Recorded at the Young Researchers Conference 2020.
Over recent years, the use of explosives for malicious attacks has become increasingly more common. The scale of these attacks varies. There are larger scale attacks such as in Oslo (2011) and Oklahoma City (1995) and smaller scale, more targeted attacks such as at the Manchester arena (2017) and the 7/7 bombings in 2005.
Explosions from terrorist attacks are killing people, and as engineers we have a duty to design and build resilient structures that can provide adequate security against a threat to life. To do this, it is crucial we have a good understanding of the pressure load following a high explosive detonation.
We currently have physics-based solvers (computational fluid dynamics codes) that can provide highly accurate solutions but will take several hours to days to complete. We also have very simplistic methods that allow us to do risk-based design approaches, but these are not accurate and can only give very vague approximations. They do not give the level of confidence required in the context of saving human lives.
This project proposes a data-driven modelling approach of explosive events through novel machine learning techniques. It sets out to answer the question: can we find a way to match the accuracy of physics-based models but in a way that runs as quickly as these approximate methods, such that we can perform risk-based engineering with far more accurate information?
Understanding loading distributions from explosive events