Global sensitivity analysis approaches applied to parameter selection for numerical model-updating of structures

Abstract

Purpose This paper aims to measure the sensitivity of the structure’s deformation numerical model (NM) related to the various types of the design parameters, which is a suitable method for parameter selection to increase the time of model-updating. Design/methodology/approach In this research, a variance-based sensitivity analysis (VBSA) approach is proposed to measure the sensitivity of NM of structures. In this way, the contribution of measurements of the structure (such as design parameter values and geometry) on the output of NM is studied using first-order and total-order sensitivity indices developed by Sobol’. In this way the generated data set of parameters by considering different distributions such as Gaussian or uniform distribution and different order as input along with, the resulted deformation variables of NM as output has been submitted to the Sobol’ indices estimation procedure. To the verification of VBSA results, a gradient-based sensitivity analysis (SA), which is developed as a global SA method has been developed to measure the global sensitivity of NM then implemented over the NM’s results of a tunnel. Findings Regarding the estimated indices, it has been concluded that the derived deformation functions from the tunnel’s NM usually are non-additive. Also, some parameters have been determined as most effective on the deformation functions, which can be selected for model-updating to avoid a time-consuming process, so those may better to be considered in the group of updating parameters. In this procedure for SA of the model, also some interactions between the selected parameters with other parameters, which are beneficial to be considered in the model-updating procedure, have been detected. In this study, some parameters approximately (27 per cent of the total) with no effect over the all objective functions have been determined to be excluded from the parameter candidates for model-updating. Also, the resulted indices of implemented VBSA were approved during validation by the gradient-based indices. Practical implications The introduced method has been implemented for a circular lined tunnel’s NM, which has been created by Fast Lagrangian Analysis of Continua software. Originality/value This paper plans to apply a statistical method, which is global on the results of the NM of a soil structure by a complex system for parameter selection to avoid the time-consuming model-updating process.