Rapid simulation of building physical processes using the coarse-grain method

Abstract

The paper is concerned with an artificial neural network (ANN) based coarse-grain method (CGM) for rapid simulation of physical processes in buildings, such transient heat-flow, fire propagation, and the circulation/dispersal of fumes or pathogens in a building. A primary objective of the approach is to facilitate real-time and accelerated-time visualization of such processes. Specifically, this paper reports on recent developments to the CGM approach for modeling transient heat-flow in buildings. CGM was originally developed as an alternative to conventional fine-grain modeling techniques (such as the Finite Difference Method (FDM) and Finite Element Method (FEM)) to radically increase simulation speed and to ease the tasks of model development and experimentation. An earlier study has shown that CGM can provide reasonably accurate simulations at a processing speed several orders of magnitude faster than FDM or FEM. This paper describes and demonstrates two refinements to the CGM approach: (i) the use of a hybrid linear regression model with an ANN to represent each coarse-grain modeling element (the hybridization of the ANN halved its complexity and doubled its processing speed); and (ii) a linear calibration of the ANN-based coarse-grain modeling elements to account for an observed positive bias in their predictions (the calibration improved the accuracy of the CGM model to a level comparable with FEM. The improved approach is demonstrated in a study of a two-dimensional model of a bay in a research building located at the University of Florida.