Problem statement
We have many 3D printers and occasionally there will be a problem with the build job which is not identified until after the job is complete. These jobs can run 24 hours or more which leads to a lot of lost production.
General Solution Idea
Ideally there should be a process to identify when there’s a problem as soon as it occurs. As it turns out the equipment takes a picture of the work surface after every layer is complete and then store it on the local machine.
We’re working on a method to automatically export those images, either to a network share, database, Snowflake, etc. for analysis.
Tulip product suggestion
What if within automations we could use Tulip vision to process the images automatically in the background and then alert if there is a problem found?
Long term we could even see about stopping the job if a problem is found once enough trust is built up.
Hi @Richard-SNN,
I’m not familiar with the equipment that you’re using but I have a few ideas here. I hope that they can help you.
If you could save each image to table in Tulip. You can use an automation that triggers off of the new row add or update to the image field in that table. The automation could use image over connectors to send to a vision model (if it’s hosted elsewhere) and could alert based off the return value of that connector.
In addition, we’re working to integrate one of our vision partners into the platform. Once you have a model trained and deployed, you could use automations to alert folks if there is a problem found.
I’m curious to get your thoughts here.
Sincerely,
Jake
Yeah, job number one is getting access to the images. They natively support MQTT so I’m hoping I can get them published. They’re not big images so not a huge issue for the broker.
I’m not sure I want 1600 images from each job being stored in Tulip forever 
I may end up using Highbyte to pipe the images into Snowflake instead.
The actual solution for my situation is mostly a thought experiment for the time being, it just made me think about the possibility of using images you already have for Tulip Vison, but this may be a better fit for Landing.ai.
1 Like
Hi @Richard-SNN,
I think we have some work in flight that will support the Landing.ai use case.
Sincerely,
Jake
1 Like