摘要：AbstractKnowledge of the critical region of an input pdf (tail, near-tail, central region) can provide valuable information towards understanding and improving a model through additional modeling or testing. A localized probabilistic sensitivity method was developed and demonstrated that will determine the region of a probability distribution that most affects the response. The methodology is based on discretizing the random variable pdfs using linear interpolation then sequentially injecting a localized disturbance into the pdf. A partial derivative of the probability-of-failure or the response moments (mean and standard deviation) with respect to each localized disturbance can be determined using sampling methods. The methodology can correctly locate the important region given sufficient samples.
