The Tolerance Stack-Up That Causes the Fault

Why this matters

The hardest faults to find are the ones where every part tests good and the system still fails. You meter component after component, each one inside spec, and walk away stumped. The answer is often a tolerance stack-up: several parts each at the far edge of their acceptable range, all leaning the same direction, adding up to a system that falls out of spec even though no single part does. A tech who only tests parts one at a time never sees it, because the fault does not live in any one part. It lives in the sum. This article teaches you to find the fault that hides in the margins.

What a tolerance stack-up is

Every component is built and rated with a tolerance, an acceptable band around its target value. A part can sit anywhere in that band and still pass. The trouble starts when several parts in one system all sit near the same edge of their bands and the errors add instead of canceling.

Picture a chain where each link is allowed to be a little long or a little short. If they vary randomly, the lengths average out. But if every link happens to run long, the chain is too long overall, even though no single link is out of spec. That is a stack-up: in-tolerance parts summing to an out-of-tolerance system.

It shows up across every trade: supply voltage at the low edge feeding a motor with windings at the high-resistance edge driving a load at the heavy edge. Each acceptable. Together, a no-start.

How to recognize one

A stack-up has a signature. Suspect it when:

• Every component you test reads in spec, but the system underperforms or faults. This is the classic tell. No single bad part, yet a bad result.
• The fault is marginal and condition-sensitive. It works most of the time and fails only when conditions push the stack one more notch: a hot day, a low-supply moment, a peak load. The extra push tips an already-loaded margin over the edge.
• The fault appeared after one part was replaced with another in-spec part. The old part may have sat at one edge and the new one at the opposite edge, changing the sum. A "good" replacement can create a stack-up that was not there before.
• Identical systems behave differently with all parts testing good. Two units, same parts, one faults: the faulting one likely has its tolerances stacked unfavorably.

Finding the stack instead of the part

The method is to stop hunting for one bad part and start measuring the sum.

1. Measure the system-level result, not just the parts. Read the delivered output, the total drop, the actual end-to-end value. The stack-up shows at the system level even when each part hides it.
2. Read each contributing part's actual value, not just pass or fail. Note where in its band each one sits. You are looking for several parts all leaning the same direction.
3. Add the directions up. If supply is low, resistance is high, and load is heavy, the errors compound. If they point different ways, they cancel and the stack-up is not your cause.
4. Find the part with the most margin to give back. You usually cannot replace everything. Find the one part sitting furthest toward the bad edge and bring it back toward center; that often recovers enough headroom to clear the fault.

Fixing a stack-up

You do not have to make every part perfect. You have to move the sum back inside spec.

• Re-center the worst offender. Replace or adjust the single part sitting furthest toward the bad edge. Moving one part from the far edge back toward the middle can recover the whole system's margin.
• Correct anything actually out of spec or marginal. A stack-up plus a genuinely marginal part is easier to fix at the marginal part (see the related marginal-component method).
• Address the condition that pushes it over. If the stack only fails under heat or peak load, reducing that stress can keep the sum in range even with tight margins.
• Restore design intent. A stack-up often traces to a substitution or a setting that drifted. Putting the system back to its designed values re-centers the margins.

The judgment to bank

A system where every part tests good but the whole thing fails is not telling you the parts are fine. It is telling you to measure the sum, not the parts. The fault is real; it is just distributed across the margins instead of sitting in one component. When you find yourself replacing good part after good part with no improvement, stop swapping and start adding up where each part sits in its band. The stack-up is the fault that wears no single part's name.

References

• Manufacturer tolerance and specification data (the basis for each part's acceptable band).
• ISO 13379-1 (Condition monitoring and diagnostics of machines; system-level vs component-level faults).
• Trade-standard practice for system-level measurement and design-intent verification.
• See related: The Marginal vs Failed Component Decision Tree; Baseline vs Faulty Comparison Diagnosis Method.