Glass Plant Transformer: How to Size the 33 kV Supply for a 600 t/d Float Glass Line, Why Electric Boosting Sets the Harmonics and the Furnace Sets the Reliability

The answer first: a 600 t/d float line is a 15,374 kW, 17,015 kVA load, and two 20 MVA 33/6.6 kV units carry it — but the furnace, not the transformer, decides what “carry” means

Start from the pull, not from a kVA rule: 600 tonnes a day is 25 t/h of glass, and a modern regenerative float furnace needs about 5.9 GJ per tonne of melt, so the gas input is 147.5 GJ/h — 40.97 MW thermal for a 4 mm ribbon running at 12.6 m/min on a 3.3 m width. The electrical side of that plant is 18,620 kW connected and 15,374 kW at the design peak, which is 17,015 kVA at 0.904 power factor once the 6 MW of electric boosting and the 9,374 kW of motor load are added with their real demand factors. Two 20 MVA 33/6.6 kV ONAN/ONAF units at 10.5% impedance serve it: 42.5% loaded each in parallel, 85.1% on one unit after a transformer outage with boosting capped at 4 MW. The interesting engineering is not in that arithmetic. It is that a glass plant buys reliability for the furnace campaign, not for the load — an 8 to 12 year campaign, a USD 15–25 million cold repair, and 219,000 tonnes a year of output that stops the day the melt freezes — while the transformer itself is a USD 232,000 item whose failure is inconvenient rather than existential. Three consequences follow. Electric boosting is 2.6 times the cost of gas per useful gigajoule, so it is bought for pull and quality and never for energy, and it is the only load on the site that needs its own transformer type. The boost control method, not the kVA, sets the harmonic distortion: thyristor phase-angle control puts 139 A of fifth harmonic into the 6.6 kV bus where tap-step control puts 8 A. And 45 °C ambient air next to a 41 MW furnace rewrites the derating, the paint system and the breather, long before it touches the rated power.

Key takeaways

• 600 t/d = 25 t/h of glass, 5.9 GJ/t of furnace input, 40.97 MW of gas, 12.6 m/min ribbon speed at 4 mm on 3.3 m. Electrical connected load is 18,620 kW, design peak 15,374 kW, 17,015 kVA at 0.904 PF.
• Two 20 MVA 33/6.6 kV YNd11 ONAN/ONAF units at 10.5% impedance: 42.5% each in parallel, 85.1% on one unit with boost capped at 4 MW. A single 40 MVA unit saves USD 84,000 of capital and converts a transformer failure into a 30-day furnace outage.
• Electric boosting costs 21.9 USD per useful GJ against 8.42 USD for gas — 2.6 times — because a regenerative furnace converts gas at about 45% while a boosting electrode puts 95% of its energy into the melt. 2.8 MW of average boost costs USD 1.68 million a year in power and displaces USD 645,000 of gas; the USD 1.03 million net buys 5% of pull, worth USD 2.2–3.3 million of contribution.
• Four 1,800 kVA 6.6/0.42 kV boost transformers with on-load tap-changer stepping hold fifth harmonic at 8 A; the same 6 MW delivered by phase-angle thyristors at 63% voltage injects 139 A, and total demand distortion at the 6.6 kV bus goes from 2.4% to 13.9% against the 5.0% that IEEE 519-2014 allows at Isc/IL = 9.5.
• At 45 °C ambient the transformers must be rated 20 MVA at 50 °C, not 40 °C, and the soda-ash dust from the batch house forces IP54, a sealed conservator with a diaphragm, 200 W anti-condensation heaters and a C4 paint system — specification items that cost USD 9,000 per unit and prevent the two failure modes that actually occur.

What does a 600 t/d float glass line actually draw? The load schedule behind 17,015 kVA

Float glass is a continuous process with a flat electrical profile, which is the opposite of the rolling mill and the arc furnace that dominate most industrial transformer advice. The furnace runs at constant pull for years; the tin bath, the lehr and the cold end run at line speed; the only large swing load is electric boosting, and even that is controlled to a setpoint rather than cycled. So the load schedule is built from installed drives and their true demand factors, not from a coincidence factor on a nameplate total, and the result is a plant that sits at 83% of its connected load for 8,000 hours a year.

The furnace side dominates the auxiliary load. Combustion air and exhaust fans on a 41 MW furnace move roughly 60,000 Nm³/h of combustion air and a similar volume of flue gas through the regenerators, and the reversing fans run at 90% demand factor because they are the load that cannot be shed. Tin bath atmosphere is smaller in kW but not in consequence: a 95/5 nitrogen-hydrogen mixture at 8–12 mbar over the bath, generated on site at about 300 Nm³/h, protects the ribbon from oxidation between the bath exit at 600 °C and the lehr; the nitrogen plant, the hydrogen supply and the bath heaters together are 680 kW. The cold end — cutting, edge grinding, stacking, packing — is 1,680 kW of drives and robots but runs at 0.65 demand factor because the lines are interlocked and never all at full torque.

The reactive side is what fixes the kVA. Motor load at 0.87 power factor contributes 5,312 kvar; the boost transformers, which see a resistive melt, contribute 1,975 kvar at 0.95; the total 7,287 kvar against 15,374 kW gives 17,015 kVA and a plant power factor of 0.904. That number matters twice. It sets the transformer rating, and it sets the size of the filter bank that the harmonic mitigation needs anyway, which is why a glass plant usually ends up at 0.98 power factor without buying a single capacitor for power factor alone. The total cost of ownership calculation for these units should start from this load duration curve, not from the nameplate total, because the plant operates at 66% average loading around the clock and the loss split between no-load and load loss is decided by that 66%.

Why does the furnace, not the transformer, set the reliability requirement?

Ask what happens when the 33 kV feeder opens. In an arc furnace shop, the breaker opens, the arc goes out, and production restarts in twenty minutes. In a float glass plant the melt stays hot from the gas burners, so a 30-minute outage is survivable; but the tin bath is the problem, because 1,100 °C glass floating on molten tin at 8–12 mbar of protective atmosphere cannot tolerate a loss of the bath atmosphere or the bath cooling, and the lehr cannot tolerate a stop with a 4 mm ribbon inside it. A ribbon that freezes in the lehr is cut out and scrapped, which costs a day. A bath that oxidises costs a week of contaminated glass and a tin change.

That asymmetry, not the transformer price, is what the electrical design has to answer. It produces three requirements that a generic industrial substation would not have.

• The critical bus is the tin bath and cooling water, not the furnace. Combustion air fans, cooling water pumps and the N₂/H₂ plant — 3,851 kW of demand between them — are split across the two 6.6 kV bus sections and backed by a 2,000 kW diesel generator with an AMF panel that closes in under 10 seconds. The cold end is deliberately not on that generator: 1,092 kW of cutting and stacking can wait.
• The transformer outage is a boost-shedding event, not a plant stop. One 20 MVA unit carries 85.1% of the 15,374 kW peak, which is inside the rating, but the 4 × 1,800 kVA boost group is interlocked to drop from 6,000 kW to 4,000 kW while the plant runs on one transformer. That interlock is worth more than a third transformer, and the arithmetic is in the next section.
• The furnace campaign sets the spare-parts list. HV bushings, an OLTC motor-drive unit, fan motors, gaskets and one set of bushings current transformers are held on site, because the alternative to a USD 6,000 spare is a 5-day air freight and 12,000 tonnes of lost output.

Put a number on the downside. A cold repair on a 600 t/d furnace is USD 15–25 million and 4–6 months; the lost contribution is 219,000 tonnes a year at USD 200–300 per tonne, so a 30-day unplanned outage costs USD 3.6–5.4 million. Against that, the difference between a 2 × 20 MVA scheme and a single 40 MVA scheme is USD 84,000. This is the same argument that makes a cement plant buy two transformers for a kiln it cannot restart, and it is the reason the redundancy decision in a glass plant is made by the furnace engineer and only costed by the electrical engineer.

How much does electric boosting cost, and why is it 2.6 times the price of gas per useful GJ?

Electric boosting is the single most misunderstood load in the plant, because it looks like an efficiency measure and is not one. A boosting system passes current between molybdenum electrodes immersed in the melt, usually at the bottom of the furnace, and delivers about 95% of the electrical energy directly into the glass. That is far more efficient than the furnace itself, which converts gas at roughly 45% — 5.9 GJ of input for 2.65 GJ of theoretical melting energy, the rest leaving in the flue gas at 450–500 °C after the regenerators. So one gigajoule of electricity displaces about 2.1 GJ of gas, and engineers who stop there conclude that boosting saves energy.

Price it and the conclusion reverses. At USD 0.075 per kWh, electricity costs 20.83 USD/GJ; at 95% efficiency that is 21.9 USD per useful GJ. Gas at USD 4 per MMBtu costs 3.79 USD/GJ, and at 45% efficiency that is 8.42 USD per useful GJ. Boosting is 2.6 times more expensive per unit of heat delivered to the glass, and it always will be wherever gas is USD 3–5 per MMBtu and power is USD 0.06–0.09 per kWh. The plant buys it anyway, and the reasons are metallurgical rather than electrical.

What 5% of pull is worth explaining: boosting raises the bottom glass temperature without raising the crown temperature, which improves fining, reduces seed and blister counts, and lets the same furnace deliver more tonnes before the refining limit is reached. It also controls the redox state for amber and green glasses, where the alternative is a different batch chemistry. In a plant making clear float, 4–6% of pull is the normal gain, and it is worth 2–3 times what the electricity costs. What boosting does not do is reduce the energy bill, and a specification that justifies the boost transformers on energy savings is wrong in a way that shows up in the first year of operation.

One design consequence follows directly. Because the load is a controlled process variable rather than a motor, the boosting transformers are sized for a continuous 1.1 pu during furnace heat-up, when the cold melt has low resistance and the electrodes run at full current at reduced voltage for two to three days. A transformer selected on the steady-state 6,000 kW alone will be 10% short exactly when the plant has no alternative, which is the first week of a campaign.

Which harmonics do glass plants inject, and why is phase-angle boost control the wrong answer?

Two loads generate harmonics in a float plant, and they behave differently. The variable-speed drives — two 800 kW compressors, four cooling water pumps, the cold end and the coating line exhaust, about 2,600 kW in total — are six-pulse front ends with DC link chokes, so they produce a stable spectrum of 30–38% current THD dominated by the fifth and seventh. The boost transformers are the difficult one, because how much harmonic current they inject depends entirely on how the power is controlled.

Phase-angle control with thyristors is the cheap way to vary boost power, and it is a harmonic generator by construction. At 40% power the firing angle sits near 76°, the voltage is 63% of rated, and a resistive load chopped at that angle produces a fifth harmonic around 35% of the fundamental. On this plant, 4 × 1,800 kVA at 63% voltage means a fundamental of 396 A at 6.6 kV and 139 A of fifth harmonic — 7.9% of the 1,749 A transformer base current, before the drives are counted. Add the drives and the 6.6 kV bus reaches 13.9% total demand distortion.

That number has to be judged against the right limit, and this is where most glass plant harmonic studies go wrong. IEEE 519-2014 sets limits by the ratio of short-circuit current to load current at the point of common coupling. At the 6.6 kV bus, 16,657 A of short-circuit current against 1,749 A of load current gives Isc/IL = 9.5, which falls in the most restrictive band: 4.0% for the fifth and seventh, 2.0% for the eleventh and thirteenth, and 5.0% total demand distortion. The 33 kV point of common coupling is far more forgiving at Isc/IL = 41, where 7.0% and 8.0% apply. A study that checks only the 33 kV bus will therefore pass a design that is 2.8 times over the limit at the bus where the drives actually are.

The specification answer is to remove the harmonic at its source instead of filtering it afterwards: boost transformers with an on-load tap changer stepping in 1.25% voltage steps, combined with group switching of the four electrode pairs. Voltage steps of 1.25% give power steps of 2.5%, and four groups give coarse 25% blocks, so the plant gets 6,000 kW in 36 usable increments with a sinusoidal current. Fifth harmonic drops from 139 A to 8 A at the transformer terminals, the 6.6 kV bus settles at 5.7% TDD, and a three-branch passive filter of 2.0 + 1.5 + 0.8 MVAr tuned to the 5th, 7th and 11th brings it to 2.4% — inside every individual limit with margin, and it corrects the power factor from 0.904 to 0.978 as a side effect.

The filter is not optional even with tap-step control, because the drives alone exceed the 4.0% fifth harmonic limit at this Isc/IL. But its size and cost change: 4.3 MVAr of tuned branches at USD 96,000, instead of a 6 MVAr active filter at USD 310,000 or a 12-pulse boost scheme that would need two extra transformer windings and a phase-shifting pair. The harmonics guide covers the spectrum arithmetic; the glass-specific point is that the control method is a specification decision worth USD 200,000, and it is usually made by the furnace supplier before the electrical engineer sees the single line.

How do you size the two 20 MVA transformers, and what happens with one out?

The kVA number is 17,015, and the temptation is to buy 2 × 20 MVA and be done. The sizing actually has four constraints, and only the first is kVA.

Normal loading. 17,015 kVA across two units is 8,508 kVA each, 42.5% of rating. That is lower than most plants, and deliberately so: a float line runs at 83% of connected load continuously, so 42.5% is a real continuous duty, not a peak that passes in ten minutes.

Single-unit loading. With one transformer out, the survivor carries 17,015 kVA, which is 85.1% of 20 MVA — acceptable for the days or weeks a replacement takes, but only if the ambient derating does not eat the margin and only if the boosting load is capped. Dropping boost from 6,000 kW to 4,000 kW brings the load to 14,927 kVA, 74.6%, which leaves room for the 45 °C ambient, the loss of one cooling fan and a 5% overload during a cold repair.

Ambient. A 20 MVA unit rated at 40 °C ambient carries 18.6 MVA at 45 °C, a factor of about 0.93 from the IEC 60076-7 loading tables. That turns the 85.1% single-unit figure into 91.5%, which is too thin for a plant that intends to run on one transformer for a month. The fix is not a bigger transformer but a different rating point: specify 20 MVA at 50 °C ambient, which adds roughly 6% to the iron and copper and costs about USD 14,000 per unit. This is the cheapest reliability decision in the project, and it is exactly the kind of derating question covered in the temperature derating guide.

Future pull. A 600 t/d furnace is usually designed to go to 650–700 t/d after its first campaign, which needs 8,000 kW of boosting rather than 6,000 kW and takes the load to about 19,500 kVA. At that point the 20 MVA ONAN rating is not enough, but the same transformer on ONAN/ONAF at 26 MVA is, and the cooling fans are already specified. Buying ONAF capability now costs USD 11,000 per unit and defers a USD 464,000 replacement.

The 3 × 12.5 MVA option deserves a sentence because it is the one a reliability engineer will propose. It does keep full boosting with one unit out, which is a real operational benefit during a cold repair when the plant wants every tonne it can make. But the probability of a second transformer failing during the same outage is around 1 in 400 per event at a 0.5% annual failure rate, and the 20 MVA scheme already holds the critical tin bath and cooling water loads on a generator. The USD 76,000 buys insurance against a scenario that costs less than the insurance.

Should both transformers run in parallel at 42% load? The 11.4 MVA crossover

Two transformers at 42.5% each look like a candidate for “run one, keep the other as standby”. The loss arithmetic says no, and the crossover point is easy to calculate. For a transformer with no-load loss P₀ and load loss P_k, one unit at load S gives P₀ + P_k(S/S_r)², while two identical units share the load and give 2P₀ + 0.5P_k(S/S_r)². Parallel operation wins when 2P₀ < 0.5P_k(S/S_r)², which for a 20 MVA S11 unit with P₀ = 15.5 kW and P_k = 96 kW means (S/S_r)² > 0.323, or S > 11.4 MVA.

This plant peaks at 17.0 MVA, so parallel operation is the lower-loss mode by a wide margin: 65.7 kW of total loss instead of 85.0 kW at the peak, a saving of 19.3 kW that is worth 154,400 kWh and USD 11,580 a year at 8,000 hours. The standby philosophy would save 15.5 kW of no-load loss and spend 34.7 kW of extra load loss to do it. The parallel operation guide covers the conditions — same vector group, ratio within 0.5%, impedance within 7.5% — and the practical point here is that the two units should be ordered as a matched pair with measured impedances within 2.5% of each other, because a 0.5% ratio mismatch with 10.5% impedance circulates 2.4% of rated current between them continuously, which is 42 A at 6.6 kV and a permanent 3 kW of loss.

One more number decides the operating procedure. At an average load of 66.4% of one unit, the annual transformer loss for the pair is 115.6 kW, or 924,800 kWh and USD 69,360 a year. That is 0.9% of the plant’s electricity bill, which is small enough that transformer efficiency is a secondary decision at this site — but large enough that the grade should still be evaluated honestly. Upgrading both units from S11 to S13 saves 7,200 USD a year against USD 32,000 of extra capital, a 4.4 year payback; amorphous cores save 9,840 USD a year against USD 64,000, a 6.5 year payback. The loss grade comparison explains why the ranking flips at low load; here the plant runs at high load, so load loss dominates and the premium for amorphous core does not pay.

What does 45 °C ambient and soda-ash dust do to the specification?

The transformer room on a glass plant is not a normal electrical room. Radiant heat from the furnace, the lehr and the 41 MW of flue ductwork raises the ambient at the outdoor transformer bay to 45 °C on a 35 °C day, and the batch house 60 m away produces a fine, alkaline, hygroscopic dust from soda ash, dolomite and limestone that finds every gasketed joint that is not actually sealed. Both conditions are routine, and both are specified away for about USD 9,000 per transformer.

• Rating point. 20 MVA at 50 °C ambient, with a 60 K top-oil rise and 65 K winding rise, verified by a temperature-rise type test at the specified ambient rather than by calculation.
• Enclosure and paint. IP54 with gasketed covers on the 6.6 kV cable box and the OLTC compartment, C4-M paint at 240 µm for the outdoor units, hot-dip galvanised hardware, and stainless steel rating plates because painted plates disappear in two campaigns.
• Oil preservation. A sealed conservator with a diaphragm or a membrane bag, not a silica-gel breather. A breather on a glass plant pulls alkaline dust into the oil, the desiccant turns into a paste, and the oil absorbs moisture at exactly the rate the plant cannot detect between annual samples.
• Anti-condensation. 200 W heaters on both the LV and HV cable boxes, thermostatically controlled, plus a terminal-box gasket that is actually compressed — condensation in a 6.6 kV box after a cool night is the most common cause of a first-energisation flashover in humid climates.
• Cooling. ONAN/ONAF with 2 × 50% fans and a fan-supply changeover, so a single fan failure costs capacity rather than a trip. OFAF is not needed at 20 MVA and adds a pump that will fail before the fans do.
• Noise. 72 dB(A) at 1 m, measured, because the units sit 30 m from the cold end control room and the plant has a boundary limit.

Two of these are worth arguing about in the tender. The 50 °C rating costs 6% of the transformer price and removes a 8.5% shortfall in single-unit operation, so it is not a specification nicety. The membrane bag costs USD 3,500 per unit and is the difference between a 30-year oil life and a 12-year one in a dusty plant, which is why the same item appears on desalination and cement specifications and almost never on a general-purpose datasheet.

What is inside a 1,800 kVA boost transformer, and why is it not a distribution transformer?

Four 1,800 kVA units at 6.6/0.42 kV look like small distribution transformers on the drawing and are nothing of the kind. Each one feeds a pair of molybdenum electrodes through a busbar that carries 2,470 A at 420 V, and each one is controlled by an on-load tap changer that operates several times an hour for years. The differences that matter are electrical, thermal and mechanical.

The impedance choice is the one that gets questioned. A distribution transformer at 6% would give 41,200 A at the electrode, and the busbar, the electrode holders and the furnace shell insulation would all have to be rated for it. At 8% the current is 30,900 A, which is still a large fault but is inside the 2 s thermal capability and inside the 50 kA short-circuit rating of the low-voltage switchgear that isolates a failed group. Going higher, to 10%, would cost more load loss than it saves in switchgear, because the transformer runs at 40–60% load for 8,000 hours.

The DC component is the failure mode that surprises people. Electrodes wear at different rates, the melt resistance becomes asymmetric, and a DC current of 2–5% of rated appears in the LV winding. That is only 50–125 A, but it saturates the core in one half-cycle, adds a DC flux that the design did not budget, and heats the tank and the structural steel. The design answer is a magnetic shunt between the windings and a DCCT on the LV busbar with an alarm at 2% and a trip at 4%; the operational answer is electrode position trimming every shift. A boost transformer specified without either will fail in the third or fourth campaign, and it will look like an insulation failure rather than a DC bias problem.

The arc furnace transformer guide describes the other extreme of the same family: 20–30% impedance, 20% cyclic overload, and a vacuum tap changer because the duty is violent. A boost transformer is the opposite — steady load, low impedance, gentle duty — and buying an arc furnace design for a glass furnace is a common and expensive mistake in the other direction.

What does the 33 kV switchgear, earthing and protection scheme need?

The supply is a single 33 kV feeder from a 750 MVA source with an X/R of 12, giving 13,120 A of short-circuit current at the incoming terminals. On the 6.6 kV side, each 20 MVA transformer delivers 16,657 A and the pair delivers 33,300 A, which with motor contribution reaches about 36,000 A — so the 6.6 kV switchgear is specified at 40 kA for 1 s, not the 31.5 kA that a smaller plant would use.

The earthing arrangement is decided by the transformer vector group. YNd11 puts a delta on the 6.6 kV side, so the 6.6 kV system is unearthed and a single phase-to-earth fault does not produce fault current — which sounds attractive until a second fault appears on another phase and the resulting phase-to-phase fault trips the plant. The standard answer is a zig-zag earthing transformer with a 400 A, 10 s resistor on each 6.6 kV bus section, giving limited earth-fault current that the 51N relay can detect and coordinate. The grounding system guide covers the LV side; the MV side rule is that an unearthed 6.6 kV system on a continuous process is not a design choice, it is a latent outage.

• 33 kV: 36 kV rated, 31.5 kA/3 s, vacuum or SF₆, 170 kV lightning impulse, two transformer bays plus one incoming and one metering bay.
• Transformer protection: 87T biased differential, 87N restricted earth fault, 51/51N backup, 49 thermal image, 63 Buchholz and sudden-pressure, 26/49 oil and winding temperature with trip at 105 °C, OLTC out-of-step and tap-position monitoring.
• 6.6 kV: 7.2 kV, 40 kA/1 s, 60 kV impulse; 50/51 and 51N on each feeder, 87M on motors above 1 MW, arc-flash relay with a maintenance switch on the two bus sections.
• Power quality: a class A analyser on each 6.6 kV section, logging harmonics, unbalance and sags, because the filter performance guarantee is measured against it and because the boost transformers are the first place a developing electrode problem shows up electrically.

One transient requirement deserves its own paragraph: energising a 20 MVA transformer from a 750 MVA source produces an inrush of 8 times rated, 2,800 A at 33 kV, and a voltage dip on the 33 kV bus of 15–21% for three to six cycles. That is enough to trip every VFD on the plant and to commutate-fail the boost thyristors, which means a transformer changeover during a campaign can stop the cold end even though the transformer itself is healthy. The inrush current guide gives the physics; the operating answer here is either a controlled switching relay with point-on-wave closing, at USD 18,000 for both bays, or an interlock that blocks energisation while the cold end drives are running. The relay is cheaper than one false trip.

What does the electrical package cost, and is the loss evaluation worth doing?

The package for this plant comes to USD 1.54 million at 2025–2026 Chinese FOB prices, of which the transformers are 40%. That ratio is the reason the loss evaluation matters less here than in a data centre: the transformers are not the dominant cost, and the plant runs at high load, so load loss dominates and the difference between grades is 4–7 years of payback rather than 18 months.

For the tender evaluation, the loss capitalisation values follow from the operating pattern: 8,000 hours a year, an average load of 66.4% of one unit, and USD 0.075 per kWh. No-load loss is worth 600 USD/kW and load loss at that duty is worth 265 USD/kW. On those values the two S11 units evaluate at USD 69,480 of capitalised loss, the S13 pair at USD 62,200, and an amorphous pair at USD 59,620 — a spread of USD 9,860 against a capital spread of USD 64,000, so the S11 or S13 wins on total cost and the amorphous core is the wrong purchase at this load factor. Getting this wrong in the other direction is common: a specification copied from a data centre, where the plant runs at 30–40% load and no-load loss dominates, will specify amorphous cores for a glass plant and add USD 64,000 for a 6.5 year payback.

Two commercial terms are worth more than the loss evaluation. A loss guarantee with a penalty of 600 USD/kW for no-load and 265 USD/kW for load loss, measured at the factory acceptance test, converts a paper promise into a number. And a matched-pair clause — measured impedances within 2.5%, ratios within 0.25% — costs nothing at the order stage and prevents the circulating current that would otherwise run for the life of the plant.

What should the datasheet and the factory tests contain?

A glass plant datasheet is a standard IEC 60076 document with six additions, and each addition maps to a failure that has happened on a float line. Write them as requirements with numbers, not as notes.

• Rating at 50 °C ambient, with the temperature-rise type test performed at that ambient or corrected to it per IEC 60076-2. A 40 °C rating is the default and will be offered if the datasheet does not say otherwise.
• Impedance 10.5% ±5% and a measured value reported for each unit, with the pair matched to 2.5%. IEC 60076-1 allows ±7.5% on impedance, which is enough mismatch to matter in parallel operation.
• Harmonic loss factor and K-factor 4 for the boost transformers, plus a stray-loss measurement at 1.1 pu current, because these units run at 47% load with a distorted current for years.
• DC bias capability: magnetic shunt, DCCT provision, and a design that tolerates 5% DC component without saturation of the tank or structural heating.
• Sealed conservator with diaphragm, IP54 gasketed cable boxes, 200 W anti-condensation heaters, C4-M paint at 240 µm, stainless rating plate.
• Routine tests per IEC 60076-1: ratio, winding resistance, impedance and load loss, no-load loss and current, applied and induced voltage, insulation resistance and tan delta. Type tests: temperature rise, lightning impulse at 170 kV, sound level. Special tests: zero-sequence impedance, harmonic loss measurement, and a frequency response analysis fingerprint for both main and boost units, because the FRA baseline taken at the factory is the only reliable way to detect a winding movement after a through-fault in year six.

The pre-shipment inspection checklist and the commissioning checklist cover the process; the glass-specific items are the 50 °C rating evidence, the matched-pair impedance report, the DC bias design review and the FRA fingerprints. The vector group guide explains why the main units are YNd11 and the boost units Dyn11, and why mixing them in one plant is normal but requires the earthing transformer on the 6.6 kV side.

None of this is exotic. A 33 kV, 20 MVA ONAN/ONAF unit with OLTC, a 50 °C rating and a sealed conservator is a standard build for a Chinese factory that supplies cement, steel and glass plants in Southeast Asia, the Middle East and Africa; the only item that needs a real engineering conversation is the boost transformer with its OLTC duty, DC bias allowance and 1.1 pu heat-up rating. QDTB supplies both types on the same order with matched impedance reports, which is the practical reason to buy the main and boost transformers from one factory rather than two: the pair matching and the FRA baselines are only meaningful if the same test bench produced them.

Sources / 资料来源

Source: IEC 60076-1:2011 — Power transformers: General.
Source: IEC 60076-2:2011 — Temperature rise for liquid-immersed transformers.
Source: IEC 60076-3:2013 — Insulation levels, dielectric tests and external clearances in air.
Source: IEC 60076-5:2006 — Ability to withstand short circuit.
Source: IEC 60076-7:2018 — Loading guide for mineral-oil-immersed power transformers.
Source: IEC 60076-11:2018 — Dry-type transformers.
Source: IEEE 519-2014 — Recommended practice and requirements for harmonic control in electric power systems, Table 2 current distortion limits for 120 V–69 kV.
Source: IEEE C57.12.00-2021 and C57.12.90-2021 — General requirements and test code for liquid-immersed distribution, power and regulating transformers.
Source: GB/T 6451-2015 — Technical parameters and requirements for three-phase oil-immersed power transformers.
Source: ISO 16293-1:2008 — Glass in building: basic soda lime silicate glass products.
Source: ISO 12944-2:2017 and ISO 12944-5:2019 — Corrosion protection of steel structures, corrosivity categories and paint systems.
Source: IEC 60071-1:2019 — Insulation co-ordination, definitions, principles and rules.
Source: IEC 62271-200:2021 — AC metal-enclosed switchgear and controlgear for rated voltages above 1 kV and up to and including 52 kV.
Source: Industry experience — float glass furnace specific energy of 5.5–6.5 GJ per tonne of melt for regenerative furnaces, 45–48% thermal efficiency against 2.6–2.7 GJ/t theoretical melting energy, 2015–2025.
Source: Industry experience — electric boosting at 10–20% of furnace energy input, 95% energy transfer to the melt, and 4–6% pull gain at constant crown temperature on 400–900 t/d float lines.
Source: Industry experience — measured fifth-harmonic content of 30–40% of fundamental at 60–65% output voltage on thyristor phase-angle controlled boosting, and below 3% with tap-step group control, from harmonic surveys at commissioned float plants, 2018–2025.
Source: Industry experience — transformer, switchgear, filter and generator price ranges for glass plant electrical packages, FOB China, 2025–2026.
Source: Industry experience — industrial electricity tariffs of USD 0.06–0.09 per kWh and natural gas at USD 3–5 per MMBtu used in the boosting and loss calculations.
Source: QDTB Engineering Team — load schedules, harmonic studies and transformer specifications for float glass, container glass and fibreglass projects in Southeast Asia, the Middle East and Africa, 2019–2026.