«Architectural Design as Combined Modeling A dissertation submitted to ETH ZURICH for the degree of Doctor of Sciences presented by HAO HUA M.Arch., ...»
DISS. ETH NO. 21516
Architectural Design as Combined Modeling
A dissertation submitted to
for the degree of
Doctor of Sciences
M.Arch., Southeast University, China
citizen of P. R. China
accepted on the recommendation of
Prof. Dr. Ludger Hovestadt
Chair of Computer-Aided Architectural Design, ETH Zurich
Prof. Dr. Li Biao School of Architecture, Southeast University, China Prof. Dr. Urs Leonhard Hirschberg Institute of Architecture and Media, TU Graz Abstract This work investigates the potentials and the limitations of models in ComputerAided Architectural design (CAAD). The research focuses on the symbolic models instead of the analogue models that resemble their referents. The symbolic model can represent architecture with
signs and produce design solutions through formal operations. Over the past five decades, more and more models have been available for architecture and have been practical with the help of computers. We regard the activity of modeling essential in coupling symbolic models with architectural design. However, modeling architecture is still a challenge mainly because of two reasons. First, design problems are usually not well defined. For instance, Rittel (1973) concluded that design problems have no definite formulations, no stopping rules, and that they are inherently unique. Thus, there is no general model of architectural design that is widely accepted. Second, there are dilemmas in modeling. Previous studies have demonstrated that models are essentially wrong or partial. Besides, the nature of modeling can result in multiple inconsistent models of the same subject matter.
In order to solve the problem of multiple models, two common views are investigated. One holds that a single coherent view on the object is necessary in modeling.
The relevant approach is metamodeling, i.e., making the model of models. The other view concedes that multiple inconsistent views are inevitable. Therefore, a single model is unnecessary and is not rewarding. This work finds out that the alternative solution, combining multiple models, could be better than developing a single model or metamodel. We consider combined modeling as a natural response to the dilemmas in modeling. In particular, the fictional combination of models could be fruitful in architectural design. Architects usually modify, add, and transform multiple issues in design. By combining multiple models, the designers can free themselves from the epistemic framework of each model and subsequently make their own position by organizing the interrelationships between the models.
This dissertation discusses on both modeling with computers and without computers, including a brief survey on the historical employment of mathematics and geometry in architecture. The proposed modeling methods are tested in a series of experiments with computers. They mainly focus on the topological and geometrical design of buildings. All experiments are implemented in the Java programming language. The results imply that the combination of multiple models could be productive in architectural design.
! iii Zusammenfassung Die Arbeit untersucht das Potential und die Grenzen von Modellen im Computerunterstützten Architektonischen Design. Die Forschung konzentriert sich auf symbolische Modelle anstatt auf analoge Modelle, die sich an ihren Referenzen orientieren. Das symbolische Modell kann Architektur mit abstrakten Zeichen repräsentieren, und Entwurfslösungen durch formale Operationen produzieren.
Während der letzten 5 Jahrzehnte, sind immer mehr Modelle für Architektur verfügbar und durch die Hilfe des Computers anwendbar geworden. Wir betrachten den Vorgang des Modellierens als essentiell an bei der Kopplung von symbolischen Modellen und architektonischem Entwerfen. Aber hauptsächlich durch zwei Gründe bleibt es eine Herausforderung Architektur zu modellieren. Zum einen sind Entwurfsprobleme normalerweise nicht sauber definiert. Rittel (1973) erklärt, das Entwurfsprobleme sich nicht definitiv formulieren lassen, keine Endbedingung haben, und immer einzigartig sind. Dadurch gibt es kein generelles Model für Architektur, das eine breite Akzeptanz findet. Zum anderen gibt es ein Dilemma im Modellieren. Vorherige Studien haben gezeigt, dass Modelle essential falsch oder unvollständig sind. Des Weiteren kann die Natur des Modellierens zu multiplen, inkonsistenten Modellen zum gleichen Thema führen.
Um das Problem von multiplen Modellen zu lösen werden zwei herkömmliche Standpunkte betrachtet. Der eine setzt voraus, dass eine einzige, kohärente Betrachtung auf das Objekt nötig ist um es zu modellieren. Der relevante Ansatz dazu ist das sogenannte Meta-modelling, also das Entwickeln eines Modells von Modellen. Der andere Ansatz geht davon aus, dass unterschiedliche, inkonsistente Standpunkte unvermeidbar sind. Deswegen ist ein einfaches Modell unnötig und nicht zielführend. Diese Arbeit kommt zu dem Schluss, dass die alternative Lösung, mehrere Modelle miteinander zu kombinieren besser ist, als ein einziges Modell oder ein Metamodell zu entwickeln. Wir betrachten kombiniertes Modellieren als natürliche Antwort auf das Dilemma des Modellierens. Die synthetisierten Kombinationen von Modellen können fruchtbar im architektonischen Entwurf sein. Normalerweise modifizieren, addieren, und transformieren Architekten verschiedene Varianten im Entwurf. Beim kombinieren von unterschiedlichen Modellen können die Entwerfer sich von dem epistemischen Rahmen jedes Modells befreien, und ihre eigene Position beim in Beziehung setzen der Modellen bestimmen.
Die Dissertation betrachtet zugleich Modellieren am Computer und ohne Computer, und beinhaltet einen kurze geschichtliche Übersicht über Anwendung von Mathematik und Geometrie in Architektur. Die vorgeschlagenen Modellieriv methoden sind in einer Serie von Experimenten mit dem Computer getestet worden. Diese fokussieren hauptsächlich auf dem toplogischen und geometrischen Entwurf von Gebäuden. Alle Experimente sind in der Programmiersprache Java programmiert. Die Ergebnisse implizieren, das die Kombination von verschiedenen Modellen produktiv im architektonischen Entwurf sein kann.
Introduction “A model is a work of fiction” (Cartwright 1983, 153) Forty years ago, Horst Rittel (1973) suggested that design problems are “wicked problems.” In the contemporary context of Computer-Aided Architectural Design, how can we approach the wicked problems with an increasing number of computational models? Over the past five decades, there have been a great number of works that applied modeling and computing to architectural design. We have observed that people do not only model architectural design, but that they also transform the design itself through modeling and computing. Although computational models have become more and more powerful, some researchers revealed that the models are essentially wrong (Box and Draper 1987) or falsifiable (Stevens 1990). In order to examine the problems of modeling, this work investigates various notions of modeling as well as their applications to architecture. In particular, we find out that the design activities of architects usually involve multiple models instead of one model. Thus managing multiple models is essential for the synergy of modeling, computing and architectural design.
1.1 Modeling, Computing, and Architectural Design 1.1.1 Modeling Innumerable researches from different disciplines have discussed the notion of
modeling. For instance, Rothenberg (1989) alleged that:
“Modeling in its broadest sense is the cost-effective use of something in place of something else.” It suggests that the model always represents something with something else. There are different kinds of models according to their representational functions. Ackoff et al. (1962) made three categories: iconic models, analogue models, and symbolic models. Iconic models look like their subject matter, such as photos and scale models. Analogue models make certain abstraction of the subject matter. For example, maps and bubble diagrams are analogue models. Symbolic models are the most abstract. They are written in mathematics or in other formal languages. This ! 1 thesis focuses on symbolic models. Nowadays, symbolic models are pervasive. If we shoot a scene with a cell phone, we get a photo in the form of pixels. Each pixel could be represented by a set of RGB color values (the three components are red, green and blue). For instance, (255, 0, 0) denotes red color, (0, 255, 0) green, and (0, 0, 255) blue. The RGB system is a symbolic model of color. The image that we can watch on the screen is an iconic model of the scene, while the digital photo (pixels) is a symbolic model of the scene. However, what is a 3-dimensional model in a CAD (Computer-aided Design) software? This question could be ambiguous, since the 3-d model that we can see on the screen, as a pictorial rendering of the underlying data, is an iconic model of the building1. Yet the data structure of the 3d model pertains to a symbolic model.
Besides the classification of models, some researchers put emphasis on the language of the models. Kleppe et al. (2003, 16) stated that “a model is a description of (part of) a system written in a well-defined language.” The well-defined language must have a clear syntax and semantics. Logics and mathematics are often employed as the language of modeling. In software engineering, the Unified Modeling Language (UML) is a common language for modeling (Bézivin 2005).
Mitchell (1990) demonstrated the first order logic as a critical language for design.
In this language, each sentence is either true or false. For instance, a two-place predicate parallel (Line, Line) examines whether two lines are parallel. The language can enumerate all possible states of the corresponding well-defined design world.
In Allgemeine Modelltheroie, Stachowiak (1973) defined the three features of models: 1) Abbildung: the model represents its original; 2) Verkürzung: the model only represents part of the aspects of the original; and 3) Pragmatismus: the model has its own purpose, relatively independent of the original. His formulation implies that the relationship between the original and the model is essential in modeling;
however, this relationship has been controversial. One mindset believes that the model should make faithful representation of its subject matter. For instance, Ashby (1957) held that we should establish the isomorphism between the model and the referent. However, many researchers realized that the model must be a partial representation of the original. Besides, the similarity between the two is subject to a certain epistemic framework. As Broadbent (1973, 89) said “No model can ever be complete, correct and universal in its application. We build a model because we want to focus down on certain aspects of a problem.” Moreover, a few studies addressed that models do not necessarily represent something real. For instance, Barberousse and Ludwig (2009) stated that: “models are fictions, that is, precisely, !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
It is either a virtual building under planning or a real building.
It has been widely agreed that the model only catches a part of the aspects of the original. Bender (1978) regarded the model as an abstract, simplified construct of the target. According to Smith (1985), the model is essentially partial. Since the models are always incomplete or partial, it is reasonable that all models are falsifiable (Stevens 1995). However, we sometimes verify a model through trying hard to falsify it. Carson (2004) explained the procedure of model verification: 1) suppose the model is correct; 2) try hard to prove the model is wrong or bad in some situation; and 3) if the model is still valid under these tests, it is verified. Of course, we are not able to test all possible situations; we are only interested in those relevant and important situations instead. Usually, people have little idea about the exceptional situation in which the model fails. For example, the computer program of an American missile warning system mistook the lunar reflection for a Soviet attack on October 5, 1960 (Smith 1985). In this case, the model was not adequate in the real environment, though many experts had carefully verified it.
Yet a number of studies put emphasis on the usefulness of models instead of their representational roles. Hence, the problem switches from “what the models are” to “what are the models good for” Kurpjuweit and Winter (2007) articulated: “A model is created by a modeler and interpreted by one or more users with respect to
a certain purpose”. Rothenberg (1989) considered cost-effectiveness as the essential attribute of modeling. In accord with them, Pinsky and Karlin (2011, 1) wrote:
“In the final analysis, a model is judged using a single, quite pragmatic, factor, the model’s usefulness.” Besides the usefulness, people care about the costs. For example, if a model is too complex to manipulate or it is too expensive to collect the required input data, then the model is not a wise choice. In practice, people have to negotiate between the cost and the actual effectiveness of the model.
1.1.2 Computational models Since commercial computers came into use in various fields, the number of computational models has been rapidly increasing. Theoretically, models can be made without computers, however, the burst of “computational” models is obviously due to the synergy between modeling and computing. Many of these models make no sense without computing. The studies of modeling and computing have been highly correlated. The theory of computing has been heavily attributed to Alan Turing.