C A PA B I L I T Y S T U D I E S
How To Perform A Machine Capability Study
PLAN, EXECUTE, ANALYZE AND AVOID PITFALLS.
ew topics have suffered more debate and misunderstanding in industry than machine capability studies. With no known standards for analysis, capability study methods vary greatly from company to company and even plant to plant within the same corporation. Automotive standards, such as QS-9000, ISO/TS 16949 and VDA (the German Automobile Industry Association), provide us requirements for capability studies and facilitate limited understanding, but they fall short in the how to category. Nevertheless, every Green Belt and Black Belt should know how to perform a machine capability study and thoroughly understand several pitfalls associated with it. Definition The best place to start with machine capability is to define it. In recent time, a number of definitions have evolved: • Inherent variation of a machine.
By Gregor y Roth II
• Calculation of Cp and Cpk. • Assessment of variations in a manufacturing process compared to a given tolerance. • A measure of the inherent ability of a production machine to reproduce its results with accuracy and precision. • A measure of performance. • Analysis of the random variability in a process. • Inherent reproducibility of parts produced by the process. • Machine capability equaling 6 sigma. • Inherent machine capability. None of these completely defines machine capability. Determination of the inherent variation of a machine is the most popular definition, but while this is an excellent theoretical definition, it is rarely achieved in industry. The preceding may seem like a harsh statement, but examine the facts. Data from a machine (current running condition) is collected and analyzed, and the results are published. If the results of the analysis are unacceptable, something is done to make the machine better. If the results of the analysis are acceptable, no action is taken on the machine. The reality of the situation is acceptable machine performance can be improved: A worn feed arm or bearing can be replaced. Either action can improve the capability of the machine. Yet, the inherent variation of the machine still has not been determined. To determine the inherent variation, as much play (variation) as possible must be removed from every moving part in the machine, and the optimum raw mate-
N O V E M B E R
2 0 0 3
W W W . A S Q . O R G
If .002. but it does not determine the inherent variation of a machine. the variation (standard deviation) remains reasonably constant.502 and . . maintenance cost and quality of product by studying various speeds. feeds and other aspects to get the optimum settings to maximize yield. . This machine spread level represents the delicate balance between output and cost factors. Every time a sample is taken from the population.) The sample average may be adjusted by machine setting.00051 = 1.501. Other helpful information includes adjustments for manufacturing and location of a machine or tool head with respect to part station. The gage on the production floor may or may not be adequate for the study. The next consideration is sampling the population to determine process parameters. If you study a given machine over a long period (several different studies). Steps for Machine Capability Study Based on the previous information. Additional considerations may be prediction of percentage out of tolerance for scrap or rework. . fixture or machine are data required? Step two—measuring.96 to 1. This data yields an average of . the estimated average changes slightly with each consecutive part produced. the population average and standard deviation could be calculated.504 (the next data point) is added to the data set. (Cpk = . and it would never change.499 are a sample from a machine. cycle time and maintenance cost.” Most companies or plants are comfortable with this study method.96. A new. then the smallest gage increment is at least . This entails selecting the measuring instrument.H o w To P e r f o r m a M a c h i n e C a p a b i l i t y S t u d y
rial properties for input must be found. Another way of stating this is. The analysis of the data will provide some insight into the operation of the machine. No one in industry determines the inherent variation of older. cost. the data valS I X S I G M A
ues . A machine capability study is performed to answer questions about the machine.752 divided by .496.503. Ensure the gage is calibrated
F O R U M M A G A Z I N E
N O V E M B E R
2 0 0 3
. This level of determination provides optimum production output.7524 divided by . The study can be categorized as capability or diagnostic in origin. This .753 . .504. This involves the selection of variables to be studied and requires asking several questions: • Is the study to determine capability or for diagnostic reasons? • For what specific question is the study being performed? • Is there an essential dimension at the heart of the problem? • Is there a key dimension that will cover machine guidance or movement? • Is the study to predict percentage out of tolerance? • Will more than one dimension be required? • Is the sample average needed from multiple cavities or stations? • From which cavity. then gage increment should be . .XXXX.5007. debugged machine running for the first time is about as close to the inherent variation of the machine as you can get. it will determine the ability of the machine to hold a given tolerance by comparing the 6-sigma spread (commonly called process spread) to the tolerance spread. It is not cost effective or necessary to producing acceptable product.0004 shift can change a Cpk value from 1. For example. For academic purposes.501. cost estimations.0004 yields . The average has now shifted . not all the measurements from the population are ever collected because machines are retired from service before running the entire population. . interrelationship of sequential processing steps and determination of the location of the sample average. In reality. The following describes the seven basic steps necessary to perform a capability study: Step one—planning. it is important that planning is completed before embarking upon a machine capability study. In theory. . existing equipment.00051 = 1. thus. This is one number.500.. quantity. If part tolerance is .18. In fact.753 . But remarkably. a different estimate of the population (process) average and the population standard deviation is obtained. but the population average is fixed. the average changes to .18. Shifting the average . if all the measurements from a population (machine ran until it could not produce any more parts) were collected. spindle. the estimated average is a dynamic entity. Any one or all of these may be needed to direct corrective actions for a machine not performing up to expectations. the results will show the sample average is constantly moving.0004 with the addition of one data point..02.5003. The gage should follow the 10 to one rule for product measurement to minimize measurement error. if the part tolerance is . let’s call this “acceptable operational level machine spread. Usually. An operating level (variation output) of the machine that is satisfactory is found by meeting expectations for quality.497.XXX.
Ideally. • Time for unforeseen circumstances to affect the process (Murphy’s law). • Speed. Step six—recording. O R G
. the supervisor may have good rationale on why the study should not be performed at this time or provide information that might need to be included in the study. the correct answer is not in any book or a technical paper. Pieces taken over time (for example. • Tool type. This represents the best the machine can produce at a point in time. and thus variation was doubled. ambient temperature. • Environmental changes—for example. No one wants to be in the embarrassing position of reporting results of the study to management only to have the area supervisor discredit the results. • Supplier changes in incoming material (both internal and external suppliers). Which is the correct method to use? Surprisingly. Product configuration should also be recorded by part number. • Material (batch) changes. This involves collection of the raw data in an unbiased manner. • Process drift. configuration. each shift for two weeks) will provide long-term data. can the customers live with this output? Step five—approval. twice a shift. humidity. machine settings should be recorded along with the raw data: • Feed rate. A S Q . Step three—personnel. • Ambient temperature. Order of manufacture must be preserved for analysis. For example. Historical data covering several months or years—for
example. from shop floor statistical process control (SPC)—will provide long-term data.
N O V E M B E R
2 0 0 3
W W W . • Coolant temperature. • Barometric pressure. You must obtain approval from the area (machine) supervisor before the study is started. In addition. • Relative humidity. to the center onethird of the control limits) for capability study and the setup approved by the area supervisor. • Head pressure. • Cycle times. Before data collection begins. the machine should be set up properly (in other words. and day and night operation. pollution from neighboring plant. This involves selection of personnel to operate the machine and measure the parts. Odd readings cannot be discarded from the study results without solid statistical rationale. • Operator shift changes. time and operator. Don’t use personnel who are temporary or do not have indepth knowledge of how to operate the required machine. product supplier. Just because something doesn’t fit with the rest of the data is not sufficient rationale to discard it. If that is not possible. This covers determination of the sample size and selection of the samples. This will certainly demonstrate what the machine is sending to the next operation over time.H o w To P e r f o r m a M a c h i n e C a p a b i l i t y S t u d y
and functioning correctly before any measurements are made for the study. • Coolant type. Don’t forget to verify calibration after the study is accomplished to ensure the gage did not drift during use. shift. • Others factors as required by a responsible authority such as a quality management or a manufacturing standard. Beyond courtesy. • Nonconsecutive pieces over time—longer-term results. Step four—sample size. use the person who normally does these things. revision level. This recorded information will be helpful in the future for making realistic comparisons between old and new studies. • Tool life. It also suggests the question. • Machine location. Factors for consideration are: • Consecutive pieces—short-term results. use personnel who are knowledgeable in the operation to be studied. Fifty consecutive pieces will provide useful short-term data (this is a suggested minimum quantity for manufacturing factories). the coolant chiller was being repaired at the time of the study. The answer leads what the person performing the study is trying to learn: • Are you looking for a short-term or long-term answer? • Is the capability study performed for diagnostic reasons or to determine capability? • How much risk in a wrong answer is acceptable to you? The answers to these questions drive the timing and quantity of samples required for the capability study. • Tool changes.
8350 2.8360 2. Descriptive statistics. process that is known to produce a normal curve. and the data gathering picture of the process (computers are helpful with technique should be reviewed. it is distribution is skewed but the data is taken from a obvious the machine is not capable of meeting tolerance.8352 2.8330 2. The next tool to use is an Negative skew Positive skew X bar and R chart (see Fig0 0 ure 3. facturing processes do not yield a normal (or Analysis can begin with just looking at the raw data.8377 2.8320 2.834
2.8342 2. approximation of a distribution curve.8330 2. 1.8345 2.8350 2. bar will never look exactly alike. Gaussian) distribution. Histogram. is equivalent to one dot on the black line.8353 2. Because of this.8337 2.832
2. as shown in Figure 1.8350 2. which is a A skewed distribution. p.8350 2.835
2. It will answer 0 10 20 30 40 50 60 70 80 60 70 80 90 100 110 120 130 whether the process is in staC3 C2 tistical control.8345 2. this).8340 2.8350 2. Skew One is normal. X bar and R chart.837
2. shown in Figure 2.838
Note: Numbers in bold are the minimum and maximum
LSL=lower specification limit USL=upper specification limit
The dotted line is a computer interpretation of how the raw data actually appears.8350 2.8350 2.8340 2. The histogram itself is in bars. Comparison of distributions Comparison of distributions it suggests a skewed or non300 normal distribution of some 300 type. The width of a bar. this indicates the data Figure 2.8370 2. can be measure of variation.8338 2.H o w To P e r f o r m a M a c h i n e C a p a b i l i t y S t u d y
Table 1. Data Collection
Figure 1.8344 2.836
2.8355 2. Results of data collection from an actual machining A histogram provides the shape of the curve indicapability study will serve as an example.8340 2.8345 2. 26). the dot and the 2.8337 2.8357 2. Data is shown cating whether the data appears normal.8332 2. If the range exceeds the tolerance readily recognized from a normal histogram.8332 2.8352 2. As these values move away from one another. this The next step is to draw a histogram to provide a may suggest a bias in sampling. Step seven—analysis of the data. therefore.8355 2.8355 2.8377) and tion should be determined for each machine study.8348 2.8355 2.8377 2. mode (most fretogram shows how a normal process should appear.8332 2. The solid line on the hisIf the mean (arithmetic average). but it does offer an 3.8339 2. minimum (2.8340 2.8340 2.8363 2. Analysis is divided into three different parts:
S I X S I G M A F O R U M M A G A Z I N E
N O V E M B E R
2 0 0 3
2.8320) yields a range (.8356 2. Some manuin Table 1. If the spread (upper tolerance minus lower tolerance). quent value) and median (data midpoint) are close in value.8365 2.8354 2.8345 2.8347 2. The correct type of distribuThe difference between the maximum (2.8359 2.833
2.8335 2.8339 2. It needs to be clearly understood that tools 200 200 designed for use on a normal curve will not provide correct 100 100 answers on nonnormal data.0057).
833 0 0. The next part is the calculation of descriptive statissame graph.836 Sample mean 2. data will be smaller. but it will result of the laws of chance—the natural variation of provide a logical approach to analysis. The following method may be new to you.002620 3.000 R = 0.0 SL = 0.001 0.H o w To P e r f o r m a M a c h i n e C a p a b i l i t y S t u d y
This chart indicates the process Table 2.83825 are evenly distributed about the Nominal = 2.839) is compared to maximum of actual data (2. the order of manufacture.005 0.008 Range = 0057 6-sigma spread = 00701 an X bar and R chart to be effecNote: Each line is to be read horizontally across and each parameter is to be comtive.833 2.35
N O V E M B E R
2 0 0 3
W W W .006 Sample range 0. When raw data is within analysis at this point. (Note the cause is repaired or removed as appropriate. This is referred to as a stable or in control. there is always nonconforming product (ideally.837) and then to the theoretical (average + 3 referred to as an unstable process or out of control. take the steps necessary to bring tolerance and the average plus 3 sigma (theoretical
Figure 3.83825) of the data. If the process is not in statistical control.0 SL = 2.831 Minimum = 2.83124 trends or shifts.0 SL = 2.8350 Mean = 2.836 2.000 Note: SL = statistical limit -3.003 0.834 -3. and there are no runs. Starting with the top line of Table 2. standard maximum in each case is a logical approach.004 0.837 3.835 2. A process in control means all the variation is the tics. while tolerance limits are individual on a normal curve will not provide correct answers on data points.83474 centerline.0 SL = 0. O R G
. stop the less than the tolerance limit). Using Table 2 will the process.839 Maximum = 2. they should not be shown on the nonnormal data.8377 Average + 3 sigma = 2. Xbar/R Chart for study
2. Control limits represent an average of Again. by engineering.) If the deviation of the machine or process will become raw data is above the upper tolerance limit. That means no plots (or points) are Part tolerance Raw data Statistical analysis outside the control limits.005540 1 2 3 4 5 6 7 8 9 10 X = 2. This is one of the few absolutes in statistics. the upper tolA process not in control has variation from the laws of chance plus variation from an assignable cause. facilitate analysis and ensure all aspects are covered. the data must be plotted in pared to the other parameters on that line. Control limits are statistical limits and cannot be interchanged with tolerance limits. When a process is out of control and the assignable sigma) maximum (2. which are set the process in control and repeat the seven steps. It is not well-publicized that for Tolerance spread =. Calculation of Descriptive Statistics is in statistical control. remember that statistical tools designed for use a subgroup size. A S Q .835 Mode = 2.8320 Average – 3 sigma = 2.002 0. plots Upper tolerance = 2. Thus. Lower tolerance = 2. This is erance (2.
008 Range = 0057 6-sigma spread =. there is always nonconforming material. Remember.03). Curve One
Figure 5. The snapshot in time yields valuable information about the process but is only as current as when the data was gathered. It is a snapshot in time. By the time the data is evaluated. Would your company run a machine with a Cp of 1. Curve Two
Lower tolerance limit Upper tolerance limit Area out of tolerance Area out of tolerance
2. If the mean. the 6-sigma spread (process spread) should be 75% or less than the tolerance spread (.00026) is precisely how far the process needs to be adjusted to bring it back to the nominal of the tolerance. As such. In this example. This example is off center by . The last line of Table 2 provides a summary of the
S I X
process. Comparison of the mean to the nominal shows exactly how far the process is off center. The second line of Table 2. this is a personal decision.835 Mode = 2. The problem is how to use this information to
F O R U M M A G A Z I N E
S I G M A
N O V E M B E R
2 0 0 3
. mode and median are close together in value. If the process is producing nonconforming product. The difference between the two numbers (.H o w To P e r f o r m a M a c h i n e C a p a b i l i t y S t u d y
Figure 4. and appropriate corrective action steps are needed or nonconforming product should be expected at some time during the production run. 6 sigma is not larger. The third line of Table 2 addresses the opposite end of the process curve from the first line but is read in the same manner. It is possible to get a different perspective by substituting the median for the mode for comparison. If the raw data is below the lower tolerance limit (2.00702) is larger than the tolerance spread . This example yields a Cp (1. it is an indication the data is producing a normal curve. the required information to calculate Cp and Cpk is available. nonconforming product should be expected at some time during the production run.83474
maximum) is above the upper tolerance. A Snapshot in Time A machine spread study is equivalent to taking a picture of a car race. nonconforming product should be expected at some time during the production run. Process centering and process variations are two separate and distinct process characteristics. Figure 4 shows an example of a process exceeding the upper tolerance limit.13 if correcting it meant serious dollars or lost schedule? Is the control system (operator) sufficient to keep the process adequately centered? Obviously. not process centering. The same holds true if the 6-sigma spread (.00026 as discussed in the nominal line.14) and Cpk (1.83124) is below the lower tolerance.008). If the range is larger than the tolerance spread. When raw data is within tolerance and the average minus 3 sigma (2. Ideally.8350 Mean = 2. Cpk calculates for the position of the mean in the tolerance zone.(008). it is in the past when compared to the process time. comparing the nominal to the mode (most frequent value). you can utilize the Z score statistical tool and a table of area under a normal curve to determine the extent of the out of tolerance conditioning. they must be dealt with individually.831). and the process continues to move with time just as the car race continues after a picture is taken.831 Tolerance spread =.839 Nominal = 2. Figure 5 shows an example of the process exceeding the lower tolerance limit. the process is not capable of meeting engineering tolerances. If the process is producing nonconforming product. At this point. you can utilize the Z score statistical tool and a table of area under a normal curve to determine the extent of the out of tolerance conditioning. yields an indication of how well the process is centered.
People believe many major myths associated with machine capability studies. long term analysis.8320 Average – 3 sigma = 2. shipped the accumulation of all the curves including 3. A machine capability study is usually performed with short-term data.835 Mode = 2. the process is not capable After years of study.03 dard deviation for each subgroup. If you chose to run the process using a machine The migration of the average inflates the apparent spread study as indicated. O R G
.83474/2. Compare these numbers to the tolerances. This calculation will yield a smaller per1. it is inflated because sigma is inflated due to continual Pitfalls to Capability Studies sample average shifting. The user must change hats and look at the process from the process point of view.8377 Average + 3 sigma = 2. Good work on a correct machine capability study show the process capable of meeting specifications. 5. and nonconforming for a specific process remains reasonably constant product (and possible scrap and rework) will occur over time. tomer) tolerate occasional nonconforming product? When percentage out of tolerance is calculated. Long-term data consists of many short-term averages added together to yield one long-term overall average. This will demonstrate whether the process is truly capable of meeting specifications over time because the predicted movement of the sample averages has been included.839 Maximum = 2. but they calculate stanmethod with the example in Table 3 yields a Cp of 1. If out of tolerance conditions are predicted. Then calculate the average plus and minus three standard deviations from the confidence interval limits to achieve the results shown in Table 3. traditional SPC methods do not provide the information required for modern. Part tolerance Raw data Statistical analysis Bottom line. high volume production lines. (assuming it is not sorted out). Figure 6 shows the average shifting. This the process average migrates. Tolerance spread =. moving the whole curve out of tolerance for There are six pitfalls to capability studies: a brief period. but the process will generate nonconformance in the long run if the Table 3.1 Control charts illustrate how and a table for area under a normal curve. long-term analysis. 1. Sometimes it is necessary to review long-term data (as with SPC data). Depending on the user’s circumstances. the percentage out of tolerance may be calculated as a Z score
can be rendered useless by measurement error. Refer back to the second line in Table 2 (where the mean was calculated) and substitute the confidence interval for the mean (95% in this example).8350 Mean = 2. Process Prediction Results process average is not carefully controlled. Unfortunately.8350 plete the information needed for Lower tolerance = 2.0078
realistically predict the future. In this case. A short-term study will 4. Could your company (your cus. you can comNominal = 2.33 Cp/Cpk. Understanding the range through which the process average (short-term averages) migrates may be more beneficial to problem solving or process control. for the same basic (with confidence inter val data added) calculations required to determine Upper tolerance = 2. The irony of this approach is the customer is 2.8308 Table 2 or Table 3 and gain more knowledge about the process. The 50 pieces from Table 1 represent short-term data. Table 3 will demonstrate one method of process prediction.H o w To P e r f o r m a M a c h i n e C a p a b i l i t y S t u d y
6. They provide short-term vs.
N O V E M B E R
2 0 0 3
W W W . The formula used to calculate standard deviation any product that is not conforming to specification could change the results of a capability study.87.831 Minimum = 2. A S Q . This can be utilized to estimate long-term data and predict the future.8386 capability indices. an overall average may not be what is required to diagnose a problem with the process. centage out of tolerance than will long-term data. and Cpk of 0. around some central point (central limit theorem). They provide short term vs. nonconforming product value of standard deviation for calculation of longwould be produced. but the process average migrates in a range sometime in the production run. I have found standard deviation of meeting a goal of 1.008 Range = 0057 6-sigma spread = .term data as shown in Figure 6. the analysis approach needs to change. Process potential software can be misleading. Read Table 3 in the same manner as Table 2 was read. Over time.
B. B. the computer estimated standard deviation will be as small as calculated. 2. It seems a bit illogical to have a computer calculate standard deviation using any shortcut method.1 Sample mean Predicted variation C Actual variation
1. 3. it is visually apparent some software will show a process capable of meeting tolerance while it is actually producing parts out of tolerance. the software shows all assignable causes removed or repaired. but also welcomed with open arms because of the simplicity of calculation.H o w To P e r f o r m a M a c h i n e C a p a b i l i t y S t u d y
Capability indices are tricky in correct interpretation and controversial in their application. However. almost everyone has access to a computer. Some software packages are taking the data strings from A.0
B 0 5 10 Sample number 15
This fact alone will cause variation in indices and add to the confusion and misunderstanding. However. Formulas are available to calculate long-term standard deviation. not predictors of the process. The user must clearly understand the use of each statistical tool to obtain correct analysis. The formula used to calculate standard deviation could change the results of a capability study. Actually. Constants in formulas are estimators. this software misleads the user since it fails to warn that its computer analysis displays only what the process is capable of performing when all problems are corrected—not what the process performance actually is. Calculated standard deviation is smaller than the true standard deviation of the process. and C contain the same average. Users may utilize any number of methods for computing standard deviation. Unfortunately. Figure 7 is an example. Many statistical software companies are now providing process potential as part of their analysis of data. This is a misleading situation. however. such as but not limited to the following:
mula or method for calculation of standard deviation to bring consistency to its organization. data from an actual machining process provides a view of misrepresentation of process potential. the calculation method remains a personal choice of the user. They tend to fall into two categories of error: 1. Process Standard Deviation Example
A 1. Today. Inconsistencies are not limited to the brand of software used. Migration of Average
Figure 7. If A. in the formula
the d2 value is a constant. The result is the user mistakenly thinks the process is capable of meeting the specification and is puzzled as to why all the scrap and rework is being generated. including sample average shifts and any associated nonconformance. which masks the true percentage out of tolerance. B and C in the figure. which produces a standard deviation smaller than the actual process standard deviation. forcing a common average between them and then calculating standard deviation from the forced average. and the customer of this machine is seeing the full extent of the variation. It would be prudent for management to standardize one forS I X S I G M A F O R U M
M A G A Z I N E
N O V E M B E R
2 0 0 3
. Before the advent of computers. and C do not share a common average. Most processes do not conform to the constant used in the formula. Process potential software can be misleading. the shortcut (R bar divided by d2 method) calculation for standard deviation was not only needed. In it. reality is A. It is accepted by industry that the sum of the squares method provides consistent results. While Figure 7 is extreme. 2. For example. It is difficult
0012 . Since these just happen to be the only two things that prevent us from producing a perfect part every time.001168 1.001143 Cp 1. A number of different methods are in practice today for process prediction. If you ask the wrong question (use the wrong statistical tool).39 1.0011738 . A machine does not understand: • Part tolerances. take the following quote to heart: “It should be remembered that the greatest designs of mankind are doomed to failure if the component pieces cannot be measured adequately and the measurement is always accomplished by utilizing a sophisticated rubber ruler to estimate dimensions. it behooves us to learn from the machine. however. 5. 4.001167792 . Think of the machine as a lower life form with limited intelligence. A big hammer against the side of a machine yields bad parts. This method will yield a better prediction of the out of tolerance conditions and a realistic assessment of just where the process average will actually migrate. change any variable (such as coolant. The user must change hats and look at the process from the process point of view. The analysis of a machine may be simplified by keeping one thought in mind: Use the right tool to ask the right question. • Latest ballgame scores. O R G
. temperature or material hardness) and the machine will react to the change. If the effect the measurement error has had on the data (and process predictions) has not been evaluated.”2 I have seen measurement error (via gage R&R) to be much as 150% of the part tolerance. Data is only as good as the measurement error associated with it. you get the wrong answer.43 1. It is important to remember that it is the machine itself that one is attempting to understand. To illustrate this. • Capability indices. This effectively yields shortterm data. resulting in different Cp index calculations. All computer software uses different algorithms to calculate standard deviation.27 1. the preceding may seem arrogant.0013143 .5 sigma. At first. Better stated. Most of us view the machine as a nonentity. one standard set of data was selected and different software packages were utilized to calculate standard deviation from the same standard data set. none of these things is part of the natural world of the machine. The standard deviation from the table is also used to calculate Cp (tolerance spread held constant at . a great deal of understanding can transpire. The problem with this approach is many of today’s processes contain less than a 1. The machine only understands natural variation (laws of chance) and assignable causes. If that thinking is reversed. • People expectations. gage R&R studies are an integral part of performing a good capability study. The user must recognize when this is occurring.42 1. Perhaps a better solution to this problem may be to calculate a confidence interval for the mean and calculate ±3 standard deviation from the confidence interval (as shown earlier).43 Standard Deviation . A S Q . Another feature of this method is all the calculations utilize the actual current process sample data (data driven) and not a developed constant to fit all processes.
Table 4. then calculates a 3-sigma limit from this shifted point. This clearly shows different types of software calculate standard deviation differently.H o w To P e r f o r m a M a c h i n e C a p a b i l i t y S t u d y
to imagine a process that follows a constant.010) to demonstrate how an index will vary depending on the sigma calculation method used.5 sigma shift. • What you want it to do. Sample Calculations Of Standard Deviations
Method Longhand Rbar/d2 Software company one Descriptive statistics Long term Six pack Software company two .46
N O V E M B E R
2 0 0 3
W W W . and you will get the right answer. as determined by a gage reproducibility and repeatability (R&R) study or a gage drifting out of calibration. • Company politics. Good work on a correct machine capability study
can be rendered useless by measurement error. The machine will react to a stimulus. Statistical tools are a way to ask the machine a question and obtain an answer. • Engineering changes. Another method arbitrarily shifts the average 1. Thus. Cp was then calculated from the standard deviations and developed into Table 4 (calculations and terminology in the table are copied directly from the individual software packages). One method is to calculate capability by subtracting the additional mean variation (calculated separately) from the 6-sigma spread.
• Measurement of the performance potential of the process is needed. booklet No. 2000. “Reality-Based SPC. 16. People believe many major myths associated with machine capability studies. No. Bad workplace habits that are part of the company culture are hard to change. One should not allow myths to stop the performance of a capability study to obtain the truth about a process. many of these pitfalls have become a way of life in many companies.
S I X
S I G M A
F O R U M
M A G A Z I N E
N O V E M B E R
2 0 0 3
WHAT DO YOU THINK OF THIS ARTICLE? Please share your comments and thoughts with the editor by e-mailing godfrey@asq. 2. • The process needs to be improved. March-April.
REFERENCES 1. • The process is perfectly centered. so a discussion of them would not be complete without listing the major myths: • All the parts are well within tolerance.” Quality In Manufacturing. July 1991. • This is as good as the process can do. The problem with these myths is many people seriously believe them. Thomas A. when the exact opposite can be true. The best advice to follow in overcoming these pit-
falls is to be aware of any errors discussed in this article and to demonstrate consistency in the performance of all studies using standardized methods and formulas. Steward. Gregory Roth II. Vol. “Why Is My Measurement Different Than Yours?” Quality Engineering. • Capital is necessary to improve the process.edu. 9. Machine and Process Capability of Machining Facilities. second edition..
BIBLIOGRAPHY Robert Bosch Corp. p. 5. Technical Statistics.H o w To P e r f o r m a M a c h i n e C a p a b i l i t y S t u d y
6. Unfortunately. 1.