This is our new labeller. Beckhoff C6030 IPC, TwinCAT 3.1 4024, 30 EtherCAT slaves, six AX5000 axes and the rest EL terminals. Our commissioning engineer put the EtherCAT and NC task at 250 us because the master offered it and the axes looked smoother on the bench.
At the customer the machine jitters, and it's worse when the HMI is busy. The NC task Exceed counter climbs a few hundred an hour and the axes pick up a small position lag that the customer can see on the label.
I've tried 500 us, less jitter but it still exceeds, so no luck there, and we turned off the Windows update service and the antivirus IT had put on. The vendor reckons we need a faster IPC, the customer reckons he wants a smaller invoice. Do we really need one, or was a 250 us task on 30 slaves never going to hold on this IPC anyway?
What does the motion actually need? On Kinetix I run most packaging axes at 2 ms coarse update and nobody can tell. What's the fastest thing on that labeller, a registration mark, and what's its tolerance? 250 us is a number the tool offered you, not a number the label asked for.
> what is its tolerance
Registration on the label is 0.2 mm at 60 m removed link The marks come in on an EL1252 timestamped input, so I'm not sure the cycle matters much for that part. The cam follower on the applicator is the tightest thing we've got. And honestly, 250 us came straight off the drop down.
Then start at 1 ms and prove it needs less. At 250 us with 30 slaves you're spending a fair chunk of the cycle just getting the frame on the wire, plus the NC calculations, plus whatever Windows helps itself to on a shared core. Two things to read before you touch anything. Real-Time tab, Settings, is the TwinCAT core isolated or shared with Windows? Shared is the default. And on the EtherCAT master, Online, the frame time. Post both.
Shared. One core of four given to TwinCAT and Windows allowed on it as well. Frame time reads about 90 us for the cyclic frame. I set 1 ms, isolated the core, Set on target, rebooted. Exceed counter stopped dead. But now the customer says the applicator looks worse, slower to catch the mark.
The exceeds were the machine. The applicator is a tuning question that 250 us was hiding. That cam was tuned at 250 us and it's now following the master on the same gain at four times the interval, so it might be lagging a fraction more. Either retune the applicator axis Kv at the new cycle, or run only the NC task at 500 us on the isolated core and leave the PLC task at 1 ms, they don't have to match. Keep the frame time well under half the cycle. There's a writeup on the cycle time side here: https://plctr.com/utilizing-ethercat-in-high-performance-plc-systems/
Ended up with the NC task at 500 us, PLC at 1 ms, core isolated. Retuned the applicator axis Kv and the customer signed off the registration. Exceed counter 0 over a week, Real-Time CPU around 45 percent.
So it was 250 us picked because the tool allowed it, on a core Windows was allowed to use. The HMI stole the core and the cycle exceeded. A cycle the machine can actually justify, 500 us for the NC, and an isolated core sorted it. No bigger IPC.
The bit I own up to is that 250 us is our commissioning template, every labeller we've shipped has got it. Three other machines run it without a word of complaint and I don't know why yet. Thanks tobik23, mattr.
The other three haven't had antivirus put on them yet. Give it time.