Application of an intelligent model developed from experimental data to building design for fire safety

Research output: Journal Publications and Reviews (RGC: 21, 22, 62)21_Publication in refereed journalpeer-review

4 Scopus Citations
View graph of relations

Detail(s)

Original languageEnglish
Pages (from-to)493-506
Journal / PublicationStochastic Environmental Research and Risk Assessment
Volume23
Issue number4 SPEC. ISS.
Publication statusPublished - May 2009

Abstract

Computational fluid dynamics (CFD) techniques are widely adopted to simulate the behavior of fire. However, CFD suffers from the shortcoming of requiring extensive computer storage and a lengthy computational time. In practical applications, although comprehensive field information on velocities, temperatures, pressure, and the fractions of different constitutes can be obtained from CFD simulations, the user may only be interested in few important parameters that index the performance of a compartment design in the event of a fire. The height of the thermal interface (HTI) is one such key index, and refers to the average height above floor level inside a fire compartment at which the temperature gradient is highest. In practice, a fire compartment is considered untenable when the HTI drops below the respiratory level of the occupants, and in optimizing the design of a fire system, another set of design parameters (e.g., the width of the door opening) must be examined if the HTI of a fire compartment design is evaluated by CFD as being too low. This trial and error exercise then continues until a close to optimum set of design parameters is achieved. This approach is theoretically feasible, but requires lengthy computational time. This paper proposes the application of an Artificial Neural Network (ANN) approach as a fast alternative to CFD models to simulate the behavior of a compartment fire. A novel ANN model named GRNNFA has been specially developed for fire studies. It is a hybrid ANN model that combines the General Regression Neural Network (GRNN) and Fuzzy ART (FA). The GRNNFA model features a network structure that grows incrementally, stable learning, and the absence of the noise embedded in experimental fire data. It has been employed to establish a system response surface based on the training samples collected from a full-scale experiment on compartment fire. However, as the available training samples may not be sufficient to describe the behavior of all systems, and especially those involving fire data, this paper proposes that extra knowledge be acquired from human experts. Human expert network training has thus been developed to remedy established system response surface problems. After transforming the remedied system response surface to the problem domain, a Genetic Algorithm (GA) is applied to evaluate the close to optimum set of design parameters. © Springer-Verlag 2008.

Research Area(s)

  • Artificial Neural Network, Building design, Compartment fire, Computational fluid dynamics, Genetic algorithm, Optimization