aluminium conductor running the length of a 1 km potroom. A short circuit anywhere on it is fed by every rectifier group at once. The prospective current from one 90 MVA group at 8.5% impedance is about 500 kA; six groups give a bus fault in the region of 2–3 MA, limited by busbar impedance and cleared by tripping all six AC breakers. Six individual DC disconnects and six AC breakers is the only arrangement that lets you isolate one group for maintenance while the potline keeps running.

N-1 at reduced current. Lose one group and the potline drops from 420 kA to 350 kA rather than to zero. Aluminium cells tolerate reduced current for days; they do not tolerate an open circuit. That single property — 5/6 operation — is worth more than the extra switchgear costs, because a potline trip is measured in weeks.

How much current imbalance can parallel rectifier groups tolerate?

Current sharing between groups is set almost entirely by impedance mismatch, and it is the same physics that governs any parallel transformer operation. At 420 kA total, a group carrying 5% more than its share runs 5% hotter in winding and 10% hotter in eddy losses, and its insulation ages at roughly twice the rate. The relationship is close to linear and it is brutal:

This is the single most important sentence in a rectiformer purchase order: impedance tolerance ±2.5% of the bank mean, measured at the tap that will be used for base load. IEC 60076-1 allows ±7.5% on the declared value for a single unit, and a supplier who quotes to that tolerance is compliant with the standard and useless for a six-unit bank. The same logic applies to ratio tolerance (±0.25%) and to the secondaries within one unit, where a 1% mismatch between the delta and star windings turns into a 12-pulse bridge imbalance and a 6th-harmonic component that no filter is designed for.

What happens when one group trips — and why does it not stop the potline?

At 420 kA the potline voltage is 1,224 V. Lose one of six groups and the DC voltage collapses toward 1,020 V while the cells, still full of molten bath at 960 °C, keep conducting. The remaining five groups pick up the current within 20–50 ms; the current-regulating system then holds 350 kA and the potline continues, hotter at the cathode and less efficient, until the group is restored. Cathode lining life is consumed faster at lower current because the side-wall freeze profile moves, which is why a smelter will restore the group within hours and why a spare rectiformer on site matters more than a spare main transformer.

The scenario that must never happen is the loss of all groups, or a DC open circuit. Below about 200 kA the bath starts to freeze against the side walls; after 4–6 hours the cells are solid and the restart becomes a mechanical project: digging out frozen bath, re-bricking, re-starting pot by pot over 30–45 days. That asymmetry — 5/6 is survivable, 0/6 is not — is why the rectifier arrangement, the 220 kV connection and the protection scheme are designed together rather than separately.

How do you specify a rectifier transformer for a 420 kA potline?

A rectiformer is not a distribution transformer with a bigger tank. It is defined by IEC 61378-1, it carries a converter duty that IEC 60076-1 does not describe, and its load loss must be measured and guaranteed for a non-sinusoidal current. The specification sheet below is the one that produces comparable bids.

Why should the on-load tap changer control voltage, and the thyristor only trim it?

A line-commutated rectifier has two ways to reduce DC voltage: reduce the AC voltage with a tap changer, or delay the firing angle. The first is nearly lossless and keeps the current in phase with the voltage. The second is neither. Firing at angle α draws current with a displacement angle of roughly α/2 plus commutation overlap, and each 1% of phase-angle control below rated voltage costs about 0.6% power factor and injects additional low-order harmonics that the filter bank was not sized for.

So the specification should read: coarse control in ±1.25% tap steps, fine control in the thyristor for ±2% only, and the tap changer sized for 300,000 operations with a vacuum interrupter in oil. Smelters move taps constantly — anode changes shift cell voltage by 20–40 mV, pots are cut out and re-lined, line current is ramped at 5 kA/h during start-up — and an off-circuit tap changer simply cannot be used on this load — the general trade-off is set out in on-load versus off-load tap changers.

Why is 8.5% impedance the right number?

Impedance in a rectiformer does three jobs at once and they pull in opposite directions. It limits the DC bus fault contribution, it sets the commutation reactance that determines the overlap angle and reactive consumption, and it sets how much of the voltage regulation you can buy with taps. The table shows the trade at 90 MVA:

At 8.5%, one group contributes about 500 kA to a DC bus fault — a number that must be written into the rectifier fuse and breaker specification, the busbar bracing, and the arc-flash study; the method is the same one used in transformer short-circuit current calculation. It also means the bank consumes about 62 MVAr per group, 165 MVAr in total for 567 MW, which is what sizes the filter bank. Reduce impedance to buy efficiency and you spend the gain three times over on switchgear, fuses and filters.

What else differs from a distribution transformer? The converter-duty items

Three items are routinely missed by suppliers whose experience is in distribution units. First, loss measurement: IEC 61378-1 requires the load loss to be guaranteed and measured with a current waveform representative of the converter duty, including the harmonic loss factor — a transformer that meets 320 kW on sinusoidal test can dissipate 380 kW in service. Second, insulation under repetitive pulses: the LV winding sees commutation notches several hundred times per cycle, and inter-turn insulation that passes a 1.2/50 µs impulse can still fail in five years if the chopped-wave and repeated-pulse requirement is not stated. Third, oil flow electrification and static charging: forced-oil cooling at high flow velocity in large units generates static charge; the specification must require the tank and pipework to be bonded and the oil-flow velocity limited, which is the one requirement that separates a converter transformer from a power transformer of the same rating.

How much harmonic distortion does a 420 kA potline inject?

A 12-pulse group produces characteristic harmonics of order 12k ± 1: the 11th, 13th, 23rd, 25th and so on, at amplitudes of roughly 1/h of the fundamental. Six groups staggered by 5° cancel most of those, leaving a residual dominated by the 5th and 7th from transformer saturation and by the 11th/13th from imperfect staggering. At the 220 kV point of common coupling, IEEE 519 allows 1.5% total demand distortion for a plant of this size; a potline without filtering lands at 8–14%.

Note the shape of the problem: staggering does most of the work, and the filters clean up the residual plus supply reactive power. That is why the filter bank is specified in MVAr, not in amperes — its primary job at a smelter is power factor, and harmonic attenuation is the second benefit. Two 60 MVAr C-type filter banks on the 33 kV bus supply 120 MVAr of the 165 MVAr the rectifiers need; the 220 kV grid sees about 45 MVAr, well inside the 0.95 pf requirement most transmission operators impose.

Why are the filters on the 33 kV bus and not at 220 kV?

Cost and switching. A 60 MVAr filter bank at 33 kV is a conventional, repairable, site-assembled capacitor installation; the same MVAr at 220 kV needs a fully insulated platform, 220 kV switchgear and a much larger footprint. Putting the filters at 33 kV also places them electrically close to the rectifiers, so the harmonic currents circulate locally and never reach the main transformers, whose eddy losses would otherwise rise. The main transformers then see a load at 0.99 power factor and carry 5–8% less current for the same MW — which is worth about 4 MW of loss across the two mains, or USD 2 million a year at smelter energy prices.

How do you size the 220/33 kV mains for N-1 at a smelter?

Total demand is 590 MVA. Two 400 MVA 220/33 kV mains each carry 295 MVA, 74% loaded. Lose one and the survivor faces 148% — a transformer that will not survive 148% for more than a few minutes, and a potline that cannot simply be switched off. The industry answer is not a third 400 MVA unit; it is a contractual and operational current-reduction protocol backed by transformer overload capability.

The arithmetic: 400 MVA at 0.96 pf is 384 MW. Auxiliaries are 45 MW and cannot be shed. The rectifiers therefore get 339 MW, which at 1,224 V and 98.5% rectifier efficiency is 275 kA — 65% of line current. A potline at 275 kA is uncomfortable but stable for days: bath temperature falls a few degrees, current efficiency drops two points, and the cells stay molten. Reducing from 420 kA to 275 kA at the maximum ramp rate of 5 kA/h takes 29 hours, which is why the protocol is written as a staged plan (420 → 380 kA within 1 hour by anode-effect suppression, → 320 kA within 8 hours, → 275 kA within 30 hours) and why the mains are specified for 130% continuous overload with forced cooling and 150% for 30 minutes.

Why is oversizing the mains the wrong answer?

Three 400 MVA mains would give true N-1 at full current, and they would cost about USD 6 million more, occupy a third bay, and add roughly 180 kW of no-load loss each — 1.6 GWh a year for the third unit, USD 95,000 a year forever, for a scenario that occurs once every five to ten years. Worse, a third main does not fix the case that actually happens: loss of the 220 kV line that feeds two of the three. Smelters therefore invest in the things that reduce outage probability and duration instead: dual 220 kV lines on separate towers, gas-insulated 220 kV switchgear, online DGA on every main and rectiformer, and a spare 90 MVA rectiformer on site. That package costs less than the third main and addresses both the line outage and the transformer outage.

What does a potline outage cost, and how does it change the transformer specification?

A potline that loses current for more than 4–6 hours freezes. Bath solidifies against the side walls, the metal pad contracts, and restarting means digging out 288 cells, replacing cathode blocks and side lining, and re-starting at 50–100 kA with a slow ramp — 30–45 days and USD 20–40 million, plus the lost production of 336,000 t/yr divided by twelve. Against that number, condition monitoring on a USD 1.3 million transformer is free.

The specification consequences are concrete: online dissolved-gas analysis on every rectiformer (hydrogen and acetylene trends catch a hot spot weeks before failure), fibre-optic winding temperature rather than calculated hot spot, OLTC operation counting and motor-current monitoring, and a busbar differential protection scheme that trips the six AC breakers in under 80 ms; the protection philosophy is summarised in transformer protection devices. On the mains, the requirement is a 220 kV connection that can be transferred between busbars without dropping the potline, which means the 33 kV side is designed for a make-before-break transfer and the rectifier control can ride through a 300 ms voltage dip without losing current regulation.

How often does the tap changer really move?

More than most specifications assume. A 420 kA potline drifts 30–60 V a day as anodes are changed and pots are cut out; the current-regulation loop holds 420 kA by moving the tap one 1.25% step at a time, which at 1,224 V is 15 V per step. Add start-up ramps, anode-effect suppression, and nightly load-following where the smelter has an interruptible supply contract, and 30–60 operations per day is normal — 20,000 a year, 300,000 over a 15-year transformer life. A vacuum OLTC rated for 300,000 operations is therefore a life-of-asset component, not an accessory, and its failure mode (a stuck tap) is manageable only if the rectifier can still regulate ±2% on the thyristor. That is the reason the fine-control requirement exists.

How do you size the auxiliary substation?

Forty-five megawatts of auxiliaries sounds small against a 514 MW potline, and it is the part of the plant that stops production when it fails. Fume treatment losing a fan means the pots must be covered and current reduced; cast house losing a holding furnace means metal backs up in the pots.

The auxiliary package is two 25 MVA 33/11 kV transformers at 90% each, split-bus so that one can be maintained, plus 11/0.4 kV distribution transformers of 1,000–2,500 kVA throughout the potrooms and cast house. Because most of the auxiliary load is VFD-driven fans and compressors, the 0.4 kV units should be K-factor rated or fed through active front ends — the harmonic problem in a smelter is not only at the rectifiers, which is where K-factor transformer selection applies.

Where does cast resin belong at a smelter, and where does it not?

Cast resin belongs indoors in the potline switchrooms and the cast house, where a mineral-oil fire next to 960 °C metal is unacceptable and where the 11/0.4 kV units are small. It does not belong at 33 kV and 25 MVA: a 25 MVA cast-resin transformer is at the top of the dry-type range — the boundary is discussed in oil-immersed versus dry-type transformers — and it costs 40–60% more than an oil-immersed unit, and needs a Class-F insulation system with a 100 K rise to stay inside a reasonable frame. The standard arrangement is oil-immersed 33/11 kV outdoor units with firewall separation and a 6% bus-tie reactor, and cast resin from 11 kV down.

What does the DC bus itself demand from the transformer and rectifier package?

The 420 kA busbar is the part of the plant that most often surprises a transformer supplier. It is aluminium, water-cooled at the rectifier end, and carries current that produces a 2–3 MA prospective fault if it is ever shorted. Three requirements follow.

Fault clearance in tens of milliseconds. The rectifier's own protection must detect a DC overcurrent and trip the AC breaker in under 80 ms; the transformer must withstand the resulting electromagnetic forces, which is why the winding is specified with continuously transposed conductor and a short-circuit test at the actual bank impedance rather than a generic 25 kA test.

Ground fault detection on an unearthed DC system. The potline is not earthed — a single ground fault must not stop 514 MW. Insulation monitoring on the DC bus, plus a transformer LV winding that is insulated for full DC voltage to earth (1.2 kV DC plus transients), are both required. Suppliers who treat the LV as a 690 V distribution winding fail this test in the first year.

Common-mode noise. The rectifier's switching produces common-mode voltages that travel back through the transformer capacitance into the 33 kV system and the control electronics. The electrostatic shield specified above is what stops it; without the shield, the rectifier's own current-regulation electronics become the largest source of nuisance trips in the plant.

Which losses should you capitalise at 8,000 hours a year?

A smelter runs 8,500 hours a year at 96–98% availability, so loss capitalisation is not a tender formality. On this bank:

Over 20 years at 6% discount, USD 1.05 million a year is a present value of about USD 12 million — roughly the purchase price of the six transformers. That is why a bid evaluation at a smelter weights losses at 30–40% of the score, and why quoting 340 kW instead of 320 kW load loss can lose an order even at a lower price. It is also why the loss must be guaranteed under converter duty: a supplier who guarantees 320 kW on a sinusoidal test and delivers 380 kW on rectifier load has given the smelter a USD 320,000-a-year surprise that no warranty clause will cover.

Should you buy a more efficient main transformer instead of a third one?

Yes. The two 400 MVA mains at 180 kW no-load and 900 kW load loss each cost about USD 1.1 million a year in losses. Moving from a conventional grain-oriented core to a high-grade or amorphous core on units this size is not practical, but specifying the lowest-loss economic design and evaluating at USD 0.06/kWh typically buys 15–20% of the loss back for 3–4% of the price. On a 400 MVA unit, 150 kW of saved load loss is USD 72,000 a year, a payback of under two years.

What should the transformer specification sheet contain?

Fourteen items, in the order that decides the bid:

What are the five specification mistakes that cost the most?

1. Impedance tolerance quoted to IEC 60076-1 (±7.5%). Six units from three factories, or even one factory with normal manufacturing spread, will not share current. The symptom appears in year two as one unit running 12 K hotter than its neighbours; the fix is a re-wind or a permanent current limit on the bank.

2. Loss measured on sinusoidal current. Converter duty adds 15–20% to load loss in the winding and more in the structural parts. A guarantee that says 320 kW without naming the spectrum is not a guarantee.

3. No electrostatic shield. Saves about 1.5% of price, and produces years of unexplained rectifier control trips and 33 kV common-mode noise that nobody can trace.

4. Thyristor-only voltage control. Cheaper by the cost of an OLTC, and it gives away 1–2% of power factor and a filter bank sized for harmonics the tap changer would have avoided. At 567 MW, 1% of power factor is USD 300,000 a year in reactive charges at many tariffs.

5. Standard distribution-transformer short-circuit rating. The DC bus fault is 500 kA per group and 2–3 MA in total; a winding designed and tested at the generic IEC level will deform on the first real fault, and the deformation will be invisible until DGA shows acetylene two years later.

How do you commission a rectiformer without stressing the potline?

Commissioning happens before the potline is started, so the bank is tested into a water rheostat or a short-circuiting busbar with staged current: 10%, 25%, 50%, 75% and 100% of rated DC current, each held for the time needed to stabilise winding temperature. The tests that matter are the ones that reveal a bank problem rather than a unit problem.

The sharing test is the one to insist on at the factory as well as on site, because it is the only test that proves the bank rather than the unit. A supplier can meet every individual guarantee and still deliver six transformers that do not run in parallel, and the re-work is done in the potroom, at altitude, in a fluoride atmosphere, with the potline waiting.

Sources / 资料来源

Source: IEC 61378-1 — Converter transformers, Part 1: Transformers for industrial applications.
Source: IEC 60146-1-1 — Semiconductor converters: general requirements and line commutated converters.
Source: IEC 60076-1 — Power transformers: general.
Source: IEC 60076-2 — Power transformers: temperature rise for liquid-immersed transformers.
Source: IEC 60076-3 — Power transformers: insulation levels, dielectric tests and external clearances in air.
Source: IEC 60076-5 — Power transformers: ability to withstand short circuit.
Source: IEC 60076-7 — Power transformers: loading guide for mineral-oil-immersed power transformers.
Source: IEC 60076-11 — Power transformers: dry-type transformers.
Source: IEEE 519 — Recommended practice and requirements for harmonic control in electric power systems.
Source: IEC 61936-1 — Power installations exceeding 1 kV AC and 1.5 kV DC: common rules.
Source: GB/T 6451 — Technical parameters and requirements for three-phase oil-immersed power transformers.
Source: QDTB Engineering Team — oil-immersed and cast-resin transformer supply to metals, industrial and infrastructure projects, 2015–2025 (cell voltages, demand factors, impedance spreads, loss capitalisation rates, price bands and lead times quoted here are industry experience values, not published standards).