Abstract

‘Data-Centric Engineering’ is a term used to reflect the new methods of collecting and analysing data that are now available to engineers. Combining sensing systems with scalable probabilistic (Bayesian) modelling can, in principle, facilitate transparent, replicable risk quantification. Decisions, such as whether to collect data or to mitigate risk, can then be optimised. However, the standards that direct engineering workflows are often not directly compatible with these approaches. In this talk I will argue that a paradigm shift away from ‘demonstrating safety’ and towards ‘quantifying reliability’ may be required for some industries to most effectively benefit from computational statistics and machine learning. I will provide examples from the field of structural integrity management of successes and challenges in the use of probabilistic (Bayesian) methods to support decision making under uncertainty.

Notes

• Personal website can be found here .