Completed: Sustainable advanced manufacturing via machine learning-assisted exploitation of sensing and data infrastructure

In collaboration with Robinson Brothers

Machine learning (ML) and data science underpin research disciplines that are prominent in numerous manufacturing sectors, and are already achieving measurable impacts in high value, low volume sector such as aerospace and automotive. The applications supported through research in ML and data science are typically around process modelling, monitoring, optimisation, control and decision making.

This project aimed to develop integral data acquisition and analytics frameworks that leverage expert process knowledge with ML techniques to generate informed data infrastructure design guidelines that will accelerate uptake of ML in the foundation industries.

This project demonstrated that regardless of the amount of data that is captured, data science and ML tools can help improve the analysis and optimisation of manufacturing processes within foundation industries. While traditionally manufacturing within these sectors tend to operate in data-scarce regimes, data driven approaches that leverage first-principles understanding of the physical processes can enhance the analytical approach.

The projects has proposed a mathematically rigorous quantitative framework that enables stakeholders to leverage their data with the following tools and methods:

• A stochastic model that can adapt to different levels of uncertainty, complexity, and data availability regimes.
• A data acquisition framework that also produces information metrics describing the evidence captured by different data sources about the parameters determining process performance.
• An ML framework that adapts to the amount of available data and incorporates the evidence provided by the information metrics into the task.

Dr. Iñaki Esnaola

University of Sheffield

esnola@sheffield.ac.uk