From DXF to a busbar the machine can actually cut

A DXF file is geometry. A CNC busbar machine needs instructions. Between the two sits a translation step that most shops discover only when the first bar comes out 4 mm short and nobody can say whose fault it is.

The gap is not about file formats, though formats cause plenty of trouble. A drawing describes a shape; a machine needs an ordered list. Nothing in a DXF says which closed loop is a hole and which is the outside edge. Your CAM has to infer it, and inference fails in specific, predictable ways.

What the machine actually needs

Strip away the software and a punch-shear-bend cycle needs six things: blank length and section, with material and temper, since springback and punching force both depend on them; a hole table of X, Y and diameter from a datum; a bend table of position, angle, direction and inside radius; the operation sequence, because punching after bending is often impossible while bending after punching can distort a hole too close to the tangent line; the datum itself, meaning which end is zero and which face is up; and the flat length, which is not the sum of the leg lengths on the drawing.

A DXF gives you the hole table well, the section partially, the bend table only if someone annotated it, and the last three not at all. Everything that goes wrong lives in that shortfall.

Layer conventions, and what a good template looks like

The only reliable way to tell CAM what a piece of geometry means is to put it on a named layer and map that name in the importer. Colour is not a substitute: it survives conversion badly, colour-by-block overrides layer colour silently, and nobody in your drawing office agrees on what red means.

A workable layer set for busbar:

Rules that make the template work in practice:

• One part per file, or one part per block with a flat namespace. Nested blocks defeat most importers, and a block reference scaled non-uniformly turns circles into ellipses.
• Model space, 1
  , millimetres. Set $INSUNITS to 4 (millimetres). A drawing that arrives unitless gets guessed at, and the guess is often inches.
• Explode nothing on the outline, explode nothing into hatch. Hatch on the part face is decoration that CAM may try to cut.
• Holes are CIRCLE entities, always. Not polygons, not arcs joined into a loop, not blocks containing a circle unless your importer resolves them.
• No dimensions or centre marks on manufacturing layers. A centre mark on BB_HOLES becomes two tiny slots.
• Datum at a stated corner, conventionally the lower-left of the flat with the reference face up, and stated in the template so it never has to be negotiated per job.

Write this down as a one-page drawing standard and issue it to everyone who sends you drawings, your own engineering department included. The cost of not having it is a phone call per job.

DXF R12, R14 and DWG

DXF version matters more than people expect, because entity types were added over time and an export to an older version has to fake anything that did not exist yet.

DXF R12 (AC1009) has no SPLINE, no ELLIPSE and no LWPOLYLINE. It does have LINE, ARC, CIRCLE and the heavyweight POLYLINE, which for busbar work is usually enough: outlines are lines and arcs, holes are circles. R12 is a safe interchange format provided the part was drawn that way. Draw a slot end as an ellipse or trace a profile with a spline and R12 export silently converts it to a many-segment polyline, which is where recognition falls over.

DXF R13 and R14 (AC1012, AC1014) introduced the SPLINE and ELLIPSE entities and the lightweight polyline. Curves survive as curves. In practice a mid-range ASCII DXF version, R14 or AutoCAD 2000, is the best default: new enough to carry true arcs and ellipses, old enough that every CAM reader on the market handles it.

DWG is Autodesk's native binary format and a moving target. Every release changes it, third-party readers reverse-engineer it, and it carries far more than you need: xrefs to files you were not sent, proxy objects from vertical applications, viewport geometry. Accept DWG when you must, convert it yourself, and inspect the result.

Prefer ASCII DXF over binary. ASCII is a text file you can search. When a part refuses to import, open it in a text editor and count entity types. A rectangle with eight holes that shows forty POLYLINE records and zero CIRCLE records has told you what happened before you open any CAM software.

Why exploded splines and polygonised arcs break hole recognition

Hole recognition works by pattern matching. The importer looks for a CIRCLE entity, or a closed loop of arcs sharing a centre, and calls it a hole with a diameter it can match to a punch. When a circle arrives as a polygon, that test fails.

The failure is not only categorical, it is dimensional. A circle of radius R approximated by an inscribed polygon of n segments has its vertices on the true circle and its edge midpoints inside it, at radius R·cos(π/n). For an 11 mm clearance hole (R = 5.5 mm) approximated with 16 segments, the mid-chord radius is 5.5 × cos(11.25°) = 5.394 mm, so the polygon's inscribed diameter is 10.79 mm. That is 0.21 mm undersize. A recogniser that fits to the mid-chord will pick the wrong punch. With 32 segments the error drops to about 0.05 mm, which is exactly the positioning accuracy published for the IMAC-CENTER 80, so you are now spending your entire geometric budget on a rounding artefact.

Some importers do not attempt a circle fit at all and treat the closed polyline as a profile, which on a punching machine means an error or a nibbled contour taking sixty hits where one would do.

One distinction matters. An arc stored as a bulge value inside an LWPOLYLINE is a true arc and survives perfectly; the bulge is tan(θ/4) for the segment's included angle, and importers reconstruct it exactly. It is linearised arcs, produced by exploding, by R12 export from spline geometry, or by a trip through an illustration package, that destroy the information. Splines are the usual culprit, because nothing downstream can represent them and every export path flattens them somewhere. Setting a tighter chord tolerance only trades geometry error for segment count. The real fix is upstream: draw holes as circles.

Flat-pattern development and where the neutral axis actually sits

A bent bar is longer on the outside than the inside. Somewhere between the two surfaces there is a fibre that neither stretches nor compresses, and the length of that fibre through the bend is the material you have to allow for in the flat.

The standard relation is:

BA = (π/180) × θ × (R + K·t)

where θ is the bend angle in degrees, R the inside radius, t the thickness, and K the K-factor, defined as the distance from the inside surface to the neutral fibre divided by t.

The naive assumption is mid-thickness, K = 0.5. It is wrong, for mechanical reasons. As the bar wraps the die the outer fibres stretch and thin while the inner fibres compress against the die, so the neutral fibre migrates towards the inside face, and it migrates further the tighter the bend. For copper busbar bent flatwise over a generous radius, R/t of 2 or more, K typically lands between 0.42 and 0.48; as R/t approaches 1 it commonly falls to around 0.35. Those are indicative, not constants. They shift with temper, die geometry and whether the machine wipes or wraps.

So measure yours. The procedure takes twenty minutes and pays back forever:

1. Cut a coupon of the section you use, at a length L₀ you have measured to 0.05 mm.
2. Bend it 90° over the die you use in production, in the same orientation.
3. Measure the two outside leg dimensions A′ and B′ to the theoretical outside apex.
4. Bend deduction BD = A′ + B′ − L₀.
5. Outside setback OSSB = (R + t) × tan(θ/2), which for 90° is simply R + t.
6. Bend allowance BA = 2 × OSSB − BD.
7. K = (BA / θ_radians − R) / t.

Worked through: t = 10 mm, die inside radius R = 10 mm, θ = 90°. If your measurements give K = 0.45, then BA = 1.5708 × (10 + 4.5) = 22.78 mm, OSSB = 20 mm, and BD = 40 − 22.78 = 17.22 mm. A part with outside legs of 200 and 300 mm has a flat length of 500 − 17.22 = 482.78 mm.

Now suppose your CAD was carrying a default sheet-metal K of 0.33, inherited from a steel template that nobody changed. BA becomes 1.5708 × 13.3 = 20.89 mm, BD becomes 19.11 mm, and the flat comes out at 480.89 mm. A 1.9 mm error per bend, from a parameter buried three dialogs deep. On a four-bend U-shaped riser that is up to 7.6 mm of accumulated error, and it will show up as a hole that misses its stud.

Edgewise bending does not unfold by the same relation, and many general-purpose CAD unfolders quietly apply the flatwise rule to it. If your parts include hard-way bends, verify the unfold on a coupon before you trust it; the difference between edgewise and flatwise bending is not a detail.

Bend-line annotation that a machine can read

A bend line on a drawing must carry four things, and each has a convention trap.

Position. Is the line at the bend centre, at the inside tangent, or at the apex of the outside surfaces? Most CAM expects the centre of the bend zone in the flat. State it on the template.

Direction. Up or down relative to a named face, not relative to "the view", because the view flips when someone mirrors the part. Two layers, BB_BEND_UP and BB_BEND_DOWN, remove the ambiguity entirely.

Angle, and which angle. Bend angle (departure from flat) and included angle (between the legs) are supplementary. At 90° they are equal, which is why the confusion survives: everybody's first test part is a right angle and both conventions agree. At a 45° bend angle the included angle is 135°, so a drawing that says only "45°" is a coin toss. Write BEND UP 45 (INC 135).

Inside radius. The bend allowance depends on it, so it is not optional. In practice the radius is set by the die you have, so the drawing should name the die or the radius, and engineering should know which radii exist on the shop floor before drawing a bend that needs one you do not own. Copper temper constrains this too, since a half-hard bar will not take the radius a soft-annealed one will; see the guidance on minimum bend radius by temper. Check the section and radius against the machine's capability with the bending force calculator at design time rather than at the press.

One more annotation to carry: hole-to-bend distance. A hole whose edge sits inside the deformation zone will distort when the bend forms. Keeping the hole edge at least 1.5 to 2 times the thickness clear of the bend tangent is the usual working rule, and it is a rule your CAM will not enforce for you.

The tolerance stack, end to end

Add up the contributions between a nominal CAD dimension and a finished bar, for a four-bend part 2 m long in 100 × 10 mm copper:

The pattern is stark. Machine contributions sit in tenths of a millimetre; the unfold contribution sits in millimetres. A more accurate machine does not fix a wrong K-factor, and a shop that buys ±0.05 mm positioning and then unfolds with a steel default has spent its money in the wrong place. The quiet term is measurement temperature: copper expands about 17 µm per metre per kelvin, so a 6 m bar measured 5 K off reference reads half a millimetre out. That accounts for a good share of the arguments between a fabricator and a customer's goods-in.

Pin hole position and pitch conventions for bolted joints to a standard instead of negotiating them drawing by drawing. DIN 43673-1 covers drilled holes and screw connections for rectangular busbars, giving hole location and size together with base data for the screw connections. Note that DIN 46433 is not a busbar standard, whatever a search engine tells you; it covers drawn rectangular wire dimensions and is one of the most persistently miscited references in this trade.

What buyers actually draw in

CAD in switchgear is fragmented, and the format problem differs in each corner of it.

AutoCAD. Still the most common source of busbar geometry. It produces flats directly, because the drawing is the flat. That cuts both ways: there is no unfold step to get wrong, but no model to check against either, and the bend allowance was applied by hand upstream by someone who may or may not have used the right K.

SolidWorks. Good sheet-metal unfolding with K-factor and bend-table support, and a clean flat-pattern DXF export. The trap is the default: sheet-metal templates ship with steel-oriented gauge and K tables, and copper busbar at 10 mm thick is not sheet metal. Build a busbar-specific set of sheet-metal parameters, one per section and temper, and lock it.

EPLAN Pro Panel with the Copper module. The one purpose-built for this job. It plans the busbar system in 3D alongside the panel, produces the unfold, and exports NC data for bending, drilling and punching machines, taking edgewise bends into account. If your panel design already lives in EPLAN, this removes a whole class of transcription error. Schematic-only EPLAN Electric P8 produces no copper geometry; you need the 3D panel product.

Elecdes Design Suite and SEE Electrical (IGE+XAO) both cover panel and enclosure layout, and both turn up in EPC and OEM work. As with EPLAN, what you get depends on which modules are licensed; schematic tools export schematics.

Whatever the source, check the geometry once against a measured first article and record the result per section and temper. Once per combination, not once per job.

The file transfer problem

None of the above is the most common cause of a wrong bar. The most common cause is revision drift.

The pattern is familiar. Engineering issues rev B. It reaches the machine by email, or on a USB stick, or through a shared folder the machine cell cannot see so someone copies it locally. The operator edits the program at the HMI to clear a clash, and that edit exists nowhere else. Engineering issues rev C. Nobody tells the machine. Two weeks later a batch is cut to rev B plus an undocumented local edit, and the only record of what was made is in an operator's memory.

This is what AIRLINK addresses on machines such as the EMAC-BB-S12 and the punch-shear and processing-centre lines: real-time drawing and file sharing between office and machine, so the machine pulls the current revision instead of holding a copy, and the office can see what the machine is running. With whole-project nesting in SMARTNEST BUSBAR and 3D-to-program conversion in SmartDraw behind it, the drawing stops being a file that gets carried around and becomes a record that gets referenced.

That is a workflow claim rather than a geometry one, and the two should stay apart. File hygiene will not fix a wrong K-factor, and a correct K-factor is worthless if the machine is running last month's drawing.

Fix the drawing standard first, because it costs nothing. Then measure your own K-factors, because that is where the millimetres are hiding.