How to Measure Flow When You Cannot Install a Meter

Why this matters

Most of the time you are not allowed to cut the line. The system is in service, the customer will not pay for a shutdown, the pipe is buried, or the only place a meter would fit is the one place nobody will let you weld. So the flow number has to come from outside the pressure boundary, and the usual reflex is to reach for a spot instrument and accept whatever it says.

There is a second route, and on a lot of calls it is the better one: stop trying to read a rate and go read a total. Watch a quantity you can already see change over a measured window, then divide. You are trading hardware for time, and the trade is good, because the errors that wreck a spot reading are mostly instantaneous and the errors that wreck a totalized reading are mostly bookkeeping, which is something you control. The limit on the method is not the instrument. It is whether the system did anything representative during the window you watched.

Before you start the clock

Nothing here requires breaking into a line, and that is the point. The hazards that remain come from where you have to stand and what you have to touch to watch the observable.

• Measuring level in a tank, sump or pit: measure from outside, using a sight glass, an external level indicator, a tape from the lid or the system's own gauge. A tank, vault or meter pit meeting the definition of a permit-required confined space is governed by 29 CFR 1910.146, and a flow question is never worth an entry.
• Watching a hot vessel or a hot line: the surface is the hazard before the water is. Where you need contact, use an insulated surface probe on a cooled spot or a non-contact reading, and keep hands off unlagged pipe above skin-burn temperature.
• Disturbing pipe insulation to reach a surface: thermal system insulation in a building constructed no later than 1980 is presumed asbestos-containing material under 29 CFR 1926.1101, which covers construction and maintenance work, and general industry has its own standard at 29 CFR 1910.1001. This route is inhalation, so the control is a respirator selected under a program meeting 29 CFR 1910.134, not gloves - and on most service calls the right answer is to find an already-bare section instead.
• Clamping a current probe to infer run time: if the conductors are reachable outside the enclosure, clamp them there and you have not done electrical work. If the reading requires opening an energized enclosure, 29 CFR 1910.333(a)(1) requires the parts be de-energized before you work on or near them unless de-energizing introduces additional or increased hazards or is infeasible because of equipment design or operational limits, the lockout and tagging procedure is 29 CFR 1910.333(b)(2), and the instrument gets proved live-dead-live per NFPA 70E-2021, 120.5 in the edition your employer's program adopts. Check the leads for cracked insulation before every use, and confirm the meter's measurement category, marked under the IEC 61010 series through its listing, covers the point you are landing on.
• Reading a gas meter in a pit or a vault: if you smell gas, nobody reads anything. Everyone leaves, no switches or lights are operated, no phone is used inside, and the call goes out from outside.

Step 1: Decide which question you owe

Three different questions get asked in the same words, and they need different windows. A rate right now needs a short window at a stable operating point. A total over a period needs a long window with nothing else drawing on the boundary. A comparison to a requirement needs whichever window reproduces the condition the requirement was written for.

Skipping this step is how a tech spends two hours totalizing an overnight window and answers a question nobody asked.

Step 2: Find an observable that flow moves in one direction

You need something that changes with flow, changes monotonically, and can be read without opening the system. The common ones:

• Level in a vessel of known geometry. A drop of known depth in a known cross-section is a known volume.
• Run time on a device of known displacement. A positive-displacement pump, a compressor, a metering pump: run time times displacement per unit time is volume.
• Cycle count on a fixed-volume event. Each pump-up of a pressure tank moves the tank's drawdown volume. Each fill of a known sump moves its working volume.
• A meter that already totalizes. The utility meter, the building's own submeter, the equipment's internal counter. It was installed for someone else's reason and it works for yours.
• A temperature difference across a load whose duty you know. A last resort here, because the error grows as the temperature difference shrinks, and the sibling card on flow methods carries that arithmetic.

Step 3: Pin the conversion constant from documentation

Every one of those observables converts to volume through a constant that belongs to a specific piece of hardware: tank cross-section, drawdown volume between cut-in and cut-out at a stated pressure setting, displacement per revolution, meter dial constant. Read it off the nameplate or the manufacturer's literature, and read the conditions it is stated at.

Skip this and you have imported someone's memory of a similar unit into your arithmetic, and it will not announce itself, because the number you produce will look exactly as clean as a right one.

Step 4: Close the boundary

Everything that can add to or draw from the measured volume during the window has to be shut, and you have to know it is shut rather than believe it. Isolate branches at valves you operated yourself. Ask what runs on a timer. Ask who is in the building.

This is the step that kills most attempts, and it fails silently: an ice maker, a landscape zone, a flush valve, a makeup line. The tell is a totalized answer that comes out high and does not repeat.

Step 5: Size the window from resolution, not patience

The rule: the observable must move at least 20 times the smallest increment you can read, per window. That holds the reading error under about 5 percent from resolution alone, and it applies per window rather than per shift, so a window that fails the test is extended rather than averaged with another.

If your stopwatch technique is good to half a second, a 48-second event passes with room to spare. If a meter dial reads to a tenth of a gallon, you need the dial to advance at least 2 gallons. If a sight glass reads to a quarter inch, you need 5 inches of level change. Work the arithmetic before you start rather than discovering afterward that the window was too short to say anything.

Step 6: Log both ends the same way

Same instrument, same point, same reader, wall time to the second at both ends, and a photograph of the meter face or sight glass at each. A totalized measurement is two readings and a duration, and a sloppy duration corrupts it exactly as fast as a sloppy volume.

Step 7: Convert, then bound it with the two terms that dominate

Two error terms outweigh everything else in this method, and both are yours rather than the instrument's:

• Resolution on the observable, which step 5 sized.
• Boundary leakage, which step 4 tried to close and which you should now measure rather than assume, by running a second window with everything deliberately off.

Report the answer with the band. A number without one invites a decision it cannot carry.

Step 8: Ask whether the window was representative

A total over a window describes that window. Generalizing it to a day, a season or a design condition is a separate claim, and it needs the duty cycle, the weather, the occupancy or the production rate that applied while you watched. Write the conditions down next to the number, because the person who reuses the number in six months will otherwise assume they were typical.

Worked example: a pressure tank and an overnight window

The complaint. A light commercial customer says water pressure fell off over the last month. No meter exists on the pump discharge and the owner will not authorize cutting one in.

Window one: how much the pump moves. The tank's documentation gives a drawdown of 12.0 gallons between cut-in and cut-out with the pressure switch at its 40/60 psi setting. Measured cut-in and cut-out at the tank gauge: 40 and 60 psi, so the documented figure applies. All fixtures isolated at the branch valves the tech closed himself, ice maker unplugged. Draw down through one hose bibb until the pump starts, then close it and time the refill.

Timed refill: 48 seconds, on a stopwatch good to about half a second at each end. Resolution check: 48 divided by 0.5 is 96, well past the 20 the rule asks for.

Convert: 12.0 gallons over 48 seconds is 0.25 gallons per second, or 15.0 gallons per minute.

Band: the two stopwatch errors are independent, so they combine as a root-sum-square rather than adding, giving about 0.7 seconds on 48, or 1.5 percent, roughly plus or minus 0.22 gpm from timing. Adding them instead gives 1.0 second, 2.1 percent and plus or minus 0.3 gpm, which is a hard worst case rather than a likely band; say which one you reported. The drawdown constant carries its own error, and the direction is knowable: if the tank's air precharge has fallen below the specification of 2 psi under cut-in, the real drawdown is less than 12.0 gallons and the computed flow is overstated. Measured precharge with the tank drained: 38 psi, which matches the specification, so that term stays small.

What the 15.0 gpm claim covers. It is the pump's delivery against this tank's 40 to 60 psi band, at the pump, with everything else closed. It is not the delivery at the far fixture, which sees additional pipe and fitting losses, and it is not the delivery at a higher system pressure, where a centrifugal pump moves less.

Window two: is anything leaking. Same boundary, opposite purpose. Building empty overnight, all fixtures closed, utility meter read at 9:40 pm and again at 5:40 am, a window of 8.0 hours. Meter dial reads to 0.1 gallon; observed advance 6.4 gallons. Resolution check: 6.4 divided by 0.1 is 64, past 20.

Convert: 6.4 gallons over 8.0 hours is 0.8 gallons per hour, or about 0.013 gpm. Against the pump's 15.0 gpm that is under 0.1 percent of capacity.

The call. Leakage is real but it is nowhere near large enough to explain a pressure complaint, so the loss is downstream of the pump in the distribution, or the complaint is about pressure rather than flow and belongs to the pressure switch setting and the piping, not the pump. The two windows together closed the pump question and eliminated a leak theory without a single line being opened.

The failure mode. Run window two with the ice maker still plugged in and the meter advances several gallons, the tech reports a significant leak, and the shop spends a day chasing a plumbing fault that is a refrigerator. Nothing about the arithmetic would have flagged it. Only the boundary discipline in step 4 catches it, and only the repeat in the verify list below catches a boundary you thought you closed and did not.

How to verify you got this right

• Run the window twice. Two windows that agree within your stated band are a measurement. Two that do not agree mean the boundary moved, and the difference between them is a lower bound on how much.
• Run a zero window. Everything off, watch the observable for the same duration. Anything that moves is boundary leakage, and you have now measured it instead of assuming it away.
• Cross-check with a physically different observable. If the tank window says 15.0 gpm, catch and time a full-flow discharge into a container of known volume and see whether the two land inside their combined bands. Two methods that share no assumptions beat the same method run twice.
• Re-read your resolution check. The most common quiet failure is a window that was too short, and it produces a confident number rather than an obviously bad one.

References

• 29 CFR 1910.146 (permit-required confined spaces); 29 CFR 1910.333(a)(1) and (b)(2) (electrical safe work practices)
• 29 CFR 1926.1101 (asbestos in construction and maintenance work) and 29 CFR 1910.1001 (general industry); 29 CFR 1910.134 (respiratory protection)
• NFPA 70E-2021, 120.5, as adopted through your employer's electrical safety program
• Equipment and tank manufacturer documentation for drawdown volume, displacement and meter dial constants
• See related: The Flow Measurement Methods and What Each One Assumes; How to Estimate Flow Without a Flow Meter; Measuring Flow: Generic Methods