Once a data center relies on more than one generator set, the electrical problem changes character. A single set either carries the load or it does not. A group of sets must agree on voltage, frequency, phase relationship, and share of the load, and it must keep agreeing while the load moves, while machines start and stop, and while one of them fails. Synchronisation and load management are the disciplines that make that possible, and they are where most multi-set installations succeed or disappoint.
This article covers the control and switching architecture behind multi-set standby power: how sets are paralleled, how load is shared, how the facility sequences its own recovery, and what must be tested before the arrangement can be trusted. It assumes the number of sets and the redundancy target have already been decided, and focuses on making that decision work in operation.
Why synchronise at all
Synchronising allows generator sets to operate as a single source. That delivers three things a data center needs. The first is capacity: several sets can supply a load no single unit could carry, which is often the only way to serve a large facility within the practical limits of equipment that can be delivered, lifted, and housed. The second is maintenance: with sets paralleled, one can be taken out of service while the others carry the critical load, provided the remaining capacity is sufficient. The third is response: a group can accept a load step by sharing it, so each machine sees a smaller disturbance.
Those benefits are not automatic. They depend on the switchgear being able to operate in the required configurations, on the controllers being able to share load accurately, and on the sequence of operation being defined for every state the plant can be in — including the states nobody planned for, such as a set tripping while the bus is live.
Bus architecture: the decision that constrains everything else
Before controllers are selected, the bus arrangement determines what the plant can physically do. A single bus with all generators and all load connected to it is simple and permits any combination of sets to supply any load, but it also means a bus fault affects everything. A split bus with a tie breaker allows the facility to be divided into sections, each supplied by its own generator, with the tie closed only when conditions permit. A ring or duplicate-bus arrangement extends that logic further, at the cost of more switchgear and a more demanding control philosophy.
The choice interacts directly with the facility's electrical topology. If the data center is arranged as two independent power trains, the generator bus should normally mirror that arrangement, so that a fault in one train does not propagate through a shared generator bus. If the load is arranged as a single block, a split generator bus adds complexity without adding resilience.
Whatever arrangement is chosen, the drawing set must show every permissible configuration, including which breakers are closed in each, and the control system must enforce those configurations rather than trusting operators to remember them.
What the controllers actually do
A paralleling controller performs four distinct functions, and it is worth separating them because procurement often conflates them. The synchroniser matches the incoming generator's voltage, frequency, and phase angle to the live bus and closes the breaker at the right moment. The load-sharing function distributes real power between running sets by trimming their speed references. The reactive-sharing function distributes reactive power by trimming excitation. The protection and interlock function prevents closure when conditions are unsafe and trips when conditions become unsafe.
Load sharing accuracy is where installations differ most in practice. Two machines of the same rating should share in proportion to their capacity; two machines of different ratings should share in proportion to rating rather than equally. The control mode — droop or isochronous — determines how the group behaves when it is islanded and how it behaves when it is paralleled with the utility, and the two modes are not interchangeable. A plant intended to parallel with the utility must be able to operate in a mode that keeps exported power under control, which generally means a droop or base-load characteristic with a defined setpoint rather than pure isochronous control.
Reactive sharing deserves equal attention and usually receives less. Unequal reactive loading causes one machine to run hotter and to reach its excitation limit first, which reduces the group's effective capacity. The cause is often a mismatch in excitation characteristics or in the voltage reference settings, and it is resolved at commissioning by measurement rather than by adjustment by feel.
Sequence of operation: the document that prevents most failures
The sequence of operation describes what the plant does, in order, for each initiating event. It should cover utility loss with all sets available, utility loss with one set unavailable, a set tripping while carrying load, a set failing to start, restoration of utility, and a planned transition to and from generator supply. For each case it should state which breakers operate, in what order, with what time delays, and what the load does in response.
Two states are routinely under-specified. The first is the degraded case: with one set out of service, the plant must either carry the full critical load on the remaining sets or shed a defined portion of it. Whichever is chosen must be written down, because the alternative — discovering at 3 a.m. that the plant cannot carry the load — is the failure mode that redundancy was supposed to prevent. The second is the resynchronisation case: after the utility returns, the plant must decide whether to transfer back with a brief interruption or to synchronise and soft-load. Each has implications for the UPS and for the mechanical plant, and each requires the appropriate switchgear and control capability.
Load management belongs in the same document. The facility should define the order in which mechanical load is reconnected after a transfer, the maximum block that may be added at each step, and the conditions that trigger a shed. Cooling restart is usually the largest and most disruptive element, and coordinating it with the generator's capability is often more valuable than adding generator capacity.
Paralleling with the utility
Some data centers operate generators in parallel with the utility for peak shaving, demand response, or a controlled transfer. Each of those uses requires protection against unintended export or import, a defined interface with the utility's protection, and typically an approval from the network operator. The technical requirements are well established: synchronising check, reverse power protection, directional overcurrent, and a means of ensuring the generator cannot energise a de-energised utility network.
The commercial and regulatory side is often the longer path. Approval timelines, metering arrangements, and the conditions under which the facility may island or reconnect should be established early, because they can determine the switchgear configuration rather than merely the paperwork. A plant designed for island operation only and later required to parallel will need equipment it does not have.
Protection and earthing in parallel operation
When generators operate in parallel, fault current contributions add, and the protection settings that were correct for a single set may be incorrect for the group. The protection study must be run for the maximum and minimum configurations — all sets running, one set running — because a setting that detects a fault in one case may be blind in the other.
Earthing is the other area where parallel operation changes the analysis. Multiple generators, multiple neutral points, and a transfer switch that may or may not switch the neutral combine into arrangements where circulating currents or unwanted earth fault paths can appear. The neutral treatment must be decided once, documented on the single-line diagram, and reflected consistently in the protection study and the transfer switch specification.
What commissioning must prove
Synchronisation is easy to demonstrate and hard to prove. The tests that matter are the ones that reproduce the conditions the plant will face. At minimum, commissioning should demonstrate: correct synchronising on each incoming set, including refusal to close when conditions are outside limits; load sharing within an agreed tolerance across the operating range, including at part load; reactive sharing within tolerance; a load step applied with all sets running and with one set out; a deliberate trip of one set while the group carries load, with the remaining sets stable and the critical load maintained; and a full transfer and retransfer cycle including the cooling restart sequence.
Each of these should produce a record. A commissioning certificate that states the plant "operates correctly" without measured values leaves the owner unable to detect drift later, because there is no baseline to compare against.
The configuration logic in one list
- Bus architecture that mirrors the facility's power-train topology, with every permissible configuration drawn and enforced.
- Controller functions separated and specified: synchronising, real power sharing, reactive sharing, protection and interlock.
- Control mode stated for both islanded and parallel-with-utility operation, with the setpoints defined.
- Sequence of operation covering normal, degraded, restoration, and failure cases.
- Load management rules: reconnection order, maximum step, shedding thresholds.
- Protection study run for maximum and minimum generation configurations.
- Neutral and earthing philosophy consistent across single-line, protection study, and switchgear.
- Commissioning tests with measured values recorded as a baseline.
Frequently asked questions
How many generator sets can be paralleled in a data center?
There is no fixed limit, but practical limits arrive quickly: switchgear size, control complexity, bus fault level, and the space available. Most data center installations use a modest number of sets in a defined topology rather than many small units, because the control and protection burden grows with each additional machine.
What happens if one generator trips while the group is carrying load?
The remaining sets must absorb the lost share or the plant must shed load. Which of those occurs is a design decision, expressed in the sequence of operation and enforced by the control system. If the remaining capacity is insufficient and no shedding is configured, the group will collapse — a scenario that should be tested deliberately at commissioning rather than discovered in service.
Is isochronous control suitable for a plant that parallels with the utility?
Not on its own. Paralleling with the utility requires a control mode that holds a defined power setpoint, typically a droop or base-load characteristic. The plant may use isochronous control when islanded and switch modes when paralleled, but that transition must be designed and tested.
How is load shared between generators of different ratings?
In proportion to rating, not equally. The controller needs the rating information and the correct sharing characteristic. If this is set incorrectly, the smaller machine will run at a higher percentage load and reach its limit first, reducing the group's usable capacity.
Does paralleling improve fuel efficiency?
It can, because the number of running sets can be matched to the load rather than running one large machine at low load. Whether that benefit materialises depends on the load profile and on the control logic that decides when to start and stop sets, which should include hysteresis to prevent excessive cycling.
What documentation should the owner retain after commissioning?
The as-built single-line with all permissible configurations, the sequence of operation, the protection settings with the study that produced them, the controller configuration file, and the measured results of every commissioning test. Together these allow a later fault or drift to be diagnosed against a known baseline.
Review your paralleling and load management design
Shanghai Outevo Machinery Co., Ltd. supplies generator sets configured for single-unit and multi-unit operation, with the control and paralleling arrangement defined against the facility's topology. Configurations typically cover 10–3,750 kVA when applicable, and the interface between generator control and site switchgear is agreed at specification stage rather than left to commissioning.
If your project involves multiple sets, send the single-line diagram, the redundancy target, the critical load and its step profile, and the load management intent. Our engineering team will respond with the control philosophy implications, the interface responsibilities, and the tests that should be written into the commissioning plan.
