A Practical Example
Imagine working on a project to reduce your energy bills at home. There are three possible approaches you might take: (1) minimize the use of all the appliances in your home simultaneously, (2) minimize the use of appliances that you think use the most energy, or (3) build a Pareto chart showing the energy consumed each type of appliance, and focus on the high-energy appliances. The first approach will result in a great deal of sacrifice but will only produce marginal results, because focusing on all appliances equally takes focus away from the vital-few appliances that use the most energy. The second approach is a roll of the dice, and wont deliver a high succe ss rate, on average. The Pareto chart that could be used for approach (3) is below (source data from Clark Public Utilities):
The Pareto chart above brings immediate focus to the project: 5 out of 22 appliances consume 75% of home energy costs. Focus on those appliances and forget about the rest. A resulting action list might be Buy a more efficient water heater. Negotiate the heating oil contract in the summer months to obtain a better price.
Get rid of the water bed who would have guessed that a heated water bed used so much energy? Switch from incandescent to fluorescent light bulbs. Enjoy the TV, microwave oven, computer, etc. they use an insignificant amount of energy compared to the top 5
appliances in the house. The Pareto Principle is at work everywhere, and the real goal of any Six Sigma project is to find the top two or three factors that make up the true Pareto chart for a given problem, and then control those factors to achieve breakthrough performance. Its that simple.
This is a standard format for Pareto charts the count is noted on the left y-axis: this person was late eight times due to his alarm clock not sounding. The blue line and right-hand axis show the cumulative percentage for all reasons noted on the Pareto chart: in the case, the top two causes (alarm clock not sounding and sleeping through the alarm) account for 73% of the reasons for being late over the last year.
When data does not already exist, a carefully planned data collection effort will be needed. Pareto categories (that will go along the bottom axis) should be anticipated ahead of time if possible. 4. Summarize and sort the data in descending order Here is a data set, ready to be plotted -
5. Create the chart Use this Excel file if you need a starting point, or check out www.paretochart.org.
Now we are finally getting down to some actionable items, like buying an alarm clock with a backup battery in case of a power loss.
Wrap Up
There are many situations where a process has been neglected over time (i.e. not supported), and the team running the process on a daily basis truly knows the reasonsbehind the poor performance. Weve seen this with scrap reduction projects basic tooling refurbishments and preventive maintenance have reduced scrap rates by more than 80%. The Pareto chart is an excellent tool for documenting the current state and setting priorities in cases like this.
Flowchart
Developing a process flowchart early in the DMAIC methodologyhas several benefits 1. Not all team members are familiar with the entire process at the start of the project. Developing a process flowchart in a group session gives all team members a full appreciation for the inputs, outputs, controls, and value-added operations. 2. A good flowchart helps structure the Analyze phase, as team members consider possible sources of variation. Its easy to overlook major causes of variation, and a complete process flowchart will help minimize this risk. 3. During the Control phase the team must decide on process controls and mistake proofing measures for long term control. Having a flowchart makes this process easier, especially as the team tries to work as far upstream as possible when implementing process controls. The following symbols are typically used in process flowcharts
After the initial flowchart is completed, we recommend adding the inputs (xs) and outputs (ys) of each process step. Here is a sample flowchart with the xs and ys noted below and above eac h process step:
Powerpoint file There are a number of software tools available for creating process flowcharts, but Microsofts Powerpoint package is available to most professionals and contains the symbols necessary to create a process flowchart. Here are a few tips for planning a productive flowcharting session
Identify a facilitator who is not an expert on the process being flowcharted. Define ahead of time where the process starts and ends. Start by listing out all of the process steps in a simple list. Then take a step back and see if anything is missing, prior to creating the flowchart
Gage R&R
Gage R&R (Gage Repeatability and Reproducibility) is the amount of measurement variation introduced by a measurement system, which consists of the measuring instrument itself and the individuals using the instrument. A Gage R&R study is a critical step in manufacturing Six Sigma projects, and it quantifies three things1: 1. 2. 3. Repeatability variation from the measurement instrument Reproducibility variation from the individuals using the instrument Overall Gage R&R, which is the combined effect of (1) and (2)
The overall Gage R&R is normally expressed as a percentage of the tolerance for the CTQbeing studied, and a value of 20% Gage R&R or less is considered acceptable in most cases. Example: for a 4.20mm to 4.22mm specification (0.02 total tolerance) on a shaft diameter, an acceptable Gage R&R value would be 20 percent of 0.02mm (0.004mm) or less.
Plan for randomization its important to randomize the sequence of measurements (who measures which part, and when), to block the effects of uncontrollable factors that might otherwise be attributed to specific operators. Each operator should measure each part three times, so with ten parts a total of 90 measurements will be taken.
A word of advice to quality leaders: always ask to see a GR&R study before accepting the results of a process capability study. For an in-depth review of Gage R & R and overall measurement systems analysis (MSA), purchase a copy of the Automotive Industry Action Groups MSA guide this is an outstanding publication.
Process Sigma
Process sigma (also referred to as sigma level) is a measure of process capability: the higher the process sigma, the more capable the process is. A Six Sigma process has a short-termprocess sigma of 6, and a long-term process sigma of 4.5 (seewhy not 4.5 sigma?). The theoretical defect rate for a Six Sigma process is 3.4 defects per million (DPM). Simply put, the process sigma indicates how many standard deviations (Sigmas) can fit inside the gap between the process average and the nearest specification limit:
Note that the above example shows a histogram for a particular CTQ, so the process sigma of 4.5 applies to the specific CTQ being studied. If an overall long-term defect rate is available for all defects, it is possible to state the process sigma for the entire process (all CTQs and their associated defects) by locating the defect rate on the Sigma Conversion Chart and finding the corresponding sigma-level. Typically, however, a Six Sigma project will be sufficiently narrowed to focus on one or two CTQs , which will be evaluated separately for their process sigma levels. In cases where both lower and upper specification limits exist and the process is not capable on either side of the distribution, the process sigma can be calculated by adding the theoretical DPM levels on each side of the distribution (using the Sigma Conversion Chart) and then finding the corresponding process sigma for the combined DPM level. A more common measure of process capability is Cpk, which is equal to the process sigma divided by 3. So a Six Sigma process has a Cpk of 2.0.
What are the defect levels associated with various process sigma levels? A Sigma Conversion Chart provides the theoretical defect rates associated with various sigma-levels. The conversion chart assumes that the underlying data is continuous and normally distributed, as in the histogram above.
Fishbone Diagram
Example and Template
Fishbone diagrams, also known as cause-and-effect diagrams, are about organizing possiblecauses behind a given problem. Of course, all possibilities will need to be proved or disproved during theAnalyze phase of the project. The following Fishbone Diagram example looks at possible causes of employee turnover, based on an excellent article from Sigma Assessment Systems, Inc. Constructing a Fishbone Diagram Visit fishbonediagram.org for a thorough review of fishbone diagrams and Excel/PPT templates.
5-Why
5-Why is a simple approach for exploring root causes and instilling a Fix the root cause, not the symptom, culture at all lev els of a company. Invented by Japanese Industrialist Sakichi Toyoda, the idea is to keep asking Why? until the root cause is arrived at. The number five is a general guideline for the number of Whys required to re ach the root cause level, but asking Why? five times versus three, four, or six times is not a rigid requirement. What matters is that we fix recurring problems by addressing true causes and not symptoms - this is true progress.
5-Why Example
(Here is the 5-Why Powerpoint file used for these graphics)
Note the improvement in corrective action effectiveness as each deeper Why is addressed above:
Responding to the first Why in the 5-Why process is almost counterproductive: we are retraining the stock pickers in our warehouse, because we assume that they pulled the wrong item from our inventory. In reality, the stock pickers performed their jobs perfectly, and the real cause was mislabeled parts coming from the supplier.
Addressing the third Why (having the supplier check their stock for other mislabled products) is much more effective that addressing the first Why, but this action will have no lasting effect beyond fixing the current inventory situation. Addressing the fifth why is powerful, because it focuses on the true cause: mistakes being made in the label application process.
World class companies routinely address systemic causes like the 5th Why? above, eliminatin g reactionary problem solving and shifting resources to prevention activities over the long run.
This format is also very useful for explaining the causes behind the top bars on a Pareto chart, as in the case above where a team has collected data and built a Pareto chart on top-level reasons for downtime. Instead of simply showing a Pareto chart with no further insight into each Pareto bar, the team selects the top two items (material shortages and downtime on machine ABC) and uses the 5-Why format to explore the root causes of each.
Limitations
5-Why is useful for straightforward problems with systemic causes like the case noted above, where poor preventive maintenance is the systemic cause for unplanned equipment downtime. In cases when the root cause is not readily apparent, 5Why by itself will not solve the problem. For example, if a toy manufacturer needs to improve color consistency in a product, they will need to understand which factors influence color the most (otherwise they might not need a Six Sigma project to begin with). In cases like this, structured analysis methods like multi-vari, correlation analysis, and DOE may be necessary to actually learn the physical relationships between the input variables (process settings, raw materials, etc.) and output variables (in this case, color). If your team is attacking a number of product variation challenges, then read Keki Bhotes World Class Quality for a highly effective approach
Hypothesis Testing
Hypothesis testing is used in the Six Sigma Analyze Phase for screening potential causes. A hypothesis test calculates the probability, p, that an observed difference between two or more data samples can be explained by random chance alone, as opposed to any fundamental difference between the underlying populations that the samples came from. So hypothesis testing answers the question, what is the probability that these data samples actually came from the same underlying population? If this probability, known as the p-value, is small (typically below 0.05), then we conclude that the two samples likely came from different underlying populations. For example, a p-value of 0.02 indicates that there is only a 2% chance that the data samples came from the same underlying population. Here are a few situations where hypothesis testing assists us in the problem solving process Evaluating a proposed process improvement to see if its effect is statistically significant, or if the same improvem ent could have occurred by random chance. Evaluating several process factors (process inputs, or xs) in a designed experiment to understand which factors are signif icant to a given output, and which are not. Understanding the likelihood that a data sample comes from a population that follows a given probability distribution (i.e. normal, exponential, uniform, etc.). An individual untrained in basic statistical knowledge might naturally question the need for a hypothesis test: Why cant we simply compare the average values of a given CTQ, before and after a process change, to determine if the change we made actually made a difference? The answer is that the supposed improvement we obs erve might have nothing to do with the change we made to the process, and might have everything to do with chance variation. In other words, the two data sets might actually have come from the same underlying population. Statistical Significance Vs. Practical Significance There are many situations where a process change has a statistically significant effect on a CTQ, but an insignificant effect in real world terms. For example, an individual working to improve his or her vehicles fuel economy might run a hypothesis test comparing fuel economy at driving speeds of 60 mph and 70 mph on the highway. The result might show that driving at the lower speed has a statistically significant effect on the CTQ, which in this case is miles-per-gallon fuel economy. However, the actual improvement in fuel economy might only be 0.5 miles per gallon, which might be deemed not worth the extra time it will take to get to work each day
Regression Analysis
The goal of regression analysis is to determine the values of parameters for a function that cause the function to best fit a set of data observations that you provide. In linear regression, the function is a linear (straight-line) equation. For example, if we assume the value of an automobile decreases by a constant amount each year after its purchase, and for each mile it is driven, the following linear function would predict its value (the dependent variable on the left side of the equal sign) as a function of the two independent variables which are age and miles:
Variables value,age,miles; Parameters price,depage,depmiles; Function value = price + depage*age + depmiles*miles; Data; {data values go here}
Once the values of the parameters are determined by NLREG, you can use the formula to predict the value of a car based on its age and miles driven. For example, if NLREG computed a value of 16000 for price, -1000 for depage, and -0.15 for depmiles, then the function
NLREG is a very powerful regression analysis program. Using it you can perform multivariate, linear, polynomial, exponential, logistic, and general nonlinear regression. What this means is that you specify the form of the function to be fitted to the data, and the function may include nonlinear terms such as variables raised to powers and library functions such as log, exponenti al, sine, etc. For complex analyses, NLREG allows you to specify function models using conditional statements ( if, else), looping (for, do, while), work variables, and arrays. NLREG uses a state-of-the-art regression algorithm that works as well, or better, than any you are likely to find in any other, more expensive, commercial statistical packages. As an example of nonlinear regression, consider another depreciation problem. The value of a used airplane decreases for each year of its age. Assuming the value of a plane falls by the same amount each year, a linear function relating value to age is:
value = p0 + p1*Age
Where p0 and p1 are the parameters whose values are to be determined. However, it is a well-known fact that planes (and automobiles) lose more value the first year than the second, and more the second than the third, etc. This means that a linear (straight-line) function cannot accurately model this situation. A better, nonlinear, function is:
value = p0 + p1*exp(-p2*Age)
Where the ''exp'' function is the value of e (2.7182818...) raised to a power. This type of function is known as "negative exponential" and is appropriate for modeling a value whose rate of decrease is proportional to the difference between the value and some base value. Here is a plot of a negative exponential function fitted to a set of data values.
Much of the convenience of NLREG comes from the fact that you can enter complicated functions using ordinary algebraic notation. Examples of functions that can be handled with NLREG include:
Y Y Y Y Y Y
= = = = = =
p0 p0 p0 p0 p0 p0
+ + + + + +
p1*X p1*X + p2*X^2 p1*X + p2*Z + p3*X*Z p1*exp(X) p1*sin(p2*X) p1*Y + p2*exp(Y) + p3*sin(Z)
In other words, the function is a general expression involving one dependent variable (on the left of the equal sign), one or more independent variables, and one or more parameters whose values are to be estimated. NLREG can handle up to 500 variables and 500 parameters. Because of its generality, NLREG can perform all of the regressions handled by ordinary linear or multivariate regression programs as well as nonlinear regression. Some other regression programs claim to perform nonlinear regression but actually do it by transforming the values of the variables such that the function is converted to linear form. They then perform a linear regression on the transformed function. This technique has a major flaw: it determines the values of the parameters that minimize the squared residuals for the transformed, linearized function rather than the original function. This is different than minimizing the squared residuals for the actual function and the estimated values of the parameters may not produce the best fit of the original function to the data. NLREG uses a true nonlinear regression technique that minimizes the squared residuals for the actual function. Also, NLREG can handle functions that cannot be transformed to a linear form. < --> Error: #include file specification missing closing quote <-/tbody>
Histograms
Frequency histograms are the ultimate tool for visualizing process capability. The height of each bar on a histogram shows how often a given range of values occurs in our data. Histograms can be made manually, but there are a number of software tools available to get the job done easily. Consider each bar on a histogram a bucket that is assigned to a given range of values, with the height of each bar representing number of data points that fall into each bucket. The following histogram shows 50 tire pressure readings from an outgoing product audit at a motorcycle factory:
Note that the x-axis is divided into buckets or cells, and the y-axis shows how many data points fall inside each bucket:
Finally, we can add specification limits to graphically show how capable the process is:
PFMEA (Process Failure Mode and Effects Analysis) is a structured approach that assigns quality risk levels to each step in a process (manufacturing or transactional). PFMEA is a powerful prevention tool, since it does not wait for defects to occur, but rather anticipates them and implements countermeasures ahead of time. PFMEA is normally used in the Analyze Phase of Six Sigmas DMAIC Methodology, and theAutomotive Industry Action Group (AIAG) publishes a comprehensive PFMEA workbook that is well worth the cost for any team seriously contemplating PFMEAs. Also, our free Six Sigma Excel Templates page has a link to a PFMEA document.
PFMEA risk factors Risk Description Factor Severity How severe a given defective condition would be to the customer Occurrence The best estimate of how often the defective condition will occur in the process The likelihood that the defective condition will be detected prior to reaching the Detection customer. The current or proposed control plan should be referenced when assigning detection values.
Prior to conducting a PFMEA, the facilitator should have a set of ranking criteria for Severity, Occurrence, and Detection, which should be shared with the team ahead of time. An example of Severity, Occurrence, and Detection ranking criteria would be as follows
10 9 8 7 6 5 4 3 2 1
Hazardous, without warning Hazardous, with warning Very High High Moderate Low Very Low Minor Very Minor None
Severity levels of 9 and 10 are typically reserved for hazardous situations, with a severity of 10 reserved for hazards that the customer is not warned about ahead of time. For example, an electrical shock hazard would typically be rated with a severity of 10.
2 1
So a Detection level of 10 indicates that is virtually impossible to detect a given defect once it occurs, and a Detection level of 1 indicates that the process is absolutely guaranteed to catch the defect should it occur.
PFMEA Example
While PFMEAs are most often applied to production processes, they can be applied to any process that follows a sequence of steps to achieve a desired outcome. The attachedPFMEA example shows two steps in a process that most people are familiar with refueling an automobile. Additional comments are noted in red font. PFMEA Countermeasures
Countermeasures are actions planned by the team to lower high-RPN (high risk) process steps, and are considered to be the value added outputs of PFMEAs. The best countermeasures utilize error proofing, which is an important step in the Control phase of DMAIC. Lets look at another process that takes place millions of times around the world each day: homeowners using their automatic garage door openers. A few years ago, it was possible to lower an automatic garage door onto a vehicle. Lets use PFMEA logic on the process step: Lowering the automatic garage door Process Step: Lowering the Garage Door Potential Failure Mode: Garage door lowers onto and damages vehicle Effect: Significant damage to vehicle Severity: Occurrence: Detection: door) RPN = 9 X 6 X 3 = 162 9 (hazardous with warning) 6 (approximately 5 out of every thousand times the homeowner will start to close the garage door when the 3 (high in most cases the homeowner will realize what they have done and click the remote again to stop the
vehicle is underneath)
RPNs over 100 are generally considered unacceptable, and given the severity of lowering a garage door onto a car, a solid countermeasure is needed in this case. Garage door manufacturers might get some benefit from adding warning labels to the garage door remote (the equivalent of training an operator in a manufacturing plant), which might lower the Occurrence rate a point but wouldnt significantly lower the overall risk of dama ging a vehicle. A truly effective countermeasure in this case, and the one actually implemented with most garage door systems, would be a photoelectric sensor to detect the presence of anything underneath the door. If the door is being lowered and the
sensor detects an object in the doors path, the circuit that powers the garage door will be automatically opened, causing the door to stop. Planned countermeasures are noted in the far-right columns of the PFMEA, and there is also a place for the anticipated RPN values following countermeasure implementation. In this case, the adjusted RPN values might be: Process Step: Lowering the garage door (photoelectric sensor implemented) Potential Failure Mode: Garage door lowers onto vehicle Effect: Significant damage to vehicle Severity: Occurrence: Detection: 9 (the effect of a garage door lowering onto a car doesnt change) 6 (homeowners will be just as likely to press the down but on the remote at the wrong time) 1 (certain)
RPN = 9 X 6 X 1 = 54 The team has successfully lowered the risks associated with garage doors damaging vehicles, and can know that they have made a true difference for their customers and their companys product liability costs.
Conducting a PFMEA
Just like team selection is crucial to a successful Six Sigma project, it is equally important to holding an effective PFMEA session. The following preparation steps can make all the difference conducting a worthwhile PFMEA session
Find a good facilitator who can keep the process moving. Teams tend to get bogged down deciding on Severity, Occurrence, and Detection values this can make the PFMEA process painfully long and cause the participants to avoid future sessions. Our rule of thumb is to go with the higher Severity, Occurrence, and Detection values when the team is in doubt its better to assume the risk is higher and therefore address it with countermeasures.
Bring design engineering and process engineering knowledge to the PFMEA session design engineering to help with Severity levels, and process engineering to help with occurrence and detection levels. Pass out hard-copies of the Severity, Occurrence, and Detection criteria (as noted above) at the start of the meeting and review these criteria with the team. Make sure a Process Flow Diagram is done prior to the PFMEA.
For more information on pitfalls to avoid with PFMEA, read the top ten reasons why PFMEAs fail. PFMEA spreadsheets are typically sorted in descending order of Risk Priority Number (RPN) at the end of the session, and time is spent brainstorming and planning countermeasures for the high RPN values. Many businesses have a maximum allowable RPN value for any process, such as 100 or 200, and process development teams (or Six Sigma Teams for existing processes) are charged with implementing effective countermeasures on all high-risk process steps. Most PFMEA spreadsheets include additional columns that show revised risk levels based on planned countermeasures. Lastly, the PFMEA should be maintained as a living document and kept up to date as more knowledge is gained around true Severity-Occurrence-Detection values and countermeasure effectiveness.
DOE
Designed of Experiments (DOE) is a structured approach for varying process and/or product factors (xs) and quantifying their effects on process outputs (ys), so that those outputs can be controlled to optimal levels. For example, a DC motor manufacturer might wish to understand the effects of two process variables, wire tension and trickle resin volume, on motor life. In this case, a simpletwo factor (wire tension and trickle resin volume), two level (low and high values established for each of the two factors) experiment would be a good starting point. Randomizing the order of trials in an experiment can help prevent false conclusions when other significant variables, not known to the experimenter, affect the results. There are a number of statistical tools available for planning and analyzing designed experiments.
ANOVA
Analysis of Variance (ANOVA) determines whether or not statistically significant differences exist between multiple sample groups, by comparing the variation or noise levels within sample groups to the average differences between sample groups.