Mechanical design in the age of AI (with a little bit of help from stochastic methods)
Antoine Jerusalem (University of Oxford)
Thursday 19th February 14:00-15:00
Maths 311B
Abstract
Complex mechanical design endeavours often leverage computational methods to circumvent costly experimental programmes. However, the need to consider a wide range of conditions and system parameters often impedes the use of advanced computationally expensive simulations on a large scale. Instead, in a seemingly ever-increasing number of cases, AI has shown promises that seem to bypass the need altogether of mechanistic understanding. Unfortunately, the sacrifice of causation for correlation comes at a price and – as we all know – AI is not without limitations. In this talk, we propose a range of methods allowing for the consideration of complex expensive systems at scale, while retaining mechanical insights. We first present the so-called Galerkin Stochastic Finite Element method, allowing for the simultaneous simulation of a range of systems. After highlighting the advantages and drawbacks of this method, we then turn our attention towards the use of coupled mechanics/machine learning schemes in optimisation frameworks to alleviate the cost of the underlying problem. We present these methods in a variety of design problems; both in Materials Sciences with examples in metamaterials and composite materials, and in Engineering applications with examples in aeronautics and bioengineering designs.
Add to your calendar
Download event information as iCalendar file (only this event)