Artificial Neutral Networks: Analysis of Architectural Spaces

Amine Benoudjit

Details of proposed research

The use of Artificial Neural Networks (ANNs) in architecture is relatively new. The following is a short review where they have been used to tackle some architectural problems: M. J. O'Neill (1992) used ANN to develop a way finding program. I. Petrovic and I. Svetle (1994) used ANN to build a program (PDP-AAM) that generates prefabricated housed out of a list of semantic attributes. C. Derix (2001) used ANN of type Self Organised Map (SOM) for the generation of implicit 3D surfaces that represent the antipoetic model of an architectural space. L. Diappi et al (2002), A. G. Yeh and X. Li (2003), used ANNs with Cellular Automata (CA) for the analysis and the forecasting of urban growth. M. A. Benoudjit (2004) used ANN of type SOM for the analysis and the classification of architectural spaces.

The present PhD aims to build a working prototype (an architectural space benchmark) for the analysis of existing and projected architectural spaces. The working prototype will embedded ANNs for extraction of spatial qualities out of different spatial configuration, and then derived from those qualities a classification map that shows similarities and differences between the spatial configurations under study.

The originalities of the proposed research subject are: (1) The present PhD will be the first time where ANNs are used for the analysis of spatial configurations, and it will be the first time where ANNs are used for the classification of spatial configuration. Provided with such a tool, an architect could for instance tell if an architectural project is similar to another one or completely different from it. (2) Precedent spatial analyses were always based on a schematization of existing spaces through two-dimensional means (ex: B. Hillier's J-Graph). By using ANNs, the proposed working prototype could analyse the space in the third dimension. As the ANNs are dynamic models, even the time could be included as an additional dimension.

The proposed working prototype/benchmark will be either a proper Computer Aided-Design software (CAD) or just a plug-in that can be added to one of the following software: AutoCAD (via VBA Macros or objectARX extensions), 3D Studio Max (MaxScript macros or a proper Max plug-in) or Rhinoceros 3D (RhinoScript macros or a Rhino plug-in). The proposed working prototype will use a mathematical model known as artificial neural networks (ANNs).

Evolutionary Generative Design System: An Algorithmic Method for Housing Conceptual Design

Afaf Dib

PhD Proposal:

Evolutionary tools have been used extensively in design, particularly in engineering, but mainly in order to optimize existing design 1. There are fewer actual examples of generative design tools. Some of them are listed below:

Bentley (1996, 1999) used Genetic Algorithms GA in the elaboration of GADES. A generic evolutionary design system. Coates and Jackson (1999) explored 3D design world using lindenmayer system and genetic programming. Shea (2000) used shape annealing, a method combining shape grammar and simulated an annealing in the development of EIFORM, a generative structural system capable of generating free forms; The program still under development.

 To move from optimisation to generation of new designs, the system must be capable of modifying every part of the design 2. The present PhD aims: (1) to distinguish between the uses of evolutionary Algorithms in order to generate design or to satisfy some optimisation process. (2) To explore the prospect of developing a new generic housing design method starting from scratch. The working prototype will combine aspects from two strands of conceptual design. One is a generative system (shape grammar), which deals with formal syntactic approaches to analysis and generation. The second is an evolutionary system (EA) dealing with evolution and self-organisation. (3) To work on the development of an evaluation criterion, which involves the use of a fitness function. This fitness function will measure the performance derived from emergent global structure criteria. (4) To investigate the different types of representations in architecture and their influence on the development of the fitness function.

The generated working prototype will be a computer program coded in VBA or C++. This study will cover the problem of designing creative evolutionary algorithms and hopefully will contribute to the development of new representation of an architectural entity with extended fitness function.

Computational Perception and Self-Organising Space

Renee Puusepp

Details of proposed research

I propose to work in the field of generative architectural design, exploring the urban environment using artificial perception. I hope that my work brings theoretical understanding of space into practical design tasks.

The proposed study rests upon three hypotheses:

  • Studies from various authors (e.g. Gibson, 1979; Ittelson, 1973) suggest that some qualities of space are capable of triggering certain conduct, for instance affecting navigational decisions of human beings. My task would be to find out the nature of such spatial characteristics and to expose how these qualities can contribute to contemporary design routines.
  • The complexity of large-scale architectural tasks can be controlled efficiently using simulations and predictive modelling.
  • Lastly and perhaps most importantly, environment is a consequence but also a stimulus and a constraint of social behaviour.

Social systems are in constant interaction with their environment. Simulating the system's behaviour could greatly help us to predict the consequences that our design decisions would cause. According to many theorists in cybernetics (e.g. Forrester, 1975; Skyttner, L., 1996), computer simulations provide practical help tackling complex problems associated with systems.

The way individual units of a social system perceive their environment influences directly their behaviour and, consequently, the way they change this environment. Therefore, artificial cognition (Pfeifer, R., Scheier, C. 2001) is the focal point of proposed research.

Perception of space is generally involved in design processes implicitly only via participation of the human designer. My research tries to introduce computational perception as a viable concept into architectural practice.

Understanding Architecture as a Complex System: A Synthetic Model of the Relation Between Space and Human Occupation

Choesnah Idarti

Details of proposed research

The development of artificial systems has led people to engineer systems inspired by the natural world. In architecture, the object of research has begun to shift from analysing common facts to the things that may cause them  This shift opens the possibility of understanding human occupation in space as the cause of architecture, to seek understanding of both how human occupation and environment as systems interact with each other. Hillier (1996) summed up this notion; which is based on human occupation in space, as socially functioning space within his space syntax concepts, with "…A configuration of space can be influenced by, or influence, a configuration of people… "

I aim to generate a synthetic model of Hillier's idea above. There are two aspects to this project: Firstly, to identify all possible simple behaviours in the human use of space, and then to use them to program a computer model of human occupation (agents). The main guidance to do this are projects sourcing from the Space Syntax Lab and projects in the field of artificial societies. Secondly, to model the generation of spatial configuration. We have built an application of the voronoi diagram in AutoCAD .

The voronoi application will lead to a re-interpretation of Paul Coates's model of an alpha syntax generator. By doing the study, I wish to have a better understanding of architecture as a complex system. I will have two systems interacting where all elements are abstractions of natural phenomena and hope to see emergent complexities. In other words, I am seeking for new regularities arising from the model as a new interpretation of the human occupation of space.