Hospital Transformer: How to Size a 300-Bed Hospital's 11 kV Supply, Why Medical IT Systems and N+1 Redundancy Change the Spec

The answer first: a 300-bed hospital is a 4,249 kW, 4,828 kVA load, and four 2,000 kVA 11/0.415 kV units carry it — N+1 at transformer level, 2N only downstream

Build the number from a load schedule rather than from watts per square metre: 6,380 kW connected becomes 4,249 kW at the 15-minute peak after diversity, which at 0.88 power factor is 4,828 kVA, and the bank that serves it honestly is four 2,000 kVA 11/0.415 kV Dyn11 units on two independent feeders with a normally-open bus tie — 60% loaded in normal operation, 87% of their derated rating with any one unit out. The redundancy that hospitals actually argue about is not at the transformer: it is 2N on the critical branch, which means two 600 kVA UPS, two four-pole ATS and two LV boards each sized for the full theatre and ICU load, and that decision — not the kVA number — is what doubles the LV switchgear and rewrites the earthing. Three things separate a hospital specification from a factory specification of the same size: group 2 medical locations need 18 single-phase medical IT systems with insulation monitoring, MRI and CT pulsed loads decide the LV design more than the bed count does, and a building that never closes evaluates transformer loss at 4,050 equivalent full-load hours a year, which makes an S13 core worth USD 2,600 more than an S11 and makes an amorphous core worth nothing at all.

Key takeaways

• 6,380 kW connected × 0.67 diversity = 4,249 kW demand = 4,828 kVA at 0.88 PF; four 2,000 kVA units sit at 60% normally and 87% of their 45 °C rating with one unit out, so the spare is real rather than nominal.
• Hospital load factor is 0.68, not the 0.35–0.45 of a factory: no-load loss is worth USD 1,051 per kW per year and load loss USD 444 per kW per year, so the four transformers burn USD 44,800 of electricity annually — 20% of their purchase price, every year.
• 2N applies to the critical branch, not to the transformers: two 600 kVA UPS, two four-pole ATS with switched neutral and two LV boards each rated for the whole theatre/ICU load. Transformer-level redundancy is N+1, and pretending otherwise adds USD 700,000 of switchgear for no measurable availability.
• IEC 60364-7-710 drives 18 medical IT systems of 8 kVA for 12 theatres, 2 cath labs and 4 ICU clusters, each with an insulation monitoring device that must alarm at 50 kΩ, an equipotential bonding system at or below 0.2 Ω, and a 0.5 s supply restoration that only a UPS can meet because the generators need 10 s.
• Harmonic current from UPS rectifiers, imaging and chiller VFDs is 9.1% TDD at the 11 kV point of common coupling against an IEEE 519 ceiling of 8.0%; 4% line reactors plus a 300 A active filter bring it to 5.6% for about USD 92,000, and leaving it alone costs more in capacitor failures and neutral overheating than the filter costs.

How much power does a 300-bed hospital actually draw, and why is watts per square metre the wrong starting point?

Design offices in different countries quote hospital demand anywhere from 40 to 120 W/m², and the spread is not a difference of climate — it is a difference of scope. A 45,000 m² building at 90 W/m² is 4,050 kW, which happens to be close to the right answer for a full-service hospital with heavy imaging, and completely wrong for the same building with no MRI, no cardiac cath labs and a rented chiller plant. The load schedule below is the version that survives a change of scope, because every line can be traced to a room, a machine or a drawing, and because a manufacturer quoting against it can see where the diversity came from.

The second thing to understand before the table is that a hospital's peak is not a shift peak. Factories peak at 10:00 and drop 40% at night; a hospital peaks in the late afternoon when the chillers, the theatres and the evening visiting load overlap, and falls only to about 55% of peak at 03:00 because wards, ICU, imaging on call, IT and the sterilisation plant keep running. That flat profile is the reason the transformers in a hospital are evaluated differently from the transformers in a plant of the same kVA.

Two observations follow from this table. First, the central plant is the largest single block at 944 kW, and it is the block most often left out of a client's own estimate because it sits in a separate contract — which is how hospitals end up with a transformer bank sized for the building rather than for the building plus its chillers. Second, the imaging group is only 333 kW of demand, 8% of the total, yet it dictates more of the LV specification than any other 8% in the building, for reasons set out below.

Annual consumption follows from the profile: 4,249 kW at a load factor of 0.68 is an average of 2,889 kW, or 25.3 GWh a year, and at USD 0.12/kWh that is USD 3.0 million of electricity — against which the transformers' own losses of 373,000 kWh, about USD 44,800, are 1.5%. Small in percentage, large in absolute terms, and fixed for twenty years. The sizing arithmetic itself is the same as any other building; if you want the one-line formula that turns kVA and load factor into a budget figure, it is in the sizing and budget formula guide.

Why is the transformer bank N+1 when the critical branch is 2N?

Because the two redundancies buy completely different things. A transformer that fails takes eight hours to replace and a mobile unit takes two days to arrive, so what you need at transformer level is a spare capacity path, not a duplicate path: four units where three can carry the load. A UPS that fails, or an ATS that fails to transfer, takes the operating theatre dark in milliseconds, and there is no mobile substitute for an operating table in use — so the critical branch needs a second, independent, fully rated path that is energised and carrying current at all times. That is 2N, and it is a different animal.

The practical consequence is that the money goes into LV switchgear, not into a fifth transformer. The design used here has two 11 kV feeders from two different 110/11 kV substations, four 2,000 kVA transformers, two LV main boards of 3,200 A each, and a 3,200 A four-pole bus tie that is normally open and closes automatically on loss of either feeder. On top of that sit the 2N elements: two 600 kVA static UPS each capable of carrying the whole 467 kVA critical load, two four-pole ATS on the generator side, and dual feeders from the UPS output to every theatre's medical IT panel, so that either UPS can be taken out for battery replacement without a single Theatre going dark.

What 2N actually changes in the specification is worth stating plainly, because tender documents often say "2N" and then buy 1N hardware:

The last row is the point. A 2N transformer bank doubles the transformer capital, doubles the no-load loss and doubles the floor space, and buys availability that the N+1 bank already delivers, because the failure mode it protects against — simultaneous loss of two transformers — is not a credible scenario in a building with two independent feeders. Spend the 2N money on the branch where the failure is instantaneous. This is also why a hospital with a beautifully redundant transformer bank and a single ATS is less reliable than the reverse, and why parallel operation rules matter here only for the brief moment the bus tie closes: the two transformers that end up in parallel must have matched impedance and the same tap position, or the tie will carry a circulating current before it carries any load.

Why must the units be rated at 45 °C when IEC 60076-2 assumes 40 °C?

Because IEC 60076-2 defines the rated power at a maximum ambient of 40 °C, a monthly average of 30 °C and an annual average of 20 °C, and a hospital plant room in Lagos, Jakarta, Chennai or Jeddah spends four months a year above 40 °C. A transformer ordered at 2,000 kVA and installed in a 45 °C room is a 1,860 kVA transformer with a nameplate that says otherwise, and the gap only appears on the day the second unit is out for maintenance.

The derating is not linear and it is not the same for every cooling class, which is why the specification should ask for the rating at the site ambient rather than apply a rule of thumb afterwards. For an ONAN unit with a 105 K average winding rise and a 78 K top-oil rise, each degree of ambient above 40 °C costs roughly 1.2% of rated power at constant hot-spot; the numbers below are the ones to put in the tender, and the same arithmetic is worked through for altitude in the altitude and temperature derating guide.

At 50 °C the design fails: 94% of derated rating after a single failure leaves nothing for load growth, nothing for the extra chiller that gets added in year four, and nothing for the 3–5% annual growth that imaging refresh brings. That is the argument for stating 45 °C and for specifying the cooling stage that goes with it — ONAN at 1,860 kVA with ONAF available at 2,325 kVA, fans interlocked to winding temperature rather than to a manual switch. The difference between ONAN and ONAF ratings, and what each does to noise and to fan failure risk, is set out in the cooling methods and temperature rise guide; here it is enough that the fans are the cheapest 25% of capacity you can buy, and they only help if the control scheme actually starts them.

The second thermal question is the one nobody asks in the tender: how hot does the plant room get when the ventilation fails? 35 kW of transformer loss in a room with no airflow raises the air temperature without limit until the units trip on winding temperature, and the hospital then loses half its supply in the middle of an operation. The fix is mechanical rather than electrical — a room with 10,500 m³/h of forced ventilation, an intake at low level, an exhaust at high level above the radiators, and a high-temperature alarm that annunciates in the engineering office rather than on the transformer itself.

What does a medical IT system ask of the transformer, and why can it not simply be an RCD?

IEC 60364-7-710 classifies medical locations into groups 0, 1 and 2, and group 2 — operating theatres, cath labs, intensive care, and any room where a mains failure during an invasive procedure is life-threatening — must be supplied through a medical IT system. An IT system has no direct connection between the live conductors and earth, so a first fault to earth does not flow current, does not trip anything, and does not plunge the theatre into darkness. Instead, an insulation monitoring device measures the resistance to earth and alarms at 50 kΩ, giving the staff time to finish the procedure and move the patient before a second fault turns the first one into an outage.

That architecture changes the transformer from a bulk device into a small, single-phase, deliberately isolated one. Each group 2 room or cluster gets its own 8 kVA single-phase transformer built to IEC 61558-2-15, with a shielded winding, a leakage current low enough not to saturate the insulation monitoring, and an inrush characteristic that does not trip the 16 A MCB every time the theatre is energised — inrush is the usual reason a medical IT system nuisance-trips, and it is a transformer selection problem, not a breaker problem, as the inrush current guide explains.

This design uses 18 systems: one per theatre (12), one per cath lab (2), and four serving 12-bed ICU clusters. The other requirements follow from the same clause and are what a hospital's electrical inspector will actually test:

• Equipotential bonding at 0.2 Ω or less, measured between any two exposed conductive parts in the room, including the operating table, the anaesthetic machine and the metalwork — a value that is easy to design and hard to achieve if the bonding is daisy-chained rather than star-connected to a single reference bar.
• Alarm annunciation in a staffed location, not only at the transformer: the IMD must repeat its alarm and its test button at the theatre control desk and at the engineering office, because the whole point is a warning that someone acts on.
• 0.5 s supply restoration for safety services in group 2 locations. The generators take 10 s to accept load, so the 0.5 s is met by the UPS, and the medical IT transformer sits downstream of the UPS — which means the UPS is not optional redundancy but a code-compliance device, and its 2N arrangement is what allows battery maintenance without closing a theatre.
• No RCD on the IT side. A 30 mA RCD on a theatre circuit is a nuisance-trip generator and defeats the purpose; RCDs belong on the group 1 circuits, on socket outlets outside the IT zone, and on the imaging feeders where the manufacturer requires one.

One more transformer-level consequence: the 18 medical IT transformers are single-phase and they are scattered across the building, so their load is not naturally balanced. The LV board design has to assign them to phases deliberately — each theatre's load is roughly 8 kW, so six theatres per phase is balanced, five is a 1.6 kW imbalance that becomes a neutral current and a small negative-sequence component at the transformer. It is not dangerous, but it is measurable, and it is the kind of thing that makes a commissioning report look careless.

Why do MRI and CT decide the LV design more than the bed count does?

A 3T MRI draws about 45 kW continuously for its compressors, chillers and electronics, and then draws 90–120 kVA for two to five seconds during a scan sequence. A CT draws 50–80 kVA in half-second pulses, sixty times an hour. A DR room draws 30–50 kVA for 100 ms. None of these numbers is large against a 4,249 kW hospital, and all of them are large against the impedance of the path that feeds them, which is why imaging rooms are specified by voltage dip and not by kVA.

The arithmetic is worth doing once. A 2,000 kVA transformer with 6% impedance has a source impedance of 0.0052 Ω at 415 V, and its rated current is 2,782 A. A CT pulse of 80 kVA at 0.9 power factor is 124 A, so the dip attributable to the transformer alone is (124/2,782) × 6% = 0.27%. Add 60 m of 4 × 185 mm² cable at 0.08 Ω/km and the dip becomes 0.6%. Feed the same CT from a 400 A board that also carries two lifts, a kitchen and a laundry, and the measured dip during a scan reaches 2.4% because the other loads pulse too — and a imaging vendor who specified "less than 3% transient dip" is now one lift start away from a service call.

The specification that follows is short and non-negotiable: every imaging room gets a dedicated feeder from the LV main board, not from a floor distribution board; the MRI gets its own 250 kVA cast-resin transformer or a dedicated 400 A feeder with a K-13 rating; the K-factor matters because the imaging rectifier's current is rich in 5th and 7th harmonic and a standard transformer's eddy-current loss rises faster than its rated capacity, which is exactly what the K-factor selection guide quantifies; the neutral is sized at 200% of phase because the same rectifiers produce triplen harmonic that adds in the neutral instead of cancelling.

Two physical requirements are routinely forgotten until the equipment arrives. The MRI needs a quench pipe to outside air, sized by the vendor and routed before the structural slab is poured, and it needs a 3 m exclusion zone for the magnetic field that no transformer, UPS or generator may occupy. The CT and DR rooms need the transformer and the feeder to be outside the shielded enclosure, because the radio-frequency shield is a room-sized Faraday cage and a cable entering it needs a filtered penetration. These are drawing issues, but they decide where the substation goes, and a substation that ends up 40 m further away than planned costs more in cable than the transformer costs in total.

How much harmonic distortion does a hospital actually produce, and does it really exceed IEEE 519?

Yes, and the surprise is where it comes from. Hospitals are not arc furnaces; the distortion is the aggregate of three ordinary things — UPS rectifiers, imaging rectifiers and chiller VFDs — each of which is individually acceptable and collectively over the limit. At 415 V the total fundamental current at peak demand is 6,717 A. Of that, about 2,578 A (38%) flows through non-linear front ends, and each type has its own harmonic signature.

At the 11 kV point of common coupling the fundamental current is 253 A and the short-circuit current is 13.1 kA, an I_sc/I_L ratio of 52. IEEE 519-2014 Table 2 allows 8.0% total demand distortion for a 1–69 kV system in that ratio band, so 9.1% is a genuine exceedance — not a rounding error, and not something a utility will ignore if it starts getting capacitor bank failures on the same feeder.

The remedy is deliberately cheap compared with the alternative. Four-percent line reactors on every VFD take the drive contribution from 35% to 22% of drive current, and a 300 A active filter on each LV main board cancels the residual 5th and 7th at source; the measured result is 5.6% TDD, comfortably inside the 8.0% limit and, more usefully, inside the 5% that most imaging vendors quietly require. Capacitor banks, if the hospital wants power-factor correction (and at 0.88 it should), must be detuned at 7% — a plain bank on a 38% non-linear load resonates with the 5th harmonic and fails within two years, which is one of the most expensive lessons in the whole electrical package and is covered from the transformer side in the harmonics guide.

One number to add to the tender that is almost never there: the transformer's own derating under non-sinusoidal load, per IEEE C57.110. A 2,000 kVA unit carrying 9.1% TDD does not need derating at that level, but a 2,000 kVA unit carrying the imaging load alone — K-13 duty, 40% harmonic content — does, and the manufacturer should state the K-factor rather than leave the hospital to assume 1.0.

What happens to the transformers when the generators take over?

Nothing violent, provided the transfer is designed before the switchgear is bought. The essential load here is 1,560 kW — 380 kW of life safety (fire pumps, smoke extraction, egress lighting, alarms) and 1,180 kW of critical (theatres, ICU, imaging on essential, IT, blood bank, NICU) — and it is served by three 1,000 kW diesel generators in N+1, each loaded to 52% in normal standby and to 78% when one is out for service. The generators do not carry the chillers: a 700 kW chiller is a 630 kW motor load that would double the generator size for a load the hospital can live without for two hours, so only one 250 kW scroll chiller is on the essential bus, enough for theatres and ICU.

The transfer sequence is what determines whether the operating theatre stays lit. On loss of both feeders the ATS at the LV main board transfers in 10 s, which is what NFPA 110 Type 10 requires and what a diesel needs to start, reach rated speed and accept load. Ten seconds is an eternity for a ventilator, so the UPS carries the critical branch for those ten seconds and for the 10 minutes it takes to confirm the generators are stable — 2 × 600 kVA of battery, sized for 10 minutes at 78% load because that is long enough to start the generators and short enough to keep the battery room small. Then the generators accept a 1,560 kW block load: at 52% of installed capacity this is a single-step load acceptance for a set with a permanent-magnet exciter, and a two-step acceptance if the sets are older or the load includes the 250 kW chiller starting direct-on-line.

Three details decide whether this behaves:

• Four-pole ATS with switched neutral. The medical IT systems are separately derived sources, and the generator is another one. If the neutral is not switched, the neutral current of the two systems shares a path, the insulation monitoring sees an earth fault that does not exist, and the theatre alarms at 3 a.m. for no reason. The pole costs USD 4,000 and the callback costs more.
• No transformer back-feed. The ATS must be on the LV side and mechanically interlocked so that the generator can never energise a transformer that the utility has isolated — a back-feed onto a 11 kV feeder is an event that ends careers, and the interlock is a mechanical device, not a PLC logic line.
• Earthing that stays TN-S through the transfer. The hospital's main equipotential bonding bar is the reference for the medical IT systems, the UPS and the generators; if the generator is earthed separately and the neutral is switched, the bonding bar must still tie the two earth systems together at one point only, or the 0.2 Ω equipotential requirement in the theatres becomes unachievable. The arrangements and the pitfalls are set out in the LV grounding system guide.

Fuel is the last sizing decision and the one that gets cut in value engineering. 1,560 kW for 72 hours at 0.21 kg/kWh is 23.6 t of diesel, or about 28,000 litres, which means two 15,000 litre tanks in a bunded room with a fire rating, not a 5,000 litre day tank and a promise. A hospital that runs out of fuel at hour nine has bought the same outage it was trying to avoid.

Oil-immersed or cast-resin dry-type for a hospital?

Cast-resin dry-type, in almost every case, and the reason is siting rather than electrical performance. A hospital's load centres are in the basement and on the fourth floor, next to theatres and wards, and an oil-filled transformer indoors requires a fire compartment, an oil containment pit, a 3-hour fire separation and, in most jurisdictions, a suppression system — all of which costs more than the price difference between the two technologies and consumes floor area the hospital does not have. A cast-resin unit to IEC 60076-11 with an F1 fire class, an E2 environmental class and a C2 climatic class will not propagate fire, will tolerate 100% humidity condensation, and will run at −25 °C, which is why it is the default indoor choice.

The economics only favour oil when the transformers can live outdoors. For this design the plant-room units are cast-resin 2,000 kVA at 11/0.415 kV, and the comparison at that rating is instructive: an ONAN oil unit at S13 losses costs about USD 54,600 and a cast-resin unit with equivalent losses about USD 92,000, so the dry-type premium is roughly USD 150,000 across four units. Set against that, the oil option needs a detached substation building, a fire wall, a bunded oil pit and a longer LV cable run — typically USD 180,000–250,000 in construction — and the dry-type wins on total cost while also removing an oil-handling maintenance item from a facility that has no electrical maintenance team of its own. If the hospital has a yard and can put the bank outdoors, the calculation flips and oil is the better buy; the decision rule is siting, not technology preference. The oil versus dry-type selection guide works through the same trade-off for industrial buildings.

Two caveats belong in the specification. Cast-resin units have a lower short-circuit withstand per unit of price and a higher impedance tolerance, so the through-fault rating must be stated against the 46.4 kA available at the 415 V terminals — not assumed. And cast-resin transformers are noisier per kVA than oil units at the same flux density, which is the next question.

How loud can a transformer be before it disturbs a ward?

Quieter than the standard allows. A 2,000 kVA ONAN oil unit measures about 62 dB(A) at 1 m; the equivalent cast-resin unit is 3–5 dB higher at 66 dB(A). The WHO guideline for hospital wards is 35 dB(A) at night and 30 dB(A) in rooms where patients sleep, so the transformer cannot be treated as an outdoor plant item with a 1 m rating and forgotten. Distance and mass do the work: sound pressure falls 6 dB per doubling of distance, and a 200 mm concrete wall between the plant room and a ward gives 40–45 dB of attenuation, which turns 66 dB(A) at source into something below the ventilation noise floor.

What that means in practice is a siting rule rather than a product requirement: no transformer immediately below or beside an ICU, a NICU or a patient room; the plant room on a structural slab with an acoustic separation detail; flexible connections and spring isolators under the units, because structure-borne vibration at 100 Hz travels further through a concrete frame than airborne noise does through a wall; and a specification that calls for 58 dB(A) at 1 m rather than the IEC 60076-10 limit, which is a maximum and not a target. The measurement method, the effect of core flux density and the difference between sound power and sound pressure are set out in the noise level guide; the hospital-specific addition is that noise complaints in a hospital come from staff on night shift, and a transformer that is acceptable at 10:00 is not acceptable at 03:00.

Ventilation and noise then pull against each other: the plant room needs 10,500 m³/h to remove 35 kW of loss at a 10 K rise, and an air path that size is an acoustic leak. The resolution is a lined plenum with two 90° bends and attenuators sized for the fan, not a louvre directly to the corridor — a detail that costs USD 15,000 at construction and USD 150,000 as a retrospective fix.

What does round-the-clock operation do to loss evaluation?

It makes no-load loss the dominant term, and it makes the usual "amorphous core saves energy" advice wrong. A hospital runs 8,760 hours a year at a load factor of 0.68, so the equivalent full-load hours for load loss are 8,760 × 0.68² = 4,050, while no-load loss accrues for every one of the 8,760 hours. At USD 0.12/kWh, no-load loss is worth 8,760 × 0.12 = USD 1,051 per kW per year, and load loss USD 444 per kW per year — a ratio of 2.4:1 that no factory can match.

Read the second and third rows together, because they are the counter-intuitive part. The S13 saves 0.45 kW of no-load and 0.7 kW of load loss against the S11, worth USD 813 per unit per year, which repays its USD 2,600 premium in 3.2 years — a clear buy. The amorphous core saves another 0.6 kW of no-load, worth USD 631 per unit per year, but it adds 1.4 kW of load loss, which at 4,050 equivalent hours costs USD 680 — a net loss of USD 49 per unit per year. In a hospital, the amorphous premium of USD 5,500 per unit never repays, because the load factor is high enough that load loss matters as much as no-load loss. In a substation feeding a residential feeder at 0.25 load factor the answer would be the opposite, which is why the loss grade comparison has to be re-run for every load profile rather than copied from a brochure.

The tender language that captures this is a loss-evaluation clause, not an efficiency class: state the capitalisation values (USD 1,051/kW for no-load, USD 444/kW for load) and let each bidder's losses be priced. A bidder offering a cheaper core then loses the evaluation on arithmetic rather than on opinion, and the buyer does not have to argue about which efficiency class is "enough". The same method, applied to a full ownership period rather than to losses alone, is in the total cost of ownership guide.

What should the transformer specification actually say?

Twelve lines, each of which has a reason, and none of which is a standard clause repeated for decoration. A specification written this way is also short enough that a bidder cannot answer it with a brochure.

Two clauses that are not about the transformer but belong in the same document: the four units must be interchangeable, which means matched impedance and the same tap range, and the 11 kV cables must be sized for the tie current rather than for each transformer's own rating — because the moment the bus tie closes, one feeder carries three units' worth of load through a cable that was ordered for one. That mistake appears in commissioning, when the tie is tested under load and the cable temperature rises faster than the calculation said it would.

What does the electrical package cost, and which four mistakes cost the most?

For a 300-bed hospital of this scope, the complete electrical infrastructure — transformers, 11 kV switchgear, LV boards, generators, UPS, medical IT systems, filters and cabling — comes to roughly USD 1.45–1.60 million, about 1.2% of a USD 120 million construction budget. The table below is the version a client can check line by line.

The four mistakes that cost the most are not exotic, and three of them are invisible until the building is occupied:

• Sizing at 40 °C. It looks like a 7% saving on transformer capital and it removes 140 kVA of capacity per unit, which is exactly the margin the N+1 arrangement needs. In a 45 °C plant room it turns a design that runs at 87% with a unit out into one that runs at 94% and trips on a hot afternoon.
• One 11 kV feeder. A second feeder from a different substation costs USD 250,000–400,000 depending on distance, and it is the only thing that protects against a utility cable fault. A hospital with one feeder and three generators is a hospital that tests its generators for real, at 02:00, in the middle of a delivery.
• One ATS with a three-pole neutral. Saving USD 22,000 here produces nuisance insulation alarms in every theatre, an electrician who cannot find the fault, and eventually a bypassed alarm — the worst possible outcome, because a bypassed medical IT alarm converts a monitored system into an unmonitored one.
• A plain capacitor bank. On a 38% non-linear load, a 600 kVAr un-detuned bank is a resonant circuit tuned close to the 5th harmonic. It will fail, it will fail in the first two summers, and it will take contactors and fuses with it. Detuned reactors cost 12% more and turn the bank into a filter.

None of these is a manufacturing problem, and none can be fixed by buying a better transformer. They are all decisions taken in the four weeks when the single-line diagram is drawn, which is why the most useful thing a hospital can do is hand the manufacturer a load schedule and a redundancy statement rather than a kVA figure — and why the questions in the supplier audit checklist are worth asking before the purchase order, not after the factory acceptance test.

The design in this article is deliberately conservative: four units where three would carry the load, a 45 °C rating where the standard allows 40, a UPS that is larger than the critical load, and a filter that costs more than the harmonic limit requires. Every one of those choices buys something a hospital cannot buy later — time. A transformer bank that runs at 60% and a critical branch that can lose any single component without a theatre going dark is not over-engineering; it is the minimum configuration for a building that is never allowed to close.

Sources / 资料来源

Source: IEC 60076-1 — Power transformers: General (rating, tolerances, routine tests).
Source: IEC 60076-2 — Power transformers: Temperature rise for liquid-immersed transformers (40 °C maximum ambient, 30 °C monthly, 20 °C annual reference).
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 60076-10 — Power transformers: Determination of sound levels.
Source: IEC 60076-11 — Power transformers: Dry-type transformers (fire class F1, environmental class E2, climatic class C2).
Source: IEC 60076-20 — Power transformers: Energy efficiency.
Source: IEC 60364-7-710 — Low-voltage electrical installations: Requirements for special installations or locations — Medical locations (group 0/1/2 classification, medical IT systems, insulation monitoring, equipotential bonding).
Source: IEC 61558-2-15 — Safety of transformers: Particular requirements and tests for isolating transformers for the supply of medical locations.
Source: IEC 60909-0 — Short-circuit currents in three-phase AC systems: Calculation of currents.
Source: IEC 61000-3-6 — Electromagnetic compatibility: Assessment of emission limits for distorting loads in MV and HV power systems.
Source: IEEE 519-2014 — Recommended practice and requirements for harmonic control in electric power systems (Table 2, 1 kV < V ≤ 69 kV).
Source: IEEE C57.110 — Recommended practice for establishing transformer capability when supplying nonsinusoidal load currents.
Source: IEEE 602 — Recommended practice for electric systems in health care facilities.
Source: NFPA 99 — Health Care Facilities Code (essential electrical system, type 1 and type 2 arrangements).
Source: NFPA 110 — Standard for emergency and standby power systems (Type 10, 10-second transfer).
Source: WHO Guidelines for Community Noise — recommended night-time limits of 35 dB(A) in hospital wards.
Source: Industry experience — connected load densities of 4–6 kW per bed and 55–70 kW per operating theatre for full-service hospitals in Southeast Asia and the Middle East, 2019–2025.
Source: Industry experience — measured transient voltage dips of 2.0–3.0% on shared LV boards serving CT and lift loads, against 0.5–0.8% on dedicated imaging feeders, from hospital commissioning records in Indonesia, Vietnam and Saudi Arabia, 2021–2025.
Source: Industry experience — cast-resin and oil-immersed price differentials and project loss-evaluation values of USD 400–1,100 per kW, transformer tenders for healthcare projects, 2024–2026.
Source: QDTB Engineering Team — hospital load schedules, substation layouts and transformer specifications for healthcare projects in Southeast Asia, the Middle East and Africa, 2019–2026.