Multivariate statistical sensitivity analysis of the completed structural state for Long-span Cable-Stayed bridges

Conventional statistical sensitivity analysis approaches can be prohibitively expensive or impracticable computationally for the long-span cable-stayed bridge involving many parameters. In this study, a novel statistical sensitivity analysis strategy is proposed to enhance the feasibility for the construction process and the completed structural state of long-span cable-stayed bridges. A uniform test design is first incorporated into the equivalent marginal distribution problem in statistics to characterise the parameter variations, and parameter significance testing is employed to identify the significant structural parameters. Moreover, the parameter substructure is introduced to the determination of random variables during the test design process to balance the computational efficiency and structural parameter variation. Thus, the structural components fabricated or constructed in the same batches or under similar construction conditions are grouped into a single parameter substructure, and each structural parameter in the substructure is defined as one random variable. When the number of structural parameters becomes so large that the uniform test design is difficult to perform effectively, the structural parameters are then grouped to ensure the uniform test design is implementable. After the significant parameters and surrogate models of the sub-group test designs are determined, the integrated surrogate model including all of the significant parameters is obtained based on the property of the normalised regression coefficients in multivariate stepwise regression. Furthermore, t-tests are used to determine the significant structural parameters for the structural responses. A long-span cable-stayed bridge with a composite/hybrid girder is considered as a case study to verify the proposed multivariate statistical sensitivity analysis. The discussion reveals that the impacts of structural parameter variations on the structural responses are not only related to the variation ranges of structural parameters but also to the relative position between the section to which the structural parameter belongs and the section to which the structural response belongs.