How to Decide Between a Repair and a Material Change
Why this matters
A large share of what shops call component failure is not a defective part at all. It is a material sitting in a condition it was never suited to: a metal beside the wrong metal, an elastomer against a fluid that attacks it, a fastener in an environment that eats it. Replace in kind and you have bought a countdown, not a repair. The trouble is that each individual visit is small and defensible, so a shop can repair the same joint four times over three years without anyone ever standing back and asking whether the material is the defect. This is the rule that makes you ask, and the discipline that keeps the answer honest.
Before you open the joint
A joint that has been holding is a joint with energy in it, and the energy is what hurts people, not the diagnosis.
- Pressurized or liquid-filled path: isolate at the nearest upstream and downstream valve, open a vent or drain at the low point, and confirm at a gauge that the pressure reads zero before a fastener moves. Isolation of mechanical and stored energy, including a locked isolating device and a tag, is the duty in 29 CFR 1910.147; a valve you merely closed is not an isolating device until it is locked and you have proved the line is dead at a gauge or a cracked fitting.
- Anything electrical: open the disconnect, lock and tag it, and prove dead with the live-dead-live sequence, meaning you verify your tester on a known live source, test the conductors, then re-verify the tester. Electrical work is carved out of 1910.147 and lives at 29 CFR 1910.333(b)(2); the proving sequence itself is NFPA 70E-2021, 120.5.
- Rooftop or elevated work: get the fall protection in place before the tool bag, not after you decide the joint needs opening.
- Thermal: a line that carried anything hot gets touched with the back of a gloved hand or an infrared thermometer before a wrench, not with a bare palm.
The gate, stated exactly
Change the material when both of these are true at one specific location:
- Two or more failures at the same location with the same failure signature. Unit of analysis is the individual joint, fastener position, or seal seat, not the equipment and not the model. Two leaks on the same rooftop unit at two different joints are two data points about the unit, and zero data points about either joint.
- AND the observed interval between failures is under half the expected service life of that material in that duty. Use your own records for the expected figure if you have them. If you do not, use the equipment's expected service life as a conservative ceiling.
The Boolean is AND, and it does real work. Two failures at the same spot with different signatures is two separate faults that happened to share an address. One failure at half-life is a defect in that individual part until a second one says otherwise.
When the gate is met, change exactly one material variable per attempt and set a review at the interval that failed. One variable means the alloy, or the elastomer class, or the coating, or the isolation, not three of them on the same visit. Changing three and having it hold teaches you nothing, and you will carry all three forever because nobody knows which one mattered.
Fixing the unit of analysis
Before anything else, write down which location you are talking about, precisely enough that another tech would put a wrench on the same fitting. "Condensate drain" is not a location. "The nipple threaded into the pan boss on the return side" is.
This sounds fussy and it is the step that decides whether the gate can ever fire. Shops that log at equipment level accumulate a history that reads as a run of unrelated small repairs. The same history logged at joint level reads as one joint failing over and over, which is a completely different conclusion from the same facts.
Naming the signature
A signature is what the failure looked like, not what it did. "It leaked" is a symptom shared by every failure mode there is. Write the signature down in the same words every time:
| Signature | What it usually names |
|---|---|
| White or grey powder at the interface of two different metals | Galvanic attack, the anodic metal sacrificing itself |
| Uniform thinning with a clean surface | General corrosion, the environment is simply aggressive |
| Small deep pits in an otherwise clean surface | Chloride or chemical attack, thinning fast at a point |
| Swollen, soft, or spongy elastomer, thicker than the groove | Fluid attack, the elastomer is absorbing what it touches |
| Shrunk, hard, cracked elastomer | Heat aging or plasticizer extraction |
| Cracks radiating from a stress point in a plastic | Environmental stress cracking, a chemical plus a load |
| Bright wear track with the mating part polished | Motion in a joint that was supposed to be static |
Photograph it before you clean it. The pattern is the evidence, and once it is wiped you cannot re-read it. See the related article on reading corrosion patterns for how to trace each of these back to a source.
Case A: two failures, one signature
A packaged rooftop unit, cast aluminum condensate pan, a galvanized steel nipple threaded into the pan boss.
- Month 0: original install.
- Month 8: weeping leak at the nipple. White powder in the threads. Cleaned, re-taped, re-made. Visit took 2.5 hours including roof access.
- Month 14: same joint, same weeping, same white powder. Another 2.5 hours.
Run the gate. Same location: yes, the same nipple in the same boss. Same signature: yes, white powder at an interface between two different metals both times, which names galvanic attack, not a bad seal. First Boolean met.
Second Boolean: the interval between the two failures is 6 months. Packaged rooftop equipment is commonly planned around a 15-year service life, so use 120 months as a deliberately conservative expected life for a threaded connection that should outlive the box it is in. Half of that is 60 months. Six months is far under 60. Second Boolean met.
Gate met, so change one material variable. The variable here is the galvanic couple itself: steel threaded directly into aluminum, wet continuously, is a metal beside the wrong metal. The change is to break the couple, either with a fitting whose material sits close to aluminum in the galvanic series or with a dielectric isolating fitting, plus a sealant that does not wick moisture into the threads. Not both plus a torque change plus a different pan. One variable.
The hours, honestly. Two visits at 2.5 hours each is 5.0 hours already spent. The material change is a 4.0-hour job once, because sourcing the isolating fitting and cleaning the boss threads takes longer than a re-tape. That is 1.5 hours more than simply doing the next repair, so the change does not pay for itself against the next failure. It pays back against the one after that. On the observed 6-month interval, the third failure would land around month 20 and the fourth around month 26, so a change made at month 14 is even by roughly month 26. All four figures are the shop's own labor hours, so they compare directly.
What would flip this. If the unit is already scheduled for replacement inside that payback window, repair in kind and note the couple on the replacement spec so the new unit does not inherit it. The gate says the material is wrong. It does not say the fix is worth doing on equipment that is leaving.
Case B: two failures, two signatures
Same building, same quarter, a circulating pump on the hydronic side. Mechanical seal replaced twice in 20 months.
The address is identical: the same seal seat on the same pump. Under a rule that only counted failures per location, this fires. It should not.
- First failure, month 0 to month 9: the seal faces came out scored, with grit embedded in the softer face.
- Second failure, month 9 to month 20: the faces came out heat-checked, with a dry-run pattern and discoloration.
Two signatures, not one. Grit embedded in a face names a filtration problem. Heat checking with a dry-run pattern names loss of flush or flow to the seal. Those are different faults that both ended at the same part, which is what a seal does, because it is the softest thing in the assembly and it records everything upstream of it.
Gate not met, because the first Boolean fails on signature even though location and interval both pass. The 11-month gap is well under half of any reasonable expected life for a mechanical seal, so an interval-only rule would have fired here too. That is exactly why the signature test is ANDed in front of it.
Changing the seal face material here would have been the expensive wrong answer: a harder face pair costs more, takes longer to source, and would have run dry just the same. The real findings were a strainer that went back in without its screen after a service visit, and a balancing valve throttled below the pump's minimum continuous flow. Both fixed, neither of them a material.
What has to be in the record for the gate to work
The gate reads history, so it is only as good as what the last tech wrote down. Three fields, on every repair of a joint, seal, or fastener:
- Location, written specifically enough to be unambiguous to someone who has never seen the unit.
- Signature, in the vocabulary from the table above, so it can be matched rather than interpreted.
- What was put back, including the material and any sealant or lubricant used, not just the part description. "Nipple" does not tell the next tech whether the couple changed. "Galvanized steel nipple, PTFE tape" does.
Without the signature field, every repeat failure looks like the same failure, and the gate fires on Case B. Without the material field, you cannot tell whether someone already tried the change you are about to try. Both of those are common, and both turn the fourth visit into the first visit again.
Verifying the change actually took
A material change is a hypothesis, and it is confirmed by an interval, not by the joint being dry when you leave. Dry when you leave is what every in-kind repair also looked like.
Set an explicit review at the interval that failed, in this case 6 months, and put it on the schedule rather than in your head. At that review, look for the signature specifically: is there any powder at the new interface, any staining below it, any weeping under a paper towel pressed to the joint. A clean joint at one full failure interval is real evidence. A clean joint at three weeks is no evidence at all, because the old one was clean at three weeks too.
If the signature reappears at roughly the same interval, the variable you changed was not the variable that mattered. Change the next one, single, and keep the review at the same interval. If it reappears much sooner, look at what you did on the visit rather than at the material.
References
- 29 CFR 1910.147, control of hazardous energy, for isolation and stored-energy release on mechanical and pressurized paths
- 29 CFR 1910.333(b)(2), for de-energizing electrical equipment before work, with NFPA 70E-2021, 120.5 for the live-dead-live proving sequence
- Manufacturer documentation for expected service life and for approved materials at a specific joint
- See related: Reading Rust and Corrosion Patterns; The Galvanic Series as a Field Tool; How Seals and Gaskets Fail