This paper presents the development of a CAD-based expert framework aide for the decision making on choosing optimal design methods. The proposed framework is based on a comparative study of gradient enhanced meta-models using response surface modeling (RSM) and Kriging techniques, the two most promising meta-modeling techniques studied extensively in design optimization schemes for approximating complex systems. However, current and past research shows that the performance of these methods mandates significant improvements due to the fact that a large number of sample points are needed from within the design space for a considerable accurate model. This makes the designers face a dilemma between accuracy and computational cost. To overcome this so called curse of dimensionality problem, a proper framework with which the models can be validated and its reliability can be evaluated is proposed and established in this research work. Our approach is aimed at forming a GUI-based software framework (utilizing DACE) in MATLAB® that will aid the designers to decide which Meta-model would be most suitable for a certain modeling for design. It will help designers to plan the design of experiments and train the models using the gradient information available at low cost and consequently, validate the model with existing data, as well as produce error analysis graphs and performance criterion charts of all the models for a final decision making. A comparative study of performance of the gradient enhanced RSM and gradient enhanced Kriging models are conducted for six different test equations of various degrees of non-linearity using this proposed framework. A Pro/Engineer® model of a bracket is taken as a test case for the verification of the efficiency and effectiveness of the proposed expert framework as a decision-making tool for designers of engineering products. All the tested Metamodels performed satisfactorily within the specified error margins.

This content is only available via PDF.
You do not currently have access to this content.