The points of an X-chart (not an individuals chart) are all averages, right ? What should one do when 1 individual measurement falls outside the specs and/or control limits but the average (X-value) is still OK ? Continue with the process that is still (according the control chart) in control or investigate and maybe even alter the process ? Will I be kidding myself if I ignore such an individual point ?
This individual point should generate an outlier on your Range chart and should be investigated and addressed accordingly.
I am sorry, but I do not agree with your comment. Just because a point is out of control on the x-chart does not necessarily mean that the same point plotted on the R chart will also be out of control. The 2 charts are complementary and should work hand in hand. It is also possible to have a point out of control on the R-chart and be in control on the x chart. These different combinations tell different stories about the process and that needs to be kept in mind when troubleshooting.
My 2 cents.
Yes. That’s right. You will see an outlier in the Range chart. What you would need to see is if there is a possibility of measurement error OR a sudden process error that has impacted this. Keep a watch but for the sake of control chart remove the outlier and recalculate the X bar and R bar and plot.
BUT WATCH out for another outlier and you have to make sure the process is stopped and the OUT OF CONTROL analysed immediately.
I still don’t agree. I will explain my position with a quote from Grant and Leavenworth’s book section 4.4.3: “If the shifts in universe average occur between subgroups and never happen to fall within a subgroup, the R chart will continue to show control regardless of the variations shown on the x chart.” The shift in average may or may not be in control but from the above statement, the R chart will be in control as long as the shift is between subgroups and not within. If you think about it, that statement makes good sense.
Scott, I agree with what you are saying, but I think you are not addressing what the original poster asked. I believe the situation is the point on the X-chart (the average of the individuals in the sample) is in control, but if he looks at the actual individual observations he sees one individual that is outside. If this is true then the difference should show up in the range chart. So I too recommend using both an Xbar and R, looking at both charts.
Individual points should fall outside the control limits of a x-bar chart. The relationship between the standard deviation of the process (sigma) and the estimated standard deviation of the mean of the sub-groups (s) is as follows:
Sigma Process = s x-bar / square root of the sample size
This is from the central limit theorem. This is not only normal but expected. If the subgroup size is 4 the Sigma process will be twice as large as the s x-bar. The control limits were based on s x-bar (R bar * A2 or R bar /square root (n) / D2)
As far as out of spec, control chart has nothing to do with spec limits. SPC is used to determine when there is a special cause of variation. The Upper and Lower Spec limits are not used when calculating control limits. If the feature was out of spec and there was not a violation of the control limits, then the process is probably not capable. Continue the control charting and perform a capability study, you will probably discover that the process is stable but not capable.
Good question, but first you need to use the correct terminology. An x-chart is for individuals data, an (x-bar)-chart or m-charts is for subgrouped data.
Now to your question. The process is “in-control” with respect to the mean by defintion because the mean is inside the control limits. You should also look at the R-chart to see it it is “in-control”. If both are OK, then the process can be considered “in-control”.
John J. Flaig, Ph.D.
Applied Technology (e-AT-USA.com)
PS- The comments that the R-chart will automatically be “out-of-control” are incorrect.
One of the primary uses of the control chart is to serve as the proverbial fork in the road for identification of sources of variation. The control chart will provide, based on the assumption of rational subgrouping, a direction to look at x’s and noise variables that are changing either between the subgroups or within the subgroups. The average section of the control charts provides you with an answer to a key question. Is the variation that the average of the subgroups displays more than that predicted from the variation estimated from within subgroups? So, based off of that answer the follow up analysis is to ask:
what x’s/noise/factors are changing and/or not changing within subgroup.
what x’s/noise/factors are changing and/or not changing between subgroups.
In your scenario, if there were individual data points that were so extreme (say within a given subgroup) it would show up as an ‘out of control’ point in the Range section of the control chart and depending how large was the shift-it could also influence the associated average reading. So, if the point shows up as out of control the question is whether you are seeing a definitive signal (or an action limit) or is it noise. This goes to the comments about gauging the potential impact of alpha risk. Our assumption is that the point should warrant attention since the limits have been derived to be +/- 3 sigma to either side of the mean. Will there be a potential to get bit by alpha? Yes, of course, but if we’ve conducted the analysis correctly and used rational subgrouping the risk has been taken to an acceptable level. The power of the control charts is that it is generating the control limits based off of a within-subgroup methodology of obtaining the estimate of standard deviation. This has been shown to be remarkably robust method to obtain an estimate.
I disagree with a post that recommended simply removinig the point from the analysis. This should only be done when there is a defintive tie between the event observed and the source of the excessive variation. With that event identified, corrected, and elimintated now you can remove it from the control chart-as an event that potentially influences the control limits-you should still have it graphically displayed with appropropriate commentary around actions taken and issue that caused it.
I tend to leave the long threads around stats alone because there are a lot of people who give goods stats advice. I may be wrong but it looks like you may not be getting the response you asked for.
Because a point falls inside the control limits does not mean your process is in control. There are tests for trends, runs, shifts and cycles which an increasing number of people choose to ignore. It is difficult to really understand a process if you are not listening to everything it is saying.
Because a point falls outside a control limit does not mean a process is out of control. By definition if you run a process with 3 sigma limits there will be a very few points which will fall outside the limits. Since you cannot tell which ones are which, you need investigate any point outside the limits.
I agree with the idea of notating it. What did the investigation turn up. I don’t agree with eliminating it. Just because something has an an assignable cause does not mean it isn’t part of the long term variation. It is always nice to have it there for reference particularly if it is a recurring thing – eliminate it and you will be able to analize this how?
R charts can run independently from the x or x bar chart. There are variables which affect the mean and variables that affect the variance. They are typically different things. That will make the charts operate independently.
One more note of caution. Spec limits and Control Limits are two different things. Being in control doesn’t mean you are building good stuff. When the spec limits and the control limits are in close proximity and you get a point on your x/x bar chart that is close to a control limit and therfore close to a spec limit you need to calculate the estimate for the std dev and plot what the distribution looks like around the point on the x/x bar chart. You can have everything in control and still be building defects.
Sorry about the length.
Annonymous, Scott and Jaime:
I think that you three are out of focus. The original poster says nothing about a point out of control, neither in the Xbar notr in the R chart. It’s about one individual out of spec. You can have a lot of individuals out of spec and still have a fully stable process that shows no out-of-control signal in any chart. Such a process would be a “stable but non-capable” one.
The original poster is not saying that an individual is beyond the control limits, but beyond the specification limits (even when the average is within control limits)
Even when Mike said something alike, I think you still didn’t get the answer to your question: If the Xbar is in-control and one individual is beyond specs, should you take action?
I am afraid that the answer is, it deppends. But ussually NO.
But the main thing is that you are asking that because you don’t understand what a control chart is for.
The control chart is to asses whether the process is behaving in a stable way, i.e. as it allways behaves. More technically, it is to detect when the process is affected by variation due to a special cause. But, clearly, a control chart IS NOT INTENDED TO ASSESS PRODUCT CONFORMITY TO SPECIFICATIONS.
Now, if you have proved before that “the way the process always behave” is such that the product meet the specification (for example, with a good result in a process capability study), and you find that the process keeps stable (by means of a well deffined set of control charts), then the absence of out-of-control signals is an indication that the precess is behaving as it allways does, i.e meeting the specifications.
How is it possible, then, that you have a non conforming part and no out-of-control signal? This situation is not possible with the two conditions previously mentioned (a capable process and a good charting strategy), because any non conforming part in the sample will lead to an out-of-control signal (which may not be in the x-bar chart but in the R chart). Then, there are two possibilities for the case of your question:
a) The charting strategy is not a good one (for example, you are using an Xbar chart alone, without an R chart). In this case, you should take an action. Improve the charting strategy! (of course, you should take an action on the process too).
b) The process is not capable (it doesn’t have a good ability to meet the specifications), even when it is stable (it behaves allways the same). If you are aware of this situation from the beginning, a non conformng part in the sample should not lead to a specific action to investigate the root cause for this specific case and take a corrective action, because it is the way the process allways behaves. It is normal that the process delivers non conforming parts.
We have a few processes like this. The process is not ccapable, it delivers a proportion of non conforming parts (let’s say 3%), we have a 100% cehck downstream to sort out those parts. We chart the process, but we do not react when a part in the sample is out of spec, only the chart(s) show an out-of-control signal. Why do we chart such a proces at all? For two reasons: a) We don’t want the process to drift and turn this 3% scrap in 30% scrap. b) We want to asses the evolution and effectiveness of the improvement activities we are introducing. The improvement plan is not the result of neither a non conforming part nor an out-of-control signal, but it was a dessicion due the poor capability of the process. In fact, if they are effective, those improvements also display as out-of-control siignals, because the process behaves better (not as allways) and this is unstability by deffinition.
Did this helped?
I agree with most of what you said.
Picture this. You have a control chart witha control limit sitting directly on top of the mean. You plot a point and it hits on the control limit. If you are assuming that the points represent a distribution that are somewhat normal – you are in control but about 50% of what you just built was out of spec (that is what my reference was to the two being in close proximity). When a person blindly follows the “its inside the control limits” idea and loses track of the relationship between the mean and the variation around the mean you can walk into problems. The logical answer is that if your capability is high you will be so far inside the Spec Limits it probably isn’t an issue. Not that many processes are that good.
The part I didn’t agree with was the “it depends” for checking on points. Regardless of any technical reason the people doing the charting will take it seriously as long as you react. When people don’t react or eliminate points they frequently will view it as a waste of time or another management game. You won’t get jerked around looking at stuff you shouldn’t and when you do it is really just an investment in your SPC program.
Just my opinion. I could be wrong.
My piece of suggestion. Ask yourself what is the objective of control charting? well, I would say donot allow a part out of spec going to the customer. Segregate the part. Look at the lot, if you are sampling then may be you might have some more parts going out of spec. Also, look why this is happening. May be your control limits are too close to your specification limits. Thats where a minimum of 1.33 cpk value comes in. Calculate the process capability (and dont forget to include this particular subgroup). Try reducing variation, this is what might be causing the problem.
Hope this helped.
Hi, Mike. I am not really back. I am too busy both in work and personal life, but time to time I like to check what is goig on.
About “the part you don’t agree”, read again my message because I feel that we DO agree, only that you didn’t understand what I ment (probably because I was not clear enough).
I do not mean at all that there is one single reason to not to react to an out-of-control signal on the chart. You MUST react to EVERY out-of-control signal. May be it was a false alarm, but this must be a conclusion of an investigation, not an assumption.
Read again what I said (here you have an bastract):
If the Xbar is in-control and one individual is beyond specs, should you take action? It deppends. There are two possibilities for the case of your question:
a) The charting strategy is not a good one.
b) The process is not capable even when it is stable. If you are aware of this situation from the beginning, a non conformng part in the sample should not lead to a specific action to investigate the root cause for this specific case and take a corrective action, because it is the way the process allways behaves. It is normal that the process delivers non conforming parts.
Of course, in the b) case you should take action (probably a long term improvement plan) to improve the process, but not as a reaction to one part out of spec but because the process is not capable. You still have to react to an out-of-control signal, which is not the case of the original question.
Do you agree now?
By the way, in part from the questions but more from the answers I ussually see in iSixSigma, I feel that many of we who teach SPC must be doing a pretty bad job. It seems to mee that there is a great missunderstanding on the concepts. May be we ar putting too much focuse on the useage of the tool and not in the fundamental concepts. At last, we can see from these messages that a wrong concept leads to a wrong usage of the tool.
Just here bellow, after all this thread, you can see a new message form Hemanth that begins:
“Ask yourself what is the objective of control charting? well, I would say donot allow a part out of spec going to the customer. Segregate the part”
I think that few things can be as far from the objective of control charting as that. And the problem is that this concept seems to be a generalized one, not an isolated case.
I would appreciate your feedback on this.
Grabriel, The original poster does not mention using a range chart, but asks what he should do if he has either an individual point out of the spec limit and/or an individual point beyond the control limit. My suggestion was simply to examine a range chart, for it could indeed show that the range is OOC (maybe not as you point out). If this is not true or investigating the range is not a proper thing to do, please help me understand why.
Sorry Jaime, you are right.
I noticed that he was asking abuot one individual out of specs but missed that he was asking about an individual beyond control limits too. In the Xbar chart, if you compare the individuals values against the control limits for the average (which is absolutely a wrong and nonsense thing to do) you WILL ussually find individuals beyond the control limits. If not, the limits are not correctly set (or the process has reduced its variation since when the limits were set).
I also adress to the use of an R (or an S chart) chart everytime you use an Xbar chart (or a Medians chart). When charting variables, one chart for the location of the distribution and one chart for the spread should allways be used together. With that said, the R chart should ALLWAYS be examined for OOC, regardless of what the Xbar chart may show or what may happen to an individual of the sample.
What I want to be clear is that you can have a sample where one individual is out-of-specification but the Xbar chart shows the average in-control, and this doesn’t mean that the R chart will be out-of-control. Both charts can be in-control and still one or more (or every) individuals in the sample can be out-of specification. It is as simple as the fact that charting is not related to specifcations.
Hope it is clear now
I find this thread very disturbing. The questions the poster is asking are very basic to SPC. There are actually correct answers to these questions. The correct answers are not subject to anyone’s opinion. I would ask the poster to please read Grant & Leavenworth, Douglas Montgomery or Wheeler before using these tools. However, you should realize that no amount of training and reading are a substitute for experience and, until such experience has been gained, working under the guidance of a seasoned professional.
I attended my first SPC/DOE seminar in 1976 and have been using the tools ever since. I get the sense following this thread that a few of you have had some cursory training and now consider yourselves experts. I apologize for being so harsh but as has so often been said, “A little knowledge is a dangerous thing.”
If this thread is an example of the quality of statistical training that black belts are receiving, then our businesses and Six Sigma are in deep trouble. We will be guilty of making wrong, costly decisions and management will be justified in pronouncing six sigma another useless, expensive fad.
To set the record straight:
The points of an X-chart (not an individuals chart) are all averages, right ?
The proper terminology is an Xbar chart. Xbar means average or mean. So if we are talking about an Xbar chart then the answer is yes, the plotted points are averages. If you are talking about an X chart, then what you should be plotting are individual measurements.
What should one do when 1 individual measurement falls outside the specs and/or control limits but the average (X-value) is still OK ?
First, as has been pointed out, do not confuse control limits with spec limits. If the average is within the control limits and if none of the other rules for control have been violated (as Mike Carnell pointed out) then there is a high probability that the process is control or stable and no action should be taken on that account. If one of your measurements is out of specification, then you should invoke your control plan’s reaction measures. This usually involves placing the product on hold and determining its fitness for use.
Continue with the process that is still (according the control chart) in control or investigate and maybe even alter the process?
If the process is in control and capable, then continue running it. If it is out of control and/or incapable, take the appropriate measures to correct it.
Will I be kidding myself if I ignore such an individual point ?
Sr. Q.A. Engineer
The question possed was regarding a single mean point that plots within the control limits but some of the individual values are outside the control limits. There is NO issue of trends or any of the other “Runs Tests” to be applied. The chart is “in-control” by defintion (Shewhart 1931), Western Electric (1956), Nelson (1983). Your technically correct just becasue a point plots outside the control limits does not imply that the process is ACTUALLY out-of-control as there is a small probability that it is due to chance causes. However, the correct statement is that if a point falls outside the control limits, then the process is by defintion “out of statistical control”.
In this case Jo is comparing the individual value to the control limit that applies to mean values and you are telling him he should investigate these. This is incorrect advice. As I said in my first post the process is in-control if the mean and range charts are in-control and no action is required.
John J. Flaig, Ph.D.
Appled Tecnology (e-AT-USA.com)
You’d get a A in my class.
John J. Flaig, Ph.D.
Applied Technology (e-AT-USA.com)
“If the process is in control and capable, then continue running it. If it is out of control and/or incapable, take the appropriate measures to correct it.”
So what tools do you give an operator to determine this in a production setting? The orginal poster appears to be using only an Xbar chart (or lets assume this). Do you recommend him do anything else to detect the problems you discuss.
I understood exactly what the original post stated. The comment you made was that if a point was inside the control limits it was in control and that is an incorrect statement. That is why we have test for trends, runs, shifts and cycles.
As far as my advice being incorrect. That is your opinion. Part of the problem with Control charting is you need to be able to read them using some common sense which means that there will be cases where they will both be incontrol and your process is in trouble which is the example Gabriel and I discussed.
There is more to running a good process than memorizing rules and speaking in absolutes.
I think you are correct. There is a good chance I was reading and not comprehending what you wrote. Thanks for the clarification.
I agree with you about the teaching of SPC. To many people doing it that haven’t actually had to runa process and deliver good parts. I was fortunate enough to learn from a guy named John Lupienski who has has a great combination of academic understanding and real life experience. John was probably as influential as anyone at Motorola on how the SS program went. He has spent the last 5 years moving his factory from 5.8 sigma to 5.2. That is the type of hard work, focus and dedication it takes to make this work.
The point I wanted to make was that in the case of the original post, where a single point fell outside the Control limit. The operator will see that and may or may not understand why the plotted point isn’t considered out of control. If they flag it and you blow it off there easily could be a person who just stopped believing in the chart. Once they are turned off it is difficult to turn them back on. For a few minutes of time looking you can avoid the risk.
There is a lot more to doing a successful SPC program than these people who run around espousing statistical dogma. People are what make it work or fail. I would mich rater react to a point that is technically in control than lose the effectiveness of my program because I wanted to stand on the fact that I was statistically correct.
I don’t know if I got you what you wanted in terms of feedback.
Glad to see you are busy.
Why would you not look at the trend of data points ?
The situation described in the original question was not at all a complete description of the process.
“but the average (X-value) is still OK ”
does not draw a good picture in my mind as I didn’t understand the ”OK” part of it - did you?
What if Jo had a bunch of points hugging a control limit but within the limit and one point was out of control? Would you not want to know about that situation in detail then? I would, if I have to render advice that Jo could use to improve something.
I guess your point of view is “given the facts we have” one doesn’t need to worry. I worry, that we do not have all the facts, and that calls for caution, doesn’t it?
I would much rather ask someone to investigate (like Mike did) and convince him/herself that there are not pitfalls in the decision making process.
What’s wrong with that?
Spoken like a true practitioner.
As always, I appreciate your comments. Thank you much.
The issue that is demonstrated here runs deeper. I will take my case. It is safe to assume that if I am critical about myself, the end result is not a very long thread of discussions :-)
I had training in statistics (college level)… and I feel, I have moments of failure when sometimes I delve more into statistical nitty gritty of problem solving than the practicalities and ask myself - am I here to glorify statistics or use statistical thinking to realize business goals.
I call that a failure as, our customer (the reason we have a business) doesn’t care if the data is normal, non-normal, bi-modal, skewed - whatever. We do have different tools, so it helps to know some details so that we can choose judiciously. Yet, our tool, Minitab’s next version perhaps will not make statements like “John’s spraying thickness is greater than Jane’s” when presented with the data. Ultimately we are required to make correct practical judgement based on data and bring in positive change. That is the Goal! And that is why we need to exercise caution about pitfalls.
Statistics (and a good dose of common sense) could be the vehicle for the journey. But who drives the vehicle? I would like to believe, it is the leader in us.
I don’t know if you are familiar with the models around adult learning. Somebody actually figured out it mattered because Blooms Taxonomy is part of the ASQ BOK for the certified BB test. Basically it is about 6 levels of learning. The first 2 are Knowledge and Comprehension and they come in the classroom. The people stuck here are great at regurgitating rules, assumptions, formulas, etc. but can’t get it translated to application.
The next is Application and now you have what the BB training is built around. Working projects. As you move up through the next 3 levels it requires an intuitive level of understanding that only comes through experience with use.
You are correct that the entire SS thing has been sold on the basis of results – business results. As a friend of mine says “we are out here to build SS companies not SS statistics.”
We have seen the influx of people who make careers of pontification of concepts that are of little importance. They damaged TQM with their lack of results. Now we see them creeping into SS. Trying to turn it into a training/stats program because that is much safer than delivering results.
Good luck Sam.
I would give the operator an Xbar & R chart. The Xbar chart will tell you if you have non-random variation present in the process between subgroups (time to time). The R chart will tell you if you have non-random variation present in the process within the subgroups (piece to piece).
An Xbar and R chart is always a good starting point for an improvement project. Not only will it allow you to identify and eliminate sources of non-random variation but even if the process is stable, it establishes a long term baseline from which the effects of any improvement efforts can be evaluated. I personally think this is more effective than running a 30 piece sample, making a change, running another 30 piece sample and then performing a T test.
That being said, first I would find out more about the process; it’s history, why it’s being charted in the first place, what the objective of the charting is. Is it a customer requirement? Is it the result of a customer complaint? Is the process producing unacceptably high levels of scrap or downtime? Is this primarily a throughput problem or a critical-to-cost issue? If the original poster would give us more information about this process, perhaps we can be of more specific assistance to him.
Sr. Q.A. Eng.
We heat seal bags filled with products. Our customers will put these bags in autoclaves and steam sterilize them.
We test this seal on a tensile testing machine. For this we cut out 3 samples per seal. We started sampling (3 bags per week) because we want to intercept a possible decline in machine performance. Such a decline will manifest itself in lower seal strenght. However we know that the bag material shows variation which leads to variation in seal strenght. Some people here want to take corrective actions when 1 measurement falls just below spec. In my opinion (based on technical knowledge of materials and machine and partly based on gut feeling) I say things are still ok.
Maybe I can prove my point with some statistical tools. I know however that I’m not a specialist in this area so I’m being very carefull. After all: garbage in = garbage out.
Thanks for all the comments !
Mark, Thankyou for the additional input. It was extremely helpful…particulary the questions about the process. As you stated, if the original poster has not found the answer to his question he needs to provide more information.
Also my apologies I just realized you are Marc with a “c” not Mark with a “k”.
Thanks for the feedback.
One little point: The original poster wrote about “one individual of the sample beyond specs and/or control limits”. You said “a single point fell outside the control limits”. “A single point” can be taken as one of the averages (if we are speaking of an Xbar chart).
I don’t know what you ment, but if you ment “one of the individuals” I do not think that, if the operator flags this point (when no OOC signal is pressent) you should react as if it was an OOC just to make him happy (I know you didn’t say that). You’d be making him think that he is doing the things right, when he is not. I wouldn’t just blow the flag off either. I would call the operator and together we would delete the flag (not the point), explaining him what we are doing and why, and I wouild thank him for his intention of alerting about what he thought it was an indication of a potential problem and encourage him to continue doing so, even if he is not sure.
That is ralated with what I feel about teaching and training. There is a general tendency to say how to do something, but not why. Even in our procedures and working instructions, we say A shall do this, B shall do that, if that happens, then C shal do this, this, and that. It looks like a computer program. We do not need persons to do such things and if we are using persons anyway it is either because they are cheaper than machines for some tasks or because we want persons to do something more, like thinking.
I do not mean that any operator using control chats must be a statistician, but we should theach them the fundamental concepts behind SPC (concepts, not formulas or theorems) and we should eplain the tool based on these concepts, and everytime we say what to do and how it should be together with a why, closing a llop to the concepts. Then, during the hands-on training and implementation, we should dedicate time to give feedbak. If the operator does somthing wrong we should explain him what was wrong and why and what is wright and why, and not only what is right (as in the previous example). On the other hand, if the operator flags a point and then we find the cause we should go back to the operator and say, for example “You know, there was a problem with the raw material, the hardness was to low. Thats, why the roughness went OOC. Your finding was very important because we could have been running the process many hours in this condition, which could have lead to a lot of scrap. Now we changed the batch of material and asked the supplier to investigate and correct the problem. In the meanwhile, we introduced a hardness control in the incoming inspection”. Then he will feel that what he is doing is important, and that we care. Of course, all this takes more time, effort and money. Is it worthwhile?
We say we want proactive, participative, curious people. Do we? Then, do not give them the instructions manual of the microwave oven and say “follow it”. Telling people what to do and expecting them to do it just because we said is something we should have stopped doing long ago.
Of course, all this is only what I think. I recognize that I am relatively young and without 20 years of background (neither in teaching nor in SPC). But still, it is what I think. I know that there are a lot of people who think differently and I am ready to change my mind if someone convince me. But if the number one in the subject came and told me how is the right way, he would not convince me unless he explains me why it is the right why.
Good stuff. You have beenusing this a while. It is nice to see people who have to teach it to operators and then work with them in the factory or office. Standing infront of a classrom full of people painting a world in black and white works right up until the time you leave the room.
Now the tough part. What you thought I didn’t say is exactly what I said. If an operator flaged a point which was part of the subgroup but the average of the subgroup was in control I would still look at it. Part of the investigation would be an explanation of how the numbers work. The fact that they flag it means that there is something about the class that was not taught or they did not get. If the training didn’t take then you have a correction you need to make there as well.
They can do you more damage by talking about you not reacting than the effort it takes to spend some time taking a look. I don’t know if you have heard of “the rule of 250.” – it may be out of date. There was a guy named Joe Gerard who was in the Guiness Book of World Records for selling the most cars. He said that the average attendance at a wedding was 250 people. If you piss someone off then you should assume that they have access to 250 people and they will talk about it to that many people. If you watch how fast stuff moves around in a factory it is probably a pretty good model to follow. Not being obsequious but making sure they are getting treated like equals.
Just my opinion. On my way to Toronto so I will be off line the rest of today. Have a great weekend.
We test this seal on a tensile testing machine. For this we cut out 3 samples per seal. We started sampling (3 bags per week) because we want to intercept a possible decline in machine performance. Such a decline will manifest itself in lower seal strength. However we know that the bag material shows variation which leads to variation in seal strength. Some people here want to take corrective actions when 1 measurement falls just below spec. In my opinion (based on technical knowledge of materials and machine and partly based on gut feeling) I say things are still ok.
I have encountered this type of problem many times. First, is this a minimum strength specification? Usually there is no upper tolerance with a characteristic like this. If this is the case, can the process average be shifted up?
It is really difficult to get people to understand that adjusting a process that is statistically stable because it produces parts that are out of specification actually increases the variation of the process and the product. The funnel experiment shows this very well. You may end up producing more parts out of spec than you would have if you had left the process alone.
This type of problem usually requires an engineering approach because continually adjusting the process is not the answer. You will probably need to use some problem solving tools like stratifying the data, preparing concentration diagrams and cause & effect diagrams. This will help you to determine the root cause or causes of why the process produces parts out of tolerance. Then you can develop and implement a permanent corrective action.
Maybe I can prove my point with some statistical tools. I know however that I’m not a specialist in this area so I’m being very careful. After all: garbage in = garbage out.
Well said. It is the wisest among us who recognize the limits of our knowledge and then act to educate ourselves.
Marc RichardsonSr. Q.A. Eng.
The probability you have one single X out of specification and the average (X bar) under control depends on the number of your sample and your Cpk. Generally it is so small that you can consider it as a few and far between. So,
1. You don’t accept the individual out of psecification.
2. You continue with the process if R is also under control.
DANG Dinh CungDDC92290@NetScape.Net
The forum ‘General’ is closed to new topics and replies.