A Day in the Life of a CAM Programmer using DigitalCNC – Aerostructure Part in NX
Fifteen seconds apart in NX. Five and a half minutes apart on the machine.
Two roughing strategies on an aluminium bulkhead rib in Siemens NX, ranked by CAM and machine cycle times – with decisions being made at the desk, before hitting the machine.
A large structural aluminium bulkhead rib is to be cut on a high-power, high-rpm, low-torque platform, with the torque under 40 Nm and the power under 80 kW. Two pockets to look at, one triangular and one square, with both being typical enough to read across the rest of the part program.
The aluminium bulkhead rib in NX.
What the NX cycle time is worth
Every programmer knows the CAM cycle time is not the machine cycle time, but it is hard to know how big that gap is or where it comes from.
Two roughing strategies, both programmed and posted from NX, are fifteen seconds apart in CAM, and their material removal rates are close to identical:
- Area mill into cavity mill, three-flutes at 75% radial shows it’ll take 35 seconds when modelled on NX. However, on the machine it takes 51 seconds at 62% efficiency (with efficiency here being the proportion of cycle time spent at or above 80% of the programmed feedrate).
Offset path: 51 seconds against a 35-second CAM estimate, 62% efficiency. Showing corner slow downs.
- 3D adaptive roughing, four-flutes at 30% radial, depth and feeds pushed up shows it’ll take 50 seconds, but on the machine it actually takes 6 minutes 17 seconds at 36% efficiency.
Adaptive path as programmed: 6:17 against a 50-second CAM estimate, 36% efficiency. The slow downs shown across much of the toolpath.
There’s only fifteen seconds difference in CAM, but five and a half minutes difference on the machine. Nothing on the screen shows it, because CAM does not consider the machine’s kinematic constraints and performance.
How to recover time using machine specific tolerancing
The adaptive path is not slowing at 6:17 because the cutter is overloaded. Motion stays tangential: it never fully stops, never reverses under load, and chip thickness holds all the way round. The time is going into out-of-cut repositioning and the sheer number of blocks the controller has to read.
Feed against time, with rapids clipped to keep the cutting feed readable.
This means it is recoverable, and can be achieved by changing the machining tolerance. The rib as programmed with a 30 µm CAM tolerance: fine enough to hold the semi-finish allowance, coarse enough to keep the block count down. Opening the machining tolerance from its programmed 30 µm to 100 µm took the pocket from 6:17 to 4:54 keeping everything else (the strategy, the tool, and the engagement) the same, and it took 83 seconds off one pocket, still tangential with chip thickness held. Across the full rib, the same change took 1 hr 07 min down to 49 minutes. Eighteen minutes saved per part from adjusting the machining tolerance.
That gap has nothing to do with offset against adaptive. The adaptive toolpath that would have been posted anyway is either 6:17 or 4:54 depending on the machining tolerance chosen, and nothing in CAM tells the programmer which one is posted.
The offset path barely responded to the same tolerance change, as it only moved from 51 to 50 seconds. CAM tolerance is a lever on adaptive strategies in particular, because CAM tolerance is what sets the block density the controller has to work through.
Adaptive roughing: continuous tangential motion, no reversals under load. Opening the tolerance changes the block count, not the cut.
Why the fastest option, the zigzag, was rejected
NX will also offer the alternative cut direction, a zigzag, which shows 4:20 at 47%, quicker than any other option and on cycle time alone, the obvious choice.
Then look at the reversal. Feed drops almost to zero with the cutter still fully engaged, and at these speeds, in aluminium, what follows is a chain: built-up edge packs the flutes, welds to the cutting edge, the edge breaks down and the cutter fractures, and on a spindle turning this fast it takes the holder and the bearings with it.
The cost of not taking it is 34 seconds a pocket. Without the loading view there is no argument to have: 4:20 at 47% just looks like the best strategy.
Zigzag cut direction. Quickest on cycle time, with a loaded reversal at the end of every pass.
4:20 at 47%. The number that gets picked when the loading is invisible.
The offset path, and where its cycle time can be recovered
Area mill into cavity mill, three-flutes at 75% radial, three-flutes at 75% radial, tips into a slotting condition at the corners and behaves like one. Feed collapses from nominal to below 1,000 mm/min at every acute corner, every 180° reversal out of full slotting, and every 90° step transition. If you back-calculate chip thickness at those points, it falls through the 20 µm floor, so the edge rubs instead of shearing while the tool is still carrying the load.
Three path-level fixes were applied to it. NX smoothing rounded two of the three challenging features well, and then the fine radii it added to clear the triangular pocket became the new acceleration limiter, taking the pocket out to 1:26 at 63%. Opening the tolerance from 30 to 100 µm gave 50 seconds at 66%.
What wasn’t tried is what most programmers would reach for next: drop the radial engagement, get more flutes in the cut, open the corner radius in CAM rather than asking smoothing to fix it afterwards, or change the tool, with any of those having the potential to rescue it. So the claim here isn’t as broad as “offset is the wrong strategy”. It is that the offset path as programmed is acceleration-limited on this geometry, and it’s visible before the first-off rather than after it.
Either way, that’s useful – if the read is that 75% radial was the problem all along, it does not take a machine slot to find out: change it, run the analysis again, read the loading. Minutes on the laptop, not on a proving run.
Offset path across the two pockets. The acute corners and step transitions are where the feed goes.
Smoothed version. The radii added to clear the triangular pocket became the new limiter.
1:26 at 63%. Smoothing cost time without changing the loading.
Lost cycle time: a rubbing cutter versus non-cutting moves
51 seconds against 4:54 is a big ask, and on a low-value part in a rate shop, with low cost tooling and a quick setup, the offset path and more frequent cutter changes may well be the right call. Nothing here overturns that.
However, on a high-value aero structural component, a cutter that fractures mid-pocket does not cost a cutter. It costs the part, the setup and the machine slot, and possibly a spindle. Against that, four minutes a pocket buys a path that runs at commanded feed, holds chip thickness where it should be, and can be left running on its own. In other words, it provides process stability.
Two numbers get read as the same thing and are not. On the offset path the missing 38% is a fully engaged cutter rubbing. On the adaptive path the missing 64% is air. The first scraps parts whereas the second is only time, recoverable by opening the CAM tolerance.
Making the call: which strategy, and at what tolerance
The best strategy is plain adaptive roughing with the CAM tolerance opened to 100 µm. It is neither the shortest predicted cycle time nor the strategy an untimed CAM estimate would have selected. It holds chip thickness above the floor, keeps the cutter moving tangentially, and produces a cycle time accurate enough to quote against. When someone asks why not the 4:20, the answer is on the screen rather than on the machine.
The different strategies, side by side
The same decision across the full rib: 49 minutes on the tuned adaptive strategy.
What the estimate could not see: the achieved feedrate
Nothing in the first read was wrong: two strategies, near-identical removal rates, one with the better loading. What was missing was the machine. A CAM estimate gives a cycle time, but not the actual machine time, and it shows neither feedrate drops or impact to chip thickness.
The prediction is deterministic and physics-based. There is no direct connection to the machine, no training data, no first-off to sacrifice, and it takes only five clicks on the path already programmed, inside NX, CATIA and Mastercam.
6:17 to 4:54 on one pocket. 1 hr 07 min to 49 minutes across the rib. One tolerance value. And the 4:20 that looked like the best decision was rejected before it cut a chip.
