The 1600-Amp Main That Fails on a Sunday: The Temperature Delta That Showed Up Three Infrared Scans Earlier

You bought the thermal camera in 2019. You sent two guys to level 1 thermography training. And you know exactly where those infrared reports are: in a folder on the server, one per customer per year, opened once by the person who wrote them and never again. The scans found some hot connections, you quoted the repairs, the customer approved about a fifth of them, and the rest sat.

So the objection is fair: this trade has been doing predictive maintenance by hand for twenty years and it has mostly been a marketing line on a service agreement. What changed in 2026 is not the camera. It is that the readings stopped being an annual snapshot and started being a trend that something watches for you.

The failure that costs the most when nobody saw it coming

For a commercial electrical service contractor, the expensive unplanned failure is not a lighting contactor or a rooftop disconnect. It is the main — a 1200-amp or 2000-amp service, a main breaker, a main lug section, or the bus in a switchboard that has been sitting in a damp mechanical room since 1987. When that goes, the building goes dark, and the building being dark is somebody else's business stopping.

Two things make it brutal. The first is availability: main breakers and switchgear sections for older lineups are not on a shelf, and for obsolete frames you are hunting reconditioned equipment or engineering a retrofit. The second is that the emergency version of the job is the worst version of the job — after-hours crew, a temporary generator and cable, a utility coordination call, an emergency permit, an inspector who has to be pulled out on a Sunday, and a customer who is angry at you specifically, because you are the one with the maintenance agreement.

Condition-based maintenance means you service a piece of electrical gear because its own measurements say it is changing, not because twelve months have passed since the last visit. That is the whole shift, and it is the reason predictive maintenance is now the leading AI use case in manufacturing at roughly 64 percent adoption — the plants your customers run are already buying it, from somebody.

The signals that show up before the bang

Electrical failures are unusually polite. They warn you, in three or four ways at once, for months.

A bolted connection that is loosening or corroding runs hot, and the useful measurement is not the absolute temperature — it is the rise above the two identical phases beside it, tracked across scans. Ten degrees of delta on B phase in March that is eighteen degrees in September is a different animal from eighteen degrees seen once. Current imbalance across phases creeps. Harmonic distortion climbs as a customer adds VFDs and LED drivers to a service that was sized for neither. Breakers with electronic trip units log their own trip counts and long-time pickups. Outdoor gear takes on moisture, and moisture leaves a visible signature inside the enclosure that a phone photo captures perfectly well.

None of that is new. What is new is that the readings from the power monitoring the customer already has installed can be watched continuously, and the photos and scans from every visit compared against every previous visit automatically, rather than by a service manager who would need a free afternoon he does not have.

flowchart TD
  A["Meter and thermal readings stream from customer switchgear"] --> B{"Is any phase trending away from the others?"}
  B -->|No| A
  B -->|Yes| C["Pull the last three infrared scans and panel photos"]
  C --> D["Service manager reviews and sets urgency"]
  D --> E["Quote a planned shutdown, order breaker and parts early"]
  E --> F["Saturday outage: torque, clean, retest, rescan"]
  F --> A

NFPA 70B stopped being a suggestion, and that is your opening

The 2023 edition of NFPA 70B converted it from a recommended practice into a standard — written in "shall" language, describing an electrical maintenance program with condition-based intervals rather than one interval for everything. That matters commercially more than technically. It gives your salesperson a document to put on the table in front of a facilities director or a risk manager, and it moves the conversation from "you should really let us look at your gear" to "here is what the standard expects of your program, here is your equipment condition, here is the interval it implies."

It also changes what you are selling. A time-based agreement is a fixed visit at a fixed price and the customer feels the cost every year. A condition-based program is a monitoring service plus a right-sized visit schedule, and the customer feels it as risk reduction on equipment they cannot easily replace. Different budget line, different objection, usually a better one.

Tuesday, 7:20am, the service manager's screen

Instead of a queue of scheduled PMs, the first thing on the screen is four flagged accounts.

The top one: a distribution center's 1600-amp switchboard, B phase on the main lugs running 22 degrees above A and C, up from 9 degrees at the January scan, with the two intermediate photos lined up side by side and the trend drawn. The recommendation is a planned shutdown inside 60 days. Attached is the equipment nameplate from the last visit, so the parts person can start checking availability today rather than after the failure.

The second: a school district's main switchgear where nothing is hot, but the neutral current has been climbing since the summer LED retrofit — a harmonics conversation, and a quotable one, since the fix is a study and possibly a transformer, not a torque wrench.

The third and fourth are lower priority and get folded into visits already scheduled. The service manager spends twenty minutes on the screen and has two real quotes to write, both for planned work at planned-work margins, both scheduled on a Saturday when a shutdown is cheap.

What it does to the maintenance book: a worked illustration

Assumptions for a contractor with 48 commercial maintenance-agreement customers, currently doing one annual infrared scan each, illustrative only.

The close-rate assumption is doing the heavy lifting, and it is the one to argue with. The reasoning is straightforward from experience: a facilities director shown one warm photo says "watch it next year." Shown the same connection getting steadily hotter across three visits, with the equipment's replacement lead time next to it, he takes it to his capital request. That is a sales argument, not a technology argument, which is exactly why it works.

Not modelled: the emergency callouts you stop having. Those are revenue too — expensive, high-margin, and terrible for the relationship. Trading them for planned Saturday work is usually the right trade, and you should go in knowing it flattens your emergency line.

Where the sensors and the software should not be trusted alone

No monitoring catches everything. A loose lug can go from warm to failed in a week under a load swing. Rodent damage, water intrusion after a roof failure, and a subcontractor drilling through a feeder are all instantaneous, and no trend precedes them. Sell condition-based monitoring as a reduction in unplanned failures, never as elimination, and put that in the agreement in plain words.

Thermal readings still need a qualified person with an infrared window or an appropriate procedure, working under your NFPA 70E program with the right arc-rated clothing. Nothing about 2026 makes it safer to open a 480-volt switchboard door. If anything, the trend data should reduce how often you need it open, and that is a safety win worth saying out loud to your crews.

And keep the recommendation human. The software flags a trend; a journeyman or the service manager decides whether it is a torque issue, a load issue, or a failing breaker. Sending a customer an automated "your equipment is failing" alert generated with no human review is how you lose an account and possibly get invited to a lawyer's office.

Frequently asked questions

Do our customers have to install monitoring hardware for this to work?

Not to start. Most of the value in year one comes from what you already collect: infrared scans, panel photos, torque records and test results from each visit, compared visit-to-visit instead of read once. Continuous monitoring is a second-phase upsell, and many of your larger accounts already have metering on their switchgear that nobody is reading.

How do we get five years of old infrared reports into something usable?

You feed them in as they are. Modern tools read the PDF reports your thermography software produced, including the images and the temperature tables, and can line up the same piece of equipment across years. The work is not the reading, it is the naming — deciding that "MDP" in the 2021 report and "Main Distribution Panel A" in the 2024 report are the same board. Budget a couple of days of somebody's time for that and do it once.

Does this let us cut back on annual scans and reduce our own service revenue?

It changes the visit rather than removing it. Some low-risk equipment stretches to a longer interval; the switchgear and the mains earn more attention. Priced as a program rather than per visit, contractors generally come out ahead, because you are selling the judgement rather than the truck roll.

What is the smallest way to start?

Take your ten largest maintenance accounts and their last three years of scans and reports. Line them up per piece of equipment, look for anything trending, and call the two customers with the worst trends. You will learn more from those two phone calls about whether this sells than from any pilot.

One closing note on the phone side of this: a condition-based program generates inbound calls — customers asking about the shutdown date, purchasing wanting the quote resent, facilities asking whether Saturday still works. CallSphere builds AI voice and chat agents that answer business phone lines and web chat, book appointments and capture leads 24/7, so the scheduling traffic your maintenance book creates does not land entirely on the service coordinator who is already dispatching trucks.