Control Charts Theory and Use

A control chart is a graphical display of a measured quality characteristic. The measurement is plotted on the vertical axis and the sample (subgroup, subsample or sample number) is recorded on the horizontal axis. All quality control charts have a midpoint (or centerline) which corresponds to the process average (the mean of the normal distribution). This line is usually plotted as a horizontal solid line but can be reversed and plotted as a central dotted line. Above and below this are the upper control limit and lower control limit lines which represent three standard deviations above and three standard deviations below the process mean ( 3). ymbols are used in the charts as a shorthand means of conveying information. 2 (pronounced 3 2"bar) means 3the average of3. !ontrol charts are usually used quite extensively at the beginning of a new program, process or product, for all"important operations and processes. !harts are removed as they are found to be unnecessary or the process stabilizes. As process #nowledge increases control charts for attributes tends to decrease while the number of charts for variables tends to increase. These are quality management decisions. There are at least $% types of control charts, divided into two categories& charts for variables and charts for attributes. Attributes are divided further into two categories& defective units and defects per unit. 'n quality control a variable is a characteristic that can be measured, an attribute is a characteristic that can be counted. All variable control charts must trac# only one quality characteristic of one product on the same chart. Attribute charts can only provide nonconformance information on characteristics outside of specifications. (ariable charts can show patterns within the specification limits.

)*! +,-% .*T /0+

A fundamental principal of 4! is that the emphasis is on controlling the process not the product. 'f the process is in control, it is much more probable that products produced will also be in control. The basic purpose of all quality control charts is to reduce variation. The means to this end is to provide information on the causes of variation so these causes can be eliminated (or their effects can be reduced). The first way variation is reduced is to correct or eliminate the cause of quality problems (measurements outside the limits). The second way variation is reduced is by improving the process 5 to reduce variation within the control limits so the limits get closer and closer to the chart central value. The operator and,or immediate supervisor usually accomplish the first variation reduction. 4rocess improvement can only be achieved with the involvement of top management. !ontrol charts have a direct role in these two variable reduction activities "corrective action and process improvement. Attribute control charts have two broad purposes. 6irst is to provide overall quality information to management (percent defective for a process, a product line, or even an entire plant). econd purpose of attribute charts is to provide information for the correction of problems and the improvement of quality. The nine steps in constructing a quality control chart are as follows. 1. Select the quality characteristic. 2. Develop a quality plan. 3. Select the type of control chart. 4. Choose the proper subgroup si e. !. Collect the data. ". Deter#ine the trial control li#its and chart #idpoint. $. Deter#ine the revised control li#its and chart #idpoint. %. Construct the revised control chart. &. Continue to use the chart.
)*! +,-% .*T /0+

Step 1 ' Select the (uality Characteristic 'nspect after operations that are li#ely to produce defective items to ensure that no more wor# will be preformed on faulty items. 'nspect before costly operations so that costly wor# will not be preformed on defective parts. 'nspect prior to those operations where faulty items are li#ely to cause production problems (such as brea#ing or 7amming a machine). 'nspect before operations that might cover up a defect (such as painting, assembly,etc.). 'nspect before assembly operations that can8t be undone (such as welding). 'nspect first and last pieces on automatic and semiautomatic machines. 'nspect before storage. 'nspect finished products (remember the customer is the next inspector). 9se engineering and,or operator 7udgement to determine when, where and how to inspect and use charts. Step 2' Develop the (uality )lan The purpose of quality planning should be to establish an economic balance between the cost of measurement and the value of the measurement. The general rule is to inspect as little as possible while ensuring product quality. :uality plans along with production plans are primarily documented on the wor# order. The wor# order lists all production steps, inspection points, references the drawings, specifications, procedures, instructions and standards that contain information required for production and inspection. :uality plans should contain as many of the following elements as possible (usually on the wor# order).
)*! +,-% .*T /0+

 'dentification of the various inspection points in the process flow where measurements should be ta#en. These inspection points should be shown on flowcharts as well as listed on the wor# order. !ontrol charts are frequently associated with inspection points. The type of measurement to be made and the method used to ma#e the measurement. This can be simply a visual attribute chec# recorded manually on paper to a sophisticated measuring gauge that has a computer interface for recording. This decision is made by quality management and,or engineering. The sample size or subgroup 5 subsample size. This is the number of units to be measured at any one time. ;ow often the measurements are to be ta#en (hourly, every 1-- units, four times a day, etc.). The mechanism (tool, gauge) to be used for ma#ing the measurement. )ho is to ma#e the measurement 5 operator, inspector, supervisor, etc. Step 3' Choose the type of Control Chart This step is divided into two sub steps (of course it couldn8t be simple). 6irst is determining which category of chart to use 5 variables or attributes. The second step is choosing which type of variable or attribute chart to use (there are at least $% options to choose from). Step 3*' Deter#ining the Chart Category This choice will depend on the characteristic to be controlled. 'f the characteristic cannot be measured such as color, surface defects, labeling for example then attributes charts are the only choice. 'f a great number of characteristics need control it may prove too costly to use variable charts because each variable chart can only trac# one characteristic at a time. The solution to this dilemma is to use a single attribute chart in place of a multitude of variable charts by classifying each measurement as good or bad and re7ecting the unit if any one measurement is found to be nonconforming. This method can greatly reduce the cost of controlling a multitude of characteristics.

)*! +,-% .*T /0+

Attribute charts have several disadvantages not shared with variable charts. 6irst they can provide hints, and even strong evidence of possible nonconformance in parts, but they seldom provide the detailed information needed for a complete solution. 6or example they cannot provide information on patterns that occur within the specification limits< it ta#es a variables chart to provide this detailed exposure. (ariable charts indicate shifts and potential problems before the process deteriorates to an out of control status and defects are produced. Attribute charts seldom provide definitive information needed for good on going process improvement programs. This is due in large part because of the lac# of in"control patterns from measurements of parts that fall between the specification limits. 'n general variable date can be converted to into attributes data (any measurement outside specifications is defective), but it is much more difficult to convert attributes data into variables data. Attributes can be converted into variables by use of an interval scale (7udge the effectiveness or degree of quality, on a scale of 1 to 1-, for instance)< but this practice is seldom necessary or useful. !onverting attributes may occasionally be done when using 4! for nonmanufacturing or service processes. )hen used in this manor its effectiveness is suspect because of the sub7ective =7udgement> that must be employed. 'nterval scale =7udgements require a great deal of training so that all inspectors will be 7udging from the same base, ie., the same perspectives and understandings. ?efinitions of each scale interval must be understood the same way by all (what constitutes a 1, a $, a /, etc.). @eneral @uidelines for !hart election +ariables chart' 'nstalling a new process or product or changing an old process or product. The process is obviously in trouble< it cannot produce to the tolerances on a consistent basis. ?estructive or expensive testing is being used.
)*! +,-% .*T /0+

ampling further along the manufacturing process can be reduced by a more positive control at an earlier stage.

Attributes control charts have shown a problem to exist but the solution is difficult or un#nown. There are difficult problem processes with tight specifications, overlapping assembly tolerances, expensive materials etc. )hen large subgroup sizes are desired (greater than A) and a variables chart is indicated, use X" bar,s. )hen a variables chart is indicated but the characteristic is not critical enough to warrant a large subsample size, use X" bar,R. The critical characteristic is measurable. !ustomer or contract requirements. *ttributes Chart Bperators have a high degree of control over assignable causes. Assembly operations are complex. :uality can only be measured in terms of good or bad. ;istorical information is needed for management review. .any characteristics must be measured at one time. !ost of measurement is high. 4roduction runs are large. Step 3, Choose the Type of +ariables or *ttribute Chart To Use Bnce a decision is made which category (Attribute or (ariable) of chart is to be used another decision must be made as to which variable or attribute chart is appropriate.
)*! +,-% .*T /0+

6ollowing is a list of the different types of variables control charts and an explanation of the conditions under which each may be used. +ariables Control Chart 1. X-bar. / charts. The average (X"bar) and range (D) of periodic small samples are plotted and evaluated. *ach small sample consists of successive units produced. This is the most basic and used of all variable charts and forms the model for most other charts. 0 or +. $. X-bar. s. 'dentical to the X"bar, D except that the standard deviation (s) of each sample is used in place of the range. (D). This variables chart is used to control critical characteristics where the cost of sample error is large. Earger sample sizes (greater than 1-) must be chosen and the choice must be carefully made to balance cost of sampling with the cost of possible sample error. /. X-bar./ charts using zero"base coding. This chart is the same as X"bar,D except that a coded value is used in place of each X"bar and each D (the averages and ranges of the differences between each measurement and its specification midpoint are plotted and evaluated instead of the averages and ranges of the actual measurements). This procedure allows for the use of more than one product per chart for evaluating processes where run sizes are small. This chart is probably the best one to use on short production runs. The recommended subgroup size is 0. 0. 0oving average. This is the same as X"bar,D chart except that, instead of individual subgroups of successive units produced, a moving average(2"bar)
)*! +,-% .*T /0+

mall sample sizes are used (0 to 1-) but usually

and a moving range D are calculated from individual measurements (only one measurement is made periodically instead of 0 or more successive ones). This chart is used where successive units produced would be impractical (chemical processes for example), or where variations in successive measurements are so tiny they are unmeasurable (bottling plants, for instance). 'n practice a subgroup of three is used. +. 0oving average using ero-base coding. This chart is identical to the zero" base chart explained above except that the moving averages are used in place of X"bar,D. 't is used for vary small run sizes or where measurements of successive units are impractical. The subgroup size is the present measurement plus the two previous ones. C. Charts using converted attributes data. Foth X"bar,D and moving averages charts can be used for attributes data which have been converted to variable data using a scale to estimate the degree of non"conformance. This chart is of little value in manufacturing but has some success in controlling nonmanufacturing processes. %. )re control. The tolerance band on this chart is divided into five zones& a central zone target bounded by two cautionary zones, which in turn are bounded by two re7ect zones. The chart center is the specification midpoint, and the specification (tolerance) spread determines the location and extent of the five zones (the center target, area is one"half the specifications spread and the specification limits form the boundaries of the two cautionary zones). This chart is frequently colored for ease of analysis, with the center portion (target zone) colored green, the two cautionary areas colored yellow and the two re7ect zones colored red. This is sometimes referred to as a stoplight chart. This
)*! +,-% .*T /0+

chart can be used for short production runs or long runs where the technician has a great deal of control. Although great for control this variables chart cannot be used for process analysis and improvement. A. Trend charts. 9sed for processes where wear is an integral part of the process (such as tool wear). The center and limit lines slope up or down. 'n the case of tool wear, the chart can be used for determining optimum tool replacement policies. G. Charts using individual values. Bnly one measurement is ta#en at a time and each individual measurement is used and plotted as if it were a subgroup average. This variables chart can be used in situations where most errors are li#ely to be measurement errors or where only one measurement at a time can be done. 1-. Charts using specifications li#its. 'n this variables chart the

specifications are used as control limits, so that charting can begin at the start of the process. Although this chart has the advantage of instantaneous start" up, it also has two serious disadvantages. 6irst if the process is not centered on the specification midpoint the chart can be misleading i.e. it can show a problem where none exists or not show one that does exist. econd, the limits do not show process improvements. This chart can be useful at the beginning of a product run but should be replaced with one of the more traditional X"bar,D charts as soon as possible. 11. /un charts. This is not a statistical chart li#e all the others. 't is 7ust a chart

with the sample means plotted to show trends. The central value and limits are not calculated. This chart is used at the start of a process, product or
)*! +,-% .*T /0+

measurement in order to show trends and provide for measurement control until regular statistical charts can be constructed. 1$. Constant li#it X-bar. 4lots the number of average deviations from the (2"bar) rather than the average measurements (using a rather

mean

complex formula for each plot). The limits are constants that are determined solely by the sample size. 9sed for short production runs and for charting more than one product on the same chart, as in zero based charts. This chart has disadvantages that zero based charts do not, including complex calculations for each plot and no way to show process improvements. Fecause of theses disadvantages this chart is used very little. 1/. Charts for larger sa#ple si es. 9sed when sample size is greater than pecial procedures are used to determine the control limits. This chart has

/-.

such limited use that you are not li#ely to see this outside the automotive industry. 10. 0ultivariate charts. 9sed for controlling two or more interrelated

characteristics at the same time and on the same chart. The application of this chart uses very complex and difficult calculations and has extremely limited use, it is mentioned here only to provide an awareness of its existence. *ttribute Control Charts for Defectives 6ollowing is a list of attributes control charts for defectives (defective units) along with a brief description of the chart and its use. These #inds of charts are called p charts, or some #ind of p.
)*! +,-% .*T /0+

1-

The constant sa#ple si e p chart. The basic p chart. Also called the fraction defective chart, this chart is used to control defective units, not defects per unit. The p is the fraction defective or process average and is the plotted value. This is the basic attributes chart that is used the most. 't is used to control ongoing, long" term production and is usually difficult to use with short production runs, long process times, or other special cases. .ost special cases require one or more of the special p chart adaptations listed next. The 111 p chart. This is not an actual chart of its own, but is 7ust the p chart converted to percentages. The central value, control limits and each plotted value (p) are multiplied by 1--. This form of the p chart is the most used because people tend to understand percentages better than decimal fractions. The np chart. The actual number of defectives in the sample (a whole number) is plotted, rather than a fraction or percentage. 'ts greatest value in use is to facilitate operator understanding. +ariable sa#ple si e p chart. 'n this chart, the subgroup (sample) sizes are not constant (not the same for every sample). Therefore the control limits do not form a straight line as they do in constant sample charts. *ach sample (subgroup) has its own control limits. The variable p chart is good for short production runs and long process times where only counted data are available. The average p chart. This chart is identical to the variable p chart, except that an average sample size is used to determine the control limits, so the limits will form a straight line. Theoretically this chart can only be used when the sample sizes

)*! +,-% .*T /0+

11

are within 1-H of each other, although this limit is often exceeded in actual practice with limited detriment to accuracy. The 2eighted p chart3 using a constant sample size. The counts of each type of defective are ad7usted by weights, so that more important types of defectives are given special consideration. *specially useful for nonmanufacturing applications needing process control. The 2eighted average p chart3 using a variable sa#ple si e. 'dentical to the weighted p chart, except that an average sample size is used in the computations. Again this chart is useful in nonmanufacturing applications. The constant li#it p chart. 4lots the number of standard deviations from the mean, instead of actual counts. The limits are always constant at I/ and "/. The sample sizes must be constant. Two big disadvantages here< complex calculations are required and chart limits do not show process improvements.

*ttributes Control Charts for Defects The constant sa#ple si e u chart. The basic u chart. Also called the average defect per unit chart, this chart is used to control defects per unit, not defective units. The u is the average defects per unit and is the plotted value. Just li#e the p chart, this chart is used to control ongoing, long"term production runs and li#e the p chart is difficult to use with short production runs, long process times, or special cases.

)*! +,-% .*T /0+

1$

The variable sa#ple si e u chart. As in all variable sample size charts, the sample sizes in this chart are not constant. Therefore the control limits do not form a straight line as they do in constant sample size charts. This chart can be used for short production runs and long process times where measurements are not possible, nor desirable. The average sa#ple si e u chart. The same as the variable u chart (sample sizes are not constant), except that an average sample size is used to determine the control limits so that the limits will form a straight line. Theoretically the sample sizes must be within 1-H of each other, but accurate results are achieved successfully in practice with greater diversity. The 2eighted u chart3 using a constant sa#ple si e. The counts of each defect are ad7usted by weights, so that more important defect types are given more attention. 't is used with constant sample sizes only< and it is especially useful for nonmanufacturing applications. The 2eighted u chart3 using a variable sa#ple si e. This chart is identical to the weighted u chart, except that an average sample size is used in computations. 't is used for variables subgroup sizes only and it is especially useful for nonmanufacturing applications. The constant li#it u chart. 4lots the number of standard deviations from the mean, instead of the actual counts. The limits are always a constant I/ and "/. The sample subgroup sizes must be constant. Two big disadvantages with this chart are the complex calculations required for each plot and chart limits that do not show process improvements. 'n other charts the limits become closer and closer as the process improves.
)*! +,-% .*T /0+

1/

The basic c chart. The c chart is a special case u chart where the subgroup size is always one unit. Also called the defect per unit chart (there can be no average size sample when the sample size is always one), it is used for very large and critical products. The c is the defects per unit and is the plotted value. The 2eighted c chart. The counts of each type of defect are ad7usted by weights, so that more important types of defects are given special attention. 't is especially useful for non"manufacturing applications. The constant li#it c chart. This chart plots the number of standard deviations from the mean, instead of the actual counts. The limits are always constant at I/ and "/. 'ts one big advantage is that it can be used for short production runs. ;owever it does require complex calculations for each plot and chart limits that do not show process improvements. Step 4' Choose the /ational )roper Sa#ple Si e. Actually there are two sample sizes involved in control charts& the total sample size needed to start 5up the chart, and the subgroup size. The total start"up sample size normally used is $+ subgroups. Thus if the subgroup size is 0, the total start"up sample size is 1--. 'n order to be statistically viable the start"up sample size should consist of 1-- or more measurements or observations. ome special charts do use less, which is one reason they should be used with a degree of caution. The subgroup size is the amount measured at one time, and should be measured consecutively as produced.
)*! +,-% .*T /0+

ubgroups are also statistical samples

10

but are called subgroups (and sometimes subsamples) so as not to confuse them with the total start"up sample. The actual size of the sample depends on the analysis method being used, the production run size, the amount available to count, and,or cost considerations. 'n actual practice, the subgroup is usually called the sample or sample size after the initial start up is completed. Sa#pling 4requency A good rule of thumb to use for determining sample size and sample frequency for variables data is an adaptation from .'EK T?"010. 'n general the number to inspected per lot is as follows& 5ot Si e 1. C-"/-$. /-1"1--/. 1--1"+--0. Bver +--)ercentage 1-H +H $H 1H

Lormally lot size is considered to be one day8s production< however this can be expanded to as much as a wee# under special circumstances. The main reason for limiting lot size to one day8s production is to facilitate process analysis and assignable cause identification. 4eople tend to forget the details of what happened after more than one day. 6or a daily production run of $-- (using the rule of thumb) $- units should be inspected each day. 'f the subgroup size is 0, there should be a minimum of + samples of 0 measured and recorded per day. The subgroups should be ta#en at random or as random as possible. Demember, however that the subgroup sample (in this case 0) should be measured consecutively as produced, not randomly. The reason for this is that the sampling time is often critical to correction and,or improvement in procedures.
)*! +,-% .*T /0+

1+

+ariables Charts Sa#ple Si es X"bar,/. Lormally, $+ subgroups of 0 or + are used for a total start up sample size of 1-- or 1$+. ubgroup sizes smaller than this can be used, say three and the mathematical error will not produce calculations that are significantly in error compared to the population. Earger subgroups will produce significant errors and are not generally used. This chart ignores the measurements between high and low for the range and in large subgroups this can introduce errors in the determination of the standard deviation. X"bar,s. 6or a critical characteristic it may be desirable to ta#e larger sample sizes for control. This chart is used for subgroup sizes of A or more depending on the cost tradeoffs. Again the start up rule of thumb is $+ subgroup samples. 6or a subgroup size of 1+ the total start"up sample size will be /%+ (1+ x $+). 6ero-base coding with X"bar,/. 'dentical to the X"bar,/ chart as far as sample size is concerned. 0oving average. 'n order to #eep the moving average from smoothing out the variation too much (variation is, after all what we are trying to measure), the subgroup size must be three. The minimum start up sample size is then /x$/M CC ($/ subgroups are constructed of $+ individual measurements. 6ero-base coding 2ith #oving average. 'dentical to moving average chart. Chart using attributes data. 'dentical to the chart being used. )re-Control. warning area.
)*! +,-% .*T /0+

ubgroup size is usually one but can be two if the first falls into the

1C

Trend charts. 'dentical to X"bar,/. Chart for individual values. The subgroup size is one. Therefore the minimum start"up sample size is $+. Charts using Specification 5i#its. there is no start"up sample size. ince the limits do not have to calculate,

ubgroup sizes are identical to X"bar,/.

/un. Bnly individual measurements, or subgroup averages, are plotted.

Step !' Collect the Data Any sample in order to properly represent the population characteristics, must be unbiased (random, independent and homogeneous). 'n meeting these criteria, subgroups can be selected at one time or over a period of time. The instant time method is usually preferred because it provides a time reference for pinpointing assignable causes. 'f the instant time method is used, the inspector measures several subgroups of parts at random times during the day. That is each subgroup is randomly chosen but the subgroup itself consists of consecutive units produced at the random time chosen. 'f the purpose is to gather enough data for the construction of the control chart, the measurements are recorded until enough observations are made, after which the chart is made and the subgroup averages recorded. ?ata are usually gathered by the inspection function as a normal part of their regular duties or by the operator themselves if a complete 4! program is in effect.

)*! +,-% .*T /0+

1%

Data Categori ation' 1. Continuous. this type of data is the measurable data used in most charts. $. /an7ed. (list from high to low). *ach ran#ing is assigned a value that is then used in the variables charts as if it were an actual measurement. This is one method of converting attributes data into variables data. /. 8nterval. (7udgement or weighting on a scale of 1 to + etc. how good or bad is it. This is the most used method of converting attributes data into variables data. 0. Discrete. 'n quality control all discrete data are countable< count the number of defects or defectives. This is the only type of data that can be used in attributes charts.

Step "' Deter#ine the Trial Control 5i#its and the Chart 0idpoint. !ontrol limits are values that are plus or minus three sigmas ( 8s) from the central value, or midpoint. Therefore if the midpoints and limits have been properly determined and the process is properly operating, GG.%/H of all subgroup averages will fall between these control limits. Any one average has only a .-$%H chance of being outside these limits. All statistical quality control charts have a midpoint or centerline which corresponds to the process average (the ) and an upper and lower control limit which corresponds to the three standard deviations from this midpoint or centerline. Step $' Deter#ine the /evised Control 5i#its and Chart 0idpoint. Trial control limits are calculated first for the entire sample, after which the out" of"control subgroup means (those that are either above the upper limit or below the lower limit) are discarded and the limits recalculated. Fefore any values
)*! +,-% .*T /0+

1A

can be discarded, they must first be proved to be out of control (have assignable causes). 'f not they should not be discarded for revised limit calculation. 4lease note that only $% subgroup means out of 1-,--- can be outside of the control limits. 'f several subgroup means are outside the limits and no assignable causes can, at first, be found, it is far more li#ely that the search for assignable causes had been flawed. 9nder these conditions, the search should probably be expanded and continued. 6or example the probability that two outside values in a row would occur by chance alone is .--$% 2 .--$% M .-----%$ or about % in 1,---,---. The chance they would occur on the same side of the chart is .--1/+ 2 .--1/+ M . -----1A or about $ out of 1,---,---. The chance that five outside values are outside the limits by chance alone is %.$ in one trillion. 't is for these reasons this reason that many analysts automatically assume an out of control condition when more than two high or low points are present, even though no assignable cause can at first be found. Step %' Construct the /evised Control Chart. !ontrol charts are 7ust graphs of the subgroup means (averages) with the central value shown as a solid line and each limit shown as a dotted line. All control charts are constructed from the basic normal curve model. !ontrol charts for variables use the normal curve model direct, while attributes charts use the normal curve approximation to the binomial. !ontrol charts are normally only constructed for revised limits only. Step &' Continue to Use the Charts. Bnce the revised control chart is constructed, all subgroup means (even out"of" control values) are entered on the chart, and the chart is displayed
)*! +,-% .*T /0+

1G

conspicuously at the 7ob site.

amples of the same size (subgroup size) are

measured periodically and charting is continued. The chart represents a continuous picture of the process improvement.

)*! +,-% .*T /0+

$-