Expert System for Generation and Evaluation in Architectural Design

The research field, focuses on a new research area of Knowledge Based Systems for Architectural Design. The research deals with concepts and tools emerging from Artificial Intelligence, Knowledge Based Systems and Expert Systems. The research is involved with the construction of a theoretical basis for the development of approaches and methods for the representation and control of design knowledge as a reasoning process. Key questions which attempt to reconsider representation and control in design are formulated. The following questions serve as a research framework out of which new approaches, methods and tools were developed. (1.)What are the existing ideas, methods and tools in Expert Systems? (2.) What are the performance characteristics of Expert Systems in Architectural Design ? (3.) What are the desired operative characteristics and interactions for Expert Systems in design ? (4.) How is it possible to formulate and apply the diverse forms of Architectural Knowledge in Expert Systems for design? (5.) What are the problems of implementation in the development of Expert Systems for design ? The state of the art in knowledge based systems is surveyed, while emphasizing the differences between conventional systems and knowledge based systems. Representation and control methods and the components of expert systems are reviewed. Expert systems for diagnosis, interpretation, planning and design are analysed with respect to their performance characteristics. Techniques and technologies of existing tools are defined. An expert system for the generation and evaluation of ill defined architectural design problems is develped. A formalization of the concept of'design interpretation'is proposed and developed. It is applied in the process of defining and classifying the performance characteristics of expert systems for design. This concept is based upon two sets of reasoning processes: those which enable a mapping between design requirements and solution descriptions in the generation stage of design and those between solution descriptions and performance evaluation in the evaluation stage of design. On the basis of the formalization of this concept, an expert system capable of integrating various modes of performance is proposed and developed. The system functions as a'design generator', a'design critic', or a'design critic-generator'. These modes, which integrate generation and evaluation in the same system, operate by employing both forward chaining and backward chaining inference mechanisms. As a result of the examination of desired forms of interactions, a new approach for dual direction interpretation between graphic and verbal modes is developed. This approach reflects the importance of both graphical and verbal expression in design. The approach is based upon a simultaneous mapping between symbolic-verbal interpretation and graphic interpretation. The work presents the mapping process through the concept of design interpretation, employing geometrical knowledge, typological knowledge and evaluation knowledge. A tool which provides communication between an expert system and a graphic system was developed and is presented. The importance of such a tool in expert systems for design resides in the provision of free choice to the user for interacting with the system either graphically or verbally during the design process. An additional component in the development of knowledge-based systems for design is related to the important question of knowledge definition and the representational schemata of design knowledge. A new representational scheme for complex architectural knowledge, termed'The generation and refinement scheme of a design prototype'is proposed and developed. Its operation as part of a total integrated design system is demonstrated. The scheme is based upon the structures of knowledge of design precedents which constitute typical situations and solutions in architectural design. This scheme provides an appropriate representation for the two types of knowledge which operate in a refinement process of a design prototype. Generative knowledge describes the solution space by predefined refinement stages, interpretive knowledge enables their selection. The examination of representational methods for the proposed scheme indicated that employing a single representational method lacked enough generalization and expressive power for the needs of the design knowledge structures. It was found that a way to represent complex structures is through the integration of multiple methods of representation, each one according to the knowledge characteristics. In order to represent the proposed scheme of design knowledge, a unique method was developed which integrates both rules and frames. The method consists of a rules-frames-rules structure for the representation of a design prototype. An approach is developed for the implementation of these concepts in an expert system for design. PRODS: A prototype based expert system shell for design is developed and demonstrated. The system consists of three basic components: a rule-based expert system shell, a frame system, and a knowledge base interface. All system interactions are controlled by the inference engine. It passes control between the rule-base and the frame-base inference engines, and provides communications between the rule-based and frame-based representations. It is suggested that expert system can interface with external CAD systems including graphics, communicating through a central representation. These concepts and developments are demonstrated in two implementations. The PREDIKT system for the preliminary design of the residential kitchen, the PROUST system for the selection and refinement of dwelling types. PREDIKT demonstrates the integration of rules and a graphical-verbal interpreter, in addition, PROUST demonstrates the significance of hybrid representation in the generation and refinement processes. The results and conlusions are summarized. Future research agenda within the field of knowledge-based systems for design is discussed, and potential research areas are defined