An artificial intelligence platform for design optimization and data analysis: application for fire and ventilation problems

Li, N 2019, An artificial intelligence platform for design optimization and data analysis: application for fire and ventilation problems, Doctor of Philosophy (PhD), Engineering, RMIT University.


Document type: Thesis
Collection: Theses

Attached Files
Name Description MIMEType Size
Li.pdf Thesis application/pdf 5.77MB
Title An artificial intelligence platform for design optimization and data analysis: application for fire and ventilation problems
Author(s) Li, N
Year 2019
Abstract This thesis focuses on the development of novel multi-objective software platforms to assist engineering design and investigation, especially for simulation-based indoor environment problems, which always involve multiple evaluation criteria. In addition, this thesis aims to develop new methods to reduce the computational cost associated with the design process.

In modern building design, engineers are constantly facing challenging to find an optimal design to maintain a high level of thermal comfort and indoor air quality for occupants while minimizing the system energy consumption. Over the past decades, several algorithms have been proposed and developed for optimizing the heating, ventilation and air conditioning (HVAC) system for indoor environment. Nevertheless, the majority of these optimization algorithms are focused on single objective optimization procedures and require a large training sample for surrogate modelling. For multi-objective HVAC design problems, previous studies introduced an arbitrary weighting factor to combine all design objectives into one single objective function. The near-optimal solutions were however sensitive to the chosen value of the weighting factor.

In another hand, the computational cost is very heavy in the computer-aided investigation process of reverse engineering problems. Computational Fluid Dynamics (CFD) aided fire investigation is one of the reverse engineering. With the significant growth of the world population, our cities are becoming more and more crowding. In this situation, any fire occurring would cause severe consequences, including property damage and human injuries or even deaths. In assessing the fire cause, the fire origin determination is a crucial step identifying the origin of fire outbreak and the sequential fire and smoke propagation. Traditionally, fire investigators relied upon the visible fire damages at the fire scene to determine the location of fire originated based on their own professional experience. The fire investigation process is however subject to the expert interpretation inherently embedded in the qualitative analyses.

In addition, we are living in an era of big data, where lots amount of data are generating every day, especially in engineering field. Traditional analysis methods are not suitable to handle large amount of data quickly and accurately. In contrast, new techniques such as machine learning are able to deal with big data and extract data features.

The main body of this thesis is composed of seven chapters, and the details of each chapter are as the followings:
The research background and a comprehensive literature review are described in the first two chapters where the research gaps found in the existing literatures are discussed. From Chapter 3 to Chapter 6, the main contributions of this research are demonstrated. In Chapter 3, a nondominated sorting-based particle swarm optimization (NSPSO) algorithm together with the Kriging method to perform optimization for the HVAC system design of a typical office room was developed. In addition, an adaptive sampling procedure was also introduced to enable the optimization platform to adjust the sampling point and resolution in constructing the training sample. Chapter 4 presents a Multi-fidelity Kriging algorithm to quantitatively determine the fire origin based on the soot deposition patterns predicted by the numerical simulations, which provides an unbiased and fast methodology to assist the fire investigation. A comprehensive multi-objective optimization platform of the ventilation system inside a typical high-speed train (HST) cabin is discussed in Chapter 5, where the NSPSO and the Multi-fidelity Kriging were combined together to reduce computational cost. Chapter 6 demonstrates a successful application of convolutional neural networks (CNN) in vegetation feature analysis to help cut powerline wildfire risk caused by vegetation conduction ignition. Finally, all the contributions in this research are summarised in Chapter 7.
Degree Doctor of Philosophy (PhD)
Institution RMIT University
School, Department or Centre Engineering
Subjects Interdisciplinary Engineering not elsewhere classified
Numerical Modelling and Mechanical Characterisation
Keyword(s) Multi-objective optimization
CFD
HVAC
NSPSO
Multi-fidelity Kriging
Reverse engineering solution
HST cabin
Versions
Version Filter Type
Access Statistics: 49 Abstract Views, 38 File Downloads  -  Detailed Statistics
Created: Thu, 19 Sep 2019, 13:11:19 EST by Adam Rivett
© 2014 RMIT Research Repository • Powered by Fez SoftwareContact us