WIN : a case-based design approach to the structural design of the wind systems of buildings

This thesis concerns the application of Case-Based Design (CBD) to the domain of the structural design of buildings. The aim of this thesis is to design and develop WIN, a prototype Case-Based Design Aid System (CBDAS) that has the ability to retrieve the most relevant case for a new design problem with a minimum of information. This thesis discusses background knowledge about structural design. It addresses the fact that structural design is a complex problem solving process. There is no one perfect method for performing structural design, and a good design is the result of recursively selecting methods based on a number of criteria. In design practice, human designers may start the process by remembering previous designs that are similar to a new design problem, and then select the most relevant case as the basis for the new design process. This thesis argues that using a Case-Based Design Aid System (CBDAS) for the structural design of buildings can improve the efficiency and the quality of the resulting designs because this approach utilises the advantages of combining computer memory with the human designer’s cross—domain, dynamic—reasoning ability. On the other hand, this approach can reduce the time and cost involved in the development of a CBDAS. Therefore, using a CBDAS in design practices becomes a foreseeable reality. Since structural design is a complex problem solving process, the information used to describe an existing design case is enormous. It is important to determine what the contents of existing design cases should be. This thesis claims that only the necessary information should be included as the contents of existing design cases. In order to have a better understanding of the contents of existing design cases, it is possible to split existing design cases into their situations, solutions and the knowledge used to generate the solutions. In the implementation of WIN, this thesis uses a multimedia case representation that combines text descriptions, attribute—value .pairs, Function—Behaviour—Structure (FBS) models, and graphical representations. Thus, human designers can have a better understanding of the existing design cases by reviewing their text descriptions, drawings and diagrams; the computer can use the attribute-values pairs to autonomously reason about the similarities between a new design problem and existing cases. In order to achieve an efficient case representation, it is necessary to divide a complex design case into a set of less complex subcases. An existing design case in WIN is decomposed into a root case and its subcases. The root case and its subcases are arranged in a case hierarchy. The root case represents the general information about the existing design, and the subcases represent the detailed information about the aggregations of physical components and the structural components of the existing design. After discussing the key issues related to applying CBDAS technology to the structural design of buildings, this thesis presents the case retrieval strategy used in WIN: model-based, human-controlled, progressive and iterative index elaboration. The major reasons supporting the case retrieval strategy in WIN are summarised below: - predefined models represent sets of generalised heterogeneous grouping of elements derived from similar design cases that provide the basis for the start and continuation of a design, these predefined models can be used to partition the case library; - design cases are organised in different partitions in the case library, finding the relevant cases for a new design problem can be achieved by finding a predefined model related to the new design problem, which leads to retrieving the relevant cases; - the contents of an existing design case are represented in its root case and sub— cases, finding the contents of an existing case means exploring the contents of the existing case along its case hierarchy, this exploration is a incremental process; - in each level of index exploration, it is essential to use different selection conditions for different situations, the approach of using iterative index refinement provides the opportunity for human designers to select different combinations of indices to evaluate the existing design cases; - with the human designers’ interaction, the retrieval of cases is more dynamic and efficient, on the other hand, the time and the cost involved in the implementation of a CBDAS is reduced. Finally, this thesis presents the implementation of WIN. This system contains a case library, a model library, a retriever, a selector, and a user interface. WIN is developed on SUN SPARC-stations. The existing cases and the predefined models are written in Framekit, the user interface, the retriever, and the selector are written in Common Lisp. The performance of WIN is demonstrated using an example problem. The approach of using CBDASs to find the most relevant case for a new design problem is shown to be considerably easier and more efficient than using a conventional case library. The issues discussed in this thesis are likely to be the basis of future development of CBD systems