Abstract
The paper discusses the first stage in a study concerned with the development and evaluation of a fast, accurate, and versatile method of simulating blast wave propagation within complex built environments (comprising any configuration of structures such as buildings and blast barriers). Existing methods of modeling the propagation of a blast wave each fail in at least one of the speed, accuracy or versatility requirements. Conventional computational fluid dynamic (CFD) simulations, for example, typically take several days to complete a single run. An alternative, novel method proposed here is to use a simulation approach implemented within a coarse spatial and time framework, where the mesh elements and time steps are orders of magnitude larger than those used in conventional CFD simulations. The approach requires the use of intelligent modeling techniques to capture the behavior of elements at the coarse level. Specifically hybrids of artificial neural networks and multivariate linear predictors are used to capture the characteristic behavior of a blast wave at different locations within a built environment. These can then be assembled to represent the configuration of the actual environment under investigation. The hybrid modules are trained using data generated specifically for the task from CFD simulation runs. The novelty of the approach is in the use of these hybrid modules to learn appropriate driving equations that operate at a coarse level in space and time, and to execute the simulation within a discrete-event (as opposed to continuous) simulation environment. As a proof of concept, this study focuses on one-dimensional simulation modeling. The results demonstrate the feasibility of the approach in terms of its ability to replicate the predictions of CFD simulations very closely and to complete a simulation run orders of magnitude faster than a CFD simulation (within a few seconds as opposed to hours or days).