Time-dependent Reliability Evaluation Method Based On Response Extreme Value

Structural reliability generally is a function of time and degrades over time due to the presence of time-variant uncertainties such as uncertain material properties,loading conditions and component deterioration.In order to guarantee structural safety during its service time,it is urgent to study accurate and efficient time-dependent reliability assessment methods.Besides,another challenge comes from the curse of dimension if the associated limit state function involves a large number of input random variables.The sampling-based methods can be used in time-dependent reliability,but they suffer from a huge computational effort.Existing approximate methods,mainly including first-passage based methods and extreme value based methods,have shortcomings.The limitation of the former is to ignore the correlation between the up-crossing time and the error associated with linearization,while the latter depends on the accuracy of the surrogate models.Aiming at the above issues,the main contents in this study are listed as follows:A maximum entropy approach is proposed for time-dependent reliability analysis based on response extreme value.The approach firstly discretizes the input stochastic process with a group of random variables.Then Latinized Centroidal Voronoi tessellation is used to generate a sample of inputs and the extreme values of the limit state function are obtained in the target time interval.Finally,the maximum entropy principle is employed to fit the extreme value distribution.Thus the time-dependent reliability analysis problem is converted into its time-independent counterpart.This method improves computational efficiency in practical examples.An evaluation method is proposed to solve the high-dimensional time-dependent reliability,which involves a large number of inputs in the limit state function,using the principle of maximum entropy principle with fractional moments.The maximum entropy principle with fractional moments is derived;the optimization problem of calculating the maximum entropy distribution is reconstructed with the fractional moment and Lagrange multipliers' side constraints and is solved by subset simulation optimization.Two problems with high-dimensional inputs show that the proposed method has both good accuracy and efficiency.The kernel density estimation methods are proposed to fit the extreme value distribution for timedependent reliability analysis.Compared with Monte Carlo simulation,the computational results show that the proposed methods take advantage of non-parametric estimation methods and improve computational efficiency in practical examples.