Tolerance Stack-Up Troubleshooting

Stack-Up Failures Quick Answer

Most "the stack passed but the parts don't fit" surprises come from wrong signs in the chain, measuring the wrong features, treating weld distortion like a print ±, mixing inch/mm, or applying RSS on a process that is not centered.

1. Wrong Signs — Comforting Math, Wrong Gap

What it looks like: Worst-case and RSS both look safe in the calculator. On the bench the cover hits the tube, the pin is short, or the clearance opens on the opposite side of the assembly from what you expected.

Likely causes: Each dimension must contribute + or − to the critical result. A "plus" on plate A can close a gap the same way a "minus" on spacer B does. Flipping one sign turns a real interference into a fake clearance (or the reverse).

Fixes: Sketch the chain once with arrows toward "more clearance" vs "more interference." Enter the same signs in the calculator. Re-run only after signs match the sketch — not after changing tolerances hoping the number improves.

2. Measuring the Wrong Features (or the Wrong Datum)

What it looks like: Parts fail fit check, but every individual dimension on the CMM or caliper is "in print." The stack used hole center-to-edge; inspection measured edge-to-edge of the plate. Or the stack assumed a machined face while production checked the as-welded face.

Likely causes: Stack chains are feature-specific. Measuring a related but different feature validates the wrong model. Datum shifts (inspecting from the opposite face) add a hidden thickness into — or out of — the chain.

Fixes: Write the stack as "feature A to feature B along this path." Measure those exact features with the same datums. If production cannot reach the drawing datum, change the drawing or the manufacturing sequence — do not invent a parallel stack that only exists at the bench.

3. Weld Distortion Treated Like a Print ±

What it looks like: Flat laser parts stack cleanly on paper. After welding the frame, covers bind, hole patterns walk, and flush conditions open. Re-running the same ± from the flat pattern still "passes."

Likely causes: Distortion is process capability, not a Gaussian print tolerance. Worst-case and RSS both assume the listed ± bounds the real variation. Weld pull often exceeds that and is biased (not centered).

Fixes: Measure critical interfaces after weld. Prefer weld-rough / machine-critical in one setup. Shorten the stack by creating datums on machined faces. Guide reminder: welded frames usually want worst-case or measure-after-weld — see when each method applies.

4. Formed Parts — Bend Deduction Left Out of the Chain

What it looks like: Brake-formed brackets or boxes: flange lengths look fine individually, but the assembled envelope is short/long. Hole-to-bend dimensions drift lot to lot.

Likely causes: Stack used flat blank dimensions or nominal flange lengths without the bend deduction / springback variation that actually sets the formed length. Related: wrong K-factor or BD in the flat pattern.

Fixes: Put the formed feature that controls the fit into the stack (flange length after bend, not blank). Calibrate BD with the Bend Deduction Calculator and process notes in the Sheet Metal Bending Guide. Then re-enter the realistic ± for that formed length.

5. RSS Used Where Worst-Case Was Required (or the Process Isn't Centered)

What it looks like: Statistical stack shows a comfortable band. A small percentage of assemblies still hard-interfere, or a safety cover / sealed joint occasionally fails. Or RSS was copied from a high-volume note onto a one-off weldment.

Likely causes: RSS assumes independent, centered, capable dimensions. It does not guarantee every unit. Hard no-go fits, seals, and fixtures need worst-case (or selective fit with documented criteria). Biased processes (torch cut, weld pull) violate the RSS model.

Fixes: Match the drawing method to the risk. If every unit must assemble without selective fit, switch to worst-case in the calculator and on the print note. Method fork detail stays in the guide — do not mix WC and RSS inside one undocumented chain.

6. Unit Mix-Ups and False Precision

What it looks like: Stack totals look absurd (gap of 25" or 0.025 mm when you meant the other). Or inspection rejects parts that "feel" fine because the print called six decimals on a brake flange.

Likely causes: One dimension entered in mm while the rest are inches. Or print precision exceeds process capability, creating inspection fights without improving fit.

Fixes: Convert everything to one unit before entry. Match decimal places to the process (and the guide FAQ). If capability is ±0.030", calling ±0.005" only guarantees scrap — shorten the stack or change the process instead.

Frequently Asked Questions