Statistical Process Control Explained

Statistical Process Control, commonly referred to as SPC, is a method for monitoring, controlling and, ideally, improving a process through statistical analysis. The philosophy states that all processes exhibit intrinsic variation. However, sometimes processes exhibit excessive variation that produces undesirable or unpredictable results. SPC, in a manufacturing process optimization context, is used to reduce variation to achieve the best target value.

n the mid!"#$%s, &r. 'alter (. Shewhart developed the fundamentals of Statistical Process Control )though that was not what it was called at the time* and the associated tool of the Control Chart. His reasoning and approach were practical, sensible and positive. n order to be so, he deliberately avoided overdoing mathematical detail. n later years, significant mathematical attributes were assigned to Shewhart+s thin,ing with the result that this wor, became better ,nown than the pioneering application that he had wor,ed up. The crucial difference between Shewhart-s wor, and the inappropriately perceived purpose of SPC that emerged, that typically involved mathematical distortion and tampering, is that his developments were in context, and with the purpose, of process improvement, as opposed to mere process monitoring. that is, they could be described as helping to get the process into that /satisfactory state0 which one might then be content to monitor. 1ote, however, that a true adherent to &eming-s principles would probably never reach that situation, following instead the philosophy and aim of continuous improvement. Suppose that we are recording, regularly over time, some measurements from a process. The measurements might be lengths of steel rods after a cutting operation, or the lengths of time to service some machine, or your weight as measured on the bathroom scales each morning, or the percentage of defective )or non!conforming* items in batches from a supplier, or measurements of ntelligence 2uotient, or times between sending out invoices and receiving the payment, etc., etc. ( series of line graphs or histograms can be drawn to represent the data as a statistical distribution. t is a picture of the behaviour of the variation in the measurement that is being recorded. f a process is deemed as /stable0 then the concept is that it is in statistical control. The point is that, if an outside influence impacts upon the process, )e.g., a machine setting is altered or you go on a diet, etc.* then, in effect, the data are of course no longer all coming from the same source. t therefore follows that no single distribution could serve to represent them. f the distribution changes unpredictably over time, then the process is said to be out of control. (s a scientist, Shewhart ,new that there is always variation in anything that can be measured. The variation may be large, or it may be imperceptibly small, or it may be between these two extremes. but it is always there. 'hat inspired Shewhart-s development of the statistical control of processes was his observation that the variability which he saw in manufacturing processes often differed in behaviour from

that which he saw in so!called /natural0 processes 3 by which he seems to have meant such phenomena as molecular motions. 'heeler and Chambers combine and summarise these two important aspects as follows4 5'hile every process displays variation, some processes display controlled variation, while others display uncontrolled variation.5 n particular, Shewhart often found controlled )stable* variation in natural processes and uncontrolled )unstable* variation in manufacturing processes. The difference is clear. n the former case, we ,now what to expect in terms of variability. in the latter we do not. 'e may predict the future, with some chance of success, in the former case. we cannot do so in the latter. Shewhart gave us a technical tool to help identify the two types of variation4 the control chart. 'hat is important is the understanding of why correct identification of the two types of variation is so vital. There are at least three prime reasons. 6irst, when there are irregular large deviations in output because of unexplained special causes, it is impossible to evaluate the effects of changes in design, training, purchasing policy, etc. which might be made to the system by management. The capability of a process is un,nown, whilst the process is out of statistical control. Second, when special causes have been eliminated, so that only common causes remain, improvement then has to depend upon management action. 6or such variation is due to the way that the processes and systems have been designed and built 3 and only management has authority and responsibility to wor, on systems and processes. (s 7yron Tribus, &irector of the (merican 2uality and Productivity nstitute, has often said4 The people work in a system. The job of the manager is to work on the system to improve it, continuously, with their help. 6inally, something of great importance, but which has to be un,nown to managers who do not have this understanding of variation, is that by )in effect* misinterpreting either type of cause as the other, and acting accordingly, they not only fail to improve matters 3 they literally ma,e things worse. These implications, and conse8uently the whole concept of the statistical control of processes, had a profound and lasting impact on &r. &eming. 7any aspects of his management philosophy emanate from considerations based on 9ust these notions. So why SPC: The plain fact is that when a process is within statistical control, its output is indiscernible from random variation4 the ,ind of variation which one gets from tossing coins, throwing dice, or shuffling cards. 'hether or not the process is in control, the numbers will go up, the numbers will go down. indeed, occasionally we shall get a number that is the highest or the lowest for

some time. ;f course we shall4 how could it be otherwise: The 8uestion is ! do these individual occurrences mean anything important: 'hen the process is out of control, the answer will sometimes be yes. 'hen the process is in control, the answer is no. So the main response to the 8uestion "Why S !"" therefore this4 t guides us to the type of action that is appropriate for trying to improve the functioning of a process. Should we react to individual results from the process )which is only sensible, if such a result is signalled by a control chart as being due to a special cause* or should we instead be going for change to the process itself, guided by cumulated evidence from its output )which is only sensible if the process is in control*: The ,ey to any process improvement program is the lan#$o#Stu%y#&ct cycle described by 'alter Shewhart.

lan involves using SPC tools to help you identify problems and possible causes. $o involves ma,ing changes to correct or improve the situation. Stu%y involves examining the effect of the changes )with the help of control charts*. &ct involves, if the result is successful, standardizing the changes and then wor,ing on further improvements or, if the outcome is not successful, implementing other corrective actions.

Cause-and-Effect Diagrams
;ne analysis tool is the Cause!and!<ffect )or 6ishbone* diagram. These are also called shi,awa diagrams because =aoru shi,awa developed them in "#>?. They are called fishbone diagrams since they resemble one with the long spine and various connecting branches.

The fishbone chart organizes and displays the relationships between different causes for the effect that is being examined. This chart helps organize the brainstorming process. The ma9or categories of causes are put on ma9or branches connecting to the bac,bone, and various sub! causes are attached to the branches. ( tree!li,e structure results, showing the many facets of the problem. The method for using this chart is to put the problem to be solved at the head, then fill in the ma9or branches. People, procedures, e8uipment and materials are commonly identified causes. This is another tool that can be used in focused brainstorming sessions to determine possible reasons for the target problem. The brainstorming team should be diverse and have experience in the problem area. ( lot of good information can be discovered and displayed using this tool.

Pareto Charts
The Pareto chart can be used to display categories of problems graphically so they can be properly prioritized. The Pareto chart is named for a "#th century talian economist who postulated that a small minority )$%@* of the people owned a great proportion )A%@* of the wealth in the land.

There are often many aspects of a process or system that can be improved, such as the number of defective products, time allocation, or cost savings. <ach aspect usually contains many smaller problems, ma,ing it difficult to determine how to approach the issue. ( Pareto chart or diagram indicates which problem to tac,le first by showing the proportion of the total problem that each of the smaller problems comprise. This is based on the Pareto principle4 $%@ of the sources cause A%@ of the problem. ( Count Pareto chart is a vertical bar graph displaying ran, in descending order of importance for the categories of problems, defects or opportunities. Benerally, you gain more by wor,ing on the problem identified by the tallest bar than trying to deal with the smaller bars. However, you should as, yourself what item on the chart has the greatest impact on the goals of your business, because sometimes the most fre8uent problem as shown by the Pareto chart is not always the most important. SPC is a tool to be used by people with experience and common sense as their guide. This is a Pareto chart of defect types for whipped cream cans.

;nce a ma9or problem has been selected, it needs to be analyzed for possible causes. Cause!and! effect diagrams, scatter plots and flow charts can be used in this part of the process.

Control Charts & Other Charts

'hether ma,ing mom-s recipe for spaghetti sauce or admitting patients to an emergency room, the outcome of a process is never exactly the same every time. 6luctuation or variability is an inevitable component of all systems and is expected, arising naturally from the effects of miscellaneous chance events. However, variation outside a stable pattern may be an indication that the process is not acting in a consistent manner. <vents which fall beyond expected variability or events forming a pattern that is not random, indicate that the process is out of control.

6rom a 8uality control perspective, an out!of!control service or production system is trouble. t is probably not meeting customer specifications or achieving business goals, and there is no way of predicting if it will or can. There are two general ways of detecting that a process is out of control. The first test for an out! of!control process as,s, 5 s any point falling above or below the control limits on its control chart:5. This particular test is very easy to perform by viewing the control chart. The second form of rule violations is based upon patterns of points on the control chart and can be difficult to detect. Some SPC software programs can 8uic,ly and accurately performs such tests on process data using the 'estern <lectric Cules, which are defined in (TDT-s Statistical 2uality Control Handboo,, the definitive source for rule violation standards. Statistical Process Control charts graphically represent the variability in a process over time. 'hen used to monitor the process, control charts can uncover inconsistencies and unnatural fluctuations. Conse8uently, SPC charts are used in many industries to improve 8uality and reduce costs. Control charts typically display the limits that statistical variability can explain as normal. f your process is performing within these limits, it is said to be in control. if not, it is out of control. t is important to remember what you can conclude about a system that is in control4 control does not necessarily mean that a product or service is meeting your needs, it only means that the process is behaving consistently. Here are some fundamental concepts and ,ey terms related to control charts and other charts.

Rules Testing
How do you 9udge when a process is out of control: Ey plotting a control chart of the output of a process, it is possible to spot special or unnatural causes of variability and indications that the process is drifting. &rifting is defined by the mean or range of the variation changing as the process is running. The most common indication of change is a point falling outside of the control limits, but other tests for process instability are also valuable. &ifferent rules are appropriate for variable data and attribute data. Conse8uently, choosing which rules to apply depends on the type of chart being produced.

Zones In Control Charts

7any of the standard rules examine points based on 'ones. The area between each control limit and the centerline is divided into thirds. The third closest to the centerline is referred to as Fone (, the next third is Fone E, and the third closest to the control limits is Fone C. 1ote that there are two of each of the Fones, one upper and one lower.

Fone ( is also referred to as the 5?!sigma zone5, Fone E is the 5$!sigma zone5, and Fone C is the 5"!sigma zone5. These sigma zone terms are appropriate only when ?!sigma is used for the control limits. Sigma is the Bree, letter for s and is used in this context to denote the spread of data.

Standard control limits are located ? sigma away from the average or centerline of the chart. The centerline is also called the control line. These are called ? sigma limits or ? sigma zones. The distance from the centerline to the control limits can be divided into ? e8ual parts of one sigma each. Statistical theory tell us that in normal data dispersion, we can expect the following percentages of data to be included within the sigma4

1 sigma 68.3% 2 sigma 95.5% 3 sigma 99.7%

'e can expect ##.G@ of the process outcomes to be within the ?!sigma control limits.

Control Limits
Control limits are calculated statistically from your data. They are referred to as the Hower Control Himit )HCH* and the Ipper Control Himit )ICH* on a control chart. These are set at ?! sigma by default since this is the most commonly used limit.

Control limits define the zone where the observed data for a stable and consistent process occurs virtually all of the time )##.G@*. (ny fluctuations within these limits come from common causes inherent to the system, such as choice of e8uipment, scheduled maintenance or the precision of the operation that results from the design. These normal fluctuations are attributed to statistical variability. (n outcome beyond the control limits results from a special cause. Special causes are events external to the ordinary operation of a production or service. Special causes indicate that there have been one or more fundamental changes to the process and the process is out of control. Special causes need to be investigated and eliminated before a control chart can be used as a 8uality!monitoring tool.

Subgroups
(n important factor in preparing for SPC charting is determining if you will measure every product of the process, such as measuring every part, or if you will use subgroups. Subgroups are a sample of data from the total possible data. Subgroups are used when it is impractical or too expensive to collect data on every single product or service in the process. &ecisions to use subgroups or not needs to be carefully thought out to ensure they accurately represent the data. Subgroups need to be homogenous within themselves so that special causes can be recognized, so problem areas stand out from the normal variation in the subgroup. 6or example, if you are in charge of analyzing processes in a number of facilities, a separate group should represent each facility, since each facility has different processes for doing the same tas,s. <ach facility subgroup should probably be bro,en down even further, for example by wor, shifts.

Subgroups in Variable Control Charts

(ll data in a subgroup has something in common, such as a common time of collection, all data for a particular date, a single shift, or a time of day. Subgroup data can have other factors in common, such as data associated with an operator, or data associated with a particular volume of li8uid.

Subgroups in Attribute Control Charts

( subgroup is the group of units that were inspected to obtain the number of defects or the number of re9ects. The number of defects is displayed using c charts and u charts. The number of re9ects, also called defective items, is displayed using p charts and np charts.

Rejects - Nonconforming Items Data

1onconforming items are re9ects. ( re9ect is tallied when an entire unit fails to meet acceptance standards, regardless of the number of defects in the unit. This includes defective products or unacceptable outcomes.

Defects - Nonconformities Data

1onconformities are defects. ( non!conformity is any characteristic, which should not be present but is, or a characteristic which needs to be present but is not. ( defective item can have multiple non!conformities, for example, errors on insurance forms, incorrect medication, or service complaints.

Using Process Control Charts

;=, enough tal,. Het+s do some actual control charting. 6irst you need to determine what data you have and select the correct chart for that data. Then ma,e the chart and analyze it to see if the process is in control.

Data Definitions For Proper Chart Selection

Choosing the correct chart for a given a situation is the first step in every analysis. There are actually 9ust a few charts to choose from, and determining the appropriate one re8uires following some fairly simple rules based on the underlying data. These rules are described in the flowchart below4

Control charts are divided into two groups4

Variable Charts
Jariable charts are based on variable data that can be measured on a continuous scale. 6or example, weight, volume, temperature, or length of stay. These can be measured to as many decimal places as necessary. ndividual, average, and range charts are used for variable data.

Attribute Charts
(ttribute charts are based on data that can be grouped and counted as present or not. (ttribute charts are also called count charts and attribute data is also ,nown as discrete data. (ttribute data is measured only with whole numbers. <xamples include4

Acceptable vs. non-acceptable Forms completed it! errors vs. it!o"t errors #"mber o$ prescriptions it! errors vs. it!o"t

'hen constructing attribute control charts, a subgroup is the group of units that were inspected to obtain the number of defects or the number of defective items. &efect and re9ect charts are used for attribute data.

Variable Data Charts Individual, verage, and !ange Charts

Jariable data re8uires the use of variable charts. Jariable charts are easy to understand and use.

In i!i ual Charts " I chart

The I chart is also referred to as an individual, item, i, or K chart. The K refers to a variable K.

ndividual charts plot the process results varying over time. ndividual observations are plotted on the chart, averages are not plotted on this type of chart. ndividual charts are used to plot variable data collected chronologically from a process, such as a part-s measurement over time. These charts are especially useful for identifying shifts in the process average. 'hen monitoring a system, it is expected that e8ual numbers of points will fall above and below the average that is represented by the centerline. Shifts or trends can indicate a change that needs to be investigated. The individual control chart is reserved for situations in which only one measurement is performed each time the data is collected, where it is impractical or impossible to collect a sample of observations. 'hen there are not enough data points to calculate valid control limits, an individual chart functions as a simple run chart.

#!erage Charts " $-bar Chart

(verage charts are made by plotting averages of individual measurements on the chart. The average chart is called the K!bar chart because, in statistical notation, a bar or line over the variable )K* symbolizes the average of K. /K!bar0 is a shorthand way of saying /the average of K0. (n K!bar chart is a variable control chart that displays the changes in the average output of a process. The chart reflects either changes over time or changes associated with a categorical data variable. The chart shows how consistent and predictable a process is at achieving the mean. K!bar charts measure variation between subgroups. They are often paired with either Standard &eviation )S* or Cange )C* charts, which measure variation within subgroups. &efinition4 Jariable &ata Subgroups

Subgroup data has something in common, such as data associated with a particular operator, or data associated with a particular volume of li8uid.

Range Chart " R-Chart

The Cange chart can be combined with charts and K!bar charts. The chart names combine the corresponding chart initials. Cange charts measure the variation in the data. (n example is the weather report in the newspaper that gives the high and low temperatures each day. The difference between the high and the low is the range for that day.

%o!ing Range Chart " %R Chart

This type of chart displays the moving range of successive observations. ( moving range chart can be used when it is impossible or impractical to collect more than a single data point for each subgroup. This chart can be paired with an individual chart, which is then called an ndividual 7oving Cange ) C* chart. (n individual chart is used to highlight the changes in a variable from a central value, the mean. The moving range chart displays variability among measurements based on the difference between one data point and the next.

In i!i ual #n Range Charts " IR Charts

This pair of variable control charts is often offered together for 8uality control analysis. The ndividual chart, the upper chart in the figure below, displays changes to the process output over time in relation to the center line which represents the mean. The 7oving Cange chart, the lower chart in the figure below, analyzes the variation between consecutive observations, which is a measure of process variability.

#!erage & Range Charts " $-'ar #n R Charts

Jariable and Cange control charts are often displayed together for 8uality control analysis. The K!bar chart, the upper chart in the figure below, is a graphic representation of the variation among the subgroup averages, the C chart, the lower chart in the figure below, loo,s at variability within these subgroups. The variation within subgroups is represented by the range )C*. The range of values for each subgroup is plotted on the L!axis of the C chart. The centerline is the average or mean of the range.

$-'ar Stan ar De!iation Charts " $-'ar #n S Charts

This pair of variable control charts is often displayed together for 8uality control analysis. The K!bar chart, the upper chart in the figure below, displays the variation in the means between the subgroups. The s chart, the lower chart in the figure below, loo,s at variability within these subgroups. n this pair of charts, the variation within subgroups is represented by the standard deviation. The standard deviation is plotted on the y!axis, and is a measure of the spread of values for each subgroup. The centerline is the average or mean of these sub!group standard deviations.

Lou can choose to use a standard deviation chart, i.e. an s!chart, instead of the 7oving Cange chart. The Cange chart is often used because the standard deviation is a more accurate and therefore more difficult measurement. 1ow that computers are automatically calculating the standard deviation, the s!chart can be used in all situations. This is called the K!bar S chart. ( standard deviation formula is used to calculate the differences in the data. This calculation can be used in cases where the subgroup sample size is large and sampling methods support the modeling of the data as normal distribution.

Process Capabilit( Chart " cp Chart

Process capability analysis is used to ad9ust the process until virtually all of the product output meets the specifications. ;nce the process is operating in control, capability analysis attempts to answer the 8uestion4 s the output meeting specifications, or is the process capable: f it is not, can the process be ad9usted to ma,e it capable: The process capability chart contains a normal curve superimposed over a histogram of the data, followed by several statistics. ( process is said to be capable if its output falls within the specifications virtually "%%@ of the time. Note: Specification Himits are the boundaries, or tolerances, set by management, engineers or customers which are based on product re8uirements or service ob9ectives. Specification Himits are 1;T established by the process itself, and may not even be possible within the given process. ;ne goal of Statistical Process Control is to determine if specifications are in fact possible in the current process. f the following statements are true, a process capability chart can be an appropriate tool for measuring the inherent reproducibility of the process and monitoring the degree to which it can meet specifications4

%!e process is stable and in control. %!e data are normall& distrib"ted. 'peci$ication limits $all on eit!er side o$ t!e centerline. (o" are investigating !et!er &o"r process is capable o$ meeting speci$ications.

ttribute Data Charts

(gain, attribute data represents particular characteristics of a product or system that can be counted, not product measurements. They are characteristics that are present or not present. This is ,nown as discrete data, and is measured only with whole numbers. <xamples include4

Acceptable vs. non-acceptable Forms completed it! errors vs. it!o"t errors #"mber o$ prescriptions it! errors vs. it!o"t

(ttribute data has another distinctive characteristic. n 8uality control analysis, this countable data falls into one of two categories4

Defects data is t!e n"mber o$ non-con$ormities it!in an item. %!ere is no limit to t!e n"mber o$ possible de$ects. Defects charts co"nt t!e n"mber o$ de$ects in t!e inspection "nit. Rejects data !ere t!e entire item is )"dged to con$orm to prod"ct speci$ications or not. Rejects charts co"nt t!e n"mber o$ re)ects in a s"bgro"p.

;ne way to determine what type of data you have is to as,, /Can count both the occurrences (1& non!occurrences of the defective data:0 6or example, you can count how many forms have errors and how many do not, however you cannot count how many errors were 1;T made on the form. f you can count both occurrences and non!occurrences, you have re9ects data. f the non!occurrences cannot be determined, then you have defects data. 6or example4 f you are counting the number of errors made on an insurance form, you have an example of the defects per form. There is no limit to the number of defects that can be counted on each form. f you are counting the total number of forms that had one or more errors, then you have a count of the re9ected units. This is either one or zero re9ects per unit.

Summary Of Defects vs. Rejects Data

n summary4

*e$ects c!arts are attrib"te c!arts $or cases in !ic! t!e possible occ"rrences are in$inite or cannot be co"nted. %!e& co"nt t!e n"mber o$ non-con$ormities it!in an item. +e)ects c!arts are attrib"te *ata c!arts $or t!e cases in !ic! re)ected !ole "nits are co"nted. %!ese $ig"res can be described as ratios instead o$ )"st co"nts.

Subgroup Size Constant or Changing

Subgroup size is another important data characteristic to consider in selecting the right type of chart. 'hen constructing attribute control charts, a subgroup is the group of units that were inspected to obtain the number of defects or the number of re9ects. To choose the correct chart, you need to determine if the subgroup size is constant or not. f constant, for example ?%% forms are processed every day, then you can loo, at a straight count of the defective occurrences. f the subgroup size changes, you need to loo, at the percentage or fraction of defective occurrences. 6or example4

(n organization may have a day in which M%% insurance forms are processed and M% have errors vs. another day in which only "M% are processed and $% have errors. f we only loo, at the count of errors, M% vs. $%, we would assume the M% error day was worse. Eut when considering the total size of the subgroup, M%% vs. "M%, we determine that, on the first day, "%@ had errors while, the other day, "?.?@ had errors. 1ow that we understand the different types of attribute data, let-s move on to the specific charts for analyzing them. There are four different types of attribute charts. 6or each type of attribute data, defects, and re9ects, there is a chart for subgroups of constant size and one for subgroups of varying size. Cemember4 &efects Charts count the number of defects within the inspection unit. Ce9ects Charts count the number of re9ected units in a subgroup.

Defects Charts
The two defects charts are the c chart and the u chart. The c refers to count of defects in a subgroup of constant size. The u is a per unit count within a variable size subgroup.

c Chart Constant Subgroup Size

( c chart, or Count chart, is an attribute control chart that displays how the number of defects, or nonconformities, for a process or system is changing over time. The number of defects is collected for the area of opportunity in each subgroup. The area of opportunity can be either a group of units or 9ust one individual unit on which defect counts are performed. The c chart is an indicator of the consistency and predictability of the level of defects in the process. 'hen constructing a c chart, it is important that the area of opportunity for a defect be constant from subgroup to subgroup since the chart shows the total number of defects. 'hen the number of items tested within a subgroup changes, then a u chart should be used, since it shows the number of defects per unit rather than total defects.

u Chart Varying Subgroup Size

( u chart )u is for Init* is an attribute control chart that displays how the fre8uency of defects, or nonconformities, is changing over time for a process or system. The number of defects is collected for the area of opportunity in each subgroup. The area of opportunity can be either a group of items or 9ust one individual item on which defect counts are performed. The u chart is an indicator of the consistency and predictability of the level of defects in the process.

( u chart is appropriate when the area of opportunity for a defect varies from subgroup to subgroup. This can be seen in the shifting ICH and HCH lines that depend on the size of the subgroup. This chart shows the number of defects per unit. 'hen the number of items tested remains the same among all the subgroups, then a c chart should be used since a c chart analyzes total defects rather than the number of defects per unit.

Rejects Charts
The two types of Ce9ects charts are the p chart and the np chart. The name of the p chart stands for the ercentage of re9ects in a subgroup. The name of the np chart stands for the (umber of re9ects within a p!type chart. Lou can also remember it as /not percentage0 or /not proportional0. ( mnemonic to remember that the p chart and its partner, the np chart, represent Ce9ects data is to thin, of P as a /pea0 and a canning plant that is re9ecting cans of peas if they are not "%%@ acceptable. (s p and np are a team, you should be able to recall this with the same story.

np Chart Number of Rejects Chart for Constant Subgroup Size

(n np chart is an attribute control chart that displays changes in the number of defective products, re9ects or unacceptable outcomes. t is an indicator of the consistency and predictability of the level of defects in the process. The np chart is only valid as long as your data are collected in subgroups that are the same size. 'hen you have a variable subgroup size, a p chart should be used.

p Chart Percentage Chart for Varying Subgroup Size

( p chart is an attribute control chart that displays changes in the proportion of defective products, re9ects, or unacceptable outcomes. t is an indicator of the consistency and predictability of the level of defects in the process. Since a p chart is used when the subgroup size varies, the chart plots the proportion or fraction of items re9ected, rather than the number re9ected. This is indicated by the shifting ICH and HCH

lines that depend on the size of the subgroup. 6or each subgroup, the proportion re9ected is calculated as the number of re9ects divided by the number of items inspected. 'hen you have a constant subgroup size, use an np chart instead.

"lo# Charts
(fter a process has been identified for improvement and given high priority, it should then be bro,en down into specific steps and put on paper in a flowchart. This procedure alone can uncover some of the reasons a process is not wor,ing correctly. ;ther problems and hidden traps are often uncovered when wor,ing through this process. 6lowcharting also brea,s the process down into its many sub!processes. (nalyzing each of these separately minimizes the number of factors that contribute to variation in the process. (fter creating the flowchart, you may want to ta,e another loo, at the fishbone diagram and see if any other factors have been uncovered. f so, you may need to do another Pareto diagram as well. 2uality Control is a continual process, in which factors and causes are constantly reviewed and changes made as re8uired.

$istograms
1ow you can put the data from the chec, sheets into a histogram. ( histogram is a snapshot of the variation of a product or the results of a process. t often forms the bell!shaped curve which is characteristic of a normal process. The histogram helps you analyze what is going on in the process and helps show the capability of a process, whether the data is falling inside the bell!shaped curve and within specifications. ( histogram displays a fre8uency distribution of the occurrence of the various measurements. The variable being measured is along the horizontal x!axis, and is grouped into several ranges of measurements. The fre8uency of occurrence of each measurement is charted along the vertical y! axis.

Histograms depict the central tendency or mean of the data, and its variation or spread. ( histogram also shows the range of measurements, which defines the process capability. ( histogram can show characteristics of the process being measured, such as4 &o the results show a normal distribution, a bell curve: f not, why not:

&oes the range of the data indicate that the process is capable of producing what is re8uired by the customer or the specifications: How much improvement is necessary to meet specifications: s this level of improvement possible in the current process:

Probabilit% Plots
n order to use Control Charts, the data needs to approximate a normal distribution, to generally form the familiar bell!shaped curve. The probability plot is a graph of the cumulative relative fre8uencies of the data, plotted on a normal probability scale. f the data is normal it forms a line that is fairly straight. The purpose of this plot is to show whether the data approximates a normal distribution. This can be an important assumption in many statistical analyses.

(lthough a probability plot is useful in analyzing data for normality, it is particularly useful for determining how capable a process is when the data is not normally distributed. That is, we are interested in finding the limits within which most of the data fall. Since the probability plot shows the percent of the data that falls below a given value, we can s,etch the curve that best fits the data. 'e can then read the value that corresponds to %.%%" )%."@* of the data. This is generally considered the lower natural limit. The value corresponding to %.### )##.#@* is generally considered the upper natural limit. )1ote4 To be more consistent with the natural limits for a normal distribution, some people choose %.%%"?M and %.##ANM for the natural limits.*

&catter Diagrams
The Scatter diagram, or plot, is another problem analysis tool. Scatter plots are also called correlation charts.

( Scatter plot is used to uncover possible cause!and!effect relationships. t is constructed by plotting two variables against one another on a pair of axes. ( Scatter plot cannot prove that one variable causes another, but it does show how a pair of variables is related and the strength of that relationship. Statistical tests 8uantify the degree of correlation between the variables.

n this example, there appears to be a relationship between height and weight. (s the student gets taller, generally spea,ing they get heavier.

SPC Glossary
Accuracy (ccuracy of measurements refers to the closeness of agreement between observed values and a ,nown reference standard. (ny offset from the ,nown standard is called bias. 2ualitative data that can be counted for recording and analysis. <xamples include4 number of defects, number of errors in a document. number of re9ected items in a sample, presence of paint flaws. (ttributes data are analyzed using the p!, np!, c! and u!charts. See mean. Short for average run length, (CH is the interval between out!of!control events that can be expected. 6or example, one point beyon% control limits is a common out!of!control event chosen to determine a process+ (CH. 'hen an out!of!control event appears on a control chart, an analyst can examine the interval between that event and the previous out!of!control event. f the interval matches or exceeds the process+ (CH value4 a* the process can probably be classified as still in!control, b* the violation can probably be attributed to typical process variation, and c* a search for an assignable cause can probably be considered unwarranted. The inverses of these statements are li,ely true if the interval between out!of!control events is smaller than the (CH value. The offset of a measured value from the true population value. ( discrete probability distribution used for counting the number of successes and failures, or conforming and nonconforming units. This distribution underlies the p!chart and the np!chart. ( graphical display of data that shows the median and upper and lower 8uartiles, along with extreme points and any outliers. The amount of variation inherent in a stable process. Capability can be determined using data from control charts and histograms and is often 8uantified using the Cp and Cp, indices. ( 8uality control tool used to analyze potential causes of problems in a product or process. t is also called a fishbone diagram or an shi,awa diagram after its developer. ( control chart based on counting the number of defects per constant size

Attribute data

Average Average Run ength !AR "

#ias #inomial $istribution

#ox and %his&er Plot Capability

Cause'and'E((ect $iagram c'Chart

subgroup. (lso ,nown as a Count of 1onconformities chart. The c!chart is based on the Poisson distribution. Center ine !C " The line on the control chart that represents the long!run expected, or average value, of the 8uality characteristic that corresponds to the in! control state which occurs when only chance causes are present. (n important statistical theorem that states that subgroup averages tend to be normally distributed even if the output as a whole is not. This allows control charts to be widely used for process control, even if the underlying process is not normally distributed. Problems with the system itself that are always present, influencing all of the production until found and removed. These are /common0 to all manufacturing or production output. (lso called chance causes, system causes, or chronic problems. Common causes contrast to special causes. The ongoing improvement of products, services, or processes through incremental and brea,through improvements. ( graphical mechanism for deciding whether the underlying process has changed based on sample data from the process. Control charts help determine which causes are /special0 and thus should be investigated for possible correction. Control charts contain the plotted values of some statistical measure for a series of samples or subgroups, along with the upper and lower control limits for the process. Statistically calculated control chart lines which indicate how the process is behaving and whether the process is in control. There is typically an upper control limit )ICH* and a lower control limit )HCH*. f the process is in control and only common causes are present, nearly all of the sample points fall within the control limits. Sometimes called the 1atural Process Himits for the sample size. ( measure of the linear relation between two variables. f both variables grow larger )or smaller* together, it is called positive correlation. f one variable becomes smaller as the other grows larger, it is called negative correlation. See attribute data. ( measure of the capability of a process to produce output within the specifications. The measurement is made without regard to the centering of the process.

Central imit )heorem

Common Causes

Continuous Improvement Control Chart

Control imits

Correlation

Count $ata Cp

Cp&

( measure of the capability of the process to produce output within the specifications. The centering of the process is ta,en into consideration by loo,ing at the minimum of the upper specification limit capability and the lower specification limit capability. Cp, O min )Cpu, Cpl*. ( control chart designed to detect small process shifts by loo,ing at the Cumulative SI7s of the deviations of successive samples from a target value. ( branch of applied statistics dealing with planning, conducting, analyzing, and interpreting controlled tests which are used to identify and evaluate the factors that control a value of a parameter of interest. ( mathematical model that relates the value of a variable with the probability of the occurrence of that value in the population. (n <xponentially 'eighted 7oving (verage control chart that uses current and historical data to detect small changes in the process. Typically, the most recent data is given the most weight, and progressively smaller weights are given to older data. ( graph of the observed fre8uencies versus each value or range of values for a set of data. ( histogram provides a graphical summary of the variation in the data. ( procedure that is used on a sample from a population to investigate the applicability of an assertion )inference* to the entire population. Hypothesis testing can also be used to test assertions about multiple populations using multiple samples. ( process in which the 8uality characteristic being evaluated is in a state of statistical control. This means that the variation among the observed samples can all be attributed to common causes, and that no special causes are influencing the process. ( single unit or a single measurement of a 8uality characteristic, usually denoted as K. This measurement is analyzed using an individuals chart, CISI7 or <'7( chart. ( control chart for processes in which individual measurements of the process are plotted for analysis. (lso called an !chart or K!chart. ( measure of the location or center of data. (lso called the average. The mean is calculated by summing all of the observations and dividing by the number of observations.

C*S*+

$esign o( Experiments

$istribution E%+A charts

,istogram

,ypothesis )esting

In'Control Process

Individual

Individuals Chart +ean

+edian +ixing

The /middle0 value of a group of observations, or the average of the two middle values. ( generally improper sampling techni8ue that arises in practice when the output from several processes is first thoroughly mixed and then random samples are drawn from the mixture. This may increase the sample variability and ma,e the control chart less sensitive to process changes. This action violates the fundamental rule of rational sampling. The observation that occurs most fre8uently in a sample. The data can have no mode, be unimodal, bimodal, etc. ( measure used to help calculate the variance of a population based on differences in consecutive data. Two consecutive individual data values are compared and the absolute value of their difference is recorded on the moving range chart. The moving range chart is typically used with an ndividuals )K* chart for single measurements. ( unit with one or more nonconformities or defects. (lso called a re9ect. ( specified re8uirement that is not fulfilled, such as a blemish, defect or imperfection. ( continuous, symmetrical, bell!shaped fre8uency distribution for variables data that is the basis for control charts for variables, such as x! bar and individuals charts. 6or normally distributed values, ##.G?@ of the population lies within ? standard deviations of the mean. (ccording to the Central Himit Theorem, subgroup averages tend to be normally distributed even if the output as a whole is not. ( control chart based on counting the number of defective units in each constant size subgroup. The np!chart is based on the binomial distribution. Inusually large or small observations relative to the rest of the data. (n element often introduced into a process by a well!meaning operator or controller who considers any appreciable deviation from the target value as a special cause. n this case, the operator is incorrectly viewing common!cause variation as a fault in the process. ;ver control of a process can actually increase the variability of the process and is viewed as a form of tampering. ( problem!solving tool that involves ran,ing all potential problem areas or sources of variation according to their contribution to cost or total variation. Typically, A%@ of the effects come from $%@ of the possible

+ode +oving Range

Noncon(orming *nit Noncon(ormity Normal $istribution

np'Chart -utliers -ver control

Pareto Chart

causes, so efforts are best spent on these /vital few0 causes, temporarily ignoring the /trivial many0 causes. p'Chart Percentiles ( control chart based on the proportion of nonconforming units per subgroup. The p!chart is based on the binomial distribution. Percentiles divide the ordered data into "%% e8ual groups. The ,th percentile p, is a value such that at least ,@ of the observations are at or below this value and )"%%!,*@ of the observations are at or above this value. ( probability distribution used to count the number of occurrences of relatively rare events. The Poisson distribution is used in constructing the c!chart and the u!chart. Precision of measurements refers to their long!run variation )s$*. t is a measure of the closeness between several individual readings. ( measure of the ability of a process to produce output that meets the process specifications. 2uartiles divide the ordered data into > e8ual groups. The second 8uartile )2$* is the median of the data. ( subset of the population chosen such that each member of the population has an e8ual probability of being included in the sample. ( measure of the spread of the data, calculated as highest value minus lowest value. Cange O C O xmax 3 xmin Rational Subgroups ( principle of sampling which states that the variation between subgroups or samples should be solely attributable to the common causes in the system rather than the sampling method. Cational subgroups are usually chosen so that the variation represented within each subgroup is as small as feasible for the process, so that any changes in the process, or special causes, appear as differences between subgroups. Cational subgroups are typically made up of consecutive pieces, although random samples are sometimes used. ( control chart based on the range )C* of a subgroup, typically used in con9unction with an x!bar chart. ( consecutive number of points consistently increasing or decreasing, or above or below the centerline. ( run can be evidence of the existence of

Poisson $istribution

Precision Process Capability .uartile Random Sampling Range

R'Chart Run

special causes of variation that should be investigated. Runs Chart ( simple graphic representation of a characteristic of a process which shows plotted values of some statistic gathered from the process. The graphic can be analyzed for trends or other unusual patterns. ( control chart based on the standard deviation, s, of a subgroup. The s! chart is typically used in con9unction with an x!bar chart. ( subset of data from a population that can be analyzed to ma,e inferences about the entire population. The probability distribution of a statistic. Common sampling distributions include t, chi!s8uare )$*, and 6. ( graphical techni8ue used to visually analyze the relationship between two variables. Two sets of data are plotted on a graph, with the y!axis being used for the variable to be predicted and the x!axis being used for the variable to ma,e the prediction. Control chart interpretation rules that are designed to increase the responsiveness of a control chart to out!of!control conditions by loo,ing for patterns of points that would rarely happen if the process has not changed. (daptations made to control charts to help determine meaningful control limits in situations when only a limited number of parts are produced or when a limited number of services are performed. Short!run techni8ues usually loo, at the deviation of a 8uality characteristic from a target value. ( high!performance, data!driven approach to analyzing the root causes of business problems and solving them. Six!sigma techni8ues were championed by 7otorola. The tendency of the data distribution to be non!symmetrical. 1egative s,ewness denotes more small observations, while positive s,ewness denotes more large observations. S,ewed data may affect the validity of control charts and other statistical tests based on the normal distribution. Causes of variation which arise periodically in a somewhat unpredictable fashion. (lso called assignable causes, local faults, or sporadic problems. Contrast to common causes. The presence of special causes indicates an out!of!control process. The written or engineering re8uirements for 9udging the acceptability of

S'Chart Sample Sampling $istribution

Scatter Plots

Sensiti/ing Rules

Short'run )echni0ues

Six Sigma

S&e1ness

Special Causes

Speci(ication

the output of a process. Spread Stability Standard $eviation Statistic The amount of variability in a sample or population. ( process is considered stable if it is free from the influences of special causes. ( stable process is said to be in control. ( measure of the spread of a set of data from its mean. ( value calculated from, or based on, sample data which is used to ma,e inferences about the population from which the sample came. Sample mean, median, range, variance, and standard deviation are commonly calculated statistics. The condition describing a process from which all special causes of variation have been removed and only common causes remain. ( collection of problem solving tools useful in achieving process stability and improving capability through the reduction of variability. SPC includes using control charts to analyze a process to identify appropriate actions that can be ta,en to achieve and maintain a state of statistical control and to improve the capability of the process. (nother name commonly used to describe statistical process control techni8ues. Stratification arises in practice when samples are collected by drawing from each of several processes, for example machines, filling heads or spindles. Stratified sampling can increase the variability of the sample data and ma,e the resulting control chart less sensitive to changes in the process. (nother name for a sample from the population. (n action ta,en to compensate for variation within the control limits of a stable system. Tampering increases rather than decreases variation, as in the case of over control. ;ccurs when a true hypothesis about the population is incorrectly re9ected. (lso called false alarm. The probability of a Type error occurring is designated by . ;ccurs when a false hypothesis about the population is incorrectly accepted. (lso called lac, of alarm. The probability of a Type error

Statistical Control Statistical Process Control !SPC"

Statistical .uality Control !S.C" Strati(ied Sampling

Subgroup )ampering

)ype I Error

)ype II Error

occurring is designated by . u'Chart ( control chart based on counting the number of nonconformities or defects per inspection unit. The u!chart is based on the Poisson distribution. &ata values which are measurements of some 8uality or characteristic of the process. The data values are used to construct the control charts. This 8ualitative data is used for the x!bar, C!, s! and individuals charts, as well as the CISI7 and moving range charts. The differences among individual results or output of a machine or process. Jariation is classified in two ways4 variation due to common causes and variation due to special causes. ( control chart used for process in which individual measurements of the process are plotted for analysis. (lso called an individuals chart or !chart. ( control chart used for processes in which the averages of subgroups of process data are plotted for analysis.

2ariables $ata

2ariation

3'Chart 3'bar Chart

Bibliography
1elson, Hoyd S. )"#AM*, 5 nterpreting Shewhart K Control Charts5, Pournal of 2uality Technology, "G4"">!"N. Steel, C. B. &. and P. H. Torrie )"#A%*, Principles and Procedures of Statistics.1ew Lor,4 7cBraw!Hill. 'estern <lectric Company )"#MN*, Statistical 2uality Control Handboo,, available from (TT Technologies, Commercial Sales Cler,, Select Code G%%!>>>, P.;. Eox "##%", ndianapolis, 1 >N$"#, "!A%%!>?$!NN%%.