GPGPU enabled CFD simulation for fully coupled fire and evacuation modelling

Traditionally fire and evacuation models are run independently of one another to ascertain two key building safety parameters: ASET (available safe evacuation time) determined by the fire spread; RSET (required safe exit time) determined by the evacuation model. A building can be deemed to be safe if RSET <ASET. A more advanced method is to couple the models together to give a dynamic fire environment superimposed on the evacuation. This has typically been achieved using a one way couple where the fire is predetermined prior to the evacuation. A more advanced two-way couple can be used in scenarios, where the evacuation behaviour effects the fire environment, e.g. opening/closing doors by agents, extinguishment of fire by agents etc. Presently the time taken to run these simulations is dominated by the CFD fire model. The problem with two-way coupling is that every change requires a recalculated CFD environment and as the evacuation simulation is based on Monte Carlo methods this leads to multiple calculations to achieve statistically significant results. A complete GPGPU implementation (solver, coefficients and other dependent variables) of the CFD based fire model has been developed which leads to a substantial execution speed-up. Many previously reported implementations are limited to the matrix solver and are thus limited to the speed of the host calculating the coefficients and thereby returning modest overall speedups. The speed-up gained through the parallel implementation enables the practical use of the two way coupling. The key point of the two way coupling is that the agents in the evacuation model dictates the way the CFD code calculates its values. By letting the agents directly interact with the geometry it eliminates the element of making assumptions when events happen and drastically reduces the number of required simulation runs for all permutations.

Name dc:type.qualificationname

phd

Level dc:type.qualificationlevel

doctoral

Grantor dc:publisher.institution

University of Greenwich

Year dc:date.issued

2015

Author dc:creator

• Sauter, Markus

Advisors dc:contributor.advisor

• Galea, Edwin
• Grandison, Angus

• QA Mathematics

Language dc:language

en