This paper describes the use of approximation in Collaborative Optimization (CO) method, one of the Multidisciplinary Design Optimization (MDO) techniques. The approximation is used to model the result of a disciplinary design, optimal discrepancy function value, as a function of the interdisciplinary target variables passed from system level to the discipline. The optimal discrepancy function value is used to examine the interdisciplinary compatibility constraint (discrepancy function = 0) duringthe system level optimization. However, the peculiar shape of the compatibility constraint makes it difficult to exploit well–developed conventional approximation methods. This paper introduces the combination of neural network classification and kriging to resolve this problem. In addition, for the purpose of enhancing the accuracy of the approximation, the approximation is continuously updated using the information obtained from the system level optimization. This iterative process is continued until acceptable convergence is achieved.

Abstract
1. Introduction
2. Using Approximation in CO
3. Modeling Optimal Disciplinary Design
    3.1 Introduction of Classification Neural Network
    3.2 Using Implicit Extra Point Information
4. Managing Approximation Models inOptimization
5. Illustrative Example
    5.1 Mathematical Example
    5.2 Hub frame example
6. Conclusions
Acknowledgements
References

• Chang-Kyu Park(School of Mechanical & Automotive Engineering, Gwangju University) | 박창규 Corresponding author