How to Tell a Water Quality Failure From a Mechanical One

Why this matters

A shop that misreads this question pays for it twice. Call a water problem mechanical and you replace the same part on a schedule until the customer stops calling you. Call a mechanical problem chemical and you sell a treatment program that cannot fix a throttled valve, and now the customer is paying monthly for something that will not work and you have staked your credibility on it.

The reason it gets misread is that the evidence everybody reaches for is the failed part, and the failed part is the one piece of the system that cannot answer the question. A pitted tube looks like a pitted tube whether the water is aggressive everywhere or the velocity is wrong in that one spot. The answer is not in the casualty. It is in everything that did not fail.

Before you open anything

The parts of this method that involve opening a system carry hazards that the method itself creates.

Isolate and depressurize the section, verify at a gauge on that section, and lock the isolation. A closed loop or a vessel holding pressure is stored energy under 29 CFR 1910.147, and a hot system is a scald source that reaches past the opening as the water flashes. Take a temperature reading before you break a joint, not after.

If any part of the survey involves opening a leg that has been standing still, treat the water in it as a biological hazard rather than as old water. Warm stagnant water grows organisms that become a health problem when they become airborne, so drain it low, slowly, and without spray, keep occupants out of the area during the work, and where aerosol generation cannot be avoided the respiratory control belongs under a 29 CFR 1910.134 program rather than a dust mask. The mechanism is covered in the companion card on stagnant water.

And if the survey requires the system down in an occupied building, the outage is a scheduled event with a notification, because loss of heat, cooling or process water lands on people who are not part of the job.

The gate

One rule, two conditions, and both have to be true:

Is the same process visible on at least two component types that share the water but not the duty, AND does its stage scale with each component's exposure?

Both true means a water quality failure. Either one false means a local condition, and the fault is mechanical or hydraulic even when the mechanism at the metal is corrosion.

The Boolean is an AND and it matters. Exposure-proportional failure on its own is also exactly what normal wear looks like: the pump that ran the most hours wears out first in a perfectly healthy system. That is why the first condition exists. A process appearing on a strainer basket, a gauge port and a valve seat, none of which have a wearing duty, cannot be explained by run hours, and its presence is what separates "the water is doing this" from "this part is old."

The unit of analysis is the component type, not the individual part. Two pump seals are one type. A pump seal and a strainer basket are two.

Step 1: define the population before you look at the failure

Write the list of like components on the same water before you form an opinion. All three circulators, not the failed one. Every strainer. Every heat exchanger on the loop. Every gauge port, every valve you can open, every visible internal surface.

This step exists because of order effects. Once you have examined the casualty, everything you find afterward gets read as confirmation of whatever the casualty suggested. Building the list first is how you stop that, and it takes minutes.

Skip it and you get the classic outcome: a correct description of one failure and no idea whether it is the system's story or that part's story.

Step 2: rank the population by exposure, not by position

Exposure is what the component actually saw, and it is usually run hours or throughput rather than calendar age. A standby pump installed the same day as the lead pump has a fraction of its exposure. A branch that serves a zone shut for a season did not see the year the rest of the loop saw.

Get the numbers rather than estimating them: run hour meters, control logs, occupancy schedules, valve positions. If the exposure ranking is guessed, the second half of the gate cannot be evaluated, because scaling is a comparison between numbers.

Skip this and the gate degrades into "some things look worse than others," which is true in every system ever built.

Step 3: read the survivors

Now open the population and read the components that have not failed. You are looking for the same process at an earlier stage: the same deposit, the same attack morphology, the same discoloration, less far along.

Photograph each one before cleaning anything. Deposit is evidence and it is destroyed by the first wipe. The morphology question, whether what you are seeing is erosion, uniform corrosion, pitting or crevice attack, is its own subject and the library has cards on reading it; here you only need to establish sameness and stage.

Skipping the survivors is the single most common failure of this whole method, and it is skipped for a good reason: the survivors are working and opening them costs time on a job that already has a failed part to replace. That time is the entire diagnostic value of the visit.

Step 4: put the two readings side by side

Now the gate is answerable. Two columns: exposure, and stage of the process. Read whether stage tracks exposure across the population, and read whether the process shows up on component types with no wearing duty.

Do this as a written comparison rather than an impression. The failure mode of doing it in your head is that a single dramatic casualty outweighs three quiet survivors.

Step 5: check the one confound that breaks the gate

The gate assumes the population shares the water. Check that it does before trusting the result.

Three things break it. Mixed materials: a copper component and a galvanized one in the same loop respond to the same water differently, so a difference between them is not evidence of a local condition. Compare like material to like material. A change in the recent past: a water source change, a treatment change, a system addition, a repipe. If the water became different partway through the exposure period, exposure hours no longer proportional to chemical exposure, and you need the date of the change. A leg that is not really on the loop: a dead leg, a bypass, a section isolated for a year.

Where any of these apply, the gate still works but the population has to be re-cut so that everything in it shares both the water and the timeline.

Case one: the gate says water quality

Three identical circulators on one loop, lead, lag and standby. Run hours read 8,200, 4,100 and 900. The 8,200-hour unit failed at a seal, was rebuilt, and failed again. Then the 4,100-hour unit failed. Two failures in about 14 months on a loop that had run for years without one.

Population defined: three circulators, two strainers, one plate exchanger, four accessible valve seats.

Exposure ranking: 8,200, 4,100, 900. Roughly a 9 to 4.5 to 1 spread.

Survivors read: the standby pump's seal faces show light scoring but are intact. Both strainer baskets carry a fine gritty deposit, more in the one on the higher-flow branch. The exchanger's inlet passages carry the same gritty material. Two valve seats show fine scoring.

Gate, first condition: the same abrasive process is on strainer baskets and valve seats, neither of which has a wearing duty. True. Second condition: seal condition tracks run hours across the three pumps, and deposit quantity tracks branch flow. True.

Both true, so this is water quality: suspended abrasive solids in circulation. The confirming evidence is not the pumps at all, it is the strainer, which had been cleaned twice in the same 14 months and dismissed as routine both times. The treatment answer is filtration and finding where the solids are coming from, and replacing the seals without it buys the interval the third pump's hours predict.

The failure mode if the gate is skipped: the shop rebuilds seals, and because the standby pump now becomes the lead, the next failure arrives on the schedule its new run hours set, which reads as a bad rebuild rather than as confirmation.

Case two: the same gate says local

A plate exchanger on a chilled water branch shows pitting concentrated at the inlet end. Same building, same water treatment program, same water.

Population defined: three exchangers on the same loop, two strainers, the branch balancing valves.

Survivors read: the other two exchangers are clean, including one with higher run hours. The strainers carry no unusual deposit. No valve seat shows the same attack.

Gate, first condition: is the process on a second component type that shares the water but not the duty? No. It is on exactly one component, in one place on that component. First condition false, so the gate closes here and the answer is local regardless of what the metal looks like.

The trace then goes to hydraulics rather than chemistry. Branch flow measures 62 gpm against a design of 40, which is about 55 percent over design, because two other branches were valved off during a partial-occupancy period and nobody rebalanced. Attack rate at a surface rises steeply and nonlinearly with velocity, and the exponent depends on the geometry and the material, so the honest statement is not that attack went up 55 percent; it is that a 55 percent velocity overshoot is well into the range where a material's velocity limit is the governing number, and that limit comes from the exchanger manufacturer rather than from a general rule.

So the fix is rebalancing and a check of the manufacturer's velocity limit, and a treatment change here would have done nothing. Note the shape of this one: the mechanism at the metal was corrosion, and the fault was mechanical. That distinction is why the gate asks about population rather than about morphology.

How to verify you called it right

The verification is a prediction, and you make it before the repair rather than after.

If you called water quality, predict where the process should appear next and at what stage: the standby pump at its current hours, the next strainer cleaning interval, the second exchanger. Write the prediction on the work order with a date. If the treatment change is real, the predicted progression slows; if you were wrong, the prediction still comes true on schedule and you find out cheaply.

If you called local, predict the opposite: after correcting the condition, the affected component's process should stop advancing while everything else stays as it was. Mark the component, and re-read it at a defined interval rather than waiting for it to fail again.

A diagnosis with no prediction attached cannot be checked, and the next tech inherits an opinion instead of a finding.

References

• 29 CFR 1910.147, control of hazardous energy, for isolating and depressurizing before opening a system
• 29 CFR 1910.134, respiratory protection program requirements, where a task generates aerosol from stagnant water
• Equipment manufacturer documentation for design flow and material velocity limits
• See related: How to Tell Erosion From Corrosion; Pitting, Crevice and Uniform Corrosion; Why Warm Stagnant Water Is a Biological Problem