摘要
A new and convenient method is presented to calculate the total sensitivity indices defined by variance-based sensitivity analysis. By decomposing the output variance using error propagation equations, this method can transform the "double-loop" sampling procedure into "single-loop" one and obviously reduce the computation cost of analysis. In contrast with Sobors and Fourier amplitude sensitivity test (FAST) method, which is limited in non-correlated variables, the new approach is suitable for correlated input variables. An application in semiconductor assembling and test manufacturing (ATM) factory indicates that this approach has a good performance in additive model and simple non-additive model.
A new and convenient method is presented to calculate the total sensitivity indices defined by variance-based sensitivity analysis. By decomposing the output variance using error propagation equations, this method can transform the "double-loop" sampling procedure into "single-loop" one and obviously reduce the computation cost of analysis. In contrast with Sobors and Fourier amplitude sensitivity test (FAST) method, which is limited in non-correlated variables, the new approach is suitable for correlated input variables. An application in semiconductor assembling and test manufacturing (ATM) factory indicates that this approach has a good performance in additive model and simple non-additive model.