Efficiently approximating the probability of system failure has gained increasing importance as expensive simulations begin to play a larger role in reliability quantification tasks in areas such as structural design, power grid design, and safety certification among others. This work derives credible intervals on the probability of failure for a simulation which we assume is a realizations of a Gaussian process. We connect these intervals to binary classification error and comment on their applicability to a broad class of iterative schemes proposed throughout the literature. A novel iterative sampling scheme is proposed which can suggest multiple samples per batch for simulations with parallel implementations. We empirically test our scalable, open-source implementation on a variety simulations including a Tsunami model where failure is quantified in terms of maximum wave hight.

翻译：高效近似系统失效概率在结构设计、电网规划、安全认证等领域中日益重要，因为昂贵仿真开始在可靠性量化任务中发挥更大作用。本文推导了仿真失效概率的可信区间，假设该仿真是高斯过程的一个实现。我们将这些区间与二分类误差相关联，并探讨了其在文献中提出的各类迭代方案中的适用性。我们提出了一种新型迭代采样方案，能够为并行实现的仿真每批次推荐多个样本。我们通过多种仿真（包括以最大波高量化失效的海啸模型）对可扩展的开源实现进行了实证测试。