Computational Efficiency of First and Second Order Sensitivity Analysis Methods for Dynamic Systems
Abstract First and second order design sensitivity information have become very popular in engineering design, due to the recent development of appropriate methodology and computer technology. This paper presents an empirical study of the computing effort required for computing first and second order design sensitivity information for constrained dynamic mechanical systems and compares the relative efficiency of analytical and finite difference method. Four typical examples have been solved, with the computing time being recorded in each case. The time comparison results indicate that it is generally much more efficient to produce first order design sensitivity information by a direct analytical method than by finite differencing. For second order design sensitivity analysis, the results indicate that a purely analytical solution is usually more efficient than a pure finite difference solution. However, a hybrid scheme appears to be very competitive in terms of computational efficiency.