embedded ai at the edge

real-time defect detection with NVIDIA jetson

Discussing industrial 3D printing with Joel Telling

I met up with Joel Telling (the 3D Printing Nerd) on the floor at Formnext 2022 to discuss the cutting edge of industrial additive manufacturing.

In this interview, we dive deep into the work I led at Ai Build, specifically focusing on how we used computer vision and machine learning to solve one of the biggest challenges in large-format 3D printing: real-time defect detection.

Xavier Malina discussing defect detection at Formnext 2022

Discussing industrial 3D printing at Formnext 2022

key topics

Large-format 3D printing with industrial robots presents unique challenges. A failed part doesn't just waste an afternoon—it can mean scrapping hundreds of kilos of expensive material and days of machine time.

During my time at Ai Build, I led the development of systems designed to catch failures before they become catastrophic. Here's what I built:

real-time defect detection

We developed a computer vision pipeline using cameras mounted directly on the extruder nozzle s well as around the print bed. This enabled detection of defects at the moment of deposition—not after a layer was complete, but within milliseconds of occurrence.

the "digital twin"

A key challenge in defect detection is distinguishing between intentional behavior (like travel moves) and actual failures. Our solution was to leverage the toolpath data to know when 'no extrusion' events were expected. 

multi-sensor fusion

Beyond cameras, we researched integrating a suite of non-vision sensors to improve defect detection accuracy from temperature sensors to microphones.