Generator acceptance and load bank turnover for data center standby power

Generator acceptance, and what it actually proves

Generator acceptance is the test and turnover that proves a standby or emergency engine-generator plant will start, take its load in time, and carry it when the utility drops. It is the day you find out whether the plant the design promised is the plant that got built. NFPA 110, the standard for emergency and standby power systems, gives the framework. The on-site installation acceptance test is run after the whole emergency power supply system is installed and operating, from a cold start, as if the power had actually failed.

What acceptance proves is not that the engine makes power. A factory test already proved that. Acceptance proves the installed system, in this building, with this fuel, this cooling, this transfer switch, and this control wiring, behaves the way the emergency depends on. The engine that ran clean on the factory dyno can still fail to pick up load in time because the transfer switch time delay was left at default, or trip on high coolant temperature because the louvers do not open far enough.

Treat the date as a hard gate, not a formality. Once the building takes occupancy, the generator is the thing standing between a utility outage and a dark critical load. Accept it on a short run nobody witnessed and you have a signature, not a generator.

The EPSS as a system, not a generator

The thing being accepted is a system. NFPA 110 calls it the emergency power supply system, the EPSS, and it covers everything from the stored fuel to the load terminals of the transfer switch. Accept the engine alone and you have proven one link in a chain that fails at its weakest point.

The pieces have to work together: the engine and its alternator, the fuel system with its day tank and transfer pumps, the cooling system, the exhaust and ventilation, the engine controls with the governor and the automatic voltage regulator, the starting system and its batteries, the paralleling switchgear if there is more than one set, and the automatic transfer switch that actually moves the load. NFPA 110 splits this into the emergency power supply, which is the prime mover and generator, and the rest of the EPSS that delivers that power to the load.

Each piece has its own acceptance, and the system has one too. A generator that passes its own load test can still miss the transfer time because the ATS, a separate device under a separate listing, was never timed against the real engine start. The acceptance has to cross those boundaries, because the emergency does.

What do NFPA 110 Type, Class, and Level mean?

NFPA 110 classifies every emergency power supply system three ways, and the three together set what the acceptance test has to prove. Type is time. Class is duration. Level is consequence.

Type is the maximum time, in seconds, that the transfer switch load terminals are allowed to be without acceptable power after the normal source fails. Type 10 means the system has 10 seconds to deliver power to the load. Type 120 means 120 seconds. The life-safety case is almost always Type 10, the ten-second start.

Class is the minimum time, in hours, the system is designed to run at rated load without refueling. Class 48 means 48 hours on hand. Class X is whatever the application, the code, or the owner requires, and on data centers it often lands at 72 or 96 hours of on-site fuel.

Level is how bad it is if the system fails. Level 1 is where failure could cost a life, and it carries the strictest requirements. Level 2 is where failure is less critical to human safety. A data center critical load is commonly engineered to Level 1, Type 10, with the Class set by how long the site must ride an extended outage. Confirm the exact Type, Class, and Level against the project documents and the adopted edition, because they drive every acceptance number that follows.

What is an NFPA 110 acceptance test?

The NFPA 110 installation acceptance test is the on-site test, run after the entire EPSS is installed and operating, that proves the system performs from a cold start as if the utility had failed. It is the final approval test, and the authority having jurisdiction usually wants to witness it or wants the signed record before occupancy.

The sequence in plain terms: with the engine cold and only the jacket water heater keeping it ready, simulate loss of normal power, let the system start and transfer on its own, time how long the load terminals are dead against the Type, then run the plant under load. The load test is commonly held for at least 1.5 hours on building or simulated load in the standard, and many data center specs go further to a 2-hour full-load test with a load bank making up 100 percent of nameplate kW less site derating. Verify the required duration and load against the adopted edition and the spec.

The data is the deliverable. The standard calls for recording a defined set of parameters across the run, and a tabulated list of what to log and how often appears in it. Record the time to accept load, then voltage, frequency, current, kW, oil pressure, coolant temperature, and battery voltage on a time base across the hold. A run with no trend data is not an acceptance test. It is a story about one.

Cranking, start reliability, and the crank cycle

The most basic thing the acceptance test proves is that the engine starts, every time, on its own. Start reliability is where standby plants actually fail, because a battery, a fuel solenoid, or a control-wiring fault sits invisible until the start signal comes and nothing turns over.

NFPA 110 defines the cranking limits the control has to honor. A cycle-crank scheme cranks for about 15 seconds, rests about 15 seconds, and repeats for up to three cycles, with the whole sequence bounded near 75 seconds before it locks out on overcrank. A continuous-crank scheme is bounded near 45 seconds. Those limits exist to protect the starter and to raise an overcrank alarm rather than grind the batteries flat. Confirm the cranking scheme and the limits against the adopted edition and the controller settings.

For a Type 10 system the engine has to be running and accepting full load in one step within 10 seconds, so the start has to be near-instant and clean. Test it cold, more than once, and watch the start batteries. A marginal battery that turns the engine over on a warm afternoon will not do it on the one cold morning that counts, and the battery charger and its alarm are part of what gets accepted.

How is an automatic transfer switch tested?

The automatic transfer switch is tested by taking away the normal source and watching it run the whole sequence on its own: sense the loss, signal the engine to start, wait out its time delay, transfer the load to the generator, then on restoration wait the retransfer delay and move the load back. The ATS is a separate listed device from the generator, and its timers are where the system makes or misses the Type time.

Walk the time delays one by one. There is a short delay on engine start to ride through momentary dips so a one-cycle blink does not start the plant. There is a transfer delay to let the generator stabilize before it takes load. On retransfer there is a delay, commonly a minimum of 5 minutes, so the system does not dump back to a utility that is still hunting, and that retransfer delay setting gets recorded. There is usually an unloaded engine cooldown run after retransfer.

Test both directions and the exerciser. Loss of normal proves the transfer and the start. Restoration proves the retransfer and the cooldown. The built-in exerciser clock proves the routine the owner inherits will actually run the set. A common miss is accepting the transfer one way, never forcing a retransfer, and shipping a switch whose 5-minute timer was never confirmed against real utility return.

What if the genset misses the transfer time?

When the system does not get power to the load terminals inside its Type time, the failure is in one of three places: the engine took too long to start and stabilize, the transfer switch timers held the load off too long, or the block of load was bigger than the machine could accept in one step. Sort which before anyone touches a setting.

Time the pieces separately. Clock the crank-to-running time, then the running-to-stable time, then the ATS transfer delay. If the engine is up and stable in a few seconds and the clock still blew past 10, the time delays in the ATS are the suspect, not the engine. If the engine itself is slow to start or slow to hold speed, that is fuel, batteries, or governor.

A Type 10 system has no slack to give away to a conservative timer. The fix is usually trimming the ATS time delays to the minimum the design allows and confirming the engine reaches rated speed and voltage fast enough to accept the step. Do not solve a slow start by stretching the allowed time. The 10 seconds is the requirement, and the AHJ holds it. Re-run the cold start after any change and re-record the time to accept load.

Load bank testing for acceptance

Load banking is how you put rated load on the plant when there is no IT load yet, and the method has its own guide. For acceptance the point is narrower. The load bank lets you hold the engine at full nameplate kW, and a resistive/reactive bank lets you load the alternator to rated kVA at the rated power factor, commonly 0.8. A resistive-only bank loads the engine but leaves the alternator and the voltage regulator at roughly 80 percent, so it does not fully prove the machine.

Load is applied in steps, commonly 25, 50, 75, and 100 percent, with the readings logged at each, then held at full load for the required duration. For acceptance, the criteria are what matter: the plant carries full kW and kVA at rated power factor, holds voltage and frequency steady, recovers from block load inside the spec band, and keeps every temperature inside the manufacturer limits for the whole hold.

The load bank guide covers resistive versus reactive, sizing, connection, and wet stacking in depth. Here the rule is simple. If the bank cannot reach 100 percent at the rated power factor, the test did not accept the machine. It sampled it.

What voltage and frequency dip is acceptable on block load?

Acceptance on block load is judged on how far voltage and frequency dip when the load is applied and how fast they recover, not just on the steady-state values. When the generator picks up a large step, the voltage sags while the regulator catches up and the frequency dips while the governor brings the engine back to speed. The acceptance question is whether both recover inside the band the spec and the equipment allow.

The numbers come from the project spec and the generator data sheet, not from a number you carry in your head. As a rough field frame, many specs hold steady-state voltage within about plus or minus 1 to 2 percent and steady-state frequency within about plus or minus 0.5 percent, with a transient voltage dip often within roughly 10 to 15 percent recovering in a few seconds. Treat those as ranges to watch, then hold to the contract values.

The transient is the part a slow data log misses. A once-a-minute reading shows the steady value and never sees the dip, so the block-load capture has to be fast enough to record the depth of the dip and the time to recover. That capture is the evidence the alternator, the regulator, and the governor actually ride the step the building will hand them.

Fuel system and the Class runtime

The fuel system is what makes the Class real. Class 48 means nothing if the on-site fuel and the way it gets to the engine cannot keep the set running at rated load for 48 hours. Acceptance has to prove the day tank, the transfer pumps, the bulk tank, and the return all keep up at full burn for the whole hold, not just at the start.

Size and prove the day tank against the burn rate. A day tank sized for short exercise runs will run the engine dry in the middle of a long acceptance hold, and that is a finding, not a nuisance. Confirm the transfer pumps refill the day tank faster than the engine empties it, that the float controls and low-fuel alarms work, and that the on-site bulk capacity matches the Class the spec requires. The fuel-polishing system, where there is one, keeps stored diesel clean enough to burn, because fuel that sat in a tank grows water and bugs and plugs filters at the worst time.

Fuel quality is part of acceptance on a new plant. Diesel degrades, and a tank filled at construction and left for months can be off-spec by the time the plant runs. Confirm the fuel was sampled and treated. An engine that starves on dirty fuel during the acceptance hold has just told you what it will do in a real outage.

Cooling, exhaust, and ventilation acceptance

Cooling and ventilation are where long acceptance runs fail, because the heat that was theoretical on a short start becomes real over a two-hour hold. The radiator or remote cooling has to reject full-load heat at the worst-case ambient the room or yard will see, and the louvers, dampers, and intake path have to open and move enough air to feed combustion and carry the heat away.

Watch the coolant temperature climb across the hold. An engine drifting toward its high-coolant-temperature trip is telling you the cooling or the ventilation is short, and that finding is worth more than any kW number, because it is exactly the failure a real summer outage produces. Confirm the louver actuators open on engine start, the dampers are not interlocked shut, and the discharge air has somewhere to go that does not recirculate to the intake.

The exhaust system gets checked too: backpressure within the engine maker's limit, the silencer and piping supported and sealed so the run does not load the manifold or leak indoors, and the termination clear of intakes so the plume does not feed back to the radiator or the building. Excess exhaust backpressure costs power and runs the engine hot, and it shows up on the long hold, not the short one.

Jacket water heaters and keeping the set ready to start

A standby engine has to be warm enough to start instantly and accept load, which is why the jacket water heater runs whenever the set is idle. The acceptance cold start is defined as the engine at normal ambient with the jacket water heater functioning, not a literally frozen engine, and that distinction matters. The heater is what lets a Type 10 set make full load in 10 seconds without a warm-up idle.

Prove the heater and its controls. Confirm it holds the coolant at the manufacturer's keep-warm temperature, that its alarm raises on failure, and that the start batteries are charged and the charger alarms on loss. A cold engine cranks harder, starts slower, and accepts load worse, so a dead jacket water heater quietly turns a Type 10 set into a set that misses its time on the first cold morning.

This is the part that decays fastest after handover. The heater, the battery, and the charger are the readiness chain, and they are the first things that fail unwatched. The acceptance proves they work on day one. The owner's monthly routine is what keeps them working, which is why the alarms feeding the annunciator are part of what gets accepted.

Vibration, alignment, and the mechanical turnover

The mechanical side of acceptance is alignment, mounting, and vibration, and it is the part an electrical commissioning agent is most tempted to wave through. The engine-alternator set sits on isolators sized to the unit, and those isolators have to be installed right, leveled, and not bottomed out or shimmed wrong, or the set walks and the vibration cracks fuel lines, exhaust connections, and conduit over time.

Check the flexible connections. The exhaust has a flex section, the fuel lines have flexible runs, the cooling has flexible couplings, and the electrical has flexible conduit, all there to keep engine vibration out of the rigid building piping. A hard connection across that boundary is a crack waiting to happen, and it is a common install miss. Confirm every service crossing the engine boundary is flexible and has slack.

Run the set and feel it. Excess vibration at a speed, a new noise, or a hot bearing during the hold is a mechanical finding that belongs in the record. The manufacturer's commissioning technician usually signs the mechanical turnover, and that signature should be in the package alongside the electrical acceptance, because the plant is one machine and both halves have to be proven.

Controls, the remote annunciator, and required alarms

The control system and the remote annunciator are part of acceptance because the operator has to know the plant's condition without standing next to it. NFPA 110 requires a remote annunciator, commonly at a constantly attended location, that shows the system status and the alarm conditions, and acceptance proves every point on it actually reflects the engine.

Walk every alarm and status point. Force the conditions you can safely force and confirm each one lights and annunciates: overcrank, low and high coolant temperature, low oil pressure, overspeed, low fuel, low coolant level, battery charger failure, and the run and not-in-auto signals. The required set comes from the standard and the spec, so confirm the list against the adopted edition. The point that reads correctly on the local panel but never made it to the remote annunciator is the classic wiring miss, and it is the one that leaves an operator blind during the outage.

Prove the not-in-auto signal hardest. A generator left in off or manual after a service visit will not start on a real outage, and the not-in-auto alarm is the one thing that tells the operator the plant is disarmed. If that point does not annunciate, the most common cause of a no-start in the field has no warning.

The acceptance sequence, from factory test to integrated test

Acceptance runs in a sequence, and each stage builds on the one before. The factory test proves the set on the manufacturer's dyno before it ships. The receiving and installation checks prove it arrived undamaged and got installed right. The start and transfer test proves it starts and picks up load on its own. The load bank test proves it carries rated load and holds. Then the integrated systems test proves the whole power chain rides through a real failure together.

Do not collapse the sequence. A common schedule pressure is to skip the standalone generator load test and go straight to integration, which buries a generator finding inside a system test where it is hard to isolate. Accept the generator on its own first, with its own load bank and its own record, then bring it into the integrated test where it has to coordinate with the UPS, the transfer scheme, and the cooling.

The integrated systems test belongs to the broader power-QA scope of commissioning and is covered in that pillar. The handoff from generator acceptance is clean. The generator arrives at the integrated test already proven on its own, so any failure there is a coordination or timing fault between systems, not an unproven machine nobody load-tested.

The turnover package

Turnover is the moment the proven plant becomes the owner's responsibility, and the package is what makes that handoff defensible. The acceptance test records are the core of it: the witnessed start and transfer times, the load bank trend data, the block-load dips and recovery, the temperatures across the hold, and the pass or fail against each criterion, each signed by who witnessed it.

Beyond the test data, the package carries the settings, the documentation, and the spares. Record the as-left ATS time delays, the governor and voltage regulator settings, and the controller configuration, because the next technician needs the baseline to know whether a setting drifted. Include the operation and maintenance manuals, the wiring and one-line drawings as-built, the fuel and coolant specifications, the recommended spare parts and the consumables the owner should stock, and the warranty terms with the maintenance they require to stay valid.

The sign-off is the point of it. The commissioning agent, the owner, and often the engineer of record sign that the plant met the spec and the adopted NFPA 110 edition on a witnessed test. That signature is what a future operator relies on when the question is whether this plant was ever actually proven, and the path from here runs into the integrated systems test and then into the owner's maintenance program.

What the owner has to maintain

Acceptance is one day. The reliability the building lives on comes from the maintenance program the owner inherits the moment the plant is turned over, and NFPA 110 governs that ongoing testing the way it governed the acceptance. The two get confused, so name the line. Acceptance proves the new plant once. The routine keeps proving it for the life of the plant.

NFPA 110 sets a monthly exercise and an annual load test for emergency and standby diesel systems. The monthly exercise is commonly met by running at no less than 30 percent of nameplate kW for a set period, or by loading to the engine maker's minimum exhaust gas temperature, because lighter loading does not get the engine hot enough and lets it wet stack. The annual test runs the set through load steps for a longer period to confirm capacity and clear deposits. Wet stacking, the unburned-fuel fouling that comes from chronic light loading, is covered in the load bank guide.

Hand the owner the schedule, not just the equipment. Confirm the loading and durations against the adopted edition and the AHJ, because the steps have changed between editions. A plant accepted clean and then run lightly once a month without enough load is a plant that will smoke and surge on its first real call, and the turnover is where that program gets set up to avoid it.

What to document

The acceptance record is what a future operator trusts when the question is whether this plant was ever proven. Capture enough that a reviewer who was not there can reconstruct the test and check each result against the spec and the adopted NFPA 110 edition.

Record the plant identification and ratings, the Type, Class, and Level, the cold-start time to accept load against the Type, the readings at each load step, the block-load dips and recovery times, the temperatures and engine readings across the hold, the fuel used and the hold duration, the ATS time-delay settings, the alarm and annunciator point checks, the pass or fail per criterion, and who witnessed it. If anything was adjusted and re-tested, record the before and after, because the next person needs to see what changed and that it was proven after the change.

Common mistakes

• Accepting the generator alone and never running the integrated systems test that proves the transfer.
• Testing resistive-only when the spec calls for rated kVA, leaving the alternator and regulator at about 80 percent.
• Sizing the day tank and on-site fuel for short runs, not for the Class runtime the spec requires.
• Skipping the retransfer and never confirming the ATS time delays against a real utility return.
• Accepting on a short run that skips the full-load hold where cooling and fuel problems actually show up.
• Leaving the ATS time delays conservative so a healthy engine misses its Type time.
• Confirming an alarm on the local panel and never proving it reached the remote annunciator.
• Confusing the NFPA 110 monthly exercise and annual test with the one-time acceptance test.

Field checklist

Standards and references

Several bodies govern different parts of this, and naming the right one for the point is the difference between a credible turnover and a guess. NFPA 110, the standard for emergency and standby power systems, is the central one. It classifies the EPSS by Type, Class, and Level, sets the on-site installation acceptance test, and governs the ongoing monthly exercise and annual load test after handover. The acceptance and the routine maintenance sit in different chapters, and the loading steps and durations have changed across editions, so cite NFPA 110 by topic and confirm the section and the numbers against the adopted edition and the AHJ.

NFPA 70, the National Electrical Code, governs the installation by the kind of system: Article 700 for emergency systems, Article 701 for legally required standby, Article 702 for optional standby, and Article 708 for critical operations power systems. Which article applies is set by how the load is classified, which the AHJ and the design determine. For field acceptance testing of the electrical gear, ANSI/NETA ATS gives the inspection and test requirements before energization, and IEEE covers generator and engine testing methods.

Above all of these sit the manufacturer's instructions and the project specification, which set the actual numbers: the voltage and frequency tolerances, the rated power factor, the temperature limits, the load steps, and the hold duration. When a standard and the spec disagree, the stricter controlling document wins, and the authority having jurisdiction has the final say on what is enforceable.

Units and terms

The power and timing numbers on a generator acceptance come in a few forms, and reading the wrong one accepts a plant that should not have passed. kW is real power, the work the engine does. kVA is apparent power, what the alternator and the conductors carry. They relate through power factor, and most gensets are rated at 0.8, so the alternator carries about 25 percent more current than the kW alone implies.

The NFPA 110 classification carries its own shorthand. Type is seconds to load, Class is hours of runtime at rated load, and Level is the consequence class, 1 or 2. The EPSS is the whole emergency power supply system through the transfer switch load terminals. The ATS is the automatic transfer switch that moves the load. Frequency is in hertz, 60 Hz in North America and 50 Hz in much of the world, held by holding engine speed.

kW (real power)

The power the engine produces and a resistive load bank dissipates

kVA (apparent power)

The total the alternator and conductors carry, kW and kVAR combined

Power factor (PF)

Ratio of kW to kVA; most gensets are rated at 0.8 lagging

EPSS

Emergency power supply system, the whole plant through the ATS load terminals

ATS

Automatic transfer switch, the device that moves the load between sources

Type

NFPA 110 class for transfer time, the maximum seconds the load may be without power

Class

NFPA 110 class for runtime, the minimum hours at rated load without refueling

Level

NFPA 110 class for consequence of failure, Level 1 for life safety or Level 2

Related tools

Calculators and readiness checks for this work

Compare your options

FAQ

People also ask

• How long does data center electrical commissioning take?
• How long is a generator load bank test?
• What are the conditions to parallel two generators?
• How do you size a data center generator?
• How much fuel does a generator need?