Why your standby generator failed to start

When a building owner tells us the generator did not work, three different things could have happened, and they have almost nothing in common.

It never started. The utility went away, the controller called for a start, and the engine did not come up. It started and the building stayed dark. The engine ran perfectly, in some cases for hours, while the transfer switch never moved the load onto it. It started, took the load, and then quit. Fifteen minutes in, or ninety, the set shut down and did not come back.

Each of those has a different cause list and a different fix, so the first useful question after an event is which one you had. The annunciator or controller fault log usually answers it, and if the generator room was dark and nobody could read the panel, that is itself a finding worth writing down.

What all three have in common is what predicts them. The National Renewable Energy Laboratory compiled the available reliability data on engine-driven generators (NREL/TP-5C00-76553, 2020) and the spread by maintenance quality is not subtle. A well-maintained fleet of 239 units showed a 0.13 percent probability of failing to start. A nuclear-regulatory data set showed 0.66 percent. A poorly maintained set showed 1.65 percent, more than ten times the well-maintained figure. The gap is wider once the engine is running: mean time to failure of 1,662 hours versus 61 hours, which NREL describes as over twenty times longer. IEEE 493-2007 puts the general figure at 0.0135 failures per start attempt.

Two things fall out of that. Maintenance quality, not equipment age or brand, is the variable that moves generator reliability. And even a good machine is not a guarantee for a long event: one study in Applied Energy found a single engine-driven generator is only about 80 percent likely to carry a two-week outage on its own.

Here is the part that trips up nearly every article written on this subject, and a fair number of service proposals.

Section 1203 of the California Fire Code contains no numeric generator test intervals. No monthly. No annual. No hour figures at all. What it contains is a set of duties that point somewhere else:

Emergency and standby power systems are maintained in accordance with NFPA 110 and NFPA 111 (1203.4). The system runs on an approved schedule established when the installation was completed and approved (1203.4.2). Operational inspection and testing happens under load, again per those standards (1203.5). Transfer switches are part of the maintenance and testing schedule, with contacts inspected (1203.4.4), and the transfer switch test electrically operates the switch from normal to alternate and back to normal (1203.5.2). Records carry specified content (1203.4.3). All of it happens under the oversight of "a properly instructed individual" (1203.6).

The intervals themselves come from NFPA 110, and the edition matters. California adopts NFPA 110-2025 through Chapter 80 of the fire code, not the 2019 or 2022 editions that most contractor articles are still working from, and not whatever the model international code references.

So when you read that "the California Fire Code requires a monthly 30 minute test," the interval is real and the citation is wrong. That matters for one practical reason: a correction notice cites a chain, and the chain here is Title 24 Part 9, then Section 1203, then NFPA 110-2025. If you are assembling a compliance file for a portfolio, cite what your inspector cites.

It matters for a second reason too. The approved schedule in 1203.4.2 is a document specific to your building, established at approval. It is not a generic interval you can pull off the internet, and most buildings we walk into cannot produce it.

One carve-out before going further: health care occupancies, Group I-2 and ambulatory care facilities, route to NFPA 99 instead under sections 1203.4.1 and 1203.5.1. That is a different standard with different requirements, and this guide does not cover it.

Two things set your schedule. The first is the system Level. NFPA 110 defines Level 1 as a system whose failure could result in loss of human life or serious injury, which is where egress lighting, fire pumps, and fire alarm systems put you, and Level 2 as less critical (4.4.1 and 4.4.2). Some of the intervals below split on that. The second is fuel: diesel and spark-ignited units are not held to the same loading rule, because spark-ignited engines do not wet stack.

Everything above is one rulebook. In California there is a second one, and it is the reason a schedule that looks reasonable on paper can be unaffordable in practice. A stationary diesel engine here is an air quality permit item, and the permit limits how many hours a year you may run it for maintenance and testing.

The state layer is CARB's airborne toxic control measure for stationary compression-ignition engines (17 CCR 93115.6, engines above 50 brake horsepower). A new emergency standby diesel gets 50 hours a year for maintenance and testing, and a district may allow up to 100 where particulate emissions are 0.01 g/bhp-hr or lower. Older in-use engines are much tighter: 20 hours a year where particulate emissions exceed 0.40, or 30 hours at 0.40 or below, with a site-specific ladder up to 40, 50, or 100 by emission rate. An older set in an existing Santa Clara County building may be working with a 20 hour annual budget.

The local layer is the Bay Area Air Quality Management District. Regulation 9-8-330 caps reliability-related activities at 50 hours a year "or limitations contained in a District permit, whichever is lower", with essential public services allowed 100 under 9-8-331. That phrase is the operative one: the binding number is not the regulation, it is the condition written into your building's own Permit to Operate. Local permits we have seen run at 50 hours and at 24 hours, the latter with only one engine permitted to run at a time. Read the permit. Do not assume the general figure applies to you.

Two more things about that budget. Both the monthly exercise and any load bank test draw from it, because the rule defines maintenance and testing as activity that evaluates the engine's ability to perform in an emergency, and both qualify. And emergency operation is unlimited under every subsection, with a Public Safety Power Shutoff treated as an emergency, so hours run because the utility de-energized your area do not count against you. Required compliance source testing is accounted for separately as well.

Run the arithmetic on your own building. Twelve monthly exercises at 30 minutes is 6 hours. Add 1.5 hours if the conditional annual test applies. Add the triennial Class-duration run in the year it lands. Then add the part nobody budgets: every troubleshooting run and every retest after a failure. Against 50 hours that is comfortable. Against 20 or 24 it is not, and a set that fails a monthly test and gets run four more times while somebody chases the fault has just spent a meaningful share of the year's allowance.

The causes are boring, which is genuinely good news, because boring causes are cheap to prevent and expensive only when ignored.

Our companion guide covers the testing regime for exit signs and emergency lights: a monthly activation test of not less than 30 seconds, and an annual power test of not less than 90 minutes. That annual test is written specifically for battery-powered emergency lighting equipment operating on battery power (California Fire Code 1032.10.2).

Where your egress illumination is fed from the generator instead, the code sends you here. Section 1104.5.1 states: "Emergency power for means of egress illumination shall be provided in accordance with Section 1203." Section 1203 is the NFPA 110 regime described above. Which means the proof that your egress path stays lit for its required duration comes from a load test of the generator, not from a test button on a fixture.

That is worth stating plainly because the test button is what everyone reaches for. Pressing it on a generator-backed exit sign proves the sign lights when the circuit is interrupted. It proves nothing about whether the engine will start, transfer, and carry that load for the full duration. Most buildings have both systems, with battery units in some areas and generator-backed circuits in others, and each regime applies to its own equipment. Sorting out which fixtures sit where is a half-day exercise that is much better done before an inspection than during one.

One detail from the same chapter that we have watched cause real trouble: section 1008.3 places the generator room itself on emergency illumination. The failure mode is exactly what it sounds like. The power goes out, the generator does not start, and nobody can see the annunciator panel well enough to read why.

If you have not worked through the battery-powered side of this, our guide on exit and emergency light testing covers it, including the self-testing fixtures that do not remove the monthly obligation.

There is a provision in the fire code that saves buildings real money, and almost nobody uses it correctly. The exception to section 1203.5 allows recorded run time from standby or peak shaving operation to substitute for scheduled generator set testing, "provided that appropriate records are maintained." NFPA 110 takes the same position in section 8.4.1.3.

So the six hours your generator carried the building through a genuine outage can stand in for that month's exercise. Or it can be worth nothing at all, depending entirely on whether somebody wrote it down. Unlogged run time did not happen, and that is not a technicality: undocumented testing and no testing produce identical evidence.

On retention, NFPA 110 describes what records are kept but leaves how long to the facility and the authority having jurisdiction (8.5.4), so the binding numbers come from the other two rulebooks. California Fire Code section 110.3 requires records of periodic inspections, tests, servicing, and maintenance to be kept on the premises or another approved location for not less than 3 years and made available to the fire code official. Your air district permit separately requires a non-resettable totalizing hour meter and monthly logs held 36 months. Three years covers both.

Here is the self-assessment. If you can put your hands on all six of these today, your building is in better shape than most. If you cannot, the gaps are the work.