An efficient dynamic response optimization using the design sensitivities approximated within the estimate confidence radius

In order to reduce the expensive CPU time for design sensitivity analysis i n dynamic response optimization, this study introduces the design sensitivi ties approximated within estimated confidence radius in dynamic response op timization with ALM method. The confidence radius is estimated by the linea r approximation with Hessian of quasi-Newton formula and qualifies the appr oximate gradient to be validly used during optimization process. In this st udy, if the design changes between consecutive iterations are within the es timated confidence radius, then the approximate gradients are accepted. Oth erwise, the exact gradients are used such as analytical or finite differenc ed gradients. This hybrid design sensitivity analysis method is embedded in an in-house ALM based dynamic response optimizer, which solves three typic al dynamic response optimization problems and one practical design problem for a tracked vehicle suspension system. The optimization results are compa red with those of the conventional method that uses only exact gradients th roughout optimization process. These comparisons show that the hybrid metho d is more efficient than the conventional method. Especially, in the tracke d vehicle suspension system design, the proposed method yields 14 percent r eduction of the total CPU time and the number of analyses than the conventi onal method, while giving similar optimum values.