A case-based design with 3D mesh models of architecture

  • Hao Hua
  • Published 2014 in Computer-Aided Design


This paper presents a case-based design (CBD) focusing on the segmentation of building models and the synthesis of new compositions. Our TRAMMA (Topology Recognition and Aggregation of Mesh Models of Architecture) program is developed as an application of the method. The segmentation includes the recognition of floors, stairs, and walls out of the mesh models. The aggregation process recombines the building segments according to the user-defined topology. The topology representation of multi-story buildings and the spatial synthesis of 3D building models are integrated in the program. © 2014 Elsevier Ltd. All rights reserved. 1. Case-based reasoning and architecture Case-based reasoning (CBR) has been characterized as a formal methodology for reasoning from experience [1]. The theory of CBR sheds light on the cycle of retrieving cases, evaluating/adapting the cases in new situations, and storing/retaining the cases [2,3]. Its significance increases when reasoning from first principles is not adequate in the domain [4]. Thus this methodology should benefit the field of architectural design as well. However, the subject of CBR in architecture is usually limited to the floor plans rather than the 3D models of buildings. With few exceptions, they only deal with the floor plan layoutswith orthogonalwalls. These limitations are attributed to the difficulties in indexing various information in buildings. Case indexing is regarded as a key issue in CBR, since it encodes our knowledge of the cases and substantially shapes the way we reuse the cases. People have to decide how to represent and organize the knowledge of the cases (as solutions) and the knowledge of the corresponding problems. As Maher and de Silva ✩ This paper has been recommended for acceptance by Kin-Chuen Hui. ∗ Correspondence to: School of Architecture, Southeast University, 2 Sipailou, Nanjing 210096, China. Tel.: +86 25 83794494. E-mail addresses: hua@arch.ethz.ch, whitegreen@163.com. http://dx.doi.org/10.1016/j.cad.2014.07.004 0010-4485/© 2014 Elsevier Ltd. All rights reserved. Garza [5] state, ‘‘The most important issue when defining a case representation is what information in the design facilitates its reuse’’. It is not easy to establish a general and versatile representation of architecture. There are several reasons. First, the design problems have been regarded as ill-defined problems or ‘‘wicked problems’’ [6]. More precisely, both the design criteria and the solution space are difficult to be formally defined. Second, there is no general model of architecture that has been widely accepted. Building Information Modeling (BIM) [7,8] aims to establish a standard modeling platform through integrating various models in the Architecture, engineering and construction (AEC) industry. Nevertheless, various difficulties arise when the distinct designers replace their domain models with the standard BIMmodel. Thus, some researchers (e.g. [9]) chose to develop additional software to adopt the BIM standard. Third, architectural design seldom leads to the ‘‘query-retrieval’’ procedure—a list of specifications leads to solutions [3]. Rather, the process of query, retrieval, evaluation, and adaptation are diffused in the design process. 2. Topology representation and synthesis Case-based design (CBD) has been treated as an application of CBR in design [5]. Heylighen and Neuckermans [10] and Richter H. Hua / Computer-Aided Design 57 (2014) 54–60 55 Table 1 Topology representation and synthesis in CBD programs in architecture. Representation Synthesis Topology of floor plan (2D) Topology of multi-story building (3D) Floor plan (2D) Multi-story building (3D) Archie-II [12] Noa No No No SEED-Layout [14] Yes No Yes No IDIOM [13] Yes No Yes No CADRE [15] Yes No Yesb No Rosenman [20] Yes No Yes No de Silva Garza [21] Yes No Yes No a.SCatch [17] Yes No No No Grabska [18] Yes Yes Noc No Langenhan [9] Yes No No No Merrelld [19] Yes Yes Yes Yes a The topology is described as texts rather than in a formal language. b The program adapts the cases according to geometric or topological concerns. c The program can reason about the topology in floor plans. d The approach learns topology from cases, however, the structured data of the cases are manually encoded. Fig. 1. A few samples of the mesh models in the repository. et al. [11] reviewed a set of CBD programs for architecture, casting light on the structure and organization of knowledge as well as on the reasoning process (retrieval and manipulation). This section compares a series of CBD programs from two aspects: topology representation and synthesis. Topology representation is essential for formulating the relationship between the rooms in buildings, and subsequently spatial synthesis could use the topology information to adapt and manipulate the cases. Archie-II [12] is one of the earliest CBD applications in architecture. It represents each case by floor plan and verbal descriptions. Most CBD approaches caught the geometric information of floor plans; however, they usually assumed that the floor plan is made of rectangles [13,14] or right-angle polygons [15]. Alternatively, the recent work by Dillenburger [16] employed a pixel-based representation of buildings and its environment. Besides, a number of works [13–15] succeeded in representing the topological configurations of floor plans. For instance, Langenhan [9] and Weber [17] extracted the topological configurations from BIM models as graphs, benefiting from the semantic information embedded in the model components. Moreover, Grabska et al. [18] extended the topology of 2D floor plans to the topology of multi-story buildings. Merrell [19] employs Bayesian networks to learn topological/geometric information from cases (multi-story buildings) and synthesizes 3D buildings according to the learned topology. However, the original topological/geometric information (the training data for the network) is manually encoded from cases. It is important to emphasize that most CBD approaches deal with floor plans instead of 3D models (Table 1). There have been diverse strategies for manipulating the case for new designs. CADRE [15] supports both the dimensional adaptation and the topological adaptation. IDIOM [13] and SEED-Layout [14] can make new compositions by re-combining the parts of previous floor plans. Intelligent synthesis was a big challenge to the early CBD programs. More sophisticated synthesis usually involves advanced searchmethods such as evolutionary algorithms, e.g., Rosenman [20] and de Silva Garza [21]. The evolutionary process in both programs can manipulate the cases and generate new designs satisfying user-defined objectives. Nevertheless, the synthesis of these programs is conventionally limited to floor plans (Table 1). The CBD approaches listed above indicate a need for representing and synthesizingmulti-story buildings. Thus our TRAMMA program commences with the 3D models of building. The major challenges of the work include: (1) recognition: extracting the topological information from the 3Dmesh models, in other words, breaking themodel intomeaningful parts such as floors, walls, and stairs. (2) synthesis: combining the parts of cases into multi-story buildings that satisfy specified topology. In opposite to the programs in Table 1, our research emphasizes the analysis and manipulation of 3D models of multi-story buildings rather than that of floor plans. 3. Processing 3D models of buildings

DOI: 10.1016/j.cad.2014.07.004

Extracted Key Phrases

13 Figures and Tables

Cite this paper

@article{Hua2014ACD, title={A case-based design with 3D mesh models of architecture}, author={Hao Hua}, journal={Computer-Aided Design}, year={2014}, volume={57}, pages={54-60} }