Where AI fits today
Most inspection sensors produce pictures: height maps, thermal images, camera frames. That makes defect detection a computer-vision problem, and convolutional neural networks have made vision very good, near-human or better in industrial inspection. In composites it isn't at production scale yet.
Predict · at 1:29 in the video
The network designs exist. The computers exist. So what's holding AI inspection back in fibre placement?
Labelled data. A network learns from examples, and each example needs an expert to mark every defect on it. Say you need five thousand labelled images, at three minutes each.
Automated Fiber Placement: Status, Challenges, and Evolution, Ramy Harik & Alex Brasington.
- Chapter 6, section 2.1.A
- Chapter 6, section 2.1.B
Simulations in this lesson are our own. Where the numbers are invented to show a mechanism, the video says so on screen.
