Machine Learning for Fibre Laser Materials Processing

Laser materials processing is a £10 billion global industry that continues to experience rapid growth of approximately 10% per year, with much of this driven by developments in fibre lasers. In the coming decade, the continual drive for improvements in efficiency and precision will see machine learning approaches fully integrated with fibre laser technology.
However, there are considerable challenges that are currently holding back progress, in particular the collection and processing of appropriate training data, the development of appropriate neural networks, and improving the robustness to ensure reliability sufficient for industrial applications.
This PhD project will involve working closely with technical staff at SPI Lasers in order to integrate machine learning approaches with novel fibre lasers used for materials processing. The fibres themselves are designed to collect the light that is produced from the material during machining into the fibre cladding, and then to an inline spectrometer that records the temporal spectral data.
This data contains a significant amount of structure that appears as "noise", and hence unidentifiable by human eye. The project hypothesis is that machine learning can be used to interpret this temporal spectral data, in order to enable real-time monitoring and control of fibre lasers for materials processing, hence driving innovations in precision and efficiency.
In addition, state-of-the-art developments in reinforcement learning will be applied in order to explore the potential for computerised discovery of novel manufacturing approaches. This project will involve the combination of experimental, theoretical, and machine learning approaches.