How to train your laser (invited)

Abstract

Lasers have transformed manufacturing, and they are rapidly becoming the standard industrial tool for cutting, drilling, welding, and even 3D printing. However, lasers used in manufacturing generally follow a predetermined set of instructions for each task.

For example, when machining through a sheet metal, the current approach is to over-machine the metal (i.e. use too many photons), to ensure the metal is machined completely for all natural variations in its thickness. Likewise for recycling tasks, such as laser cleaning of materials for the removal of paint or rust on a metal surface, the thickness of the contaminant layer may be unknown and varying across the material, and again, the material will need to be over-cleaned (i.e. too many photons).

As we move towards Smart Factories and the Circular Economy in Manufacturing, allowing the laser system to make manufacturing decisions without human input will be critical for achieving 100% photon efficiency, and for unlocking further improvements in productivity, reliability, and accuracy.

At the University of Southampton in the UK, we have been applying the latest in deep learning research to develop the capability for lasers that can self-optimise and self-correct for errors, whilst also being able to predict and visualise the outcome of laser machining under different parameters, all in real-time.
In this talk, I will present some highlights from our recent work in this area and discuss implications for the future of artificial intelligence in manufacturing.

More information

Identifiers

ORCID for Benjamin Mills: orcid.org/0000-0002-1784-1012

ORCID for James Grant-Jacob: orcid.org/0000-0002-4270-4247

ORCID for Michael N. Zervas: orcid.org/0000-0002-0651-4059

Catalogue record

Export record