石油和相关产品.测量方法和结果的精度.第4部分:使用统计控制图验证单个实验室标准试验方法执行的“统计控制”状态

[中文适用范围]： 本文件规定了统计控制图的构建、操作和维护的过程和方法，以评估实验室执行的标准测试方法是否处于统计控制范围内，以及如何建立和验证“统计控制范围内”地位。 它指定了最适合 ISO/TC 28 测试方法的控制图，其中主要的共因变化与长期、多个操作员条件相关。 指定用于确定统计控制内的控制图有：个体 (I)、移动极差 2 (MR2) 以及指数加权移动平均 (EWMA) 或基于区域的运行规则 [类似于 Western Electric (WE) ）运行规则[3]]作为灵敏度增强策略来支持I图。 本文件中的程序主要设计用于通过测试由同质石油和相关产品来源制备的对照样品获得的数值结果，以保持对照样品之间感兴趣的性质的均质性。 如果测试方法允许，则使用有证参考物质 (CRM) 样品作为对照样品，前提是样品成分能够代表被测材料并且不是纯化合物；如果这样做，那么实验室最好为 CRM 样本建立自己的平均值。 本文档适用于（已知）随时间稳定的感兴趣属性，以及具有足够分辨率的数据集，以支持验证数据分布可以由正态（高斯）模型近似表示的假设。 对于无法验证假设的情况，建议采取缓解策略。 [外文原描述]： This document specifies the process and methodology for the construction, operation, and maintenance of statistical control charts to assess if a laboratory's execution of a standard test method is in-statistical-control and how to establish and validate the 'in-statistical-control' status. It specifies control charts that are most appropriate for ISO/TC 28 test methods where the dominant common cause variation is associated with the long term, multiple operator conditions. The control charts specified for determination of in-statistical-control are: individual (I), moving range of 2 (MR 2 ), and either the exponentially weighted moving average (EWMA) or zone-based run rules [similar to Western Electric (WE) run rules [ 3 ] ] as sensitivity enhancement strategy to support the I-chart. The procedures in this document have been primarily designed for numerical results obtained from testing of control samples prepared from a homogenous source of petroleum and related products in a manner that preserves the homogeneity of properties of interest between control samples. If the test method permits, a certified reference material (CRM) sample is used as a control sample provided the sample composition is representative of the material being tested and is not a pure compound; if this is done then the laboratory best establishes its own mean for the CRM sample. This document is applicable to properties of interest that are (known to be) stable over time, and for data sets with sufficient resolution to support validation of the assumption that the data distribution can be approximately represented by the normal (Gaussian) model. Mitigating strategies are suggested for situations where the assumption cannot be validated.