Not available / text

Knowledge bases require a considerable investment of time and money to construct. For many applications the dual use of a knowledge base in an expert system and in a teaching system would be cost effective. This research addresses the issue of whether a single knowledge base that can be used effectively in an expert system can form the basis of a functionally effective teaching system. To understand what is meant by a "functionally effective" teaching system a review of computer-aided instruction and intelligent computer-aided instruction systems has been carried out. From this, a list of twelve functional requirements were identified that should be met by the subject knowledge contained in a teaching system, and therefore by a dual-purpose knowledge base. A review of expert systems was also undertaken, and this resulted in the generation of the hypothesis that the deep knowledge approach to the construction of expert systems could be used for the construction of a dual-purpose knowledge base. A test system, DR MECHO, was constructed from the MECHO program, which solves problems in the domain of statics. This new system included: a deep knowledge base; a meta-interpreter, to provide explanations of the reasoning process of DR MECHO; and a set of actions, referred to as teaching actions, which could replace some of DR MECHO's usual problem solving operations with student input. The performance of DR MECHO as an expert system, and its ability to meet the functional teaching requirements, were evaluated. DR MECHO performed well as an expert system. It also met most of the teaching requirements. The main underlying reasons for the latter were: its performance as an expert system, the explicit structuring of the knowledge base, and the observance of the "no function in structure" principle (the latter two being characteristics of deep knowledge systems

An expert system approach to fault diagnosis of fluid power circuits is considered with emphasis on leakage flow detection, and for valvecontrolled cylinders. Two test rigs were used, one being a solenoid-valve controlled cylinder operated directly and in an open-loop mode, the other being a servo-valve controlled actuator operated by microcomputer and in a closed-loop mode. Both systems incorporated the use of on-line dynamic data, and for the closed-loop case operation and fault diagnosis was integrated into an automated procedure. Flow leakage detection was considered a priority, and an alternative approach using displaced volumes was successfully implemented. The research work concentrated initially on the use of an expert system and the establishment of an appropriate knowledge base using a hybrid reasoning approach. This approach was found to be excellent for single-fault conditions but could not differentiate components of multiple-fault conditions, other than that they existed, due to the use of a minimum number of flow sensors. Additional techniques were then considered for the closed-loop control system utilising steady-state position error, time series analysis, and Artificial Neural Networks. It was found that the consideration of steady-state error gave information complementary to the existing knowledge base but could not give any additional information. The use of an artificial neural network was found to give more information with regards to multiple-fault conditions, resulting in a percentage probability for each fault combination.

Despite the development of computational technology and AI over the past 40 years, it is the case that the application of computers for architectural uesign, especially in the early design stage, still remains in its infancy. This thesis searches for a solution, from a theoretical framework, for the computation of design, which is different from the problem-solving paradigm often adopted. Firstly, the thesis accounts for designing as a phenomenon of design thinking-action. That is, the two activities - doing the thinking and design action - usually occur and develop simultaneously, and are characterised as a creative activity and a visual thinking process. Secondly, in order to account for the mental mechanism occurring in design thinking-action, this thesis describes the designer's mind as with a metaphor derived from current theories in image processing research in cognitive science; the ideas of mental space. In the thesis, mental space is defined as a conscious system, which has its structure and functions that can transfer external events into inner symbolic representations (design thinking) and simultaneously visualise these internal representations during the external process (design action). Thirdly, based on these theoretical assumptions, I propose a mental space computational model, which is a design computational environment to attempt to mimic the mental operations and processes in the architect's mental space. It focuses on design activities rather than design cognition; the usefulness of computers for design rather than the computability of design; and design tools rather than memorybased intelligent systems. Throughout this thesis, I try to avoid the dualistic arguments which classify architectural design as either artistic or scientific, so that I can provide an inclusive theoretical foundation in explaining design phenomena for general design studies as well as for CAAD (Computer-Aided Architectural Design).

Not available / text

No description available.

No description available.

No description available.

No description available.

No description available.