UPS topologies

A data centre promises its customers that the servers will not lose power, ever. The utility cannot keep that promise: even a very reliable grid delivers several momentary interruptions and dozens of voltage sags a year, and occasionally an outage lasting hours. Servers, on the other hand, tolerate an interruption of only about 10 to 20 milliseconds before their power supplies drop out. Backup power closes that gap with two layers: an uninterruptible power supply (UPS), which carries the load instantly from stored energy for seconds to minutes, and standby generators, which start, take over the load and run for as long as fuel lasts. Transfer switches and paralleling switchgear connect the pieces, and the redundancy architecture decides how many of each are needed so that a failure or a maintenance outage never reaches the IT load.

This course follows that chain, from the utility supply through UPS, batteries, generators and transfer equipment to protection, redundancy, testing and operations. This lesson covers the UPS itself: the power problems it must ride through, the main topologies, ratings, efficiency and bypass arrangements. The section at the end of this lesson sets out how the twelve lessons fit together. The course builds on the distribution systems of DCE-310.

What the UPS protects against

Disturbances on the utility supply range from microseconds to hours:

• Transients and surges: microsecond spikes from lightning and switching, handled mainly by surge protective devices.
• Sags (dips): drops to 10 to 90% of nominal voltage for half a cycle to a minute, usually caused by faults elsewhere on the network that are cleared by protection. Sags are the most common disturbance by far.
• Swells: brief rises, for example when a large load is disconnected or during a ground fault on another phase.
• Momentary interruptions: loss of voltage for a fraction of a second to a few seconds while utility reclosers clear a temporary fault.
• Sustained outages: anything longer, from minutes to days after storms.
• Frequency variations, harmonics and voltage unbalance, which matter more when running on generators or weak grids.

The tolerance of IT equipment is described by the ITIC curve (formerly the CBEMA curve), published by the Information Technology Industry Council for 120 V equipment and widely used as a reference for other voltages. It shows the voltage and duration combinations equipment should survive: for example 0% voltage (a complete interruption) for up to 20 ms, 70% for up to 0.5 s and 80% for up to 10 s. Disturbances outside the curve may cause the power supply to shut down, so the UPS must keep the output inside it whatever the input does.

UPS topologies

The international standard for UPS performance, IEC 62040-3, classifies UPSs by how their output depends on the input: VFD (voltage and frequency dependent), VI (voltage independent) and VFI (voltage and frequency independent). These correspond to the three static topologies used in practice.

Standby (off-line)

In normal operation, the load is fed directly from the utility through a filter. When the input fails or goes outside limits, a transfer switch connects an inverter powered from the battery. The transfer takes a few milliseconds (typically 2 to 10 ms) and the output waveform may be a stepped approximation of a sine wave. Standby UPSs are cheap and efficient (98 to 99%) but are classified VFD: the load sees every sag and frequency variation until the UPS decides to transfer. They are used for desktop computers and small network equipment, not data centres.

Line-interactive

A line-interactive UPS keeps the inverter connected in parallel with the load and adds a tapped autotransformer that boosts or bucks the voltage during sags and swells without using the battery. The inverter runs in reverse to charge the battery and takes over in an outage, with a very short transfer. Classified VI, line-interactive units suit edge sites, network closets and small server rooms up to a few kVA, and offer efficiency near 97 to 98%.

Double-conversion (on-line)

In a double-conversion UPS, the rectifier converts all incoming AC to DC, the DC bus charges the battery, and the inverter recreates clean AC for the load continuously. When the input fails, the rectifier simply stops and the battery feeds the DC bus with no transfer at all; the load cannot see the event. The output voltage and frequency are generated by the inverter, so the UPS is VFI: input sags, swells, frequency variations and generator instability never reach the load. Double conversion is the standard topology for data centres and large critical loads.

The penalty is losses: every watt passes through two conversion stages. Modern transformerless IGBT double-conversion UPSs reach about 96 to 97% efficiency at 50 to 100% load, falling at light load; older transformer-based units were nearer 92 to 94%.

Eco-mode and energy-saver modes

Most large double-conversion UPSs offer an eco-mode in which the load is normally fed through the static bypass and the inverter idles, ready to take over within a few milliseconds when the input deviates. Efficiency rises to about 99%, but the load is exposed to input disturbances until transfer, effectively making the UPS VFD or VI. Newer "advanced eco-modes" keep the inverter running in parallel to filter harmonics and correct power factor, transferring in under 2 ms. Operators weigh the savings against the risk; many use eco-mode only when the input is stable and the UPS has redundant capacity.

Rotary UPS and flywheels

A rotary UPS uses a motor-generator set or a synchronous machine with a large rotating mass instead of (or as well as) an inverter. Flywheel energy storage spins a heavy rotor, often in a vacuum on magnetic bearings, and discharges it through a power converter during an interruption. Flywheels store enough energy for only 10 to 30 seconds at full load, but that is enough to start generators, and they avoid the chemistry and replacement cycle of batteries.

A diesel rotary UPS (DRUPS) combines a flywheel or induction coupling, a synchronous machine and a diesel engine on one shaft: in normal operation the machine filters the utility and stores kinetic energy, and on an outage the engine starts and clutches in within a few seconds while the stored energy carries the load. DRUPS units of 1 to 3 MW are popular in Europe and in some large North American sites because they eliminate batteries and separate generators, at the cost of mechanical complexity and higher noise.

Ratings: kVA, kW and power factor

A UPS has two output ratings: apparent power in kVA, limited by the inverter's current, and real power in kW, limited by its thermal and DC capacity. The ratio is the rated output power factor. Older UPSs were rated at 0.8 power factor (a 100 kVA unit delivered 80 kW); modern data centre UPSs are typically rated at unity (1.0), so kVA equals kW. IT loads with active power factor correction present a power factor of 0.95 to 0.99, often slightly leading at light load, which the UPS must be able to handle.

The load on a UPS must not exceed either rating. Designers normally load a UPS module to no more than about 80 to 90% of its rating in the worst case, which leaves margin for growth, load inrush and measurement error, and keeps the UPS in its efficient range.

Inside a double-conversion UPS

Rectifier and input harmonics

Older UPSs used thyristor (SCR) rectifiers, which draw current in pulses and inject harmonics into the supply: a six-pulse rectifier has a current total harmonic distortion (THDi) around 30%, dominated by the 5th and 7th harmonics. Twelve-pulse rectifiers with phase-shifting transformers and passive filters reduced this to about 8 to 10%. Modern UPSs use IGBT active rectifiers that draw nearly sinusoidal current (THDi under 3 to 5%) at near-unity power factor. Low input distortion matters most when running on generators, whose high internal impedance turns harmonic current into voltage distortion.

DC bus and battery

The DC bus links rectifier, battery and inverter. Its voltage, typically 400 to 800 V DC in large UPSs, sets the number of battery cells in series (Lesson 3). The rectifier provides both the inverter's power and the battery charging current, so it is sized above the output rating, and the charge current is limited to avoid overloading the input or the generator after an outage.

Inverter

The inverter uses IGBTs switched at several kilohertz with pulse-width modulation, followed by an output filter, to create a low-distortion sine wave (THDv under 2% for linear loads, a few percent for nonlinear loads). Its overload capability is limited: typically 110% for 10 to 60 minutes, 125% for about 10 minutes, and 150% for under a minute. Above that the UPS transfers the load to the static bypass, which can carry very high currents briefly, for example to clear a downstream fault or supply transformer inrush.

Static bypass and maintenance bypass

The static bypass is a solid-state switch (back-to-back thyristors) that connects the load directly to a bypass supply in under a few milliseconds if the inverter fails or is overloaded. Transfer is seamless only if the inverter is synchronized to the bypass, which it does continuously whenever the bypass is within limits. The bypass can be fed from the same source as the rectifier (single feed) or from a separate source (dual feed), which keeps the bypass available if the rectifier input breaker trips.

The maintenance bypass is a set of external breakers that route power around the whole UPS so that it can be isolated, de-energized and serviced while the load stays on (raw) utility or generator power. It must be interlocked, with key interlocks or electrically operated breakers, so that operators cannot parallel the inverter output with an unsynchronized source or open the wrong breaker. Many outages in the industry's incident reports occurred during maintenance bypass switching, which is why it is a carefully scripted procedure.

Modular and parallel UPS systems

Large UPS capacity is built either from big monolithic modules (500 kVA to 1.6 MVA) paralleled on a common output bus, or from frame-based modular systems in which 25 to 250 kW power modules slide into a cabinet. Modular systems allow capacity to grow with the IT load and offer internal N+1 redundancy: a 600 kW frame with seven 100 kW modules can lose one and still carry 600 kW. They also allow failed modules to be swapped without shutting down. Paralleled UPSs share load through communication between modules and must be synchronized to each other and to the bypass.

UPS in the power chain

Large data centres place UPSs at the low-voltage level (480 V in the US, 600 V or 415 V in Canada, 400 V in many other countries) between the main switchboard and the PDUs, feeding critical loads through dedicated distribution. Mechanical loads that need to keep running through the generator start, such as chilled-water pumps, CRAH fans and controls, are often placed on a separate mechanical UPS so that cooling does not stop while the generators start. Medium-voltage UPSs (rotary or static at 4 to 15 kV) are used on some hyperscale campuses to reduce current and cable sizes.

Alternative approaches remove the central UPS altogether. Open Compute Project racks use a 48 V DC power shelf with built-in lithium-ion battery backup units inside each rack, eliminating one conversion stage and the central UPS room. Some hyperscalers place small batteries in each server. These designs trade the maintainability and monitoring of a central UPS for efficiency and simplicity at the rack.

How this course is organized

The twelve lessons follow the power path from the grid to the rack and then turn to the work of keeping it reliable:

• Lesson 1 (this lesson): UPS topologies, ratings and arrangements.
• Lesson 2: the utility supply and power quality, including sags, harmonics, ride-through and service arrangements.
• Lesson 3: UPS batteries, runtime, sizing, monitoring and battery room safety.
• Lesson 4: DC power and rack-level storage, from −48 V telecom plants and 380 V DC to high-voltage DC and battery backup units for AI racks.
• Lesson 5: generators, sizing for data centre loads, fuel and emissions.
• Lesson 6: battery energy storage systems, grid services and on-site generation such as gas engines and fuel cells.
• Lesson 7: transfer switches, paralleling switchgear and the sequence of operation in an outage.
• Lesson 8: protection and coordination when the source is a generator, UPS or battery rather than the utility.
• Lesson 9: redundancy architectures and availability calculations.
• Lesson 10: testing and commissioning, from factory tests to integrated systems testing.
• Lesson 11: operations and maintenance, procedures, monitoring and incident response.
• Lesson 12: a design project combining all of these for a mixed AI and conventional data hall.