Efficient optimization method for noisy responses of mechanical systems

Abstract

To optimize the noisy and computationally expensive mechanical system, a new conservative response surface modelling method is introduced, in which the duality theory is used to directly represent the over- or underestimation behaviours of the approximated response surface models. Although, the incomplete small composite designs are newly proposed to reduce the number of experimental runs, it may induce rank-deficiency in the normal equation. Thus, the singular-value decomposition is suggested to solve the normal equation. Then a sequential approximate optimization program is developed. To show the numerical performance two design problems are solved, such as a well-known gear reducer design problem and the tracked vehicle suspension system design. The test results show the efficiency of the proposed method.