iSixSigma

Gage RR Question

Six Sigma – iSixSigma Forums Old Forums General Gage RR Question

Viewing 62 posts - 1 through 62 (of 62 total)
  • Author
    Posts
  • #40817

    Anderson
    Participant

    Hello,I have a quick question on a gage R&R study I’m doing. I new to this type of study and wanted to clear something up.Is it standard practice to use (10) of the SAME part number to use for measurements or is it acceptable to use (10) different parts (with different measurements) in the study?I’ve done a few studies using (10) different parts that represent our normal range of measurements. The numbers for the GRR using this method have been awful. (>50%). If I would use (10) of the same part I’m sure the GRR numbers would be better, but I do not feel that it would be a good indicator of how well our measuring system works.Any comments / info would be helpful.-MikeA      

    0
    #127446

    jimmie65
    Participant

    When possible, measure the same part 10 times. What you’re looking for is the measurement error: 1) Will the same person get the same measurement every time he measures the same part? 2) Will 2 different people get the same measurement on the same part? Remember, your purpose is to determine if the measurement system is capable.The method you are using is affected by your part-to-part variation. Measure this after you are sure your measurement system is capable.(The only time you don’t use the same part is when the test is destructive, but then you want to make sure the parts are as alike as possible.)

    0
    #127450

    Haugen
    Participant

    Mike,
    You want to select 10 parts(same part #) that exibit dimensions that represent the full tolerance range.  Have each operator measure each dimension 3 times(total # of measurements = #operators x #parts x # repetitions).  That way you test the operator repeatability and the part-to-part repeatability across the full range of variability that the gage is expected to be accurate within.

    0
    #127451

    jimmie65
    Participant

    Mike –
    Ignore my previous post (at least parts of it). I misunderstood your question and was typing too fast to pay attention to my own response.
    Measure different parts that represent the range of your process, but measure each part multiple times. I generally have 2 operators measure 10 different parts 3 times each.
    Jimmie

    0
    #127456

    Mr_big
    Participant

    Where do you come up with “10 parts”?  Are you just making that up or do you have a reference? 

    0
    #127458

    CT
    Participant

    Mike A,
    1) Is this an automated gage………No operator intervention? If so I would recommend using a P/T ratio first to make sure the gage is capable of measuring to your tolerance.
    2)When conducting the R&R, you will wont to gather 10 parts that are the full range of the tolerance. Try to have them all the same part number to reduce Part to Part Variation (PV). If you would like, you may use a Known gage standard that falls with in the range of tolerance. I have seen instances where R&R’s where performed using all different kinds of parts and the R&R was horrible (ie what you explained you encountered).
    3) When performing the R&R, use 2 operators and 1 master operator(Someone who is very good with the equipment) to compare to. Especially when using gages like digital calipers and micrometers, which have a lot of human error inherent to them. Measure the parts a minimum of 3 times in complete random order. The more times you measure the more accurate the R&R will be, and will also help you determine any special cause.
    4) I have in the past sorted my R&R parts by standard deviation before performing the actual R&R. (ie leak testing). FYI
    Hope this helps
    CT

    0
    #127460

    CT
    Participant

    Mr_big
    When performing an R&R the number of parts can vary but generally as a rule at least 10 is needed to perform an accurate R&R. And they must be representive of the product tolerance.
    CT

    0
    #127461

    Anderson
    Participant

    Thanks for the replies. I’ll give you a little background on what we do. We have a line of machines (10 of them) that produce a part. The operators (one per machine) measure a dimension (usually between .250 and .300, depending on the actual part number) with a dial caliper. The tolerance is +/- .004″.
    My study is to see if our current measuring system is adequate. The first time I did the study I used (10) different parts that represented the most common range of measurement (.200 to .300), (3) different appraisers each using his own dial caliper. In this case it was (2) different calipers since two appraisers ran the same machine (1 on 1st shift and 1 on 2nd). The results I got were very poor… around 65% GRR, 60% EV and 25% AV.
    I then redid the study changing only the gage being used. I had all three apprasiers use the same dial caliper. The results were better, but still not in an acceptable range. (<20% GRR).
    I redid the study a 3rd time (to the dismay of the apprasiers… haha).  I used (10) of the same part this time. I got a GRR% of around 30%. Better… but still not good enough.
    Is the ‘proper’ of seeing if our measurement system is adequate by way doing a complete GRR for every set of calipers (10 total) for each machine? I orginally did the (10) different parts along with the different gages as a ‘catch-all’ or ‘worst-case’ type of study. It seems I’ll never get the results I’m looking for unless we change the tolerance or go to more precise gages (micrometers).
     

    0
    #127464

    Mr_big
    Participant

    I can understand you backing up your “10” with some sort of reference however once again you failed to do so.
    AIAG MSA 3rd edition manual, Chapter I – Section E, under discrimination. The explanation seems quite clear to me.  If the number of categories is 5 or more then there is no issue.  You won’t know how many distinct categories you have until you do the analysis.  Ten parts maybe enough, not enough or too many and a waste of time and money.  To state that “10” is the answer is foolhardy.
     
     

    0
    #127466

    Anderson
    Participant

    I used (10) because the excel gage R&R data sheets that were supplied by AIAG had space for (3) trials of (10) parts with (3) appraisers.

    0
    #127471

    CT
    Participant

    Mike a
    My opinion would be that you re-run the experiment using 10 parts that cover the range of the process. Check them yourself. (you will be appraiser 1,2,&3) just as a check to see if you can learn from the experiment of WHY the other appraisers vary so much. Is it the Gage? Is it the way you hold the part? Or is it simply the part? With AV of 25% I would say you have other issues than just the Calipers. at + .004″ Calipers should be reliable enough, unless you have other variables like roundness of part of flatness of part measured. Make sure you take your measurments from the same location on the part. Mark a spot if you have to. One other note; any gage should be calibrated and checked with standards before an R&R is ran.
    CT

    0
    #127473

    gs
    Participant

    CT
    Can you elaborate point # 4 ? I am interested in the leak testing .
    If the leak testing equipment is automated , do you still require a R & R ?
     
     
     

    0
    #127477

    Bill Craig
    Participant

    Mike,
    One thing to consider is if you are using the system for process control. In this case, your SPC would probably involve data from a variety of part numbers. Your gage study should include multiple part numbers in that case. If you are only using the system for product control, you could use 10 pieces of the same part #. In both cases you would want to cover a good portion of the Spec range. To get real elaborate you could use operator, part number, and part (nested within part number). Call the MBB for help with this analysis!!

    0
    #127483

    Mikel
    Member

    With all due respect, the right way would be to do a proper study on each of the part numbers.

    0
    #127488

    Mike Walmsley
    Participant

    I agree with Stan.
    Ref the AIAG MSA Reference Manual.

    0
    #127493

    CT
    Participant

    GS
    Yes, you still have to do an R&R. The way I chose to approach an automated leak tester was to first attempt to sort my R&R parts by Calculating the standard deviation for the parts after I tested them 5-6 times. This helped me eliminate any special cause, (some parts just test more consistant than others) especially since the leak test is at 500psi and vacuum decay. I am simply using a P/T ratio for the study since it is automated. I have a ton of information if you are interested; email me at [email protected]
     

    0
    #127498

    Bill Craig
    Participant

    If it is not economically feasible to do a study on all part numbers, give a 3 factor design a try. (Operator, Part Number, Part) This will allow you to decompose all sources of variation. Your selection of part numbers could be a random selection from your array of products.  Good luck!

    0
    #127499

    Mikel
    Member

    Wrong
     

    0
    #127501

    Haugen
    Participant

    Any unique part that you do not conduct a gage study on results in you  flying blind regarding the components of variation – you don’t know the true data.
    A part can fail based on something as simple as a different chamfer angle.  You can’t take a random sample of different parts and make any judgement on the parts that were not run.
    I went into an engine plant that did not believe they needed to run GR&R on each feature – I ran all the parts/features, and 60% failed, some of them would not even run in the gage.

    0
    #127516

    Ron
    Member

    The purpose for the study is to quantify the effects of part to part variaition, operator to part, and operator to operator variation, also variaition ofthe gage.
    A minimum of three parts should be used and they should be the same three parts going through the study.
    Using the same ten parts is a bit of overkill in my opinion but statistically should give more accurate results.

    0
    #127524

    CT
    Participant

    Exactly Stan……………WRONG

    0
    #127531

    Art
    Participant

    Stanley – there you go again, you have proved my point.  You are an arrogant jerk who has nothing nice to say to anyone.  I bet you live alone with 3 cats, holed up in a ratty house chatting on line to your “friends” all day and night.  Don’t ask me what I have posted, I don’t need to prove my worth to a jerk like you by copying and pasting from a stats text book.  Try being nice to people.  You reap what you sow.

    0
    #127532

    Mikel
    Member

    Mr. D A (stands for dumb art in case you are wondering),
    I gave the guy the right answer and then he needed to come back with the same wrong answer again.
    You are the one calling names. I am calling you dumb here because simple correct answers don’t make me a jerk. Curt, short, and a few other adjectives would be appropriate but not the name calling you have resorted to.
    Why don’t you explain for everyone here what the right answer is and why? Don’t know? Thought so. I think the appropriate term for this behavior is big mouthed jerk.

    0
    #127534

    Art
    Participant

    Easy now stanley, you are scaring your kitties.

    0
    #127535

    Mikel
    Member

    Do you know why the answer is wrong or are you just a big mouthed jerk?

    0
    #127537

    Art
    Participant

    No Stanley, I don’t – but zip me your BOK PDF file and I’ll look it up, same as you and regergitate it. 
    As they say in Cozumel – “take time to be kind”, I am outta here, so long, this tryst is over…g’ night to the cats.

    0
    #127538

    Brad
    Participant

    Stan,
    He is looking to be a hero for Jenny who is to busy smoking cigarettes and flatulating. Hot night at the bowling alley. Probably just his kind of woman.

    0
    #127542

    Bill Craig
    Participant

    If you want to know the “true” data, that entails a bias study. 
    I’ll try to stay out of the crossfire and just say that an R&R study is a DOE, and you don’t have to limit yourself to a canned design. If you want to do 6000  DOEs on different part numbers, I hope your boss would scream. Do your intial DOE, and optimize the system so it is robust to different part numbers.

    0
    #127543

    Bill Craig
    Participant

    Mike,
    I just read your second post, and I think you might just have a general issue with using calipers. They seem to be a little cumbersome for a +/- .004 tolerance.  The only way I have been able to overcome this sort of measurement issue was by taking 3 consecutive readings and averaging them. Try doing bias and linearity with a known spectrum of part sizes. See if the bias and linearity improves when the operator takes multiple readings and uses the average value. The next step is to invest in micrometers.  Once again….good luck. It is interesting to hear the real life dilemmas!

    0
    #127558

    Mikel
    Member

    Bill,
    Wrong again. MSA is part of a basic Quality System, not some DOE you run on a bunch of part numbers all at once.

    0
    #127560

    Gooseneck
    Participant

    Stan,
    You can lead a horse to water…..
    ps
    You da Man!!!

    0
    #127569

    Bill Craig
    Participant

    Stan,
    A gage study is not a DOE????? WRONG.
    Does the expression “operator-part interaction” suggest to you that you are running a DOE when doing an R&R study? It is a 3 x 10 factorial (fully crossed) with 3 replicates. ANOVA is used to analyze the variance components. The “Part” main effect is removed and the R&R components are made up of the “operator” main-effect and the “operator-part” interaction. Your measurements should also be randomized.
    All those R&R studies you did, and you never realized you were doing an experiment that was designed ages ago.
    Mike….good topic! I hope you can filter what you need out of all of this.

    0
    #127572

    CT
    Participant

    Since I am somewhat of open minded person. Can someone please explain to me how in the world an R&R is a DOE?
    In no way have I ever heard of that in terms of the traditional design. But would be curious as to what the Factors would be? How do you place significance on Human Variation? What is the Response?
    Just curious, but willing to learn
    CT

    0
    #127574

    Iain Hastings
    Participant

    Don’t know that I can add much to the argument but for a good explanation I would reference:
    Montgomery – “Design and Analysis of Experiments” 5th Edition, Page 519, Wiley
    “Statistically designed experiments are frequently used to investigate the sources of variation that affect a system. A common industrial application is to use a designed experiment to study the components of variability in a measurement system. These studies are often called gauge capability studies or gauge repeatibility and reproducibility (R&R) studies.”
    That is just the opening paragraph, the section then continues for about 8 or so pages. Rather than spending too much time arguing I would recommend any one interested pick up the book and make their own conclusion.

    0
    #127575

    Mikel
    Member

    I did not say a MSA was not a DOE. It is a random effects DOE to be exact.
    I said your approach was wrong. MSA is part of the Quality System and is done part by part, measurement by measurement as the parts and measurements are brought into the system. You suggested approach says we have all of these parts and all of these measurements already out there not already understood and corrected.
    MSA really should not be a tool of Six Sigma, it only is because people are not properly qualifying the Systems as they introduce them.

    0
    #127576

    GomezAdams
    Participant

    3 factor,1 level,10 replicate?????
    3 factor,2 level , 3 replicates nets 32 runs
    ????????????
     
     

    0
    #127581

    Andejrad Ich
    Participant

    Mike,
    Hi.  First ask yourself, “What do I want to be able to say when I’m done with the study?”  If the answer is, “I want to be able to say that whenever anyone uses this gage (Serial # 12345) when measuring individuals of this part number” then that directs what to include in the test.  In fact, that is the original Gage R&R format.  Part-to-part variation was introduced because, if you were to (e.g.) use a wooden yardstick to measure the thickness of 10 quarters, they would all read “1/16th of an inch” and you would have proven that a wooden yardstick is the perfect gage for measuring the thickness of quarters (which, clearly it cannot be).  Unfortunately, it would also appear to be the perfect gage for measuring dimes, nickels, and pennies – ALL being 1/16th of an inch, but completely unable to distinguish the difference between the four coins (i.e., part numbers).
    So…ask what it is you want to be able say when you are done.  Then include a mix of operators, gages, and parts as indicated by your answer.  (But note that if your different part numbers have different specified tolerances, you will find yourself in a bit of a jam trying to calculate a single Gage R&R rating/value (e.g., 28%) for the total study.)

    0
    #127591

    Bill Craig
    Participant

    Stan,
    I think we are moving towards being on the same page and agree that the RR study is a DOE. There was a question about how the RR study could be a DOE, and I saw a response referencing the Montgomery book.  Another question was raised about the design layout….
    Factor 1: Operator (3 operators chosen at random)
    Factor 2: Part (10 parts chosen at random,  spanning most of the tolerance range)
    These 2 factors are crossed because each operator measures each part. This would be considered a general factorial (3 x 10 or 30 treatment combinations). Each treatment combination is run 3 times.
    The full model would include Operator, Part, and Operator*Part.  The reduced model removes “Part” from the equation. 

    0
    #127592

    CT
    Participant

    Answer me this please, assuming that an R&R is a DOE then what do you do if the significance to the response is one of the 10 parts or are you grouping them as a whole?  And what about the gage? Is it not a factor? And you cant remove part in the reduced model, that creates bias on the other two factors. Whats the response?
    CT

    0
    #127593

    Zeek
    Member

    No.   You were wrong.   Don’t change your position when called on it and then say, “That’s what I said.”    It was not what you said and Bill was correct.    The mark of a good blunt, plainspoken antagonist versus an average blunt plainspoken antagonist is in being able to admit when you are off-base.    Right now you are sounding like an average blunt, plainspoken antagonist.

    0
    #127595

    Mikel
    Member

    You don’t know what you are talking about.
    Go look at the recommendation of running multiple parts numbers as a DOE. The recommendation is bad for the reasons already given.

    0
    #127596

    Mikel
    Member

    This was your original post –
    Mike,
    One thing to consider is if you are using the system for process control. In this case, your SPC would probably involve data from a variety of part numbers. Your gage study should include multiple part numbers in that case. If you are only using the system for product control, you could use 10 pieces of the same part #. In both cases you would want to cover a good portion of the Spec range. To get real elaborate you could use operator, part number, and part (nested within part number). Call the MBB for help with this analysis!!
    That is the only thing that I have been objecting to. Otherwise I agree to and know you are right.

    0
    #127602

    Bill Craig
    Participant

    CT…good questions. I’ll do my best to answer them.
    Q:assuming that an R&R is a DOE then what do you do if the significance to the response is one of the 10 parts or are you grouping them as a whole? 
    A: A well conducted study will have part as the most significant factor, just due to the selection of parts. The sum of squares due to parts will be the largest, but is dropped automatically from the model because it is not part of Appraiser or Equipment variation.
    Q: And what about the gage? Is it not a factor?
    Usually a gage is not a factor. The gage is part of the process you are trying the characterize.   I suppose if you looked at it like a SIPOC the inputs would be operators and parts. The “process” would be the gage. The output is the measurement.
    Q:And you can’t remove part in the reduced model, that creates bias on the other two factors. Whats the response?
    A:ANOVA calculates the total sum of squares with: SS-parts, SS-operators, SS-operator*Part,and SS-error.  Repeatability is the SS-error (same part measured by the same operator multiple times). Reproducibility is SS-Operators and SS-operator*part interaction.  The sum of squares for parts is not part of the repeatability or reproducibility. 
    Hope this helps.

    0
    #127603

    Bill Craig
    Participant

    Stan,
    You are right about my orginal post, but take a look at the question being asked. (It is pasted below). He was looking to see if it would be better to use 10 of the same parts or 10 different parts. His GRRs were >50% when different parts were used. Adding “part number” as a third factor would be a way to quantify the magnitude of the effect. The random selection of part numbers would support your statement that a gage study is a random effects DOE. I recommended this to help him explore that additional source of variation. His problem is even more complicated because he has 10 different measurement systems. (10 sets of calipers). I don’t know how many part numbers they run, but lets say it is 50. Do you think it is best to do a gage study on all 50 parts on each set of calipers? That’s 500 studies times 90 data points each.  Do you pick one set of calipers and do a total of 50 studies? I am not sure if the 50% RR is the RR/Tolerance or RR/TV in his studies.  I just think the poor numbers are due to the use of hand held calipers and a tight tolerance of +/- .004 in. 
    Tag….you’re it! Looking forward to your response! BC
    Q: I have a quick question on a gage R&R study I’m doing. I new to this type of study and wanted to clear something up.
    Is it standard practice to use (10) of the SAME part number to use for measurements or is it acceptable to use (10) different parts (with different measurements) in the study?
    I’ve done a few studies using (10) different parts that represent our normal range of measurements. The numbers for the GRR using this method have been awful. (>50%). If I would use (10) of the same part I’m sure the GRR numbers would be better, but I do not feel that it would be a good indicator of how well our measuring system works.

    0
    #127614

    Zeek
    Member

    Wrong again. MSA is part of a basic Quality System, not some DOE you run on a bunch of part numbers all at once.
    Above is what you said.   You are not only curt and rude you are incorrect in your description of your past statements.   I conclude that you make foolish statements and then lie about it.  

    0
    #127615

    CT
    Participant

    Bill,
    This is my problem with saying that an R&R is A DOE. An R&R is an MSA tool used to quantify and qualify a particular gauge and or process. Like other MSA tools it does not place significance on any one factor but rather the whole process, yes one can break it down to better learn what it is affecting the process but is in no way a DOE. I have spoken with 3 of the MBB’s in my company and they think your smoking something. Although you do make a good arguement you are wrong about the interactions and how to treat the parts. The parts, if selected over the entire range of the process should give distinctive groups that are reliable for the R&R, the way you described is not possible to check.
    DOE is a design of experiment used to calculate specific interactions and place the most significance on factors that most affect the process. the outcome of how accurate this is, is based on the Rsqrd values of the whole interaction, and thus this why you eliminate some interactions during the analization of the DOE.
    I spent about an hour trying to setup a minitab DOE to run your R&R and it cant be done in any way that means anything. And further more I dont know why you would want too.
    Good Luck with That line of Thinking.
    CT

    0
    #127617

    CT
    Participant

    EXACTLY STAN,
     

    0
    #127628

    Bill Craig
    Participant

    CT,
    You can use ANOVA for observational data, and in that case there is no DOE involved. (Just collecting data and putting into different buckets for analysis). When you run a gage study, you are intentionally changing factors in a controlled fashion. Running all the parts by all the operators makes it a full factorial. When Mike suggested that part number seemed to have an effect, it adds another source of variation that should be studied. Since there is no software that has a canned 3-factor gage study, you have to find someone who knows how to calculate the variance components. Since it is rare for MBBs to have this knowledge, you might want to consult a statistician. If you are not treating a gage study, then you are probably not randomizing the measurements. This will add operator bias to your data.
     

    0
    #127631

    HF Chris
    Participant

    Bill,
    So in ANOVA you don’t/can’t intentionally make anything constant and don’t compare intentional changes……mmmmmm.
    Chris
     
     

    0
    #127634

    CT
    Participant

    Yes he needs to do an R&R for every part if he truely wants to know if his gauge is capable for each process. Sorry but that is the way it is. and + .004″ is not a tight tolerance for calipers if used correctly.
    CT

    0
    #127668

    Bill Craig
    Participant

    Chris,
    ANOVA is used for observational data and DOE data.                       I was trying to distinguish between how the data set is generated in each case.
     DOE: Data is gathered by intentional manipulation of factors. A gage study follows this definition. If you look at page 519 of Montgomery’s book on Design and Analysis of Experiments (5th edition), it describes how a gage study is a common application of a designed experiment. A real good explanation of gage studies is given in Montgomery’s statistical quality control book.
    Observational Data: No intentional manipulation of factors. Sometimes called “happen-stance” data. Try doing a search on observational data and you will find some interesting applications. The basic principle is that the experimenter does not have control of the treatments applied.

    0
    #127680

    Mike Carnell
    Participant

    Bill,
    I’m glad to see you decided to get involved. Actually it looks like you have jumped right in the middle of a hot topic.
    I can’t wait to see you in the middle of a good GD&T string.
    Did the Keys over Labor Day. We still need to hook up and fish down there.
    Welcome,
    Regards

    0
    #127685

    Mikel
    Member

    Count me in on the fishing. I’d like to meet Bill.

    0
    #127688

    Bill Craig
    Participant

    Mike / Stan,
    Mahi Mahi fishing is great in late May – early June. Someone should schedule a six sigma forum in Key West during that time!
    Rule # 1: SS topics stay on Dry Land
    Rule # 2: If you mention sigma, DOE, RR, SIPOC, etc. while on the boat, you could get turned into chum
    Rule #3: If you say the word “six” on the boat, it must be followed by the word “pack” .
    Rule #4: If you think you can drink Guiness faster than Irish Kevin, guess again.
    Mike…. I hope my GD&T knowledge is still up to snuff. Haven’t used it since the good ole days at the big M.
     
     
     

    0
    #127691

    Mikel
    Member

    It’s okay with me if we extend rules 1, 2, and 3 to anything south of Alligator Alley.

    0
    #127724

    Mike Carnell
    Participant

    You have. PWF at McDowell Road Facility.

    0
    #127726

    Mike Carnell
    Participant

    Bill,
    After this last week I won’t go anywhere near the Keys during hurricane season (weather not football).
    I have asked Mike Cyger about a Six Sigma forum in Key West (let’s do Duck Key instead – us old guys need our sleep). We can do it like a Mardi Gras thing so the people who post under a name other than their own can wear a mask and a name tag with their screen name. Could be interesting. That close to Lauderdale Darth should be able to bring the Vadermobile. We will need to bring some decent CDs so he doesn’t play that Yanni stuff.
    I can live with the rules and Kevin can have all the Guiness. I’ll bring a nice bottle of Havana Club rum for the evening on the beach watching the sun go down.
    I’ve never met anyone who knew GD&T better than you. I’m sure you can deal with whatever shows up.
    Regards

    0
    #127727

    Mike Carnell
    Participant

    Works for me.
    Have you been in Alabama Jack’s lately? Talk about getting pure Yuppie-fied. It was embarrassing. One beer just for old times sake and we were out of there. Hog’s Breath is about the same.
    Regards

    0
    #127738

    Mikel
    Member

    Yuppies on Card Sound Road – Yikes!!!!

    0
    #127740

    Mikel
    Member

    I knew the name was familiar. Sounds like a pretty smart guy if only he did not want to throw in lots of different part types into a single MSA (it’s a joke Bill),

    0
    #127750

    Mike Carnell
    Participant

    Since the late 80’s.
    That is the northeast coast influence. Linus country.

    0
    #127751

    Mike Carnell
    Participant

    Even the bikers were Yuppies. I think the best selling drink was herbal tea.

    0
Viewing 62 posts - 1 through 62 (of 62 total)

The forum ‘General’ is closed to new topics and replies.