SPC有用统计手法.ppt

* * * * * REVIEW and reinforce definitions Select a few items in each column and discuss in more detail. * 如果出现系统特殊原因—即特殊原因出现在每个子组中，则 X, R 图的优点就消失了。例如，假设您在计算通过电话收到的订单中的错误数，而且有四个接线员在接订单。那么很自然会想到构造大小为 4 的子组，即从每个接线员手中取一份订单表。但如果一个接线员一直比其他人好或差，您可能会混淆数据中的特殊原因变化和普通原因变化。此图将变得无用，它使接线员之间的差别模糊而且使检测过程或可变性中的变化很困难。 在用下列数据构成子组时应仔细考虑： 来自不同接线员、机器、班次、设备等的值。 按星期、月或季度确定的数据。 * * * X, R 图使用子组内的变化建立子组平均值的界限。 当子组之间比子组内有更大变化时，表明存在特殊原因。 选择子组至关重要。 * 本页的上图是子组平均值图。下图显示每个子组中存在多大的变化。例如： 第一个子组的平均值大约为 13（上图）。 子组中最低点和最高点之差是 2 个单位（下图）。 第 8 个数据点的平均值大约为 11.2，最低点和最高点之间的范围是 5 个单位。 * * * * * * * * * DISCUSS: These decision rules are set up so that there is roughly the same probability of violating any rule as there is of being outside the 3 s limits (about .3%). Some sources may quote various counts of points in a row to signal a special cause - they are willing to take more (or less) of a risk of signaling a special cause when none is present than these rules allow for. INSTRUCTOR NOTE: Not all rules are used by all areas. The local management or quality team should decide which rules are appropriate for the area - the references below provide guidance in this selection. People sometimes have trouble understanding the rules based on so many successive increasing or within a band. One way to think about it is comparing it to flipping a coin. Seeing a head or a tail on one flip is nothing extraordinary; seeing 12 heads in a row may be a sign that the coin is not fair. The rules are ways of detecting if the process is fair. REFERENCE: For additional information on how the various rules were established and how they should be applied, see the articles: ? The Shewhart Control Chart – Tests for Special Causes, Vol. 16, No. 4, October, 1984, Journal of Quality Technology, by Lloyd Nelson ? Interpreting Shewhart X-Bar Control Charts, Vol. 17, No. 2, April 1985, Journal of Quality Technology, by Lloyd Nelson * EXERCISE: Have the