Abstract:In order to balance the construction cost and the prediction accuracy of the surrogate model used to estimate structural system reliability with multi-failure modes, the Largest Contribution Function (LCF) is proposed. The LCF is used to identify the best samples from candidate samples in the construction process of surrogate model. Since the best samples have great influence on the variance of system failure probability, the prediction accuracy of system failure probability can be improved. Mathematical formulas of LCFs for the series system, parallel system and series-parallel system are derived based on the structural reliability theory. The learning stopping condition of LCF is established by using confidence level and allowable relative error, it can ensure that the information about existing samples is not wasted. A multi-output Kriging model is selected as surrogate model, it can approximate multiple performance functions at the same time. The structural system reliability is calculated by using the LCF-Kriging model and MCS, the correlation between performance functions can be considered by the logical relationship of failure modes. Some examples are introduced to demonstrate the efficiency and the accuracy of the proposed method. It is found that the proposed method can achieve the satisfactory balance between the construction cost and the accuracy for structural reliability estimations of series system, parallel system and series-parallel system under the appropriate learning stopping condition.