What Inrush Does to Everything Else on the Supply

Why this matters

The load that starts is rarely the load that gets reported. A compressor kicks on and somebody two rooms away loses a controller, a backup supply clicks to battery, or the lights dip. Nobody connects those events because the machine that caused them is behaving perfectly and never complains. Meanwhile the reported device gets replaced, the replacement does the same thing, and the shop has now sold a part that was never faulty. The trick is that inrush is a current event at the starting load and a voltage event everywhere else, and where it lands is set by which conductors the two loads share.

How a start reaches a load it is not connected to

Current through an impedance produces a voltage across it. A starting load draws a large current through everything between it and the source, so every other load fed through any part of that same path sees its supply pulled down for the duration of the start. Loads fed through none of that path see nothing at all.

             transformer
                  |
             shared feeder both currents flow here
                  |
              panel bus
             /          \
        compressor office
         branch branch
        (starting)    (victims)

The office branch in that sketch carries none of the compressor's current, so the office loads see the dip that develops across the transformer and the shared feeder, and nothing more. Move the compressor onto its own feeder from the transformer and the shared portion shrinks to the transformer alone, which is the entire basis of the usual fix.

The three relationships you need, with the conditions they hold under

Contribution of the source transformer. The dip a transformer contributes at its own secondary is approximately its nameplate percent impedance multiplied by the ratio of the starting current to the transformer's rated current. That holds where the source impedance is dominated by transformer reactance and the starting current is heavily lagging, which is the normal case for a motor start on a modest transformer. It does not hold where the utility side is weak relative to the transformer, in which case the utility's contribution has to come from the utility.

Contribution of a conductor run. For a fixed conductor, the drop scales with the current through it. But a small conductor's impedance is largely resistive while a motor start's current is heavily lagging, so a resistive run develops less drop from a lagging current than a straight multiplication suggests. Treat the multiplied figure as an upper bound, not a prediction.

They add along the path. The dip a victim sees is the sum of the contributions of every element between it and the source that also carries the starting current. That is why the same start produces a different dip at every point in a building, deepest at the point electrically closest to the starting load.

One consequence worth stating plainly: a small aggressor sharing a long branch circuit with its victim can produce a deeper local dip than a large aggressor sharing only a transformer, because the impedance per ampere of a branch conductor is far higher than that of the transformer everything hangs off.

Before you make anything start

Repeated test starts are the natural instinct and they carry two hazards your own instruction creates. A motor has a permitted number of starts per hour and a permitted accelerating time, both of which come from the motor's data rather than from judgment, and exceeding them heats the rotor with no load-driven cooling to remove it, so plan the number of test starts against that figure before you begin. And on a compressor or any stored-pressure machine, do not force starts by defeating a pressure switch, an unloader or a safety control; make it start by allowing the process to call for it, or leave a logger in place and let the day produce the starts for you.

Where you must read voltage during a start, that is energized work, permitted under 29 CFR 1910.333(a)(1) only where the employer can demonstrate that de-energizing introduces additional or increased hazards or is infeasible due to equipment design or operational limitations, which a transient measurement genuinely can. Work to the boundaries and PPE from the risk assessments at NFPA 70E-2021, 130.5 and 130.7, in the edition your employer's electrical safety program adopts, and prove the instrument live-dead-live per 120.5 before you believe it. Note that the general lockout standard at 29 CFR 1910.147 excludes exposure to electrical hazards from work on conductors and equipment in electric utilization installations at (a)(1)(ii)(C); electrical isolation follows 29 CFR 1910.333(b)(2), with 29 CFR 1926.417 as the construction counterpart.

The call

A shop reports two symptoms in one sentence: the lights dip in the work bay, and the small backup supplies under the front-office desks click over to battery several times a day. The customer has concluded the utility is at fault and wants that written down.

The first useful move is to stop treating those as one symptom. Perceptible light flicker and a backup supply transfer sit an order of magnitude apart in depth, so a single cause explaining both is possible but not assumed.

The second useful move is to find a recorder that already exists. The backup supplies log their transfers with timestamps. That is a voltage event recorder nobody had to install, and it is the strongest evidence on the site.

What the logs said

Two days of logs gave eleven transfers. A current logger left on the compressor feeder gave its start times over the same two days.

Eight of the eleven transfers fell within the compressor's start window. The remaining three clustered within a few minutes of each other on the second afternoon and matched no compressor start at all, so they belong to something else and get chased separately rather than folded into the finding. Reporting eight of eleven as if it were eleven of eleven is how a correct diagnosis gets discredited by its own overreach.

Whether the numbers support the story

The values here come from the equipment: the transformer's nameplate impedance, the motor's nameplate and data for locked-rotor current, and a measured feeder drop.

• Transformer nameplate impedance: 4 percent.
• Compressor motor full-load current: 20 percent of the transformer's rated current.
• Motor data gives locked-rotor at 6 times full-load current, so the start draws 120 percent of the transformer's rated current.
• Transformer contribution at start: 4 percent times 1.20, which is 4.8 percent.
• Measured drop from transformer secondary to the panel bus at the panel's normal running current: 2 percent. The start current is 3 times that normal running current, so the shared feeder's contribution is up to 6 percent, treated as an upper bound because the start current is heavily lagging against a mostly resistive run.
• Worst-case total at the panel bus: 4.8 plus 6.0, which is 10.8 percent, leaving 89.2 percent of nominal retained.

The backup supplies are set to transfer at 10 percent below nominal, which is 90 percent retained. The worst-case retained voltage of 89.2 percent is below that threshold; the best case, if the feeder contributed nothing at all, would be 95.2 percent retained and no transfer.

That straddle is the finding, not a weakness in it. The predicted dip crosses the transfer threshold only near its worst case, and the worst case is not every start: starting current magnitude varies with the point on the voltage wave at which the contacts close and with how loaded the machine is when it starts. A cause that only sometimes crosses a threshold produces symptoms that only sometimes appear, which is exactly what the logs show, and it is why the previous tech's two visits found nothing wrong.

The flicker in the work bay needs no such straddle. Perceptible light flicker responds to dips far shallower than 10 percent when they repeat, so the 4.8 percent transformer contribution alone accounts for it. Two symptoms, one aggressor, two different sensitivities, and the customer was right that they were related and wrong about what the relation was.

Where this reasoning would have failed. If the office receptacles had been fed from a separate feeder off the transformer rather than from the same panel bus, the shared impedance would be the transformer alone, the predicted worst case would be 4.8 percent lost and 95.2 percent retained, comfortably above the 90 percent threshold, and the transfers would need a different explanation entirely. The topology is not a detail of the write-up. It is the input that decides the answer.

Choosing which end to fix

Three fixes exist and the right one depends on how many victims there are, not on which is easiest to sell.

Reduce the aggressor's starting current. This helps every victim at once and it is the correct move when a start is disturbing a whole building. It is also the fix with real consequences for the driven machine, because reducing starting current reduces starting torque, and that trade is its own subject; the sibling reference on what a starting method changes about the supply owns it.

Reduce the shared impedance. A larger or separate feeder from the source removes the shared element rather than reducing the current through it. This is the durable answer where the aggressor cannot be changed and there are several victims, and it is the only one of the three that also improves running voltage at the panel.

Harden the victim. Raising a transfer threshold's tolerance, moving one sensitive load to a supply with less shared path, or specifying equipment with real ride-through is right when there is one victim and the dip is within everything else's tolerance. It is wrong as a first move when four devices are complaining, because you will buy four fixes for one cause.

The fix that is not a fix: increasing the size of the overcurrent device protecting the aggressor. Nothing here was tripping. Enlarging a breaker to solve a voltage complaint changes the protection of the conductors and does not change the dip by one part in a hundred.

Building the evidence so it holds up

The record that makes this stick has four parts, and it is worth assembling in this order because each one narrows the next.

First, the victim's own timestamps, from whatever already logs them. Second, the aggressor's start times over the same window, from a current logger on its feeder rather than from anyone's memory. Third, the topology, drawn as far back as the source, because that is what says which elements are shared. Fourth, the nameplate figures the prediction rests on, transcribed from the nameplates rather than from a catalogue.

If the first two correlate and the last two predict a dip that reaches the victim's threshold, you have a case. If they correlate and the numbers say the dip cannot reach the threshold, you have a second cause somewhere in the shared path, most often a high-resistance connection that only shows itself at start current, and that is a different investigation with a different instrument.

References

• 29 CFR 1910.333(a)(1) energized-work gate and (b)(2) electrical lockout; 29 CFR 1910.147(a)(1)(ii)(C) exclusion of electric utilization installations; 29 CFR 1926.417 construction counterpart
• NFPA 70E-2021, 130.5 and 130.7 for boundaries and PPE and 120.5 for live-dead-live, in the edition adopted by your employer's electrical safety program
• Transformer nameplate percent impedance and motor nameplate and manufacturer data for locked-rotor current, permitted starts per hour and accelerating time
• See related: Why Inrush Current Trips Things That Should Hold; What a Starting Method Changes About the Supply; How to Decide Whether a Supply Can Start What You Are Adding