Goal:

To create a computer-vision-based 3D printer monitoring system that works with Quinly in order to make a 3D printer smart enough to manage itself with little to no human intervention. The machine learning model created will aid Quinly in aborting and restarting irrecoverable prints, give warnings about minor print defects and log reoccurring artifacts of specific printers for preventative maintenance.

Current Design:

Currently, the model is continuously being improved by training on the new data that is being acquired from a set of "failure" 3D printers I set up. A model's performance relies heavily on having a copious amount of good data. Due to them being so data-hungry, I have developed an almost completely automated data collection system by writing some Bash and Python scripts to work with Quinly.

QuinlyVision Update:

Is 360° Failure Detection Useful? Try it Out and Let Me Know - Details in comments

Untitled

cube00000817_jpg.rf.e18092e83a02c12fcb16495e04b5c87e.jpg

detect_IP74_2021-07-02_2237_0026_jpg.rf.d7545306b9d00724c4d85db1bc3f9eb4.jpg

string00000247_jpg.rf.a7271355e8b2e2ffaf19c3ab90711515.jpg

Details in LinkedIn & Medium Articles

As a high-level overview of this entire project, I have created a five-part series of articles while working at 3DQue for community engagement. The series goes into detail about machine learning in general, classification, data collection, annotation and training/validation.

The Path to Autonomous 3D Printing

Classification: How Many Ways Can a Print Fail?

Data Acquisition: Design for Failure

Annotation: Teaching a Machine

Training, Evaluating and Implementing an Object Detection Model: Finally, The Interesting Part!

Live Stream: