GD&T and datum strategy on welded structures

On a machined part from solid, the datum question is usually easy — the features exist, they are crisp, and they are where the model says they are. On a weldment none of that is reliable. The as-welded surface is a plate face that has been cut, heated, pulled by shrinkage and possibly flame-straightened. It is somewhere near nominal and it is not a reference.

Getting the datum frame right on a fabricated structure is the difference between an inspection report you can read against your drawing and one that requires a week of argument to interpret.

The rule that solves most of it

Datums on a weldment must be machined features.

Not a plate edge. Not a cut face. Not “the bottom of the base”. Not a centreline, which does not physically exist on a fabrication until someone decides where it is.

Three machined features, mutually perpendicular, forming a proper datum reference frame — primary plane, secondary, tertiary — and every functional dimension referenced to them. Anything else and each shop invents a reference, and their invention will not be yours.

The practical consequence for the fabrication route: the datum surfaces have to be machined first, in the same setup family as everything referenced from them. That means the machining sequence is partly determined by the datum scheme, which is a good reason to settle the datum scheme before the fabrication drawing is issued rather than after.

Choosing which features become datums

Three criteria, in order:

1. Functional. The primary datum should be the surface that actually locates the part in service — the face that sits on the foundation, or the face the next assembly bolts to. If inspection references a different surface than assembly does, the part can pass inspection and not fit.
2. Accessible. The datum features must be reachable for probing without dismantling anything. A datum bore in the middle of a closed box section is a datum nobody will measure.
3. Stable. Prefer large, well-supported features over small ones on thin sections. A datum on a 6 mm panel that flexes under probe pressure is not a datum.

On a typical machine base that resolves to: primary is the large machined underside or the machined top plane, secondary is a machined end face or a pair of bores, tertiary is a side face or a single bore.

Datum targets, which weldments genuinely need

A full datum plane on a large fabricated surface is often impractical — the surface is big, slightly wavy, and simulating the whole plane needs a surface table the part may not fit on.

Datum targets solve this properly. Instead of “datum A is this entire face”, you specify three small target areas on that face, dimensioned in position, and the datum plane is established from those three points. Inspection then needs three supports rather than a granite table the size of the part, and — importantly — every inspection establishes the same plane the same way.

For any machined face over roughly a metre, datum targets are the right call. They are also the thing most often missing from weldment drawings, which is why two inspection reports of the same part can disagree.

Which controls earn their place

Geometric controls cost money in steps, not smoothly, because each tightening pushes the part into a different process route. The ones that matter on machine frames:

Flatness on rail seats and mounting planes. This is usually the control that drives the whole machining strategy, and it is the one most often over-specified. The cost steps are covered in what drives the cost of machine frame fabrication.

Parallelism between the top mounting plane and the underside. Cheap when both faces are machined in a planned sequence; expensive when specified as an afterthought that forces an extra setup.

Perpendicularity of a column face to the base plane. This is the control that determines whether the spindle is square to the table, and it is worth its cost.

Position on bolt-hole patterns and bearing bores. Almost always better expressed as position with a material condition modifier than as coordinate dimensions with plus-minus tolerances, because it gives the shop the bonus tolerance it is entitled to and gives you the functional requirement you actually have.

Profile on surfaces that mate to a complex counterpart. Powerful and expensive; use sparingly.

Coplanarity across separate machined pads. This is the control that lets you replace an expensive full-face flatness callout with a few machined pads — a substitution that routinely takes a large cut out of the price without changing function at all.

The general principle is set out in ASME Y14.5 and its ISO GPS counterparts, but the engineering judgement is local: control what the machine needs, release everything else to a general fabrication tolerance such as ISO 13920 class B or C.

The general tolerance note, which does a lot of work

A weldment drawing without a general tolerance note is asking for trouble, because the default assumption differs between shops and countries.

Specify the standard and the class explicitly:

Unless otherwise stated, general tolerances for welded construction per ISO 13920, linear class B, angular class B. Machined features per ISO 2768-m. Geometric tolerances per ISO 1101, datum reference frame A-B-C as shown.

That note releases the sixty structural features you do not care about to a sane tolerance and leaves you free to tighten the six that matter. Applying a precision general tolerance to the whole drawing is the most expensive drafting habit in this field — it silently pulls every feature into the machined budget.

Inspection reporting

Require the dimensional report to be measured from your datums, and to list the geometric controls in the same order and with the same identifiers as the drawing.

A report full of measurements taken from features the shop chose — even if the numbers are good — cannot be reconciled against your drawing without redoing the work. This is a cheap thing to specify and a very expensive thing to negotiate after the fact. The full requirement set is in first article inspection on large weldments.

A short review before release

• Are all three datums machined features?
• Does the primary datum match how the part is located in service?
• Are datum targets used on any datum face over a metre?
• Is every geometric control traceable to a function, not to habit?
• Could any flatness callout be replaced with coplanar pads?
• Is there a general tolerance note naming a standard and a class?
• Does the drawing state the GD&T standard in the title block?

A weldment drawing that passes that list gets inspected once, accepted once, and argued about never — which on a large frame is worth considerably more than the hour it takes to check.

If you want a frame drawing reviewed for datum scheme and tolerance realism before it goes out to quote, send it through. The review comes back within 24 business hours and covers whether the datums are machinable in a sensible sequence.