In the field of mechanical product design, there have been difficulties in building a general model to represent a design process since the type of mechanical products is diverse and the design process requires different complex domain knowledge. This results in the need for different domain experts to design special system development schemes for different kinds of mechanical products, which greatly increases the cost of the research and development. In addition, there have been difficulties in the automation, intellectualization and acceleration of mechanical product design process, as the process is based on experiential knowledge and relies extensively on the synthetic decision of experts (SDE), which is hard to be integrated into the design process without the participation of experts in person. To cope with these two issues, a knowledge-based automated design (KAD) system for mechanical products was developed in this study. Firstly, a general feature design flow (GFDF) framework was proposed to represent the design knowledge of various mechanical products from initial design stage, including requirement analysis, to the final automated generation of computer-aided design (CAD) model. Based on parametric technology and application programming interface (API) functions, the GFDF framework decomposes the explicit and implicit knowledge of a top-down design process into four rank-correlated features and two feature correlation matrices (FCMs) respectively. Based on the GFDF framework, a requirement analysis method using analysis hierarchy process (AHP) was proposed to transfer qualitative features to quantitative parameters. Secondly, by adopting the support vector regression (SVR) machine, a feature reuse case adaptation (FR-CA) method based on case-based reasoning (CBR) was proposed to achieve the automation and intellectualization of parameter solving integrating the SDE without the actual participation of experts. The FRCA method transforms the SDE into an FCM, which can be identified during the case adaptation process, and maximizes the utilization of solved features in each adaptation process. A comparison experiment between FR-CA and conventional method for case adaptation indicated that the adaptation performance of FR-CA method is better than that of the conventional one. Finally, the KAD system was applied to the design of corn huskers. The result showed that the KAD system could improve the automated, intelligent and rapid design of mechanical products.
Read full abstract