Increasing Processing Speed for Interactive Real-Time Simulation Environments Ian Flood University of Florida Gainesville, FL 32611-5703 flood@ufl.edu Raja R.A. Issa University of Florida Gainesville, FL 32611-5703 raymond-issa@ufl.edu Caesar Abi Shedid Florida International University Miami, FL 33174 cabi@fiu.edu Abstract The paper is concerned with simulating physical processes in buildings (such as heat transfer, fire propagation, and dynamic stress response within a structure) at real-time and accelerated-time speeds, so that they may be used in an interactive 4 dimensional visualization environment. Initially, it is demonstrated that conventional computing techniques will not be able to achieve satisfactory processing speeds within our lifetime, since their rate of progress is overwhelmed by the size of these models. Alternative computing techniques are then ex- plored as means of achieving the required processing speed, including the use of parallel computers and direct mapping models. It is shown that the only current technology that has any potential of resolving this problem is the coarse-grain method (CGM). CGM is briefly introduced and a summary of its performance capabilities are presented for the problem of modeling transient heat-flow in buildings. The paper concludes with an identifica- tion of where future research needs to be focused.. Keywords Accelerated Time Simulation; Artificial Neural Networks; Coarse-grain modeling; Parallel Computing; Real- Time Simulation; Visualization 1. INTRODUCTION Virtual reality and related environments usually require the representation of dynamic visual information in real- time or accelerated time. Today’s computers can readily handle the processing involved with modeling such envi- ronments provided the image only has to adjust in re- sponse to a change of position/orientation of the viewer, or in response to similar simple shifts in the orientation of the environment or its objects. A problem arises, how- ever, where changes in the environment must be gener- ated in real-time or faster using numeric simulation tech- niques. Unfortunately, there are many situations within the A/E/C disciplines where this is the case. Examples include modeling fire and smoke propagation through buildings (NIST, 2006), modeling transient heat-flow through structures (US Department of Energy, 2006), and visualizing the distribution of stresses throughout a struc- ture (UC Berkley, 2006). In all cases, visualizing these phenomena in a virtual environment would help deter- mine the effectiveness of alternative designs, evaluate the impact on building performance of alternative design decisions and, in the case of fire propagation, help emer- gency crews determine appropriate strategies for fire fighting and evacuation purposes. Typically, a three- dimensional simulation of such problems can take several days or even weeks to process, making interactive visu- alization impossible. Even problems where visualization is required to occur at less than real-time speeds (such as the propagation of blast waves through a building) can- not be processed fast enough since the simulation oper- ates orders of magnitude slower than the required view- ing speed. Numeric models are inherently expensive in terms of computer processing time (see, for example, Chen et al., 2000). They require the system under investigation to be represented by a very large number of elements, each representing the state of the system at a discrete location in the modeled space. The state of the model is advanced in small time increments by resolving a set of driving equations for each element, derived from known physical laws. The speed of execution of a simulation is thus de- pendent on the number of elements in the model and the size of the time steps. Generally, the accuracy of the simulation improves as the sizes of the spatial elements and the time increment are decreased. However, reduc- ing the size of the spatial elements results in an increase in their total number, and consequently an increase in the processing time (this is a geometric rate of increase for models operating in two or three spatial dimensions). Similarly, reducing the size of the time increments in- creases the number that must be executed in order to ad- vance the model over a given period of time, further in- creasing the amount of processing to be executed. As a practical example, consider the problem of model- ing the flow of heat through a building. Research indi- cates (Abi-Shdid, 2005) that this type of model can