1. Introduction

This research is focused on the development of a methodology for capturing and reusing design knowledge to enable the automatic generation of multiple design configurations to support parallel lines of design development in early stages. Unlike technical domains (Amor & Faraj, 2001) architectural design entail significant amount of tacit knowledge (Woo, 2004) which makes reusability more difficult because this expertise usually is neither declared nor formalized, and it is mainly based on assumptions and heuristics. Although different designers differ in terms of style and they rely on their experience to make decisions, many common actions they perform have been well-documented. Designers integrate knowledge of different disciplines(Cross, 2004), simultaneously evolve problem and solutions(Maher, Poon, & Boulanger, 1996), recall patterns of organization and chunk of constraints (Gobet et al., 2001) and rapidly generate feasible solutions candidates without extensive analyses(Lawson & Dorst, 2009). All these actions are highly sophisticated and poorly declared yet (Eastman, 2001).

The specific focus of this research is on the ability of experts to handle parallel lines of thought and generate preliminary designs by combining features of previous solution for similar problems. Although parametric modeling technology encapsulates knowledge into parts, assemblies, rules, parameters and constraints, it provides limited resources to represent the underlying patterns of design organization. It also has limitations to support variations beyond the scope of its hierarchical binary tree data structure (Mantyla, 1988), prematurely limiting the generation of diverse possible solution candidates. In addition, the interfaces of the computational tools are conceived from the perspective of the physical components, but not necessarily facilitating the manipulation of the patterns that provide coherence to the designs.

2. Approach

The notion of topological modeling is introduced to address the problem of the generation of different design alternatives from a design domain.  It implies modeling the domain as the main repository of design knowledge and developing instantiation mechanisms to create solution candidates according to design rules declared on the domain. To achieve the necessary flexibility to produce topological changes, the separation between configuration specification and geometrical representation is adopted. This autonomy provides the flexibility to specify configurations according to the shifting nature of the design problems without dealing with the complexity of geometric models. The resulting specification is in the domain specific semantic (Eck & Schaefer, 2011) that is mapped to different BIM tools for representation.

3. Methodology for Meta-Modeling and Geometric Representation

The methodology is based on meta-modeling, understood as the model of the attributes of the design. It structures in three layers methods to create the meta-model the design domain, support topological modeling and geometrically represent the resulting configuration (Fig.1). The design domain layer formalizes physical parts, design rules and patterns of organization. The topological modeling layer addresses the problem of generation of instances from the domain, and the geometric representation layer interfaces between the configuration specification and the means of representation. The methodology relies on the System Modeling Language (SysML), a tool independent language that models class definitions and multiple types of associations (Reichwein & Paredis, 2011). From the design domain meta-model, variable number of instance with different topologies can be specified (Böhnke, Reichwein, & Rudolph, 2009) and mapped to CAD or BIM tools through stereotypes, an existing UML (Bruck & Hussey, 2007) resource, that links abstract conceptuality of the domain with the tool internals to create geometric models that can be also manually edited if necessary.

Figure 1. Methodological approach for topological modeling based on meta-models

Case Study

The design domain of façade system is the chosen to verify the validity of the approach since it involves limited number elements that can be combined in different configurations. The elements of the design domain have been defined in collaboration with an expert designer in this field to describe the taxonomy of underlying design patterns, physical parts, design rules and evaluation methods that provide feedback from the resulting configurations. Although multiple façade systems exist, they share common underlying structures to address geometrical problems, technical challenges and eventually physical parts from which multiple design variants can be generated. A domain meta-model in SysML language captured the elements and related attributes of the Pre-fab façade system of the 100 & 10^th residential tower at NYC by Atelier Jean Nouvel. Parallel to the effort to capture the attributes of the physical parts and design rules, a complete library of actual parts was developed in Digital Project, a BIM tool by Gerhy Technologies. Those parts where successfully assembled in different arrangements of pre-fab panels based on the abstract specification derived from the meta-model.

Results

The meta-models of the design domain seems to be an effective mean to collect and organize apparently disperse information, create specifications of possible topological configurations and interface with BIM tools.  Results show the flexibility of the meta-modeling approach to captures and integrate design knowledge from multiple sources in a domain represented by objects and their attributes defined in domain specific terms. Meta-models augment the ability of designers to produce variations since they define relationship and attribute of parts. It also facilitates interfacing with multiple computational tools since the stereotypes link the high-level specification with the low level BIM tool basic operations such as create project, insert part with given attributes or  execute linear transformation like move, rotate or scale parts. Meta-modeling techniques provide a flexible and extensible platform that facilitates the manipulation of the topological relationships among parts driven by design knowledge.