A standby set that has only ever run off load has proved that it starts. A load bank is the only routine test that puts working heat into the engine, the radiator and the fuel system at the same time, and it is the difference between a generator that starts and a generator that carries. This page sets out what the test proves, what a resistive bank can and cannot load, how a test is run step by step, and how to do it on a site that cannot drop its load.
The short answer: what a load bank test proves
A load bank is a controlled electrical load that is connected to the generator and dissipates the power the set produces. It replaces the building as the load, so the test can be run at any figure from a light load to full rating, at any time, without waiting for an outage and without putting a real installation at risk.
That control is the point. A monthly start with no load proves the battery, the starter and the fuel supply to the injection pump. It does not raise the combustion temperature, it does not heat the radiator to its working range, and it does not ask the fuel system to deliver at rated power. A load bank does all three, and it does them in a sequence that can be repeated and recorded.
The annual test on this site is written as 50 % to 100 % of rating, because that is the band in which the engine reaches working temperature and the cooling system is asked to reject real heat. A test that stops at 50 % is better than no test, but it is not a full-load proof.
| A load bank proves | A load bank does not prove |
|---|---|
| The engine carries its rated load without overheating, misfiring or smoking | That the changeover from mains to generator works: that is an ATS or AMF test |
| The radiator and the room reject the heat of a real load at the site’s ambient temperature | That the UPS carries the load while the set starts, or that the batteries have the autonomy claimed |
| The fuel system delivers at rated power, and the consumption matches the curve | That the cable, the switchgear and the distribution boards downstream can carry the current |
| The alternator holds voltage and the engine holds frequency through load steps | That the protection grading is correct for a fault on the installation |
| The protection settings do not trip a healthy set at rated load | That the fuel tank has the autonomy the site needs for a long outage |
| The exhaust is clean at load and the set recovers from a step without a sustained dip | That the building’s own load profile will behave like the bank |
What happens to an engine that never carries load
Diesel combustion needs heat in the chamber, and heat comes from load. At 10 % to 20 % of rating the engine runs cool enough that the fuel does not burn completely. The unburnt fraction leaves as soot and liquid fuel, some of it collects in the exhaust and the turbocharger, some of it washes past the rings into the oil, and the deposits that form on the pistons, the valves and the exhaust path reduce the set’s ability to make power later.
The industry name for the result is wet stacking. The visible sign is a set that smokes under load, or an exhaust that is wet rather than dry, or an oil level that rises between services because fuel has diluted the oil. The expensive sign is a set that passes every no-load start for years and then fails on the first real transfer, which is exactly the moment the installation existed for.
Two things follow from the mechanism. First, a monthly run should carry whatever load the site can give it: the maintenance schedule on this site puts the useful figure at 50 % or more of rating, held long enough for temperatures to stabilise. Second, a set that spends its life at 10 % to 20 % has a sizing or load-management problem, and no service interval fixes it. The oil change interval has to come down, but the cause stays.
| Symptom | What is happening | What it points to |
|---|---|---|
| Black or grey smoke when load is applied | Fuel that has accumulated in the exhaust and the cylinders burns off in a cloud | A history of light-load running, not necessarily a fault in the injection system |
| Wet exhaust, soot at the outlet, dirty turbocharger | Combustion temperature is too low to burn the fuel completely | Load factor; check the load profile before changing parts |
| Oil level rising between services | Fuel dilution through the rings, or coolant entering the sump | Fuel dilution is the light-load signature; coolant is a different fault |
| Loss of rated power over a few years | Carbon on the valves, the pistons and the exhaust path | A set that has never been loaded, or one that is oversized for its duty |
| Smoke and a stumble on the first real transfer | Deposits release when the engine is suddenly asked for full fuel | The failure the annual load test exists to find on a planned day |
Resistance, reactance and what each one loads
A generator is rated in kVA at a power factor, usually 0.8. A resistive load bank presents power factor 1.0, so at the set’s rated kW it loads the engine fully but the alternator only to about 80 % of its rated current. That is enough to prove the engine, the cooling system and the fuel system. It is not enough to prove the alternator at its rated kVA, because the alternator’s heating comes from current, and current at unity power factor is lower for the same kW.
A reactive bank adds inductive kvar, which brings the current up at the same kW and lets the set be run at its rated kVA and power factor. That is the test that exercises the alternator windings, the exciter and the automatic voltage regulator together. A combined bank does both, in steps, and is the instrument for a witnessed commissioning test on a set that matters.
Capacitive banks are used for leading power factor tests, which are a specialist check on the AVR’s stability rather than a routine site test. Most rental fleets in the Gulf are resistive, because the faults they are hired to find are engine and cooling faults; a data centre commissioning test usually needs the reactive capability as well.
| Bank type | What it presents | What it proves | Where it fits |
|---|---|---|---|
| Resistive | kW at power factor 1.0 | Engine, fuel system, radiator, exhaust, governor | Annual standby test, rental fleet checks, commissioning where the alternator is proven separately |
| Reactive (inductive) | kvar at lagging power factor | Alternator windings, exciter, AVR, voltage regulation at rated kVA | Witnessed commissioning, data centre and hospital acceptance |
| Combined | kW and kvar together, at 0.8 power factor | The whole set at its rating, in the way the site will actually load it | Factory or site acceptance where the full rating has to be demonstrated |
| Capacitive | Leading kvar | AVR stability at leading power factor | Specialist checks, not a routine site test |
How a test is run: steps, hold points and records
A load test is a sequence, not a switch. The set is started and warmed at no load, then loaded in steps, held at each step long enough for the temperatures to settle, and brought down in the reverse order with a cooling period before it is stopped. The step sizes and the hold times come from the set maker’s test procedure and the project specification; the shape below is the one that produces a usable record.
Before the first step, four things are checked: the fuel quantity is enough for the whole test with margin, the oil and coolant are at the right level, the load cables and the connection point are rated for the test current, and the exhaust and the room ventilation can handle the heat the bank and the set will reject. A load bank in a closed room adds its own heat to the set’s, and a test that overheats the room proves nothing about the generator.
| Step | Typical hold | What the hold establishes |
|---|---|---|
| Start and no-load warm-up | 5 to 10 minutes | Oil pressure, charging, frequency at no load, no unusual noise or smoke |
| 25 % of rating | 10 to 15 minutes | The first loaded step: fuel delivery, governor response, exhaust colour |
| 50 % of rating | 30 minutes | Cooling begins to work; the set should be stable and dry |
| 75 % of rating | 30 to 60 minutes | The band where a marginal radiator or a restricted air path starts to show |
| 100 % of rating | Until temperatures are stable, commonly 1 to 4 hours | Rated output, fuel rate, exhaust, protection stability, alternator temperature |
| Step down and cool | 5 to 10 minutes at no load | Turbocharger and exhaust cool evenly; the set is stopped without heat soak |
Readings are taken at every step and at fixed intervals during the full-load hold, not only at the end. The record should carry kW, kVA, power factor, voltage per phase, current per phase, frequency, oil pressure, coolant temperature, exhaust temperature, ambient temperature and fuel consumed. A single set of numbers at the end of the test cannot show a trend; a table taken every 15 minutes can show a temperature that is still climbing, which is the difference between a pass and a set that would have tripped ten minutes later.
The readings that decide pass or fail
“It ran for an hour” is not an acceptance criterion. A test passes when the set holds its rated load with the temperatures stable inside the maker’s limits, the voltage and frequency recover after each step, the fuel consumption is close to the published curve, and the exhaust is clean. The ambient temperature at the test is part of the result: a set rated at 40 °C that is tested at 48 °C in a Dubai summer is working outside its reference condition, and the comparison has to be made with the derating applied rather than against the brochure figure.
| Reading | Planning acceptance | Where the limit comes from |
|---|---|---|
| kW, kVA and power factor | The rated figure, held without a reduction other than the ambient derating | The set rating at the site ambient, not the brochure rating at 40 °C |
| Voltage, steady state | Within about ±1 % of the set point, balanced across phases | The AVR and the alternator data; a large imbalance points to the load or the sensing |
| Frequency, steady state | Within about ±0.5 % of rated | The governor; a slow drift under fixed load is a fuel or governor issue |
| Transient voltage and frequency | Dip and recovery inside the project’s step-response figure, with no trip | The specification for the site; the alternator’s subtransient reactance sets the dip |
| Coolant and oil temperature | Stable, not still climbing, inside the maker’s limit | The engine manual; stability matters as much as the absolute figure |
| Fuel consumption | Close to the published curve for the load and the ambient | The fuel consumption figures for the set size, corrected for load |
| Exhaust | Clean at steady load, no sustained smoke after the first step | The engine maker’s smoke limit and the local emission requirement |
Testing a site that cannot drop its load
The sites that need the test most are the ones that can least afford to run it on their own load. A data centre cannot drop its IT load to prove a generator, and a hospital cannot switch off a theatre to run a test. The answer is a portable load bank connected at the generator terminals or at a dedicated test point, with the changeover held in the test position so that the building never sees the bank.
Three arrangements cover most cases. At the generator breaker, the bank is connected to the set’s output terminals with the set isolated from the installation; this is the cleanest test of the machine and proves nothing about the changeover. At a dedicated test point downstream of the generator breaker, the bank is switched in while the set carries the building, which tests the set under a controlled additional load. In a full integrated test, the bank provides the base load while the changeover, the UPS and the building sequence are exercised together.
What must not happen is a bank connected in parallel with the mains. The set is not paralleled to the utility for a test, and an arrangement that depends on two sources being in step is a different, engineered operation with its own protection. If the site has a closed-transition changeover, the test procedure has to come from the changeover maker and the parallel operation has to be treated as a designed function, not as a convenient way to avoid an outage.
| Site | Arrangement | What has to be arranged |
|---|---|---|
| Data centre | Bank at the generator breaker or a dedicated test point; UPS carries the IT load | Test procedure agreed with the operator, battery autonomy confirmed, cooling available for the bank’s heat |
| Hospital | Bank on the essential bus during a planned window, or at the set with the essential load transferred | Clinical sign-off, a window that does not overlap a procedure, and a fallback supply |
| Hotel | Bank at the set, with the building on mains and the changeover in test | Guest-facing noise and exhaust routing, and a window that avoids peak occupancy |
| Industrial plant | Bank at the set, or the plant load itself if it can be staged | Production schedule, load staging, and the site’s own permit-to-work |
The sequence the load bank does not test
The bank proves the machine. It does not prove that the installation notices a mains failure and does something about it. That is a separate test, and the two belong on the same day: the bank supplies the load, the mains is opened at the incomer, and the changeover is timed from the loss of supply to the closure of the generator breaker and the acceptance of load.
What is timed and recorded: the delay before the set starts, the time to reach rated speed and voltage, the transfer time of the changeover, the voltage and frequency dip seen by the load, and the retransfer and cool-down after the mains returns. On a site with a UPS, the same test shows whether the battery carried the load for the whole of the gap, which is the one number a load bank alone cannot produce.
The load bank and the sequence test also answer different questions about protection. The bank shows that the set does not trip a healthy circuit at rated current. The sequence test shows that the changeover does not create a fault current path or a back-feed, and that the neutral arrangement is correct for the transfer. The earthing and neutral arrangement on a generator supply is set out in generator earthing and the neutral, and it is worth reading before a transfer test rather than after one.
How often, and what changes with duty
The maintenance schedule on this site puts a load bank test at once a year, alongside the protection test and the fuel quality test. That is the standby case. Other duties move the interval: a new set is tested before handover, a set that has had a major overhaul is tested before it goes back into service, and a rental set is tested between hires because the next site will not forgive a fault that the last site caused.
| Duty | Interval | Why |
|---|---|---|
| New installation | At commissioning, before handover | The first full-load proof, and the baseline every later test is compared with |
| Standby | Annually, at 50 % to 100 % of rating | The set may run only a few hours a year, so the test is the only time it reaches working temperature |
| Prime or continuous | At commissioning, after overhaul, and at a planned interval | The set runs loaded in service, so the test is a verification rather than the only load it sees |
| Rental fleet | Before and after each hire, or at the fleet’s own interval | The set changes site, load and operator, and the next customer inherits the last site’s treatment |
| After a major repair | Before return to service | Injectors, turbocharger, fuel pump and head work all change the way the set makes power |
The mistakes that cost the most
Most failed load tests are not failures of the generator. They are tests that were set up in a way that could not prove anything, or tests whose result was read without the ambient correction that makes it comparable.
| Mistake | What it costs | What prevents it |
|---|---|---|
| Testing at 50 % and calling it a full-load test | The radiator, the fuel system and the exhaust are never asked for rated output | Run to 100 % of the ambient-corrected rating, or state the test as a partial test |
| Resistive bank only, then claiming the alternator was proven | The alternator runs below rated current at unity power factor and never reaches its thermal limit | Use a reactive or combined bank when the alternator rating is the question |
| No ambient correction | A set that passes at 35 °C is recorded as passing at 48 °C, and the site believes a figure it does not have | Record the ambient and compare against the derated rating for that temperature |
| Twenty minutes at full load | Temperatures are still climbing when the test is stopped, so the result is a snapshot, not a stable point | Hold until the readings are stable, and record the trend |
| Undersized or long load cables | The voltage at the set looks poor because the cable drops it, and the set is blamed for the test rig | Size the cables for the test current and keep the run short |
| No record, or a record with one column | Nothing to compare next year, and no evidence for the insurer or the operator | One sheet per test, every step, every 15 minutes, kept with the machine |
| Bank in a closed room with no heat path | The room overheats, the set derates, and the test proves the room rather than the generator | Ventilate for the combined heat of the set and the bank, or test outdoors |
Questions about generator load bank testing
Can I use the building load instead of hiring a bank?
Sometimes, and it is the better test when the building load is large enough and can be controlled. The difficulty is that a building load is not a dial: it cannot be held at 75 % for an hour, it cannot be repeated next year at the same figure, and on most sites it cannot be switched off if the test needs to stop. A bank gives a repeatable load at a known power factor, which is why it is used for the record even where the building load is available.
How long should the set be held at full load?
Long enough for the coolant, oil and exhaust temperatures to stop rising, which is commonly one to four hours depending on the set, the ambient and the room. The number in the maker’s procedure is the one to use. A short hold at full load is still useful as a functional check, but it does not establish that the cooling system can reject the heat continuously, and that is the question a standby set exists to answer.
Is a monthly run at 50 % enough, without an annual bank test?
If the site can genuinely put 50 % or more on the set every month and hold it long enough for temperatures to stabilise, the engine sees real load and the wet-stacking risk falls. It still does not produce the annual record at 100 %, and it does not test the protection at rated current. The maintenance schedule keeps the annual test for that reason: the monthly run is routine health, the annual test is proof.
Will a load bank damage a set that has never been loaded?
A set that has been badly maintained can fail during its first real load, and that is a finding rather than a fault of the test. The safe sequence is to check the fuel, the oil, the coolant, the battery and the exhaust before the test, run the steps in order, and stop on any reading that leaves the maker’s limits. A set with a genuine defect should fail on a planned day with the bank attached, not on the day the building loses power.
What size bank do I need for a 500 kVA set?
The bank has to be able to hold 100 % of the set’s ambient-corrected kW rating, and to reach the rated kVA if the alternator is being proven. For a 500 kVA set at 0.8 power factor that is 400 kW, and a 400 kW resistive bank reaches it at unity power factor. To test the alternator at 500 kVA and 0.8 power factor, the bank needs the reactive capability as well, which is a larger and heavier unit. The set’s rating at the site ambient, not the nameplate at 40 °C, sets the figure.
Does the test replace the factory test?
No. The factory test is carried out on the set before it ships, in the maker’s conditions, and it is the reference for what the machine can do. The site test proves the same machine in the site’s ambient, with the site’s fuel, cooling air and exhaust arrangement. The two are complementary, and the site acceptance page sets out what the handover check covers beyond the load test.
Put the failure on a planned day
Tell us the set rating, the site ambient and whether the building load can be used or the test needs a bank. The reply sets out the test steps, the load figures and the readings to record, in writing.
Written and maintained by the KVA Grid sales engineering desk. The intervals and figures on this page are the ones published elsewhere on this site: a load bank test once a year at 50 % to 100 % of rating and a monthly run at 50 % or more where the site can take it, from the maintenance schedule; the 10 % to 20 % light-load band and its symptoms from fault finding and the total cost of ownership; the fuel figures of about 0.24 to 0.29 litres per delivered kWh, and 101, 80 and 60 litres an hour for a 500 kVA set at full, 75 % and 50 % load, from fuel consumption; and the 40 °C reference with derating above it from hot and humid climates. The step sequence, the hold times and the acceptance values are planning values; the set maker’s test procedure and the project specification govern the test that is actually run. Nothing on this page is a quotation for a particular project. Reviewed 2 November 2026.
