A Day in the Life of a CAM Programmer using DigitalCNC – CATIA on the Ecospeed

Three CATIA concentric strategies, one aluminium aerostructure bracket, and why the fastest option on paper is not the one you should run.

The part is an A-frame aerostructure bracket in aluminium, programmed in CATIA V5-6, on a DST Ecospeed at 30,000 rpm. If you program aluminium aerostructure work in CATIA for an Ecospeed, the feature set will look familiar: long pockets with tight corner radii, thin webs between them, and a volumetric roughing job that dominates the cycle.

The study part: an A-frame aerostructure bracket, programmed in CATIA V5-6.

Commanded feed versus achieved feed

The tool is a four flute 20 mm dynamic mill, running at 30% radial engagement. Below 50% immersion you get radial chip thinning, so the feed per tooth is compensated upward to hold a true chip thickness of 0.20 mm. Theoretically that gives 0.218 mm/tooth, a table feed of 26,160 mm/min, and a 40 mm axial reach. At that removal rate the spindle is working near its power ceiling, so any feed the platform fails to deliver cannot simply be pushed back in.

All of which assumes the machine achieves the feed you commanded.

Run the same logic on a smaller sanity-check feature, a 12 mm four flute at 3 mm radial, 22,500 rpm and 14,000 mm/min programmed, and the result is completely different. Across the cut motions the platform never reaches the programmed feed. The maximum achieved cut feedrate is less than half of the commanded, at around 6,500 mm/min.

The compensated feed per tooth was designed to hold 0.14 mm of chip, but you are cutting roughly half of that. It is still forming a chip, but a thin one: specific cutting energy rises, less of the heat leaves with the chip, and you pay for it in tool life and cycle time. The numbers were not wrong, they were just never delivered by the machine.

So the first job is not to tune anything, but rather find out what the platform will actually give you, then build the process to that number so the cutting conditions are understood rather than assumed.

Three strategies that all look feasible

•   Circular. A tangential path, kinematically smooth, but the loading profile only reaches maximum immersion and chip thickness at the midpoint of each arc. Sinusoidal loading.
•   Dynamic. Ramps to full engagement as early as possible and holds it as far across the cut as it can. A square wave rather than a sine wave. Less kinematically friendly, but it maximises the area under the loading curve.
•   Zigzag. Uses the return move as a productive cutting pass. Theoretically the fastest of the three, at the cost of flipping between climb and conventional milling.

CAM cycle time backs the intuition: dynamic and zigzag both generate a shorter ideal time than circular, therefore that is the number most shops will plan to, but when looking at the machine data it becomes clear that a shorter ideal time shouldn’t be the deciding factor.

What the machine data says

Same part, same tooling, same three strategies, analysed against the Ecospeed kinematics inside CATIA:

Two things become clear immediately:

The first is the size of the gap. Each strategy takes between three and six times longer than CAM first estimates. If you are basing decisions on ideal cycle time, you are using data that isn’t possible to achieve.

The second is the reversal. Without kinematic influence, dynamic beats circular, exactly as the theory says it should. With the platform in the loop, the order flips: the tangential circular path comes out over a minute quicker than dynamic. The heat map shows why. The complex geometry across the dynamic cusp is hard for the platform to follow at programmed feed, so you get feedrate reduction due to complex geometrical features, all the way down the pocket. Dynamic holds the ideal chip load fine, but then spends a quarter of its efficiency not moving.

Achieved feedrate across the toolpath using the Circular strategy, green for optimal and red for poor, with a predicted actual time of 4m 40s at an efficiency of 48%

Achieved feedrate across the toolpath using the Dynamic strategy, green for optimal and red for poor, with a predicted actual time of 5m 39s at an efficiency of 25%

Achieved feedrate across the toolpath using the Zigzag, green for optimal and red for poor, with a predicted actual time of 3m 26s at an efficiency of 40%

Repositioning moves were tested too. Swapping the default “stay on the bottom and traverse at machining feed” behaviour for in-level approach and retract macros at rapid traverse gave a small gain. Useful to know, but not a step change, whereas the strategy choice is where the time saving is.

The fastest strategy, the Zigzag, is the one to reject

Zigzag is comfortably quickest, at 3m 26s, but before it goes near the shop, we need to look at the feedrate plot.

Circular holds feed across the toolpath with the expected dips. Dynamic drops hard but keeps a cut velocity throughout. Zigzag hits zero feed at every reversal, in contact with material, as it flips between climb and conventional.

Feedrate against time. The zigzag plot drops to the axis at every reversal, with the tool still in cut. 

Zero feed in cut means the tool is still turning but no longer advancing into the material, so it stops cutting and starts rubbing. In aluminium that means friction heat with no chip to carry it away, and near-zero chip thickness is exactly the condition that promotes built-up edge, which chips the carbide when it breaks away. It can also leave a mark: cutting force falls with the feed, the tool springs back toward nominal, and the wall carries a step where the deflection recovers. 

So circular is the decision: the slowest of the three on paper but still the one to run, because it is the only one of the three that is both quick enough and secure. That is a judgement a good programmer makes on instinct, but the difference is that this judgement is made on evidence, before the part is on the table.

However, this is not a general rule that circular beats dynamic, but instead the answer for this feature type on this platform. Put the same three part programs against a different spindle and drive configuration and the ranking can move again. The value is not a rule of thumb, but the right answer for the specific part and machine in front of you, without needing to cut to find it.

On a machine running continuously, every minute spent proving out a strategy is a minute of production wasted. Running these three options on the machine would have meant three setups, three billets and close to fourteen minutes of cutting, and it would have put a dwelling tool into an aerostructure bracket to establish something the feedrate plot showed in a few seconds at a desk. The cheapest place to be wrong about a toolpath is in CAM. The most expensive is on the Ecospeed, and that is where problems are currently still found out.

The tolerance check

With the strategy chosen and the parameters built to a feed the machine can manage, one variable to check before rollout is the tolerance. 10 microns against 100 microns on the full roughing program: 11m 01s against 11m 38s. Ideal time and efficiency unchanged. No meaningful difference, which tells you the toolpath geometry is the limiting factor here, not the platform tolerance setting.

There is a related trap worth knowing in CATIA. Because the path generator reworks the progression through the pocket when tolerance changes, a finer CAM tolerance can produce a shorter overall path and therefore a faster cycle, which is the opposite of what’s expected. Tolerance and cycle time are not the simple trade-offs they are usually treated as, and so this check is crucial.

Ten microns (top image) against one hundred microns (bottom image), with everything else held constant. 

What changed: the decision moved off the machine

Nothing in the above required a machine, a trial cut, a scrapped billet or a week of proving out on the shop floor. It ran inside CATIA V5-6, on the programmer’s own workstation, in five clicks per analysis, against a model of the specific platform the job is going to.

That is what DigitalCNC does. It is deterministic physics-based prediction of machine behaviour, built natively into CATIA V5, NX and Mastercam, out of a decade of five-axis research at the University of Sheffield AMRC. It does not read live controller data and it does not require specialist on-machine CNC training. It tells you, at the point of programming, what your part program will actually do on your machine.

The analysis runs in the CAM session, against a model of the machine the job is going to. 

The Ecospeed rewards programming that understands its kinematics. The gap between what it can do and what a CAM cycle time says it will do is the gap this software aims to close. It’s about where the decision gets made: the Ecospeed will tell you which of the three strategies was the right one eventually. It just tells you at the point where the billet is already on the table. DigitalCNC predicts this before a part is on the table.