Overview / Usage

Using the Movidius NCS on the UP Squared board to detect errors in a 3D print job and stop it with Octoprint so it can be resumed.

How to leverage the capabilities of the Movidius Neural Compute Stick (NCS) using an UP Squared development board to extend OctoPrint to detect and control a 3D printer to stop when an error is detected. A camera is installed via USB to the the UP Squared board and the Movidius NCS is used to accelerate the detection of print failure so the print job can be paused and the user can be alerted to resolve, saving time and money by reducing print loss.

Adapted from https://thespaghettidetective.com, the model was designed to be run in the cloud. Because of the acceleration capabilities of the NCS, it can be run at the edge.

Additional Project Info:
https://www.hackster.io/bltrobotics/3d-printer-ai-up-squared-movidius-octoprint-to-catch-errors

Methodology / Approach

This problem is better solved at the edge, being able to detect errors in manufacturing and production is often restricted by access to the cloud. This project is intended to understand the limits of what is achievable given the use of current state SBC and low priced accelerators.

The next phase of this research will drive the use of AI on manufacturing floor where direct cloud connectivity to run models will not be available.

In this project OpenCV/OpenVINO was utilized. Exploration using Caffe and Tensor Flow for training of print failure models. Edge computing is the desired implementation as factory line remote connectivity is not guaranteed.

Technologies Used

Hardware:

• UP2 Development Kit
• Movidius Neural Compute Stick

Software:

• OpenCV
• OpenVINO

Repository

https://github.com/TheSpaghettiDetective/TheSpaghettiDetective

Other links

• https://www.hackster.io/bltrobotics/3d-printer-ai-up-squared-movidius-octoprint-to-catch-errors

• OctoPrint Plugin Source (https://github.com/TheSpaghettiDetective/OctoPrint-TheSpaghettiDetective)