The use of machine learning techniques for the optimisation of experimental laser machining processes.

Abstract

In recent decades, laser machining has transformed manufacturing. Pulsed laser machining, using short and ultra-short pulse durations, has become a standard technique for the fabrication of features on micro and nano scales. At these power and length scales, many interesting non-linear effects occur, making accurate modelling of these processes challenging without introducing simplifications and assumptions. This thesis presents an alternative to traditional modelling techniques, namely using machine learning, in addition to predictive regression, full experimental modelling is conducted. A broad set of machine learning techniques has been applied to the field of laser machining. This includes both analytical and generative modelling techniques, showing that neural networks can achieve very high levels of accuracy. These techniques are also extended to situations that would be impossible to model, where knowledge of the system is very limited. In these situations, the presented methods were still able to achieve high accuracy and be used for data optimisation tasks. In addition, generative methods are applied for the prediction of depth profiles from a variety of machining parameters. It is also shown that generative networks could complement the experimental process by reproducing the results found. This presents opportunities both for increased levels of data collection and for easy presentation and investigation of new ideas before committing to a full experimental process.

More information

Identifiers

ORCID for Michael David Tom McDonnell: orcid.org/0000-0003-4308-1165

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

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

ORCID for Matthew Praeger: orcid.org/0000-0002-5814-6155

ORCID for Robert Eason: orcid.org/0000-0001-9704-2204

Catalogue record

Export record