《绿带培训之控制图》由会员分享，可在线阅读，更多相关《绿带培训之控制图（28页珍藏版）》请在金锄头文库上搜索。
1、M ADI CTimken Lean6Sigma Training 铁姆肯L6S培训Introduction to Control Charts控制图简介lControl Chart Basics控制图基础控制图基础lX-Bar and S Control Charts均值和标准差控制图均值和标准差控制图lIndividuals Control Charts单值控制图单值控制图lOther Common Control Charts其他常见控制图其他常见控制图lFinal Thoughts结束语结束语 2002 North Haven Group2M ADICUnderstanding Var
2、iation Revisited理解波动-回顾Recall that we defined three categories of process variation:回忆一下，我们把过程波动定义为三种类型：Common Cause Variation: Natural background variation or process noise that is always present and is inherent to the system. This is small scale variation. 偶因波动：自然本底波动或过程噪音，它是一直出现的，并且是系统固有的。Special
3、 Cause Variation: Large scale variation that may permanently change the process due to some assignable or special sources of variation. The causes are external. 异因波动：大幅度的波动，也许会由于一些可归因的或特殊的波动源，而永久性地改变过程。原因是外部的。Systematic Variation: Large scale variation that systematically changes the process mean an
4、d is inherent to the process.系统波动：大幅度的波动，有规律地改变了过程的均值，是过程固有的。A stable or in control process is subject to only common cause variation. An unstable or out of control process is subject to common cause, special cause and/or systematic variation. 一个稳定的过程是指只存在偶因的波动。 不稳定的过程是指存在偶因，异因和/或系统波动的过程。 2002 North
5、 Haven Group3M ADICUnderstanding Variation Revisited理解波动-回顾A control chart is basically a signal detection device, which attempts to distinguish special and systematic causes of variation from common cause variation. 控制图基本上是一个信号检测手段，目的是把异波和系统波动从偶波中区别开来。If a process is performing acceptably, one want
6、s to maintain that process so that it continues to perform at that level free of special causes. 如果一个过程的表现是可以接受的话，有人想维护那个过程，以便过程能够持续地保持没有异因波动的状态。In the case of our KPIs, we would like to shift process performance from an undesirable state to a desired state. 就我们的KPI而言，我们希望过程绩效能从我们不想要的状态转到想要的状态。We ar
7、e purposely trying to introduce a special cause in our process that will move it from the undesirable level to the desired level. 我们有意地在我们的过程中引进了异因的概念，这一概念实施将有助于从不想要的状态转变成想要的状态。 2002 North Haven Group4M ADICIntroduction to Control Charts控制图简介Detecting and maintaining improvement require that we moni
8、tor our processes over time. 探测和维护改进要求我们随时监控过程。We have already discussed how project KPIs need to be plotted over time in order to provide baseline information about changes in process level. 关于过程水平改变，为了提供基线信息，我们已经讨论了如何图示KPI。This section focuses on how time series data can be monitored using control
9、 charts. 本节课程将重点讨论，如何使用控制图来控制时序数据。 There are various types of control charts. 有多种类型的控制图。Which chart is used depends on the type of data (and the statistic of interest) and ones ability to form subgroups. 使用何种控制图，取决于数据的类型和个人构成子组的能力。 2002 North Haven Group5M ADICIntroduction to Control Charts控制图简介In t
10、he following plot, the control limits are calculated from the data that was obtained prior to implementing the solution. 在下面的图中，控制限是根据实施解决方案之前收集到的数据进行绘制的。When new process data is plotted against these control limits, there is clear evidence that a special cause has occurred. 当新的过程数据加入到这些控制限中时，就会出现异因
11、的明显证据。In this case, that special cause was desirable! 在这种情况下，异因是需要的。Solution 2002 North Haven Group6M ADICSummary of Basic Control Charts基本控制图总结连续型数据连续型数据离散型数据离散型数据计数计数比例比例可以分组可以分组不可分组不可分组 2002 North Haven Group7M ADICControl Chart Basics控制图基础A control chart consists of four basic elements:一个控制图包括四项
12、基本要素：A statistic that is plotted at regular time periods. 等时间段绘制的统计量。A time axis to show the time ordering of the plotted statistic. 显示时间顺序的时间轴。A center line representing the overall process average. 代表总过程平均数的中线Upper and lower control limits that show the limits of variability, assuming only common
13、causes of variation are present.显示波动界限的上下控制限，假设只出现偶因波动。For a normal distribution, 99.73% of all values fall within 3 standard deviations of the mean.对于一个正态分布，所有值的99.73%落在均值的正负三个标准差以内。The idea is that it is unlikely that points will fall outside the control limits unless some special cause has occurr
14、ed. 观点是，点落在控制限外面的可观点是，点落在控制限外面的可能性很小，除非出现异因的情况。能性很小，除非出现异因的情况。 2002 North Haven Group8M ADICControl Chart Basics控制图基础Control limits can be used for process diagnosis or for ongoing process maintenance. 控制限可以用作过程诊断，或正在进行的过程维护。When used for process diagnosis-control limits are calculated from the give
15、n data, and the plot is studied for evidence of special or systematic variation. 当用于过程诊断时， 控制限可根据给定的数据进行计算，然后研究图形，以分析出异波或系统波动。When used for process maintenance, control limits are computed from data taken when the process is likely to be stable.当用于过程维护时，控制限是根据稳定过程中收集来的数据进行计算的。During the Improve Phas
16、e, control charts for project KPIs have an emphasis of process diagnosis (have we improved?). 在改进阶段，用于项目KPI的控制图着重强调过程诊断（过程是否改进？）During the Control Phase, the KPI for the improved process is plotted on a control chart with an emphasis of maintenance. 在控制阶段，在控制图上绘制用于改进过程的KPI，重点强调过程维护。 2002 North Haven
17、 Group9M ADICControl Chart Basics控制图基础We will first discuss control charts used when process data is continuous.我们将先讨论连续型数据的控制图。In deriving control chart limits, it is desirable to estimate variation using samples gathered over a period where it is unlikely that any special cause has intervened. 在推导
18、控制限时，需要收集没有异因干扰的一段时间的过程波动样本数据，然后再根据这些数据估计波动标准差。In a manufacturing situation, this is often accomplished by selecting samples of parts that are produced at almost the same time since they are not likely to have been affected by special cause variation. 在制造业环境中，通常选取同一时间段生产的零件作为样本，这样过程受异因波动影响的可能性不大。Var
19、iation is estimated within each of these samples, and then these estimates are combined. 波动是根据每个样本估计的，然后再将这些估计合并起来。 2002 North Haven Group10M ADICControl Chart Basics控制图基础Consider, for example, a production situation where piston rings are being produced, and where the outside diameter (OD) is of in
20、terest (Pistons.jmp). 比如，生产活塞时，活塞环的外径是重要因素。At the beginning of each hour, five consecutively produced piston rings are taken from production, and their OD is measured. 在每个小时开始时，连续测量五个活塞环的外径，并记录数据。The samples of size five are called subgroups. 这里的五个样品称为子组。 2002 North Haven Group11M ADICControl Chart
21、Basics控制图基础Each subgroup is fairly likely to be affected only by common cause variation. 每个子组可以认为只受偶因波动的影响。注：用于控制限的西格玛是基于标准差的注：用于控制限的西格玛是基于标准差的 2002 North Haven Group12M ADICControl Chart Basics控制图基础It is generally a good idea to plot averages rather than individual measurements:一般说来，图示均值要比图示单个值要好。P
22、lotting subgroup averages on control charts helps ensure that the underlying normality assumption applies. 用子组均值绘制控制图，帮助保证应用下面的常态假设。Subgroups allow the construction of control limits where the variability estimate is likely to be free of special causes. 子组可以绘制没有异因的控制图。These limits make the chart mor
23、e capable of detecting process changes. 这些界限使得控制图更有能力识别过程改变。Note: The subgroups must be logical to the process under review, not just random groupings of data.注：子组要有逻辑性，不能把数据随便分组。注：子组要有逻辑性，不能把数据随便分组。 2002 North Haven Group13M ADICRational Subgrouping合理分组The most sensitive kinds of control charts, ca
24、lled Shewhart charts, are based on the idea of rational subgrouping. 最敏感的控制图，叫作休哈特图，是基于合理分组的观点。Instead of plotting individual values on a control chart, we plot the mean and standard deviation (or range) of a subgroup of parts sampled from the process at some regular time interval. 在某一相同的时间间隔，我们把样本的
25、均值，标准差（或极差）标在控制图上，而不是把样本的单值标在控制图上面。By doing so, we minimize the possibility that the subgroup contains the effects of a special cause of variation. Rational subgroups are considered to exhibit only common cause variation. 通过这样做，我们使得子组中异波的影响最小化了。合理分组被当作只存在偶因波动。A typical approach to rational subgroupi
26、ng is to form subgroups from consecutive parts taken from the production line. 合理分组的通常方法是：从生产线中取得连续的加工零件，组成子组。When data on all parts is recorded, one often constructs rational subgroups based on criteria such as Shift, Day, or Week. 当所有零件的数据记录好了以后，通常按照标准进行分组，比如：班次，天或周。 2002 North Haven Group14M ADIC
27、X-Bar and S Control Charts均值和标准差控制图Case Study: Elapsed Time Data (file ElapsedTimes.jmp)案例分析：共用时案例分析：共用时间分析间分析The data in this file are the number of days between the initiation of an action and its completion, recorded on the completion date. 该文件中的数据是活动开始日期与完成日期之间相隔的天数，是在完成日期进行记录的。This data was col
28、lected for actions terminated in April and May of 2002. 这些数据是为2002年4-5月份结束的活动而记录的。We will construct a control chart subgrouped by day. 我们将按照天来绘制控制图。 2002 North Haven Group15M ADICX-Bar and S Control Charts均值和标准差控制图An X-bar and S chart actually consists of two charts: one is a plot of subgroup means
29、(the X-bar chart) and the other is a plot of subgroups standard deviations (the S chart).均值和标准差控制图实际上包含两张图，一个是子组均值图，另一个是子组标准差图。Note that the control limits vary in size as a result of the fact that the subgroup sizes are not constant. 由于子组大小不一致，控制限也根着变化。 2002 North Haven Group16M ADICX-Bar and S Con
30、trol Charts均值和标准差控制图The X-bar chart shows all points within the control limits. 均值图显示了所有在控制图限里面的点。There is no evidence of unusual patterns. The process averages approximately 21 days. 没有产生不正常振纹的证据。过程的平均天数大约为21天。We conclude that there is no evidence of special or systematic causes. 我们得出的结论是没有异因或系统因素。
31、 2002 North Haven Group17M ADICX-Bar and S Control Charts均值和标准差控制图The S chart, on the other hand, shows several points falling outside the control limits. 另一方面，标准差图显示了一些点落在了控制限的外面。The S chart is very sensitive to non-normality, and elapsed time data tends to be fairly skewed. 标准差图对不正常现象是很敏感的，共用的时间数据
32、趋势有明显偏斜。On this chart we are observing evidence of skewness. 在这张图上，我们看到了偏斜的现象。 2002 North Haven Group18M ADICIndividuals Control Charts单值控制图Continuous data that cannot be subgrouped can be monitored by an Individuals and Moving Range chart . 连续数据不能分成子组，可以由单值和移动极差图来监控。The centerline for the Individua
33、l Measurement chart is the average of the data collected. 单值测量图的中线是收集到数据的均值。Since individual measurements are plotted on the individuals chart, and control limits based on a normal distribution, the individuals chart will not work well if the data is not normally distributed. 既然单值测量值是绘制在单值图上的，控制限是基于
34、正态分布的，单值控制图的效果会不理想。The normality of the data should always be tested before using an individuals control chart. 在使用单值控制图之前，应该先检查一下，数据的正态性。 2002 North Haven Group19M ADICIndividuals Control Charts单值控制图Other Key Points:其他关键点：To use an individuals chart, individual observations must be independent. 为了使
35、用单值控制图，必须独立地对单值进行观测。Individual charts are not very sensitive, and so the control limits are often tightened. Unfortunately, this will result in more false signals. 单值控制图很敏感，因此，控制限通常很严格，但不幸的是，这将导致较多的错误信号。Why? With 3 limits, 0.27% of observations will fall outside the control limits by chance alone. I
36、n contrast, approximately 5% of observations will fall outside 2 limits by chance alone. 为什么？在三西格玛控制限内，只有0.27%的观测值落在控制限外面。相反，大约有5%的观测值将落在二西格玛控制限内。 2002 North Haven Group20M ADICIndividuals Control Charts单值控制图Consider the production costs in ProductionCosts2.jmp. 考虑生产成本These are weekly production cos
37、ts for a manufacturing operation. 对于制造工厂，有一个周生产成本的统计。This data cannot be subgrouped in any reasonable way. 这些数据不能以任何合理的方法进行分组。An individuals control chart is appropriate, so long as the data has an approximately normal distribution. 只要数据有一个大约的正态分布， 单值控制图是适当的。 2002 North Haven Group21M ADICIndividual
38、s Control Charts单值控制图 2002 North Haven Group22M ADICDetecting Process Shifts or Changes探测过程漂移或更改It is important to realize that control charts can result in two types of errors: 认识到控制图可能导致产生两种类型的错误是很重要的。They can signal a special cause when they should not. This is called a false positive. 没有异因时，出现异因
39、信号。这称为正错误。They can fail to signal a special cause when they should. This is called a false negative. 有异因时，没有出现异因信号。这称为负错误。To date, we have only discussed a single test for special causes, namely a point falls beyond the control limits. 到目前为止，我们只讨论了异因的单点测试，即某一点落在控制限以外。If the statistic that is plotted
40、 has approximately a normal distribution, when the process is stable, points will exceed the control limits only about 0.27% of the time. 如果绘制的统计量大体上是一个正态分布，当过程是稳定时，点落在控制限外面的机会大约0.27%. 2002 North Haven Group23M ADICOther Common Control Charts其他常用控制图Exponentially weighted moving average chart (EWMA):
41、 This control chart can be used with individuals as well as subgrouped data. Each point is a weighted average of current and past measurements at each time period. It can be useful in identifying trends since it typical signals them earlier than an X-bar or IR chart. 指数加权移动平均控制图（EWMA）：这类控制图可以用于单值，也可
42、以用于子组数据。每一点是在每一时间段上，当前和过去测量值的加权平均。既然该图会比均值或IR图，通常能较早发出异因信号，则对识别趋势很有帮助。P chart or chart for proportions: This control chart can be used with proportions data, defined when an entire item is considered to be either a success or a failure (good or bad). Examples include the proportion of input errors,
43、proportion of expedited deliveries, or the proportion of “lost” sales opportunities. P图或比例图：该控制图可用于比例数据，当整个事件被认为成功或失败（好或坏）时。样本包括输入错误的比例，加快交货期的比例，或损失销售机会的比例。 2002 North Haven Group24M ADICFinal Thoughts结束语A stable process is not necessarily a capable process - the natural common cause variation may b
44、e too large compared to the customer requirements. 稳定的过程不一定是有能力的过程-自然偶因波动与顾客的要求相比，也许太大了。Control limits are based on natural variation that is inherent to a process, while specification limits are imposed externally upon a process. 控制限是基于自然波动之上的，它是过程的固有波动。而规格限是强加在过程外部的。Also, control limits on X-bar c
45、harts refer to averages, not individual parts. Specification limits apply to measures of individual parts. 另外，均值控制图的控制限是有关均值的，不是单个零件的。而规格限是对单个零件的测量值。Do not judge process capability with respect to specification limits using control limits. You will seriously over rate your capability. 不要从规格限的方面来评价过程
46、能力，要使用控制限。你将过高评估你的过程能力。 2002 North Haven Group25M ADICSkills Practice技能练习技能练习A team was studying how long it takes to recruit an external person to fill an open position within the company. They measured the time from when the recruitment advertisement was published until an individual accepted the
47、position. 一个团队在研究从外部招收一名新员工，要花多少时间员工能够到位。他们从招聘广告公布之日起，直到有人接受这个职位结束。In the data set, that time is reflected in the column Time to Fill (Days). Each row is a unique recruitment. Columns Month, Week and Year give the timing of the acceptance. 员工到位的天数填在第一列，其他年月日则表示每一次的招聘要求。QUESTION: What type of Control
48、 Chart would you use & why:问题：你将使用何种控制图，为什么？ 2002 North Haven Group26M ADICSkills Practice技能练习技能练习A team was studying how many hours a department spends manually processing a variety of invoices. Individuals recorded their time spent on different invoices each week. 一个团队在研究某一部门要花多少时间来手工处理各种发票。每个人记录他
49、们每周花费在不同发票上的时间。In the data set, Week is the time period data was collected, Description identifies the type of invoice being processed, and Hours Reported is the sum of all individuals time spent processing that type of invoice for that particular week. 收集数据的时间间隔是周，描述一栏列明了不同类型的发票，时间一栏填的是那一周处理该类型发票的总
50、时间。QUESTION: What type of Control Chart would you use & why:问题：你将选择哪种控制图，为什么？ 2002 North Haven Group27M ADICSkills Practice技能练习技能练习QUESTIONS: 问题问题1.Does the process appear to be stable? Why or Why not?过程稳定吗？为什么？2.Do you notice any unusual behavior? If so, explain.你是否注意到任何的不正常现象？如果是，请解释。 2002 North Haven Group28M ADICSkills Practice技能练习技能练习QUESTIONS: 问题：问题：1.Does the process appear to be stable? Why or Why not?过程看起来稳定吗？为什么稳定，或为什么不稳定？2.Do you notice any unusual behavior? If so, explain.你是否发现任何不正常的现象？ 如果是，请解释。
