【作者】 李智杰；

【导师】 李昌华；

【作者基本信息】 西安建筑科技大学 ， 数字建筑， 2015， 博士

【摘要】 数字化的建筑设计方案已在建筑领域中大量存在并得到广泛应用,对其中蕴含的丰富知识财富和经验积累加以有效再利用具有重要意义。借助计算机技术构建智能化的辅助设计平台并应用于工程项目的辅助设计或建筑设计辅助教学,可为建筑设计人员从已有的数字化设计案例中获取建筑知识、学习设计经验和设计规范、激发设计灵感、辅助建筑设计构思、重用部分设计等提供工具支持,使得建筑设计及学习过程更加高效、便捷和愉悦,是提升计算机辅助建筑设计效率和质量的一种有效途径。BIM是利用数字技术表达建筑的几何、物理和功能信息,以支持建筑物全生命周期管理的信息资源共享技术。BIM模型是BIM技术的核心,随着BIM技术的蓬勃发展和广泛应用,BIM模型的数量将会越来越多,合理地利用这些资源能够有效地缩短设计周期、降低设计成本并提高设计质量。案例推理(CBR)是一种智能化设计理论,其基本原理是通过对以往类似问题的求解经验(即案例)进行推理来解决当前问题,利用该理论建立基于BIM的智能化辅助设计平台的关键问题是模型的特征选取与表达、分类与检索方法的构建。BIM模型的一种重要特征是隐含有大量的建筑空间关系信息,而目前广泛采用的以关键词作为检索条件的方法仅仅适用于语言能够描述的情况,建筑空间关系则难以用语言刻画,因此研究建筑空间关系的计算机表达方式是对BIM模型进行有效检索的前提。图论是描述现实世界事物间关系的数学及计算机理论,建筑学科以图论为基础产生了空间句法理论用来表达建筑空间关系并进行分析,因此利用空间句法理论和图论建立属性关系图可为建筑空间关系提供计算机表达方式。利用属性关系图开展BIM模型分类与检索研究的难点在于构建运行速度快、准确率高的非精确图匹配方法,而这同时也是计算机学科进行图数据分类和检索的研究热点与难点问题之一。本文以建筑设计方案的知识再利用为研究目标,以CBR理论为指导,以BIM模型为研究对象,针对BIM模型中隐含的建筑空间关系信息,运用图论和空间句法理论的基本原理,通过研究建筑空间关系的图模型形式化表达和自动提取算法,将基于建筑空间关系的BIM模型分类与检索问题转化为图匹配问题,进而研究相适应的基于图核和图嵌入的非精确图匹配方法,在此基础上研究基于功能气泡草图的BIM模型检索方法和原型系统的构建,为计算机辅助建筑设计提供更高效的服务。论文工作主要包括以下几个方面:1.建筑空间关系可以通过构造属性关系图模型进行形式化表达,由此产生的量化信息有利于在计算机中进行基于建筑空间关系的智能信息处理。本文借助图论和空间句法理论构造了融合空间句法的属性关系图模型(SSARG),在此基础上提出了建筑空间关系相关属性的形式化表达——建筑空间属性关系图模型(ASARG)及其自动提取算法,进而利用Revit API和C#语言在Revit平台上通过二次开发实现了ASARG的提取和属性计算的相关插件,最后给出了基于XML的BIM图匹配测试库的建立方法和相应的BIM图匹配测试库——BIMGM,为BIM模型分类与检索及其相关研究提供了数据支撑。2.结构模式识别领域中图匹配方法复杂度高,并且缺乏描述图拓扑性质的相关变量,难以满足基于建筑空间关系的BIM模型分类与检索所需的图匹配或通用图匹配的速度要求。本文在SSARG和ASARG的基础上提出了两种图核用于非精确图匹配,通过构造空间句法核和基于最短路径的空间句法核隐式地将图数据空间映射到特征向量空间,使得图之间的相似性度量转化为特征向量空间中的点积,进而利用SVM实现图数据的分类。与其他图匹配方法相比,该方法的复杂度低且分类识别率较高,为仅有拓扑属性的BIM模型或图数据提供了性能较好的基础图匹配方法。3.统计模式识别领域具有丰富的数学工具,而结构模式识别领域却很难对其加以利用,难以满足基于建筑空间关系的BIM模型分类与检索所需的图匹配或通用图匹配的精度要求。本文借鉴图嵌入的基本思想,在SSARG和ASARG的基础上提出了基于拓扑特征与领域特征的图嵌入方法、基于多尺度特征的图嵌入方法和基于改进K-均值聚类的图嵌入方法。通过这些方法将结构模式识别问题转化为统计模式识别问题,进而可以利用统计模式领域拥有的丰富数学方法实现非精确图匹配。与其他图匹配方法相比,上述方法均采用的是统计的方法并有效利用了基于空间句法理论构造的图拓扑特征,因此算法复杂度较低且分类识别率高,为具有拓扑属性和领域属性的BIM模型或图数据提供了性能较好的基础图匹配方法。4.BIM模型中隐含有大量的建筑空间关系信息,而传统的BIM模型检索方法却未加以有效利用,并且缺乏相应的检索接口。本文在上述方法的基础上提出了一种基于功能气泡草图的BIM模型检索方法并构造了原型系统。通过手绘和矢量功能气泡草图实现检索条件输入,进而提取草图中建筑空间的拓扑和领域属性等信息构造ASARG模型,借助基于图嵌入的特征提取方法将ASARG转换到特征向量空间,最后通过构造特征向量空间中的相似性度量函数实现了基于建筑空间关系的BIM模型的检索方法。在上述步骤的基础上构造了原型系统,并通过实验验证了方法的有效性,为构建基于功能气泡草图的BIM模型检索提供了关键技术。本研究为基于BIM的智能化的辅助设计平台提供了基础理论和核心技术,也为海量建筑信息的智能数据处理提供了基础方法。更多还原

【Abstract】 Digital architectural design plans have abounded and been widely used in the construction field. It is significant to use the intellectual wealth and accumulated experience existing in those plans. By taking advantage of the computer technology, the intelligent computer-aided design platform can be constructed and used in the construction or educational practice. It can provide technical support for the architects’ knowledge reuse process in many ways: the architectural knowledge’s acquisition, the learning of design experience and specifications, the inspiring of the design inspiration, the reuse of some parts of the design plans. As a consequence, the architectural design process is more efficient, convenient and enjoyable. The platform can also be an effective way to improve the efficiency and quality of the computer-aided architectural design.By expressing the architecture’s geometric, physical and functional information and using the digital technology, BIM becomes a kind of information resource sharing technology that supports the buildings’ lifecycle management. BIM model is the core of the BIM technology. With the rapid development and wide application, the BIM model’s number will be larger. The rational use of these resources can effectively shorten the design cycles, reduce design costs and improve design quality. Case-Based Reasoning(CBR) is an intelligent design theory. Its basic principle is to solve the current problem by reasoning on the previous problem solving experience(ie case). When taking advantage of this theory to construct the BIM-based intelligent aided design platform, the key issue includes two aspects: the models’ selection and expression, the classification and retrieve method’s construction. The aforementioned important feature of the BIM model is that it contains much implicit architectural spatial relationships. However, the widely used method that considering keywords as search criteria can only be used in the describable case and the architectural spatial relationship(ASR) is difficult to describe with words. So the study of computer architecture spatial relationships of expression is a prerequisite for efficient retrieval of BIM model. Graph theory is a theory describing the relationship between the things in the real world based on the mathematics and computer theory. Architecture discipline created the space syntax theory based on the graph theory to express the ASR and analyze these relationships. Taking advantage of the space syntax theory and graph theory, it’s possible to establish attribute relationship graph providing computer expression of the ASR. When using the attribute relationship graph for BIM model’s classification and retrieval, the difficulty is to construct the inexact graph matching method with high speed and accuracy. It is also the research hotspot and nodus when it comes to the graph data’s classification and retrieval in the computer science.In this paper, the research goal is the knowledge reuse of architecture design; the guide is the CBR theory. Aiming to the BIM models’ implicit information that revealing the architectural spatial relationship, this paper converted the ASR based BIM models’ classification and retrieval into graph matching. This progress used the graph theory and Space Syntax theory, studied the graph model expressing the architectural spatial relationship and its formal description & automatic extraction algorithm. The article studied the inexact graph matching method based on the graph kernel and graph embedding. Moreover, it studied the BIM models’ retrieval method based on the function bubble sketch and the prototype’s construction, thus it can provide more efficient services for the computer aided architecture design. The Dissertation work includes the following aspects:1. The architectural spatial relationship can be presented and formally described by constructing mathematical models, thus the quantified information can be used by the computer for ASR based intelligent information processing. This paper constructed the Space Syntax’s Attribute Relationship Graph(SSARG) based on the graph theory and Space Syntax theory, then proposed a formal description of the architectural spatial relationship’s related attributes——the Architectural Space Attribute’s Relationship Graph model(ASARG) and its extraction algorithm. Moreover, this article used the Revit API and C# language in the Revit platform and developed the relevant plug for the ASARG’s extraction and property calculation through the secondary development. Finally, the paper proposed the BIM graph matching test library’s design strategies and the corresponding BIM graph matching test library(BIMGM) based on XML, thus laid a solid data foundation for future research.2. In the field of structural graph-based pattern recognition, the existing graph matching methods have high complexity and lack related variables to describe the graph’s topological attributes. So it is difficult to satisfy the ASR-based BIM models’ matching or the generic graph matching speed’s requirements. A graph-kernel-based inexact graph matching method was proposed in this paper on the basis of SSARG and ASARG. By constructing the new graph kernels based on space syntax and shortest path, this paper could map the graph data space to the feature vector space implicitly, thus the similarity measurement between graphs was converted into the dot product in the feature vector space. Moreover, the SVM could be used to classify the graph data. Compared with other graph matching method, this method has lower complexity with higher classification recognition rate. Given that some of the graph data or the BIM models only have topological attributes, this method is a basic graph matching method providing better performance.3. In the domain of traditional graph-based pattern recognition, it is difficult to take advantage of the rich mathematical methods of the statistical pattern recognition. So it is difficult to satisfy the ASR-based BIM models’ matching or the generic graph matching speed’s requirements. Given the basic idea of the graph embedding, this paper proposed a series of graph embedding methods based on the SSARG and the ASARG, including the graph embedding method based on the topological features and domain features, the graph embedding method based on multi-scale features, the improved K-means clustering graph embedding method. With all these methods, the structural pattern recognition could be converted into statistical pattern recognition, thus the rich mathematical methods in the statistical pattern recognition field could be used to implement the inexact graph matching. Compared with other graph matching method, the above methods adopt the statistical methods and take fully advantage of the graph’s topological feature constructed based on the space syntax, thus the algorithms have lower complexity with higher classification recognition rate. These algorithms can be basic graph matching algorithms providing better performance for the graph data or the BIM models which have topological attributes and domain attributes.4. The aforementioned important feature of the BIM model is that it contains much implicit architectural spatial relationships. However, the existing BIM model retrieval methods fail to take advantage of this feature effectively and lack the appropriate retrieval interface. This paper proposed a BIM model retrieval method based on the function bubble sketch and constructed the prototype system. The retrieval conditions’ input was divided into two parts: the hand-drawn structural analysis sketch input and vector structural analysis sketch input. Moreover, the method extracted the information of the sketch, including the architectural space’s topological information and domain attribute, and constructed the ASARG model. Combining with the graph-embedding-based feature extract method, the ASARG was converted into feature vector space. Finally, this paper implemented the BIM model’s retrieve method based on the architectural spatial relationship by constructing the similarity measure function in the feature vector space. On the basis of the above steps, this article accomplished the construction of the prototype system, verified the effectiveness of this method through experiments and provided key technologies for the construction of the BIM model retrieval platform based on the function bubble sketch.Our study provides basic theory and key technologies for intelligent computer-aided architectural design and the vast amounts of information’s processing.更多还原