3/4/5-axis
CNC machine

I sized the machine for three jobs: aluminium plate parts up to 30 mm thick, one-off prototypes in mixed materials, and carving foam for molds. The trunnion adds tilted plate and block work on top of that.

The gantry spans the 600 mm side of the work area and travels along the 1000 mm one. How far a beam bends goes up with the cube of its span, so spanning 600 mm instead of 1200 mm cuts the bending by 8x.

The table stays still and the gantry moves. A moving table would have needed about 2 m of floor to travel through, and at a 600 mm span it wouldn't have made the machine any stiffer.

Every part of the machine bends a little, like a spring, and the springs add up in series, so the softest one decides how stiff the machine is. With a 6 mm tool, the softest spring is the cutter itself.

A 6 mm cutter sticking out 20 mm is softer than the target for the whole frame, so the frame stops at 8 to 12 N/um instead of going all the way to 20. The cheapest stiffness on the machine comes from shortening the stickout. Pulling the same 6 mm tool in from 30 mm to 20 mm makes it 3.5x stiffer.

The first spec called for welded steel tube filled with epoxy granite. I went with bolted 80x160 aluminium T-slot instead, with sand or resin fill for damping. At a 600 mm span the joints are weaker than the beam, so every joint gets a gusset.

The accuracy budget comes out around 0.05 to 0.1 mm, which is good for a hobby mill but falls short of a precision machine. The biggest single source of error in it is heat. An aluminium frame grows about 0.14 mm over its travel for every 5 °C.

X runs on two ball screws with one motor on each side, slaved together in software, so the machine squares itself off two home switches every time it powers on. The motors are NEMA 34 closed-loop steppers, 12 Nm on X and Y and 6 Nm on Z. Every driver's alarm output is wired back to the controller, so a stalled axis stops the job instead of scrapping the part.

The Z motor has a brake built in. The spindle and its plate come to about 30 kg, and that weight back-drives a 5 mm lead screw at only 0.26 Nm. Without the brake, the spindle would drop into the part every time the power goes out.

The Z rails are fixed to the gantry and the blocks ride on the moving plate, so Z is as stiff fully extended as it is retracted.

The machine sits on a granite slab, and the slab sits on a 700 mm stand of bolted 40-series extrusion. Granite needs support about every 700 mm, and without it a strip model of the slab sags about 0.7 mm.

The slab only touches 80 x 80 x 10 mm steel pads. They get ground flat to each other after the stand is levelled, so the frame's own flatness can't bend the slab.

Under the table there are three bays: one for the tool changer, one for 4-axis machining and one for 5-axis machining.

Modelling it showed that the 80x80 legs only have four 6.7 mm core bores and nothing in the middle, so a levelling foot can't screw into the end of a leg. Each leg gets a bolted steel base plate to carry its foot instead. As modelled, the stand is 47 parts and 137.4 kg, with zero interferences.

It runs LinuxCNC on a desktop or mini PC instead of a laptop, through a Mesa 7i96S and 7i76U. Both cards together come to $268. LinuxCNC was the only open-source controller that could handle a 5-axis mill with a tool changer, and it drives six outputs: X, X2, Y, Z, and the trunnion's A tilt and B rotate.

The spindle is a 4.5 kW BT30 tool-change spindle. It's 8-pole for 18,000 rpm, water-cooled, and runs on 220 V single phase. The E-stop is a hardwired button that drops a contactor and cuts the drives and the spindle at the same time, so it still works if the PC hangs.

As drawn it's a 3-axis mill, and the 4th axis has its own assembly started in the model. For five axes, the plan is a trunnion I build myself, tilting on A and rotating on B. The trunnion has to fit in whatever height the frame leaves, so the frame doesn't have to wait for it. The 4th axis and the 5-axis setup each get their own bay under the table.

The main limit on five axes is the CAM. LinuxCNC runs 5-axis code fine, but generating that code takes expensive software and real skill.

The tool changer gets its own bay under the table. There's no published drop-in rack for BT30 tools, and every fully documented build uses ISO20, ISO30 or TTS. I've started a tool mover assembly in CAD, and the rack spacing has to wait for the spindle nose drawing.