A transfer that happens once is not proof that the site is protected. The useful test is the one that shows where the sequence spends its time, what happens when a step fails, and whether the people who operate the plant can repeat the result and leave a record behind.
A transfer test is not a start test
An engine start and a load transfer are different events. A start test proves that the battery, starter, fuel supply and controller can bring the set to rated speed. A transfer test proves that the automatic transfer switch recognises the source that is available, moves the load in the right order, and does not create an unsafe connection between two supplies.
The distinction matters because a set can start perfectly and still fail the site: the ATS may be stuck in the wrong position, transfer while frequency is still recovering, or chatter on retransfer. None of these faults appears in a start-only test.
The transfer has to be judged against three outcomes: the correct source is connected, the time taken is acceptable for the loads that cannot ride through, and no unsafe state exists at any point in the sequence. The UPS is what covers the seconds between mains loss and generator load, so the transfer time and the UPS autonomy have to be read together. A transfer that is fast enough for the UPS but leaves the chilled-water plant unstable is still a failed test for the facility.
For the difference between the switching device and the mains-failure function, see ATS versus AMF. This article starts after that choice has been made and asks a narrower question: how do you prove the installed sequence works, and how do you prove it again next year?
| Test | What it proves | What it does not prove |
|---|---|---|
| No-load engine start | The set cranks, starts and reaches rated speed | Whether the ATS can carry load or whether the sequence is correctly timed |
| Isolated ATS mechanical transfer | The switch moves between positions and the interlocks hold | Behaviour with real load, real harmonics or a weak source |
| Loaded transfer | The set accepts the site load and the ATS carries it without a trip | Whether the retransfer will be stable when the mains returns |
| Retransfer test | The mains-return window, resynchronisation and cool-down work | Whether a failed start or a stuck switch will be detected |
| Fault injection | The controller responds to abnormal conditions as designed | Long-term wear, fuel quality or battery ageing |
| Full drill with roles and records | People, communications and documentation work under time pressure | Every conceivable failure, or the endurance of the plant |
The sequence, stage by stage
A transfer is a chain of decisions, not a single command. The controller first has to decide that the mains is genuinely unavailable rather than dipped. It then issues a start command, waits for the set to become a valid source, and only then closes the generator side. The mains side has to open before or with that closure, depending on the transition type, and the interlocks have to make the prohibited state impossible.
When the mains returns, the sequence runs in reverse but not symmetrically. The controller usually waits through a return window before accepting the utility as stable. It then transfers the load back, runs the set at no load for a cool-down period, and stops it. Each stage has a timer or a threshold, and each one can be set to a value that looks reasonable in a datasheet but is wrong for the site.
Record each stage, not only the final result. A transfer inside the expected total time can still hide a slow start, an unusually fast transfer or a load lower than the design case. Stage-by-stage evidence makes the result repeatable.
| Stage | What starts it | Evidence to capture | Common false pass |
|---|---|---|---|
| Mains failure detection | Voltage, frequency or phase loss outside the accepted window | Controller event, timestamp and measured value | The test used a clean open circuit instead of a realistic dip |
| Start command | Controller accepts the failure and calls for the set | Start signal, start delay and alarm state | The set was already running, so the start path was never exercised |
| Crank and ignition | Start command reaches the engine controller | Crank time, battery voltage, first firing | The battery was on charge and not representative of a cold standby condition |
| Voltage and frequency ready | Set reaches the controller’s acceptance band | Voltage, frequency, excitation and stabilisation time | The ATS accepted a source that was still outside the load’s tolerance |
| Transfer to generator | Ready signal and transfer command | Switch position, transition type and interlock status | The switch moved but the load was isolated by a downstream breaker |
| Load acceptance | Load is connected to the generator | Current, power, frequency dip and recovery | The test load was much smaller than the design load |
| Mains return window | Utility voltage and frequency return | Return delay, number of attempts and stability | A short window hid a supply that would chatter in service |
| Retransfer | Return window expires | Transfer command, switch position and load step | The load was dropped before retransfer, so the real step was never seen |
| Cool-down and stop | Load is off the generator | Cool-down timer, temperatures and stop event | The set was stopped immediately, hiding a hot-stop problem |
Timing: measure the gaps, not the total
The number most often quoted after a test is the total time from mains failure to load restoration. It is a useful headline, but it is a poor diagnostic. A total that meets the target can be made up of a fast detection, a slow start and a transfer that only just beat the deadline. If the next event has a colder engine or a lower battery, the same total may not be repeatable.
Record each gap separately and keep the timestamps on one clock. The ATS controller event log, the generator controller and the site power recorder should be time-synchronised before the test; otherwise a two-second argument between logs becomes a day of investigation. Where the controller cannot export events, record the screen at each stage and note the time. A stopwatch at the switchroom door is not evidence of what the controller did.
The target for each gap belongs to the project, not to this article. A data centre with a UPS may accept a longer generator start because the UPS carries the load; a site with no ride-through may need a shorter transfer. The manufacturer’s data defines what the set can do, and the project specification defines what the load can tolerate. The test compares the two, then records the difference.
Timing also has a fuel cost. The 500 kVA example burns 60 litres an hour at 50 percent load, 80 at 75 percent and 101 at full load; a 1000 kVA set at about 200 litres an hour uses about 4800 litres over 24 hours. Repeated retransfer or an extended cool-down is not free, even within the 0.24 to 0.29 litres per kWh band, which describes delivered energy rather than idle running.
| Checkpoint | Start event | End event | What to record | Failure signature |
|---|---|---|---|---|
| Detection to start command | Mains leaves the accepted window | Controller issues the start signal | Measured voltage and frequency, delay setting, event time | Start delayed by a marginal voltage threshold or a filter that is too slow |
| Crank to ready | Start signal reaches the engine | Voltage and frequency enter the acceptance band | Crank time, battery voltage, oil pressure, speed | Long crank, repeated attempts or a set that is ready only after the ATS has given up |
| Ready to transfer | Set is accepted as a valid source | Generator side closes | Transfer command, switch position, interlock status | Extra delay from a conservative setting, or a switch that receives the command but does not move |
| Close to load stable | Load is connected | Voltage and frequency recover inside the load tolerance | Current step, frequency dip, recovery time, UPS status | Load step exceeds the set’s capability or the UPS transfers back too early |
| Mains return to retransfer | Utility returns inside the accepted window | Load transfers back to the mains | Return delay, number of confirmations, source stability | A window that is too short causes chatter; one that is too long keeps the site on fuel |
| Retransfer to stop | Load leaves the generator | Engine stops after cool-down | Cool-down time, temperatures, stop command | Immediate stop, repeated restart or a controller left in the wrong mode |
Failure modes to inject, one at a time
A transfer test that only demonstrates a successful transfer proves the happy path. The failure paths are where most sites discover that the controller was configured for a different switch, that an alarm was never wired, or that the operator does not know how to reset a lockout. Inject one fault at a time, record the response, reset the system and confirm normal operation before the next injection.
Use the controller’s simulation mode, a secondary injection set or an approved test switch. Do not create a real backfeed, an actual arc or an open live conductor to make the test look realistic. If a fault cannot be injected safely, test the logic at the controller terminals with the power source isolated, and have a second person verify the isolation. The project specification and the manufacturer’s instructions decide what is permitted on the installed equipment.
The expected response should already exist in the cause-and-effect matrix. If it says only “alarm”, the test cannot tell whether the transfer was inhibited or the generator allowed to run. Write the expected state before the drill, then compare it with the observed state.
The most revealing faults leave the site on the wrong source without a trip. A stuck ATS with the correct lamp, a control supply that fails only when the generator runs, or a mains monitor seeing a backfeed can look normal at the panel. Check the actual source at the load terminals, not only the controller indication.
| Injected condition | Expected response | Evidence | Safe method |
|---|---|---|---|
| Mains undervoltage | Detection, start command and transfer per the matrix | Event log, measured voltage, transfer time | Secondary injection at the controller voltage input |
| Loss of one phase | Phase-failure alarm and no transfer to an invalid source | Alarm state, phase voltages, switch position | Controller simulation or an isolated test input |
| Frequency outside the window | Start is allowed, but transfer waits for a valid source | Frequency trace, ready signal, transfer command | Signal injection with the ATS test mode enabled |
| Engine fails to start | Start attempts stop at the set limit and an alarm is raised | Attempt count, alarm, lockout state | Inhibit starting at the controller, not by removing fuel under load |
| Generator not ready | No transfer; the load stays on the available source or is dropped as designed | Voltage and frequency at the ATS, transfer command | Simulate a low excitation or speed signal at the controller |
| ATS fails to close | Alarm, retry if configured, and a defined safe state | Command, position feedback, motor or coil current | Open the control circuit at an approved test point |
| Control supply loss | The switch holds position or moves to the defined fail-safe state | Position, alarm, restoration behaviour | Remove the control fuse with the main circuit isolated |
| Mains returns during transfer | No simultaneous closure; the sequence completes or aborts safely | Both source voltages, interlock status, event order | Controller sequence simulation, not a live utility switching test |
Load matters: no-load, load bank and the annual test
The annual load test is the point at which the generator is asked to carry 50 percent to 100 percent of its rating, as set out in the maintenance schedule. The 10 percent to 20 percent light-load band is an operating caution, not a target for acceptance: a set that spends long periods in that band may not reach the temperatures and pressures that keep the engine healthy. A transfer test at no load does not replace the annual loaded test, and the annual loaded test does not replace the transfer sequence check.
A load bank proves the generator and ATS together without putting the site at risk, provided it is connected at the right point and the ATS is tested in the real configuration. A bank connected directly to the generator terminals may bypass the ATS and prove less than it appears to.
In a data centre, the UPS covers the seconds between mains loss and generator load, but not a generator that never accepts the load. A PUE 1.4 example shows why the test load is the facility load, not the IT nameplate alone: the generators see the IT load plus cooling and losses. The same logic applies to large motor loads, where starting current and load step are what the set feels.
Test at the site’s reference ambient. A 40 °C reference is the condition the equipment will meet in service, not a laboratory condition to avoid. Capacity, cooling and transfer timing can change when the room is hot, the radiator recirculates or the enclosure is restricted. Derating is set by the manufacturer’s data and the project specification, but the test should run in the site’s actual condition.
| Tier | What it proves | What it cannot prove | Typical trigger |
|---|---|---|---|
| No-load functional check | Start, stop, alarms and switch movement without load | Load acceptance, fuel autonomy or thermal performance | Routine exercise and after a control change |
| Loaded transfer at site load | The real load can be carried and the sequence is stable | The set’s full rating or a worst-case step | Commissioning and planned shutdowns |
| Annual loaded test at 50%-100% | The set performs across the upper part of its rating | Every possible transfer fault or the full fuel autonomy | Annual maintenance programme |
| Light-load check at 10%-20% | Operation in the low band is understood and monitored | That the set is suitable for continuous light-load service | When the site load is unusually low |
| Post-change retest | The modified sequence still matches the cause-and-effect matrix | Long-term reliability of the new settings | Any change to ATS, controller, generator or protection |
The drill: roles, communications and stop conditions
A drill is more than someone pressing the test button. It is a planned exercise with a start state, a script, a person in charge and a defined way to stop. The script should cover the normal transfer and at least one failure path, and it should be written before the day so that the people involved are testing the plant rather than inventing the procedure.
Before the drill, confirm fuel, battery, coolant, oil and every alarm. Confirm the UPS is healthy and has enough autonomy for the transfer, because it covers the seconds between mains loss and generator load. Confirm the fire-system interface and fire mode: a generator required for a life-safety load must not be stopped by a routine test. The fire safety sequence and the commissioning checklist are the references.
Give one person control of the test and one person the log. Use one clock and one communications channel. The person at the ATS should call each state change; the person at the generator should confirm the engine condition; the load owner should confirm what the site actually lost. If the site has a building management system, the BMS trend is a second witness, not a substitute for the local observation.
Write the stop conditions into the script: a fuel leak, smoke, over-temperature, abnormal vibration, a chattering switch, a source that cannot be isolated or any doubt about the next step. Abort and restore the site to a known source. A drill that continues because the schedule says so is no longer a test. Record every source and switch state before anyone leaves.
| Role | Before | During | After |
|---|---|---|---|
| Site operator | Confirms load status, access and notifications | Calls the load state and watches for unexpected loss | Confirms normal supply and closes the permit |
| Electrical authorised person | Confirms isolation points, interlocks and test method | Controls switching and verifies source separation | Checks the switch positions and restores locks |
| Generator technician | Checks fuel, battery, coolant, oil and alarms | Observes start, load acceptance and temperatures | Reviews the event log and records defects |
| Load owner | Confirms which loads may be interrupted and for how long | Confirms the effect on the process or occupied space | Confirms the load is back and stable |
| Safety or fire watch | Confirms detection, suppression and fire mode status | Watches for smoke, fuel release and blocked egress | Confirms the fire system is back in service |
Acceptance criteria and the record that proves it
A transfer passes when the observed sequence matches the cause-and-effect matrix, every stage is inside its agreed time, the voltage and frequency recover within the load’s tolerance, no two sources are connected at once, the correct alarms appear and reset, the retransfer is stable and the engine completes its cool-down before stopping. The criteria should be written as measurable statements, not as “transfer OK”.
The record should let a person who was not present reconstruct what happened. Keep the controller event log export, the timing sheet, the load readings, the ambient temperature, the fuel quantity before and after, the battery voltage, the UPS status and the list of defects. If a stage was not measured, say so; a gap in the record is more honest than a reconstructed time. The report should also state which tests were not performed and why.
Retest after any change that can alter the sequence: a new ATS, a controller firmware change, a changed timer, a modified protection setting, a replaced battery, a fuel-system change or a new load that changes the step. The annual load test at 50 percent to 100 percent remains the generator anchor; the transfer drill remains the sequence anchor.
Skipping the record costs at the next event. Without the last transfer time, load and active alarms, there is no baseline for judging the next result. The record also shows the drill was safe: who authorised it, what was isolated, what was restored and who confirmed it.
Map the sequence before the next drill
A transfer test is only as good as the cause-and-effect matrix, the switch settings and the load list behind it. Send the single-line diagram, the ATS sequence table, the generator rating and the last event log. The review will identify the stages to measure, the faults to inject safely and the records to keep before the next annual load test.
How long should an automatic transfer take?
There is no universal figure. The acceptable time is set by the loads that cannot ride through, the UPS autonomy and the project specification. Measure detection, start, ready, transfer and load recovery separately so that a slow stage is visible even when the total looks acceptable.
Can the ATS be tested without shutting down the site?
Part of it can. A no-load functional check and some fault injections can be done with the load isolated or transferred to a load bank. A loaded transfer that proves the real load can be carried normally needs a planned interruption, a bypass arrangement or a test configuration agreed with the site owner.
How often should the transfer sequence be tested?
The annual load test at 50 percent to 100 percent of rating is the established anchor. Functional checks, failure drills and retests after changes are set by the project specification and the manufacturer’s instructions. Any change to the ATS, controller, protection settings or load should trigger a retest of the affected stages.
What is the difference between an ATS test and a generator load test?
A generator load test proves the set can carry load and sustain its voltage and frequency. An ATS test proves the switch recognises the sources, transfers in the correct order, holds the interlocks and retransfers safely. A complete test programme needs both, and a combined loaded transfer is the closest to the real event.
Which failure modes should be simulated?
At least a mains undervoltage or phase loss, a failed start, a generator that is not ready, an ATS that fails to close, a control-supply loss and a mains return during the sequence. Use the controller’s simulation mode or an approved test set; never create a real backfeed or an open live conductor to make the test look realistic.
Does a successful transfer prove the UPS is covered?
No. The UPS covers the seconds between mains loss and generator load, so its autonomy and transfer behaviour have to be checked against the measured sequence. A transfer that completes within the UPS ride-through can still fail if the UPS cannot accept the generator source or if its batteries are degraded.
KVA Grid sales engineering desk. Figures are published elsewhere on this site: the annual load test at 50 % to 100 % of rating and the 10 % to 20 % light-load band from the maintenance schedule; 0.24 to 0.29 litres per delivered kWh and the 500 kVA figures from fuel consumption; the 1000 kVA example, UPS role and PUE 1.4 from data center backup power; 40 °C ambient from hot and humid climates; the transfer sequence and adjustable delays from ATS versus AMF; load-bank practice from load-bank testing; UPS interaction from generator and UPS interaction; control interfaces from surge protection and control panel; earthing from generator earthing and neutral; cable sizing from cable sizing; fire interfaces from fire safety; and handover records from the commissioning checklist. Timing targets, tolerances, fault-injection limits and acceptance criteria are set by the manufacturer’s data and the project specification. Reviewed 7 November 2026.
