Efficient response surface method for high-dimensional structural reliability analysis

In common response surface method (RSM) for structural reliability analysis, performances of several experimental points must be evaluated via finite element analysis. The number of required experimental points is proportional to the number of random variables. Hence, for a high-dimensional structural reliability problem, computational cost is high, especially for structures with computationally intensive finite element models. On the other hand, the accuracy of classical RSM in estimating the probability of failure depends on the locations of experimental points. This paper proposes an efficient and accurate RSM. The efficiency is increased by using exponential surrogate model instead of quadratic one and by using experiment updating technique. In this way, the number of required experimental points is significantly reduced. Meanwhile, the accuracy of RSM is improved by choosing locations of experimental points judiciously, such that, to be close to the actual limit state surface. Five examples are solved to demonstrate the high efficiency and accuracy of the proposed method.