Artifice, Inc., P.O. Box 1588, Eugene, OR 97440
Stephen Duff, Donald Corner
Department of Architecture, University of Oregon, Eugene, OR 97403
Recent advances in computer graphics and 3D user interfaces have enabled the emergence of 3D sketch modeling as a viable approach to architectural design, especially in the early schematic phase. This paper describes how a system can be built and used which integrates the capabilities of a good structural analysis system in the user-friendly working environment of a design-oriented modeling program.
The structure of a building model as seen by finite element algorithms is a schematic idealization of the building's physical structure into nodes, elements, internal releases, boundary conditions, and loads. The more familiar architectural model used for design visualization represents spatial elements such as roofs, floors, walls, and windows.
Rather than treat these models independently, the structural model can be defined in relation to the architectural as a virtual model with inherited common characteristics and additional relational and attribute information, using feature-based geometry data structures to organize topological intelligence in the spatial model. This provides the basis for synchronous modification of structural and architectural aspects of the design.
With recent advances in computer graphics and 3D user interfaces, 3D sketch modeling has emerged as a viable and productive approach to architectural representation and design, especially in the early schematic phase. Concurrently, we have witnessed the development of effective analytical applications which support improved understandings of structural systems, energy performance, lighting and acoustics.
As currently available, design and analysis applications remain largely independent and discontinuous. Not only do they lack a common integrated interface, but their underlying data structures are mutually unintelligible. As such, the various disparate applications are not easily applied in concert, effectively restricting their use to the end of the building design process. But to be truly generative, all basic design considerations must be explored from early in the design process.
A major step toward a system capable of supporting the whole scope of the building design process would be achieved if the capabilities of the best analytical systems were made ready-to-hand in the user-friendly working environment of a design modeling program. If designers could move with equal facility between representations of the building form and analytical models of critical systems, the schematic design would emerge from a dramatically new creative context. A more comprehensive understanding of design determinates could exist at all stages of the design process.
A system uniting sketch modeling and structural analysis into a single design tool would provide a dramatic new environment for design exploration, with powerful applications in design education and professional practice. Moreover, it would contribute to a much-needed transformation of the conception of building structurefrom an apparently mundane necessity, into an inspiring core aspect of the architects creative opportunity.
This paper outlines the characteristics and capabilities of such a unified design environment and how to realize it.
2. A Fragmented Paradigm
With a few notable exceptions, the design of architectural space and the design of building structure have, in this century, been regarded as two distinct domains of practice, weakly linked only through an overall common cause.
The reasons for this unfortunate segregation of architectural and engineering design are numerous (Black and Duff, 1994), but the consequences are severe: students are educated within the intellectually isolated confines of their own fields, little inter-disciplinary cross-fertilization exists at any level, and designs are rarely conceived as integrated wholes. The typical design process is one driven predominantly by an architectural agenda, with engineers called in to make the building stand up, long after the essential form of the building has been determined.
The resulting buildings lack the unification of form and function and the integration of structure and space seen in the great iron structures of the last century or the gothic cathedrals built in medieval Europe. Contemporary projects with normative budgets have become a direct, physical manifestation of the split between the disciplines, and so we inhabit architectural spaces more and more defined by steel studs and gypsum sheathing, with little or no expression of primary structure.
Even though electronic formats have replaced traditional media, the typical design process presumes a limited interchange of information between disciplinary domains. Current design processes are built around sequential communication of limited information from one domain to another. For many years it has been possible to transfer 3D geometry from a CAD application to a structural analysis application, yet it has not become standard practice to do so. This is partly due to the cost in time and complexity of translation, amplified by the mediocrity of available out-dated data interchange formats, like DXF.
However, deeper problems arise from this sequential/translation approach. When the spatial and structural models are discrete, intractable problems of redundancy and non-concurrence become a powerful impediment to any kind of revision cycle. Typically, significant post-translation development of the structural model is required to support analysis. Such downstream structural development becomes disconnected and obsolete each time the evolving building form is revised and retranslated, and so the structural information has to be re-specified. A natural response to such redundancy of effort is to postpone structural analysis until building form has been clearly established.
Figure 1 . Translation leads to loss of data across design stages and iterations.
3. An Integrated Design Process
Although there are many differences between architecture and engineering in terms of problem domains, knowledge bases, thought processes, and design languages, there is a common underlying design process. In both fields, design development occurs through linked cycles of design and evaluation, which exhibit an overall trend in one direction toward a domain of acceptable responses, but which are also characterized by backtracking (apparent movement away from, rather than toward, increasing problem resolution) and iteration (repetition of activities with shifting focus or increasing resolution) (Zeisel 1981).
Furthermore, the discrete but related models of these two domains can naturally occupy a common Cartesian coordinate system, and share many material and geometric characteristics.
Existing impediments to integrating these differing models can be avoided if the building form and structural abstraction are joined into one unified model, in which structural information need be defined only once to support many revision cycles. Bridging the technological separation between visual and structural design in this way will allow architectural designers to perform structural analysis early and often during design.
Figure 2. A unified model preserves data across stages and design iterations.
4. An Integrated System for Spatial and Structural Design
The proposed system consists of a design-oriented modeling environment that supports structural modeling objects, a geometric inference engine to help define structural elements, and one or more structural analysis tools. An underlying unified data model and surrounding unified modeling environment ensure that any changes made to either the architectural or structural representation of the building design are automatically current to both.
Given such an integrated model, a series of analysis approaches are desirable, so that as the design becomes progressively less ambiguous, increasingly detailed and specific behavior data can be extracted. Gross analysis of the general building form, early in the spatial model and massing stage of scheme generation (when building elements are only roughly defined), would yield global behavior response values helpful in selecting building configuration, structural systems, structural density and approximate member size.
At the next stage of development, as major architectural and structural elements emerge, initial quantitative analysis of building structure becomes appropriate. By recognizing and classifying relevant spatial elements using a background geometric analysis process, necessary information for rule-of-thumb design can be obtained. If the user utilizes preliminary response data in conjunction with a menu of potential construction systems to drive the rule-of-thumb structural design, key assumptions and missing general parameters can be filled in transparently by the system.
Later, when a detailed model emerges with well-defined structural elements, comprehensive structural analysis is needed. Using more extensive and finer grained geometric inferences to identify structural elements, and using a graphical user interface for specifying connections, an idealized model required for finite element analysis can easily be created, retained and updated as one representation within the overall unified building model.
Figure 3. Multiple representations of the building model.
5. 3D Spatial Modeling
The essential idea of sketch modeling is a live three-dimensional modeling immersion environment in which a designer can quickly and easily define and modify architectural form (Matthews 1988). Such malleable, artistically potent applications are capable of supporting the tentative and exploratory processes of form definition which characterize design in the early emergent stages.
DesignWorkshop® Professional by Matthews and others (1998), a design-oriented feature-based solid modeling application, is an example of this type of sketch-modeler. A mouse-controlled cursor allows direct-manipulation in three dimensions within a perspective-oriented viewing environment, and selected solid objects show handles at vertices and edge mid-points. The system user graphically defines translation vectors with the 3D crosshair to accomplish a variety of modeling operations. The geometry selection state is used to mediate tool mode control implicitly, freeing the user entirely from menu or icon-based tool selection for many common design operations like moving, re-sizing, reshaping, and extruding solids and openings.
Context-sensitive automatic tool selection occurs, for example, when the user selects a hole in a solid, such as would represent a window opening in a wall. The crosshair automatically changes from the fully-3D default spatial mode to a special 2-1/2D faces crosshair, with reduced degrees of freedom such that the crosshair is constrained to object faces. This change of tool mode is transparent to the user, mediated by an unusual two-way interaction between the order of the selected geometry and the user interface.
The underlying geometric intelligence is embodied through feature-based object structures, according to which the hole is recognized and stored as a coherent sub-object of its parent solid (Matthews 1988). Technically, the basic constitutive entities of the spatial model are solid geometric objects, geometric feature sub-objects of solids, and non-geometric attributes attached to geometry. Such elements provide a foundational context for the definition of a unified spatial and structural model.
The sketch-modeling environment supports both the artistic immediacy of paper-based sketching and the full architectural embodiment facilitated by three-dimensional modeling. As a result, this software allows a subtly new approach to architectural design, focused on development of a computer-based spatial model as the primary locus of design intent.
6. Finite Element Analysis
Modern structural analysis programs are based on the finite element method, a numerical technique involving the matrix formulation and solution of a large linear system of equations.
An analysis model is an abstract idealization of the geometry, material properties, and construction details of a real structure, and can be represented schematically in Cartesian space. The principal constitutive entities of an analysis model are nodes , rigid bodies that occupy points in Cartesian space and have six potential degrees of freedom, elements , deformable bodies which idealize real structural members, applied external forces , modeling real loads, and specified nodal restraints , imposed on nodes to establish the boundary conditions (supports).
In assembling an analysis model, appropriate elements are connected to nodes with specified force releases (congruent with real connections), boundary conditions are imposed and applied loads are specified. These parameters are manually defined by the user, while subsequent formulation of a mathematical model is automatic. Solution gives numerical results (nodal displacements, support reactions, and element end forces or stresses), which are commonly presented in graphical form as force diagrams, stress contours, and schematic representations of structure deformations.
From these results, a complete understanding of the behavior of a structure can be obtained in terms of global displacements and deformations, relative internal stiffnesses, the distribution of forces throughout the structure, and the magnitude of stresses in individual members.
Once an initial analysis model is in place, it can be readily modified, facilitating the generation, evaluation, and comparison of structural alternatives.
7. Unification of Spatial and Structural Domains
We consider a unified model to be the essential foundation of an integrated design environment, in which form and structure become alternate, inter-dependent, and potentially simultaneous representations of a single data model containing all the information necessary for each, allowing graceful concurrence between the architectural and structural views of the overall building model. Within a common Cartesian coordinate system, we can define an active relationship between the interacting representations of form and structure.
From a software perspective, the unified model contains a superset of the data structures required for both spatial and structural design, including geometric primitives, object attributes, structural elements, and connection definitions with various degrees of freedom and releases. Rather than store independent duplicate data for each significant building component, and then erect a complex network of update indexing within the data, the use of feature-based data structures (Shah 1995) allows the idealized structural geometry of each general building component to be stored as a separately addressable sub-object of the building component itself.
The analysis model required by the FEA software is then a virtual model derived from the architectural model by means of an interaction of automatic and manual methods.
8. Unified User Interface
A unified design environment must allow seamless access to both the spatial model and the structural model, through one common set of general tools. Ease-of-use is critical to the success of any environment for exploratory design work, and when fewer tools can be used to efficiently accomplish a given task, application ease-of-use is increased.
Figure 4 . Assigning structural attributes within sketch modeler.
Tools are designed and presented in the user interface to minimize both redundancy and complexity, with a small number of tools addressing a wide range of modeling tasks. When specialized behavior of a general tool is called for in only one domain, an appropriate mode selection defining the specialized behavior can be triggered automatically according to the context state of the environment.
For instance, when the user selects a structural feature of a geometric object using a general tool, then a structural modeling mode of the tool is invoked by the software, and the tool is adapted transparently to perform the structural variant of the task. Feature-based geometry data structures are well-suited to support this kind of compact user interface functionality, although they have rarely been exploited in this way. At the same time, they naturally address data concurrency issues by structuring the nested dependency of abstract geometric surrogates to the visually more literal components of the architectural model.
In an integrated application, tools created to meet a clear requirement in one domain, can unexpectedly enrich functionality in the other domain. For example, the Hide command is a visually-oriented display control function in the DesignWorkshop spatial modeling environment, which is also be useful in the performance of structural analysis. To quickly focus analysis on only a small portion of a structure, other parts of the structure are simply hidden temporarily using this function.
Conversely, while connections between objects are not usually specified in spatial modeling applications, connections must be defined for thorough structural analysis. The resulting joints between model elements can be exploited to streamline certain general transformations of architectural geometry. For instance, once a set of struts in a truss have been connected at a node, the user can reposition all the connected struts quickly and accurately (perhaps to accommodate a revised building envelope) just by applying graphical translations to the node.
9. Internal Data Interface
In our system, the critical data interface to specialized structural object features exists internal to the overall modeler data space. This provides the bridge from the architectural model to the structural abstraction, which in turn is the basis for generating correct analysis-ready input data for the FEA or other analytical code.
Rather than externalize the structural idealization of the building model as seen by the FEA algorithms as data independent from the geometric entities of the spatial model, we specify the new geometric elements needed to form the structural idealization as sub-objects of the corresponding spatial components. Using appropriate parameter-recognition algorithms, the structural features are derived from the original spatial representation element by element. Of course, many architectural elements will not be relevant to the structural idealization, and these are ignored after labeling as non-structural.
Developing feature-based topological intelligence in the spatial model in this way, many aspects of the FEA analysis model can be derived automatically, while others may need to be defined by the user. General geometry is operated on, as it is created, by a simple geometric inference engine for identification and abstraction of structural features of objects. The algorithmic design of this inference engine draws on concepts of simple shape grammars (Mitchell 1995, p372).
Figure 5a, Figure 5b
Figure 5. Structural data within the visual modeling application
For example, at the end of the creation operation for a simple block, its overall dimensions are analyzed, and if it has a height to width ratio of 6:1 or greater, a width to depth ratio of 3:1 or less, a minimum dimension greater than 10cm, and a height greater than 2.0m, the software infers the object is a column, and assigns it this structural type (subject to user override). Similar parametric criteria allow the background identification process to label other newly created geometric objects as beams, slabs, or non-structural. While the inferences are not entirely accurate, they are sufficient to handle the bulk work of object type-casting, allowing the user to move directly into fine-tuning as soon as the structural model is consciously addressed.
The primary building blocks of this system existed prior to our current software design effort. They have not previously been successfully integrated, and substantial software design on several levels is necessary to realize the functions outlined. While high-end mechanical engineering CAD systems have embraced feature-based geometry concepts over the last decade, these concepts have not been adopted by the dominant CAD vendors in the AEC arena. The breadth of appropriate applications for feature-based technology in the world of architecture remains a matter for further research.
Building on emergent software technologies, we are working to create a sorely needed bridge between two conceptual worlds - an integrated system which allows simultaneous spatial and structural design to take place in a shared 3D design environment, united by user interface and a common feature-based data model.
Black, R. G. and Duff, S. F.: 1994, A Model for Teaching Structures: Finite Element Analysis and Architectural Education, Journal of Architectural Education 48(1), 38-55
French, M.:1994, Invention and Evolution : Design in Nature and Engineering, 2nd edition. Cambridge, Cambridge University Press
Korngold, E. V., Shepard, M. S., Wentorf, E., and Spooner, D. L.: 1989, Architecture of a design system for engineering visualizations in Ravani B. (ed) Proceedings of 15th ASME Design Automation Conference, Montreal, Sept. 17-21, ASME Press, 259-265
Matthews, K.: 1988, Three Dimensional Sketching. Berkeley, California, College of Environmental Design Library, University of California, Berkeley
Matthews, K. et al.: 1998, DesignWorkshop® Professional (software application). Eugene, Oregon, Artifice, Inc.
Mitchell, W and McCullough, M.: 1995, Digital Design Media, 2nd edition. New York,Van Nostrand Reinhold
Shah, J. and Mantyla, M.: 1995, Parametric and Feature-Based CAD/CAM, New York, John Wiley and Sons
Shephard, M. S.: 1990, Idealization in engineering modeling and design, Research in Engineering. Design, 1, 229-238
Zeisel. J.: 1981, Inquiry by Design, Monterey, CA, Brooks/Cole Publishing Co.
Additional References - post-publication
B. J. Novitski. Rendering Real & Imagined Buildings : The Art of Computer Modeling. Rockport Pub., Book and CD-ROM edition, January 1999. ISBN 1-5649-6511-2. Available at Amazon.com
Malcolm McCullough. Abstracting Craft : The Practiced Digital Hand. MIT Press, Reprint edition, September 1998. ISBN 0-2626-3189-X. Available at Amazon.com