THE USE OF NUMERICAL OPTMISATION TECHNIQUES IN COMPUTATIONAL FIRE ENGINEERING MODELS: A STUDY THROUGH EVACUATION MODELLING ANALYSES

Rodrigo Machado Tavares

2010

Abstract

Evacuation models have been playing an important function in the transition process from prescriptive fire safety codes to performance-based ones over the last three decades. In fact, such models became also useful tools in different tasks within fire safety engineering field, such as fire risks assessment and fire investigation. However, there are some difficulties in this process when using these models. For instance, during the evacuation modeling analysis, a common problem faced by fire safety engineers concerns the number of simulations which needs to be performed. In other terms, which fire designs (i.e., scenarios) should be investigated using the evacuation models? This type of question becomes more complex when specific issues such as the optimal positioning of exits within an arbitrarily structure needs to be addressed.

In the other hand, numerical optimisation techniques have been applied to a range of different fields such as structural analysis. These techniques have shown to be a powerful tool for designers, saving their time and consequently reducing costs during the process.

For this reason, the emphasis, throughout this study, is to develop a methodology that enables the optimisation of fire safety analysis of structural designs. In other words, the current research was primarily intended to demonstrate and develop this combination of fire engineering tools and techniques such as the Design of Experiments (DoE) and numerical optimisation techniques. For this purpose, a Computational Fire Engineering (CFE) tool combined with Numerical Optimisation Techniques and associated statistical methods (i.e., Design of Experiments (DoE) and Response Surface Models (RSM)) are used. The study is focused on evacuation modelling; nevertheless the methodology proposed here could equally be applied to CFD-based fire simulation tools. While the approach that has been developed is intended to be generally applicable, the techniques have been explored and demonstrated using the buildingEXODUS computational package. This fire engineering simulation tool is used worldwide, to improve the fire safety in building designs.

This study therefore intended, besides to develop a numerical methodology to allow the efficient optimisation of fire safety aspects of structural designs, to understand how the core variables impact the evacuation efficiency.

For instance, a common problem faced by fire safety engineers, in the field of evacuation analysis, is the optimal positioning of exits within an arbitrarily complex structure. This problem is usually addressed through time consuming and expensive trial and error. While a solution is usually found, to this problem, it is seldom the optimal solution, resulting in a compromise in building performance and safety.

The methodology explored in this thesis, as applied to CFE, was initially based around a relatively small set of physical variables. This approach evolved and was subsequently expanded to include more complex behavioural, procedural and environmental parameters. The methodology has also been further developed and applied to evacuation simulation.

This integrated approach is intended to help fire safety engineers and designers to develop optimal designs (i.e., safe designs) in a optimized manner. In reality, this was the motivation of this study: to introduce numerical optimisation techniques and associated concepts, well known within the operational research field, as an approach for a more efficient and systematic procedure when developing and/or improving fire safety designs.

Post comparisons between the outputs obtained, using these different DoE techniques, have been also performed in order to analyze which technique is most suitable for the optimisation of structural designs.

This thesis describes a number of analyses (of a variety of structural designs) that have been used to calibrate the optimisation technique. This included the use of the buildingEXODUS simulation tool, as mentioned previously, followed by the application of a variety of optimisation techniques (both gradient and non-gradient based numerical optimisation techniques) as well as different types of DoE (such as Latin Hypercube, Central Composite Design (CCD) and also a Random approach) in order to improve the designs according to a number of different variables. These variables have initially included physical modifications to the geometry.

The proposed methodological approach developed in this thesis is demonstrated on a variety of practical problems. These problems are represented by 4 case studies which vary from complexity to the nature of the variables. These case studies involved both types of problems, namely: unconstrained and constrained.

The results obtained have shown to be satisfactory, i.e., global minima and local minima closest to the global minima region were found. For all the cases, a gradient-based algorithm (i.e., the Fletcher-Reeves numerical optimisation technique) and non gradient-based algorithm (i.e., the Particle Swarm Optimisation numerical optimisation technique) were used to find the optimal solution. And as mentioned before, different DoE techniques were also applied.

The analysis revealed that this methodology seems to be a very powerful tool for evacuation modelling analysis.

This systematic methodology to efficiently optimise evacuation safety aspects of structural designs should be also extended to more complex designs, such as larger enclosures and open spaces.

This methodology is also intended to be applied to problems found in the field of fire simulation, such as: the sizing and positioning of smoke extraction vents and the modelling of cable fires.