Intelligent hybrid approach for integrated design

The process of Total Design consists of numerous stages, such as the formulation of product design specifications, development of conceptual designs, detail design and manufacture. Conducting a design throughout the entire process is tedious and time- consuming, due to the complexity of each stage, often requiring redesign. In order to reduce production costs and time-to-market, it is highly desirable to automate the design process, using modem artificial intelligence (AI) techniques. An intelligent hybrid approach to integrate the stages of the Total Design process within a single environment has been developed. Integration has been achieved through a combination of rule based systems, artificial neural networks (ANNs) and genetic algorithms (GAs) with multi-media and CAD/CAE/CAM, providing a powerful tool for design automation. Both design integration and application of AI in engineering are currently attractive research topics, with a number of successful applications. However, the integration of multi AI techniques with CAD/CAE/CAM for Total Design has never been reported, hence, this project is novel research. The Total Design process has been evaluated with regard to identifying stages and decision making processes required to generate successful designs. The results of the evaluation are formulated, considering the methods of knowledge representation with an emphasis on a modular structure, forming the intelligent hybrid approach. Several methodologies have been developed within the intelligent hybrid approach including: design evaluation and knowledge acquisition, knowledge encapsulation, AI integration, system structure, adaptive design selection/retrieval, GA optimisation and ANN training. These methodologies combine to form an intelligent integrated system (IIS). An IIS for the design of mechanical power transmissions has been used as an application to help develop and validate the approach. The conceptual design stage combines a rule base with a series of ANNs to generate the conceptual arrangement and method of transmission between shafts. The detail design stage takes particular advantage of the modular structure that is encouraged within the IIS to breakdown the design process. Design modules relating to individual component designs interact with each other, successfully applying AI for decision making, information manipulation and design optimisation, using the single environment to combine and exploit the strengths of different techniques. The use of backpropagation ANNs provoked an investigation into the training process. The conclusion from which indicated that no general rule exists to determine the training parameters that create high performance networks. The necessity for a method of simplifying training led to the integration of a GA to the training process which, adaptively alters training parameters, improving network performance irrespective of the application. The application of GAs to design optimisation has proved very effective at emulating expertise. The technique enables high quality designs to be tailored to applications without extensive knowledge of the particular field. Additionally, an investigation into the evolutionary process has overcome the traditional GA problems of computational expense and repeatability of results, enabling their inclusion in the approach