ABSTRACT Probability of detection (POD) is commonly used to measure a nondestructive evaluation (NDE) inspection procedure’s performance. Due to inherent variability in the inspection procedure caused by variability in factors such as crack morphology and operators, it is important, for some purposes, to model POD as a random function. Traditionally, inspection variabilities are pooled and an estimate of the mean POD (averaged over all sources of variability) is reported. In some applications it is important to know how poor typical inspections might be, and this question is answered by estimating a quantile of the POD distribution. This article shows how to fit and compare different models to repeated-measures hit--miss data with multiple inspections with different operators for each crack and shows how to estimate the mean POD as well as quantiles of the POD distribution for binary (hit--miss) NDE data. We also show how to compute credible intervals (quantifying uncertainty due to limited data) for these quantities using a Bayesian estimation approach. We use NDE for the detection of fatigue cracks as the motivating example, but the concepts apply more generally to other NDE applications areas.

中文翻译：

---

使用命中-未命中数据进行无损评估的分位数 POD

摘要 检测概率 (POD) 通常用于测量无损评估 (NDE) 检测程序的性能。由于裂纹形态和操作员等因素的可变性导致检查程序的固有可变性，因此出于某些目的，将 POD 建模为随机函数非常重要。传统上，检验变异被合并，并报告平均 POD（所有变异来源的平均值）的估计值。在某些应用中，了解典型检查的糟糕程度很重要，而这个问题可以通过估计 POD 分布的分位数来回答。本文展示了如何将不同模型与重复测量命中-未命中数据进行拟合和比较，并对每个裂缝使用不同的操作员进行多次检查，并展示如何估计平均 POD 以及二进制 (hit--错过) NDE 数据。我们还展示了如何使用贝叶斯估计方法计算这些数量的可信区间（由于数据有限而量化不确定性）。我们使用 NDE 检测疲劳裂纹作为激励示例，但这些概念更普遍地适用于其他 NDE 应用领域。