Agent Based Modelling in Practical Engineering Applications

The proposed description is straight deriving from the post advertisement; see:

Why people move the way they do when crossing a bridge? Why they evacuate (e.g., in a fire or earthquake scenario) buildings creating certain outflow patterns? Why they get in step or out of step when just walking side by side or the one behind the other? Why stampede (or crowd crush) tragedies still happen and finally how much behaviour between pedestrians approaches behaviour between car drivers?

Agent based, or crowd dynamics, simulations try to capture the underlying mechanisms of all the above, yet not always accurately. There are a number of validation studies towards trying to make such simulations more realistic but these tend to refer to quite limited scenarios with little impact to our everyday lives. The current PhD project would aim to:

i) Optimise current agent based simulation approaches using latest numerical (e.g., machine learning) techniques for establishing the interaction component the individual receives when moving inside a crowd. As a starting point, validation would take place in real-world crowd scenarios, such of this of people crossing a bridge.

ii) Blend the optimised tool(s) reproducing more realistic crowd motion and interactions between individuals, with standard design software. How good are the structural design approaches currently practiced, and what differences emerge from this blending operation are the novel research question targets.

iii) Expand the above agent based simulation approaches to topics like traffic routing (micro-simulation models in SUMO) and designing for more efficient transport infrastructure or more compatible with future trends like that of autonomous driving.

The research would be a joint effort between the School of Computing (Dr Evangelos Pournaras) and the School of Civil Engineering (Dr Nikolaos Nikitas) and would have a hybrid numerical and physical testing aspect. Off-the-shelve software like Oasys MassMotion would be used in this research and there will be close interaction with practicing engineers using agent based simulations and design software.