An efficient reliability analysis method for structures with hybrid time-dependent uncertainty

Performing time-dependent reliability analysis is an effective way to estimate the failure probability of structural system throughout its lifetime. In the engineering practices, uncertain parameters with sufficient sample and limited sample may exist simultaneously. The uncertain parameters with limited sample data are difficult to construct its precise probabilistic characteristics during estimating the accurate time-dependent reliability. To address this issue, this paper first develops a new hybrid time-dependent reliability model involving interval processes. Then, to reduce the high dimensionality, an extension method based on equivalent stochastic process transformation approach is proposed to transform the stochastic processes and the interval processes into corresponding equivalent random variables respectively. In particular, an instantaneous reliability model is constructed to envelope all potential system failures that may occur during the time interval. In order to identify the instantaneous failure surface accurately, an active learning method is proposed based on the deep neural network model and the weighted sampling method. With the constructed deep neural network model, the new hybrid time-dependent reliability can be evaluated by performing the Monte Carlo Sampling. Three numerical examples are used to verify the accuracy and efficiency of the proposed method.