Borderline Component: Replace Now vs Watch (Condition-Based) Decision Tree

Why this matters

A borderline component is one whose condition straddles the published acceptance line. The reading is at the limit, the visual is "almost," the test result sits on the boundary. These are the parts that drive callback rates. Decide too aggressively and you replace serviceable parts. Decide too conservatively and the part fails between visits. This decision tree is built on observable condition, failure mode, and consequence, with no price input.

The framing matches the condition-based logic in ISO 17359 and the failure-modes-and-effects approach in IEC 60812: judge the part on what it shows you and what its failure does, not on calendar age.

Symptom presentation

You have a measurement or observation that lands inside the borderline band:

• A capacitor reading microfarads at the lower end of the rated tolerance
• A motor winding insulation resistance reading at the published minimum
• A pump impeller with edge erosion that is visible but not deep
• A heat exchanger with a faint surface stain that may or may not be a developing breach
• A flexible coupling with rubber elements showing the first surface checking
• A check valve with a soft seat that closes but with a trickle of reverse flow
• A control board with one bulged capacitor cap among many flat ones

The part is not failed. It is not pristine. You have a judgment call.

Quick checks

1. Re-measure. Borderline readings are often instrument or technique artifacts. Use a known-good meter, a fresh probe, the correct test mode.
2. Compare to baseline. A capacitor at minimum is fine if it has been at minimum for five PMs. A capacitor that dropped from mid-range to minimum since last visit is failing.
3. Look for confirming indicators. One borderline reading is a measurement. Two independent indicators on the same part is a verdict.
4. Map the failure mode. Is failure gradual and announced, or sudden and silent?
5. Map the consequence. What does this part take down when it fails?

Decision tree

Branch A: Replace now

Take this branch if any of these are true:

• The borderline reading is moving in the wrong direction across the trend (you have prior data showing degradation)
• Two or more independent indicators are at the boundary on the same component
• The failure mode is sudden-death (insulation breakdown, seat failure under pressure, sudden bearing seizure)
• The component is in a no-fail-tolerated position: safety, life-safety, freezer/cooler load, occupied-space combustion, water-source heat reject
• Re-access cost is high and you have the unit open
• The component is part of a matched set (e.g., one of three contactors, one of two pumps) and the others are also in the borderline band; cascading failures cost more than batch replacement

Branch B: Watch (and define how)

Take this branch only if all are true:

• The borderline reading is stable across at least one prior trend point
• The failure mode is gradual with a clear next indicator
• The consequence of in-service failure is contained
• You can articulate the threshold that would move it to Branch A, and you record that threshold in the customer file
• You can articulate the visit cadence that will catch the next degradation step

If you cannot name the next indicator and the threshold, you are not watching, you are deferring. Move to Branch A.

Branch C: Replace at next planned visit

Take this branch if:

• The reading is borderline and trending wrong, but consequence is contained and you do not have the part on the truck
• The customer has a planned shutdown, season change, or inspection coming where replacement is cheaper to schedule than to react to

Order the part, calendar the visit, and leave a documented plan.

Confirming the diagnosis

A single borderline measurement is not a diagnosis. To call a part borderline-failing you need either a trend (this visit vs prior) or two independent symptom channels (electrical and visual, or thermal and acoustic, or pressure and flow). One channel alone is more often a measurement quirk than a real condition.

When you have trend data, the rate of change matters more than the absolute reading. A capacitor that loses 1 percent of capacitance per year is fine through its rated life. A capacitor that loses 8 percent between two visits is in collapse.

Next steps

Write the call into the customer record with the data that supports it: the reading, the trend, the failure mode, the consequence. The next tech inherits your judgment, not your conclusion. If you said watch, write the next threshold. If you said replace, write what you saw so future planning calibrates against real-world wear life on this site's duty cycle.

If the customer pushes back, lead with the failure mode and the consequence. The decision is theirs once they understand what they are signing up for.

References

• ISO 17359 Condition monitoring and diagnostics of machines, general guidelines
• ISO 13379-1 Condition monitoring and diagnostics of machines, data interpretation
• IEC 60812 Failure modes and effects analysis (FMEA and FMECA)
• NFPA 70B Recommended Practice for Electrical Equipment Maintenance
• PHCC National Standard Plumbing Code, component condition guidance for water-system components