Drawing revision traps that put the wrong part on the bench

A revision is approved, the PDF is re-issued, the email says "rev C supersedes rev B", and the shop still cuts the old flat pattern. Nothing in the approval chain failed. The failure is inside the revision itself: one of the drawing revision traps where the part that changed and the artefacts that describe it were allowed to disagree.

This is the companion to revision control for sheet metal, which covers the route a change takes from approval to machine. This article covers the traps in the revision content, the eight ways a correctly issued revision still produces a wrong part, and the checks that catch each one before the blank is cut.

The most expensive of them is the one that looks smallest: a bend radius edited from R2 to R3 with every dimensional callout untouched. Every dimension on the drawing still measures right. The flat pattern is 1.57 mm longer at that bend, and nothing on the sheet says so.

Which change needs a revision and which needs a part number?

The test is interchangeability, not size. If the new part can be fitted in place of the old one, with no rework to the mating parts and no change to the assembly sequence, a revision letter is enough. If it cannot, the change needs a new part number.

Interchangeability fails on four interfaces: the mounting pattern, the envelope, the mating face and the electrical or fluid connection. A hole moved 1.5 mm on a bracket that bolts to a frame is a new part number if the frame is already drilled. The same hole moved on a part that is located by two dowels and clamped is a revision.

The distinction matters because of the bin. A revision letter tells everyone that the two parts are the same part at different stages, so they may be mixed, and the oldest one may be picked first. A new part number tells everyone they are different parts, so they are stored, counted and shipped separately.

The four artefacts a change can invalidate

A drawing revision can move four things: the PDF drawing, the flat pattern, the cutting program and the checking fixture. A hole position change usually moves the flat pattern and the program. A bend radius change moves all four, because the fixture was built around the formed part.

Write that list into the revision block as a checklist. A revision that says "R3" and nothing else leaves the shop to work out whether the fixture is still valid, and the safe assumption on a busy floor is that it is.

What are the eight drawing revision traps?

The pattern in the table is that every trap is a disagreement between two objects that are supposed to describe one part. The drawing, the model, the flat pattern and the program are four descriptions, and any two of them can drift apart while the other two stay correct.

The radius trap in numbers

Bend allowance for a 90° bend is 1.5708(R + Kt), with K taken as 0.44 for a typical air bend in CRCA. On 2 mm material, an R2 bend gives 1.5708 × (2 + 0.88), or 4.52 mm. An R3 bend gives 1.5708 × (3 + 0.88), or 6.09 mm.

The difference is 1.57 mm per bend, which is 0.79 of a material thickness. On a four-bend box, the developed length grows by 6.28 mm. If the blank was nested to the old flat pattern, the box comes out 6.28 mm long, and the two end flanges are 3.14 mm out of position each.

Nobody measures the developed length of a box. They measure the outside dimensions, which are set by the bend lines, and those are correct. The error shows up at assembly, when the box will not sit inside its frame, and by then the batch is cut.

Why a thickness note is geometry

Changing 2.0 mm to 1.6 mm looks like a material substitution. It is a 20% reduction in thickness and a 49% reduction in bending stiffness, because stiffness scales with t³: 1.6³/2.0³ is 4.10/8.00, or 0.512.

It also changes the flat pattern. At the same R2, bend allowance falls from 4.52 mm to 1.5708 × (2 + 0.704), or 4.25 mm, a 0.27 mm shift per bend. And it changes springback: the estimate 3σyRθ/(Et) for 240 MPa CRCA at R2 and 90° gives 0.31° at 2.0 mm and 0.39° at 1.6 mm, so the same tool overbends by a different amount.

Three consequences, one note. A part that was designed around 2 mm and made in 1.6 mm will pass a dimensional check on the bench and flex in service, and the drawing will show that the material is correct.

The tolerance default nobody revisits

ISO 2768-1 sets general tolerances for dimensions that carry no individual tolerance. On a 400 mm dimension, class m allows ±0.5 mm and class f allows ±0.2 mm. If a revision tightens the class because one feature needed it, every other dimension on the sheet tightens too, and the price moves with it.

The reverse is the dangerous direction. A drawing that drops from f to m to reduce cost loosens every untoleranced dimension at once, including the ones the assembly depends on. Three stacked dimensions at ±0.5 mm instead of ±0.2 mm turn a 0.6 mm stack into a 1.8 mm stack, and the shim that used to be optional becomes mandatory.

Put the class in the title block, state which dimensions it applies to, and list the critical ones individually. A general tolerance is a default, not a specification.

Why the model, the PDF and the DXF drift apart

They are produced by three different people at three different times, and only one of them is usually part of the approval. The model is saved by the designer, the PDF is plotted by the checker, and the DXF is exported by whoever prepares the flat pattern, often days later.

The common failure is the DXF exported before the last edit. The shop receives a package where the PDF is rev C and the DXF is rev B, and the cutting program is built from the DXF. The router says rev C, the part is rev B, and the first article passes because it is measured against the DXF.

There is only one fix that works: release the model, the PDF, the DXF and the flat pattern as one package, named with the same revision, and refuse to accept a partial package. STEP files for bend review covers what the model has to carry for the bend check, and the drawing notes guide covers the notes that stop the shop guessing.

The scale and unit trap

A DXF has no units. A file drawn at 1:2 and exported without a scale note produces a part at half size, which is caught at the saw. A file in inches imported as millimetres produces a part 25.4 times too large, which is caught immediately.

The dangerous case is a drawing with a mixed unit: a metric sheet with one imported block in inches, so a 250 mm hole spacing becomes 9.84 mm on one view and 250 mm on the other. Put the unit and the scale on the title block, and check one known dimension against a caliper before the nesting run.

A rollback is not a revision

Going back to an earlier design is a change like any other, and it cannot be expressed by re-issuing rev B. A drawing that returns to B leaves two different Bs in circulation: the one that was superseded and the one that is current, with the same letter and the same title block.

Roll it forward instead. A rollback becomes rev D that restores the B geometry, with the description saying so. The history then reads A, B, C, D, and anyone holding a B knows it is two changes old rather than current.

The same rule applies to a temporary deviation. A part made to a concession is not at the drawing revision, and the concession needs its own number and its own effectivity. Writing "rev C, deviation" on a traveller is how a deviated part ends up in a bin of conforming ones.

What Ganesh actually runs

Every job enters the shop as a router with the drawing revision on it, and the revision on the router is the one that governs. Cutting programs are named with the part number and revision, and a program without a revision is not released to the 12 kW fibre laser or to the punch.

The bend program on the 200 T 5-axis brake carries the same revision, and the tool room checks the fixture against the new revision before the first part. Where a revision moves a bend line, the checking fixture is re-set or rebuilt, because a fixture built to the old flat pattern will accept a part that the drawing rejects.

First article inspection after a revision measures every critical characteristic against the new drawing, not against the previous first article. The design and CAD hub indexes the rest of the drawing-discipline guides, and the bending capability page sets out the brake and the tooling the bend program runs on.

What has to be in the revision block?

1. The revision letter, unique and sequential, with no reuse.
2. The zone, so the change can be found on a large sheet without reading every view.
3. A description in one line: "R2 to R3 on bends 2 and 4", not "radius updated".
4. The effectivity: serial, date or quantity.
5. The artefacts that re-issue: PDF, DXF, flat pattern, fixture.
6. The approver and the date, and whether the customer approved it.
7. A reference to the change request, so the history survives the person who made it.
8. A note on interchangeability: can this part be used in place of the previous revision?

The last one is the field that prevents mixed shipments, and it is the one most often left blank.

When a revision is the wrong instrument

A revision is wrong when the change alters the interface, because interchangeability has already failed. It is also wrong when the customer's own drawing is the master and the change is being made on a local copy, because the next customer release will overwrite it.

And it is wrong when the change is a process change rather than a design change. Switching a part from a press tool to the laser, or from spot welding to MIG, does not belong on the drawing revision at all; it belongs in a process change note with its own first article, which is a separate requalification.

Send the revised drawing with the change clouded, the effectivity stated and the flat pattern included to request a quote. We will confirm which files, programs and fixtures move with the revision before the first blank is cut.