Why Generators Fail When You Finally Need Them
The backup generator occupies a strange psychological position. It is the thing people point to when asked whether they are protected — 'we have a generator' — and it is, simultaneously, among the least genuinely tested pieces of critical equipment they own. That gap between trust and evidence is where backup power goes to die, and it dies in a small number of recurring ways that are worth knowing by name, because naming a failure mode is the first step to provoking it on purpose before it provokes itself at the worst moment.
Start with the most common and least excusable: the generator that will not start at all. Batteries are the usual culprit — a starting battery quietly ages, loses capacity, and is never load-tested, so it turns the engine over fine during a warm monthly check and fails to crank on a cold emergency morning. Or the fuel has aged in the tank; diesel degrades, grows microbial contamination at the fuel-water interface, and forms sludge that a resting tank hides until the day flow demand stirs it up and it clogs the filters. Or the coolant heater failed months ago and nobody noticed, so the engine that always started warm now refuses to start cold. None of these is exotic. All of them are invisible to a test that does not stress the system the way an emergency will.
Suppose it starts. Now the deeper failures appear, and they appear only under load, which is precisely why they are missed. A diesel engine running at light load or no load does not reach proper operating temperature, and unburned fuel and oil accumulate in the exhaust — a condition called wet stacking that fouls the engine and, over time, damages it. The irony is exact: the monthly no-load test, performed to ensure reliability, is itself degrading the machine. An engine babied at idle for years may be in worse shape than one worked hard, and the babying was done in the name of care.
Under real load, cooling becomes the frequent villain. An engine idling in a cool room sheds heat easily. The same engine carrying full building load for hours generates far more heat, and any weakness in the cooling system — a marginal radiator, a failing pump, a blocked louver, an undersized room ventilation path — shows up only then. The generator runs beautifully for the fifteen-minute test and overheats on hour two of a real outage, because hour two is the first time it was ever asked to do the actual job. The same is true of the governor and voltage regulator, which hold frequency and voltage steady as load swings; instability that is harmless at idle can trip protective devices or damage sensitive equipment when the machine is actually loaded and the load is actually changing.
Then there is the transfer itself. The automatic transfer switch must sense the utility failure and swing the load to the generator, and swing it back when utility returns, without dropping or double-feeding the load. Transfer switches fail, stick, or mis-sequence, and a generator that runs perfectly in isolation is useless if the switch never hands it the load. This component is often the least tested of all, because truly testing it means actually cutting the building over to generator power, which people are reluctant to do precisely because they are not confident it will work — a circular timidity that guarantees the first real test is the emergency.
The common thread is that every one of these failure modes is provoked by realistic load and sustained duration, and hidden by the gentle, brief, unloaded test that most facilities actually perform. The test was designed to be non-disruptive, and in achieving non-disruption it achieved non-information. It proved the engine could idle. It never asked the engine to do its job, so it never learned whether the engine could.
Faraday's read: generators do not fail randomly in emergencies. They fail in the specific, knowable ways that a real test would have exposed and a fake test was carefully designed not to. The failure was always there, waiting. The only question was whether you would meet it during a rehearsal you controlled or during a crisis you did not.