A new methodology based on covariance and HDMR for global sensitivity analysis

To execute variance based global sensitivity analysis efficiently and purposefully, a computing methodology containing two covariance methods is proposed. The first method estimates sensitivity indices by making full use of a sampling matrix and a re-sampling matrix. The second one is proposed for compensating the systematic error of the first method. Sources of error of the two methods become clear when utilizing high-dimensional model representation technique, then the application scope is obtained and verified by test examples, and it can help researchers choose an appropriate method according to the size of sensitivity index of a given variable. Compared with other sampling-based methods, the new methodology is proved to be efficient and robust by examples.