Cura time estimation using Gyroid infill

[CTotten 18]

I have been noticing over time that my S5 takes significantly longer to complete jobs than the Cura estimate says.  Over time, I have come to believe this is linked to using Gyroid infill.  The larger the print, or atleast the more infill, the bigger the difference.  For instance, I printed a large fixture this weekend.  Cura said it was 28 hours.  I came in 30 hours later, and the display said it had another 4.5 hours to go.  So, assuming that it finished when it said, that was an extra 6.5 hours, or extra 23%.  This did not make me happy, since I drove into work just to change over print jobs on a weekend.

So, today I tracked a smaller print job.  I was a large disc (about 8 inches in diameter), but was fairly thin (less than 1/2" thick).  It was a 7.5 hour job according to Cura.  Through the first 3 hours, when it was printing just the bottom layers (no infill), the time remaining tracked with the clock.  After it had completed the infill portions, the time remaining had added a total of 23 minutes.  After that point, so basically the last 2.5 hours or so, the time remaining again tracked with the clock.

 

I read somewhere that the jerk value for the Gyroid infill pattern was reduced in the code at some point (maybe from 20 to 8 if I remember right).  Is it possible that this change was not accounted for in the completion time computation, thus causing the time difference??? I plan to test large print over the next week, doing it once with gyroid then again with another infill.

 

While I understand it is an "estimate", for the most part is seems to track well, especially on smaller jobs that have little to no infill.  But people schedule things around these machines, particular if used in a business.  We try to maximize the up-time, which means deviations in the end times for jobs can cause big problems.  But when I am printing large jobs, that take 2-3 days to complete, and have multiple ones stacked up in the queue to be completed, it is frustrating when your 54 hour job suddenly become a 65 hour job...

[CTotten 18]

Yesterday I started a large print (2d, 17h, 38m estimated time) using gyroid infill at 25%.  As of this morning, so 16 hours in, I have already "lost" 1h 24m based on the time remaining showing on the display.  Once this one finishes, I have an identical print to start using tri-hex infill at 25% (2d, 16h, 45m estimated time). 

[CTotten 18]

I am about 60% through the print, and so far the estimation completion time has added almost 4 hours.  That is over the course of 40 hours.  I have a spreadsheet set up to track the estimation completion time.  Unfortunately, I did not watch the first few layers go down, so I didn't specifically log the non-infill layers at the beginning.  I am getting close to the point where the support material structure in the center cavity of the print (being done with Breakaway) is going to be done, meaning there will be significantly more gyroid infill per layer, so I am interested to see if that rate goes up.  The support material is being down in the default triangles pattern at 15%.

But 10% (so far) is a significant error.

Edited April 2, 2020 by CTotten

[CTotten 18]

Morning update.  I still have 9+ hours to go by the display, and so far the completion time has added over 7 hours to the print (from 2d 17h to over 3d).

And looking over the data, the slope (rate of time being added) appears to have went up overnight, which I thought would happen.  When I left yesterday afternoon, the print was reaching the top of the support area, so overnight the amount of infill per layer went up significantly.

Edited April 3, 2020 by CTotten

[CTotten 18]

First print, with Gyriod is complete.  Original estimate was 65.63 hours; it ended up taking just short of 74 hours to complete.  So that is about 12.5% more time than estimated.  In this case, it caused this print to run beyond the work day, so I either had to come in a night (which I did), or let the machine sit idle for 12 hours (come in a Saturday morning), or let the machine sit idle until Monday morning.  None of those are good options, especially when this job was "planned" to finish before lunch on Friday, allowing to start the next big print to run through the weekend.

[CTotten 18]

I am 12 hours into the 64.75 hour (Cura estimate) print of the same part, this time using Tri-Hex infill at 25%, and the estimated completion time is holding (in fact, it is now scheduled to end 5 minutes earlier).

Edited April 4, 2020 by CTotten

[CTotten 18]

As of this morning, I am about 90% through the second print (tri-hex infill), and the estimated finish time is tracking very well.  In fact, the estimated finish time says it will end 13 minutes earlier than originally predicted. 

[SandervG 1,521]

Thank you for sharing your updates with us @CTotten! Looks like the gyroid infill might have something to do with the inaccurate estimated print time, does it not? 

[CTotten 18]

10 minutes ago, SandervG said:

Thank you for sharing your updates with us @CTotten! Looks like the gyroid infill might have something to do with the inaccurate estimated print time, does it not? 

That is my working theory right now.  I read somewhere (can't remember where) that the jerk setting on the gyroid pattern was reduced at one point (from 20 to 8), so that could possibly explain a delta in the calculated time, if the time estimation still uses the old value.  But when I open up those settings in Cura, the default value doesn't seem to match up (shows 25 mm/s for infill jerk). 

 

Is this an issue for Ultimaker to look into, or should this go to GitHub?

 

I like using the gyroid infill pattern because it seems to run very smoothly compared to other infills.  But if I can't rely on the time estimate, then I have to stop using it because I rely on the time estimate in estimate work (cost, workflow, etc.).

[SandervG 1,521]

Just now, CTotten said:

Is this an issue for Ultimaker to look into, or should this go to GitHub?

 

 

I discussed this thread with a product expert before the weekend too, and he also suspected the infill but I didn't get a tangible reason. Developers monitor both, but to be sure we can tag @maht in here. @maht, do you know if this bug is already registered? 

[CTotten 18]

26 minutes ago, SandervG said:

 

I discussed this thread with a product expert before the weekend too, and he also suspected the infill but I didn't get a tangible reason. Developers monitor both, but to be sure we can tag @maht in here. @maht, do you know if this bug is already registered? 

Thanks.  Here are the two simple plots I made of the progress.  Whenever I could check on the machines, I tried to log the completion time estimate throughout the print.  I even backed out the time for any pauses (material changes) to keep it as consistent as possible.

The print job has a large cavity in the middle, that goes up to over half the height.  On the gyroid plot, the cavity (support area) was just about complete when I made the series of entries toward the end of the second day.  It seems like the "growth" in time got worse around that same point.  My assumption was that after the support area was done, the percentage of the print area getting the gyroid infill went up, causing the increase in the slope or the line.

And it is hard to see, but after the first 45 minutes of the gyroid print (before it started the infill pattern layers), the completion time estimate was constant.  Unfortunately it was late in the afternoon, so I didn't get a lot of data points during that part of the print.

Edited April 6, 2020 by CTotten

[Guest maht]

35 minutes ago, CTotten said:

 

Is this an issue for Ultimaker to look into, or should this go to GitHub?

 

Yes, if you consider it to be a bug, please open it on GitHub where Cura developers can see it. You can cite this discussion in there too.

[CTotten 18]

12 minutes ago, maht said:

 

Yes, if you consider it to be a bug, please open it on GitHub where Cura developers can see it. You can cite this discussion in there too.

Submitted.  Issue #7431.

[CTotten 18]

Not sure if @maht or @SandervG have checked up on the GitHub issue, but to summarize:

The gyriod infill pattern is "high-resolution" and the S5 Linux board is not keeping up, causing the printer to basically buffer during the print, causing unexpected slow downs.

So my workaround is to stop using the gyroid infill pattern; but it seems like some finger pointing saying the S5 hardware is not fast enough to keep up...

Edited April 10, 2020 by CTotten

[burtoogle 516]

You may wish to try one of my Cura builds as they support lower resolution versions of the TPMS infills (including gyroid). My releases can be installed alongside the UM releases without conflict. You can find them at https://www.dropbox.com/sh/s43vqzmi4d2bqe2/AAADdYdSu9iwcKa0Knqgurm4a?dl=0 Please read the README file there before using.