Abstract
A methodology which can automatically arrange and choose the unknown parameters for efficient back analysis is proposed and discussed. Interpreting a back analysis as estimation of the expansion coefficients of basis vector of unknown parameter space, 1) determine the basis vector from eigen mode of a posteriors covariance matrix of unknown parameters, 2) determine the dimension of subspace to back-analyze, based on the criterion with information entropy, 3) the expansion coefficients of basis vector composing the subspace are back-analyzed, then the unknown parameters are estimated. The proposed methodology is demonstrated through numerical examples, in which the dimension and basis vectors of the subspace are determined according to the quality and quantity of given observation data.
