CNC Programming That Will Hold Up on the Shop Floor

Most scrapped first articles are not caused by a wrong feed rate. They come from a sequence that leaves nowhere to hold the part, or that machines a datum after the features that depend on it. Before opening the CAM file, decide how many setups the part needs, which face is the primary datum, and which side stays untouched until the last operation.

A useful rule: establish the datum first, cut the mating features second, and leave cosmetic surfaces for the final pass. That order keeps fits predictable and keeps visible faces free of clamp marks. It also shortens the program, because each setup does one clear job instead of chasing dimensions across two fixtures.

Grouping operations also changes the tool list. When a single tool can finish every feature it reaches in one orientation, tool changes drop and the cycle gets shorter. The trade-off is that the part may sit in the vise longer, so check chip evacuation before you commit to a long roughing pass.

Write the setup sheet and the program together. Operators need to know the zero point, the clamp position, and the torque value before the first rapid move. A program with no setup sheet is a program that will be run differently by every shift.

A vise is fine until the part is thin, tall, or mostly unsupported. Clamping pressure flexes thin walls, and the part springs back after unclamping, so the measured dimension on the machine differs from the one on the inspection bench. Rough and finish in separate operations for anything below about 3 mm wall thickness.

For five-sided work, a dovetail or a machined soft jaw gives more rigidity than a standard vise. On our simultaneous 5-axis centers, a Ø400 mm rotary table handles most production parts, and the fixture is usually designed so the part is cut from one orientation. Fewer repositions means fewer datum transfers and less stacked error.

Long parts need support, not more clamp force. A 4,000 mm travel machine can hold a long extrusion, but the middle will deflect under cutting load unless you add a steady rest or a tailstock. Check the unsupported span against the tool pressure before choosing the depth of cut.

Zero-point systems pay for themselves when a part runs more than a few times. They remove the re-indication step and repeat to a few microns, which matters at ±0.005 mm. For one-off prototypes, a simple pin and clamp fixture is faster to build and just as accurate.

Pick the largest tool that reaches the corner without chatter, then add a smaller tool only for the features that need it. A Ø12 mm end mill removes material far faster than a Ø6 mm tool, and every extra tool adds a tool change plus a chance for a mismatch on a blending surface. Small tools are for detail, not for bulk removal.

Feeds and speeds come from surface speed and chip load, not from a table copied off the internet. For 6061-T6 aluminium, a coated carbide end mill runs well around 300–500 m/min surface speed with a chip load of 0.05–0.10 mm per tooth. Switch to 316L stainless and the same tool wants roughly 80–120 m/min. Titanium TC4 runs slower still, near 40–60 m/min, because heat stays in the cut instead of leaving with the chip.

Chip thinning matters on light radial engagements. When the radial depth of cut drops below half the tool diameter, the actual chip gets thinner than the programmed feed per tooth, so the edge rubs instead of cutting. Raise the feed per tooth to compensate, or the tool will work-harden the surface and wear quickly.

Use air blast or high-pressure coolant on aluminium to clear chips. Recutting a chip is the fastest way to break a small tool. On deep pockets, program a peck or a helical entry rather than plunging straight into the material.

Run the program in simulation with the actual stock and fixture models loaded. Collision checks catch the obvious crashes, but gouges and leftover material only show up when the stock shape is accurate. Import the real blank, not a bounding box, or the simulation will miss the areas that matter.

Probe the datum on the machine before the first cut whenever the part value justifies it. A spindle probe confirms the zero point and catches a fixture that moved overnight. For tight work at ±0.005 mm, this step is cheaper than scrapping one part.

When a first article is out of tolerance, measure before you change the program. Distinguish between a tool wear offset, a thermal drift, and a genuine geometry error. Adjusting the CAM file to fix a wear offset hides the real cause and repeats the mistake on the next run.

Keep a short setup log for each part number: tool list, offsets, fixture position, and any edits made at the control. That record is what turns a one-off program into a repeatable job, and it is the first thing a new operator should read.