iSixSigma

Setting action limits

Six Sigma – iSixSigma Forums Old Forums General Setting action limits

Viewing 30 posts - 1 through 30 (of 30 total)
  • Author
    Posts
  • #42487

    Jo
    Participant

    In our company there is a history of setting limits for particle measurements based on monitoring data. Action limits are set at the average + 3 sigma.
    However, when applying a normal distribution on the data set, it can go below zero. In practice this is not possible. We cannot measure below than zero. Now, some people want to apply special statistics to come to the “correct limits”.
    In my opinion, setting limits or specification must be based on what we and our customers want/need and has nothing to do with statistics. Statistics must be used to analyze data, to calculate process capability, to measure improvements …
    Can anybody give some info on how other companies set action limits for any measurement ?
    Thanks,
    Jo
     
     

    0
    #134113

    Darth
    Participant

    Control limits are set by the process using statistics, not personal opinion.  Specs are set by personal opinion.  If you are monitoring the process and taking action on points outside the traditional plus/minus 3 s.d. that’s fine.  The control limit on the low side can be truncated at zero if below zero is not feasible.  It also means that possibly your distribution will not be normally distributed due to the inability to measure below zero.  The control chart is robust to that so don’t worry about it.

    0
    #134139

    Jo
    Participant

    Thanks Darth for your comment. Indeed that’s what I meant: specs = personal opinion, control limits come from control charts and statistics.
    Is it correct to say that control limits can vary as the process goes along (I’m not an expert on statistics) ?

    0
    #134142

    Darth
    Participant

    Control limits act as a baseline of a stable process.  They only change, that is are recalculated, when the process changes and a new level of stability is reached.

    0
    #134159

    ezweld
    Participant

    I think we have all been spoiled by programs that do charts for us.
    The calculations for an Average chart are:
    Upper Control Limit= Grand Average + (Avg. Range x A2)
    Lower Control Limit= Grand Average – (Avg. Range x A2)
     
    Range Chart calculations
    Upper Control Limit= Average Range x D4
    Lower Control Limit= Average Range x D3
     
    Sub Group       A2       d2        D4       D3
    2                      1.880   1.128   3.267   0
    3                      1.023   1.693   2.575   0
    4                      .729     2.059   2.282   0
    5                      .577     2.326   2.115   0
     
    That being said I’m still going to use +/- 3 standard deviations.
     
    Has any one set up rules for a chart in Excel to flag out of control conditions?
    Such as the 3 on the same side with 2 out of the limit 2 sigma, 5 in a row rule, or the 6 in a row rule. Please post some code or functions.

    0
    #134179

    Darth
    Participant

    The question is what plus/minus s.d. calculation will you use?  Control chart limits are based on the within sample variation.  What will you be using?  There are control chart constants similar to the A2 value but are for sample s.d. instead of sample range.

    0
    #134204

    ezweld
    Participant

    I must have misunderstood the question.
    However, that is how most of us learned how to find control limits.
    I think the upper control limit and lower control limit are the best action limits.
    So I have to ask; How do you set your action limits Darth?

    0
    #134220

    Mike Carnell
    Participant

    ezweld,
    I am not sure I get your point.
    “I think we have all been spoiled by programs that do charts for us.”  I can’t argue with that and I am certainly guilty of it. That made some sense until we hit the part where you said “Has any one set up rules for a chart in Excel to flag out of control conditions?Such as the 3 on the same side with 2 out of the limit 2 sigma, 5 in a row rule, or the 6 in a row rule. Please post some code or functions.”
    I am not the most astute user of a computer but last time I looked Excel was a program somewhere on my computer. If you are willing to do your own calculations a search for code for out of control conditions it would seem much more pragmatic to use a stats software package and leverage the knowledge of the statiticians they use to write their programs. The cost of most of these programs is pretty quickly offset when someone decides that in just a few weeks time they can teach Excel to do it all. Then they start to debug it.
    Just my opinion.
    Good luck

    0
    #134224

    ezweld
    Participant

    The company I work for has given me the task of implementing SPC for 4 of the plants. They don’t want to spend the money on the software I wanted. So idiot that I am said, I will just build it in Excel and pull out of Access databases so the training for the QA Tech will just be limited to interpretation of the charts. I have been using Excel for years so I had most of the code. The only code I need to write is the rules of being out of control and how to display an out of control chart. Writing the code for the charts really is stupid when there are such incredible tools for sale, but the truth is for the simple run charts the company wants right now 95% of MiniTab or Sigma XL will not be used. I’m hoping some poor slob has had to do something like this before and would be willing to lend me some ideas.

    0
    #134225

    Darth
    Participant

    I am unclear what your definition of action limits is.  I am assuming it is the same as the control limits.  Taking action or not taking action on anything else could be resulting in tampering or under reacting.

    0
    #134237

    ezweld
    Participant

    Darth,
     
    I took action limits to means the same as control limits when I first read the original post. After thinking about it I thought if I can build some type of relationship calculation I can setup a report based on predetermined responses for the operators.
    It is true tampering will be involved but I’m hoping to minimize process tampering and increases process understanding. I am working in flexible packaging industry were Quality has not yet reach the level of AQL. The machine operators constantly make adjustment and measurements have a bias. I hope to bring some level of control to my company. If you could lend me some Ideas, I would truly appreciate it.

    0
    #134244

    Darth
    Participant

    Here is your key statement:”The machine operators constantly make adjustment and measurements have a bias.”My advice is to the control charts will help prevent that.  Only adjust upon statistical signal.  If the process is totally out of control then major intervention is needed.

    0
    #134245

    lin
    Participant

    There are inexpensive Excel based SPC programs out there that do what you want.  Search spc excel and you will see a variety of programs offered.

    0
    #134246

    ezweld
    Participant

    The Excel add-in programs will take time to train operators in other plants on how to use. I could use that time to train the operators on how to understand a chart. You guys have to understand this operation is like going back in time to before Deming was accepted. Only one operator in my plant had ever seen a run chart before. Let me put it this way, my first three projects are ISO 9000 cert, Gage R&R and SPC charts for all the plants.

    0
    #134279

    Vilfredo
    Member

    Mr. Jo,At my company we had similar problems in monitoring particle counts in controlled environments such as medical device assembly and pharmaceutical processing areas. Many of the regulatory folks like the FDA and EU like to see two levels set on trend charts: Alert and Action levels. As you suggest the Action level would be set using customer, product, or technical requirements, but not statistical methods. For regulated products the Action Levels for air borne particulate counts should be set to meet the device or API’s requirements commensurate with regulated and medical standards. The use of the EU or Americal Pharmacopeia as a start point may be useful.For Action Levels you are interested in observing changes in the trended pattern that would signal changes in the process – which could be producing higher than expected particulate levels. In this case a properly developed Shewhart Control Chart would be useful. However, the question comes to mind which type chart is appropriate? Typically, particle measures would be considered Poisson NOT Normal distributed data, and a C-Chart would be used. However, because there are typically many zero counts in particulate data the Poisson distribution is not the best one to use, and a C-Chart established UCL would produce too many Type I errors. There are two other distributions that better fit particulate data, Lognormal and Gamma. As it turns out the Lognormal distribution provides the most convenient to use of the two. My company has used the Lognormal with specific recomendations to develop control chart limits in order to monitor particulate and bioburden levels while controlling the Type I error rate to 1 in 370 typically reported for standard Shewhart Control Charts. The equation for the UCL of a Lognormal Control Chart is a bit detailed. If you’re interested in more info, then please provide me your email address.Vilfredo

    0
    #134299

    Mike Carnell
    Participant

    ezweld,
    Have these guys seen the cost of something like Minitab versus the time it is going to take you to get this done. Just put some cost numbers together so they can understand how people do a make/buy decision. One way they write a check and the other they expend a finite resource (your time) either way they pay for it.
    Maybe you should start with your guys that decided to have you implement SPC. That seems to be a dumb assignment. How do they want you to select the variables to control with SPC and how will you know the effect of the variable until someone does some analysis of the process.
    Somewhere along the way we seem to have lost some of the logic that goes with SPC – most likely with that fine group a SQE’s from the US automotive market that wanted a chart neatly hung on the wall in their SPC prophylactics (sheet protectors) so they actually felt like thay had done something besides drag their customers to Jason’s in Windsor. My rant for the day – sorry. Go look at Schewharts original stuff and take note of the word “economical.”
    If you use the logic of Y=f(x) then you understand that putting a chart on anything other than an x is a waste of time. If you have ever used an attribute chart you know that when it goes out of control you still do not know much except a Y is out of control and you now have to figure out which x is driving it. To do that you need to know which x affects it. If you know that put the chart on the x in the first place. That knowlwedgem (about the x) is an ancillary effect of a Six Sigma project.
    If you are putting in variable charts   how are you selecting variables. Some mean something and some don’t. Again the same logic that the knowledge comes from understanding a process and selecting variables. When we started doing deployments we thought we had this straightened out then we got Breyfogles book coming out with his control charts in Measure and it started all over again. The angst over process stability, in control, out of control, etc. No wonder everyone is afraid to start a project. We have created such a list of caveats that a new BB feels like a country dog in the city. If he runs he gets bite in the butt and if he stands still he gets _____.
    After all that I guess my question is: 1. Why would they make a decision to do SPC until they had further information 2. Are free to add some intelligence to the decision they have made and to do it correctly? Sorry for wandering around so much but this SPC thing just drives me crazy.
    That Columbo thing. “Excuse me sir but one more thing.” Go ask your management team 1. who is going to train people to do the charts. 2 who is going to maintain the charts. 3. Who decides when a chart is and isn’t useful and who pulls it when it isn’t doing anything. 4 what happens when a supervisor doesn’t stop a process that is out of control (or the operator stops and the supervisor signs it off and says it is ok to run). If they cannot answer these simple questions then ask them how they decided to implement SPC.
    Just my opinion.
    Good luck

    0
    #134371

    ezweld
    Participant

    Thank Mike,
     
    I wish I could say that you’re wrong about my company.
    In some ways I will have to reinvent the wheel as well as prove methods and rules to people that don’t believe in statistics.
    One Quality Engineer I worked for, at my last job, told me he did not believe in statistics, he said you can makes those numbers say anything you want them to.
    He is still  well respected in the company while my name is a curse word there.
    So the fact remains I have an impossible task in front of me.
    I would like to ask if Schewharts rules for a process being out of control can or should be used for a run chart?
    When I had my SPC training I was told to apply them to a capability study, I never put any thought into using them for a run chart.
    My boss has asked me to add to my charts the out of control logic for the 4 parameters that we can measure.
    I think his hopes are that we can shed light on our problems so maybe we can start to understand the processes.

    0
    #134506

    Jonathon Andell
    Participant

    If your action is based on CONTROL limits instead of SPEC limits you’re on the right track.If the control limits you get yield an unrealistic number (like an impossible negative), you have two options:1. Don’t fret about that side of the process, or2. COnsider something like a Box-Cox transformation. The transformed data, along with similarly transformed control limits, might be actionable.The trouble with Box-Cox is that you may not understand the underlying distribution of your data. You might also attempt a distribution ID routine, because identifying the distribution may give insight into what makes the process tick.That’s the abridged explanation…

    0
    #134508

    Mike Carnell
    Participant

    ezweld,
    You made me do some thinking with your comments. It seems to have gotten easy to walk away from the guys that are being a complete POS and find someone who wants to make a difference. What you made me remeber where the days in the early 80’s when I first worked as a Quality Engineer at Motorola. I met a guy named Gary Cone who knew about sampling, control charts, statistics, etc and had a better way. I ended up in much the same situation as you with coworkers. The good part was that where ever I went I found good people to work with. I worked with Mario Perez Wilson on a bomb fuse program and Bill Craig in a Printed Circuit Facility )I picked them because they post here) and I had a guy named John Lupienski for a mentor and a guy named Marty Rayl for a Leader. I have been blessed in my career and there is still rarely a day that goes by that we don’t deal with some type of resistance to change somewhere.
    Occasionally it all comes together and it is good thing but you need to understand that just like Camalot it won’t last forver so you enjoy it as much as you can for as long as it lasts. Then cherish the memories.
    You have to pull your satisfaction from inside you and knowing you are doing the right thing. In that case when you get some support or praise extermally it is a plus.
    I wish you the best of luck. If you hit that wall and need to talk feel free to contact me at [email protected].
    Regards

    0
    #134514

    Darth
    Participant

    My sources at Motorola said that this Gary Cone guy was pretty much of a jerk….arrogant, know it all who insulted people if they asked stupid questions.  But they did indicate he had a really smart, good looking wife who made up for it.  Was that the same Gary Cone?

    0
    #134515

    Lomax
    Participant

    The questions and replies on SPC and what seems to be managements desire to take the “short cheap route” is one that seems to be in ever small to mid size company (my experience). The responses were good although it seems that while trying to convince management or the powers-that-be to invest in MINITAB or similar products often fall on deaf ears and tight wallets. Is there a simple training presentation or an “easy to read and understand” article that explains SPC, Control Charts, etc. on the SixSigma website that anyone is aware of? Or one elsewhere? This may help answer a lot of questions to the beginning stats people and/or QE’s. Feel free to contact me at [email protected] if anyone has any info and good luck EZ.
    Neil

    0
    #134526

    Willis Major
    Member

    Ezweld,
    From what you have said, your company wants you to do the impossible but not give you the resources. I have training as a green belt, a have a Masters in Statistics and I am a Shainin Red X Problem Solving Journeyman. I mostly use the Shainin Red X tools to solve problems. They use Precontrol to keep processes under control instead of using SPC. Precontrol uses a green zone, yellow zone and red zone on the chart. The green zone is where if parts are measuring in this zone you make no adjustments to the process. The yellow is where is you get two measures in a row that are in this range you make a centering adjustment to the process. The red zone is where if you get just one measure in this zone you make an adjustment. This is a simple way to keep your process under control. If I were you I would not go through all this complicated training to teach your company SPC wher there are bigger more sophisticated companys that take the data with a computer but don’t really utlilize it to keep their process running under control. Start your company out with something simple to do, easy to learn and you will see a big benefit. If you start them out with too much complication then other in your company may be waiting for you to fail because of their dislike with statistics. Do some research on precontrol or go to the Shainin.com website and give them a call.
    Willis

    0
    #134527

    Major Minor
    Participant

    Warning! Warning!
    Anyone claiming to have a Masters in Stats who is promoting Shainin isn’t being completely truthful. Careful of any advice you take there.
    You may want to look at stuff from Box on SPC vs EPC (Engineering Process Control).

    0
    #134533

    Mr IAM
    Participant

    Hmmm…
    I think that was a misleading post and interpretation of Pre-Control.  Pre-control does not control a process the way SPC does.
    When people hear the word “control” they normally think “statistical control” which has nothing to do with Pre-Control.  Some people also think you use “Pre-Control” before using SPC.
    I like Pre-Control but it is not a process control tool in the sense that SPC is.
    Pre-Control is more like “capability control” where SPC is “statistical control”.  Those two tools are not interchangable.
    Cheers!

    0
    #134538

    Willis Major
    Member

    Warning! Warning! Beware of Major Minor Jokes!
    Remember you can receive all the education in the world on 6sigma, statistics, Shainin or what ever, but if you don’t produce the result at your company then it is all in vain. It is not only what you know but about how you apply your learnings and how you sell it to top mamnagement. I think six sigma is great, I also think Shainin Red X problem solving is great, and various statistical tools are great. But I guess after using Shainin to save my company over 8 million dollars over the last 4 years, I would have to take head to the tools that Shainin Red X problem solving uses in addition to six sigma. Actually they use some of the same tools but are named differently.
    I am not asking anyone to take my advice, because I know that words mean nothing but action means everything. So find out for yourself, gain some knowledge before you speak of things you know nothing about.
    And for your information, Shainin is based on Statistics : Non parametric Statistics to be more specific. They have a PHD in Statistics that uses statistical methods when they come up with trade marked tools. Have you ever heard of the founding father Dorin Shainin?
    Probably not Major Minor – dont be jealous, it is not a good quality!

    0
    #134543

    Willis Major
    Member

    MR Iam,
    Good Statement on the difference. SPC is looking to see if data from your process is random or non random (hence, using statical control).
    Precontrol is trying to control your process out and keep it running runnign within the specifications (Hence, specification control).
    I hope I was not giving the impression that they were doing the same thing, if so I apologise.
    What I was trying to point out, was when you are at a company that is in the stone age. You might want to think about implementing something that is simple and can be easily implemented with minimal training first.
    After all if you can make an improvement quickly in your quality and do this with minimal training. you may bring your company further along when you are up against some must resistance to SPC in the first place. The managers did not want to even buy a SPC program and wanted him to write code in Excel. They don’t know anything about SPC in the first place, and I was simply suggesting an alternitive that may help him to show significant improvements in quality of products they produce with the smallest amount of learning involved.
    It was only a suggestion. But it might just be the best thing for them to implement.
     
    Willis 

    0
    #134544

    Mr IAM
    Participant

    Willis,
    I agree that Pre-Control would be a good option for them to consider.  I just thought your original post did not provide enough detail on the differences between Pre-Control and SPC.
    Cheers!

    0
    #134545

    Willis Major
    Member

    Mr Iam,
    Thanks for adding the details so eloquently.
    Willis

    0
    #135360

    shawn
    Member

    Darth,
    I know you guys like to fool around on here, but your statement is far from the truth. I have worked with him for the last year and he is the best I have seen for teaching those who want to learn. He is completely unselfish with his time and will sacrifice his job to support those who work with him.
    That’s a tough act to follow in corporate America.

    0
    #153861

    ezweld
    Participant

    I thought I would give you guys an update on what had happened with these charts. I did build run charts that used Excel to pull data out of Access databases. The points on the charts change color for various conditions like being under spec limit and control flags. I even put in pretty dashed lines for UCL and LCL. All that work was for nothing. The sad story I’m about to tell just breaks my heart. During the time I was working on these charts we started to have unexplainable problems with our products. To put it simply during product testing we had a strange failure that past the spec but just did not look like what the product had months before. Our spec was just a made up bunch of crap that was never based on the customer’s requirements or actual product performance.  It turns out one of the engineers at the other plant decided he didn’t like the way our machine worked. He made changes to it and then just went back to his plant. I was new at the time and I had to devote all of my time to writing documents to get the plant ready for ISO certification. One of the procedures I wrote was about machine settings; operators are allowed to change their machine controls past a certain percent from the nominal settings. This procedure was signed off by the head of production (who is really a glorified operator) and our plant Engineer. Well because the machine was altered there was no way the operators could run with the SOP settings, so they just changed them and said nothing. And so the Engineer from the other plant made more changes to the machine. The operators of course just changed the settings to the machine and said nothing. We at the very least lost running capacity because the main change the operators made was to decrease line speed. Let me just say the strange failure never received a customer complaint but the plant that the Engineer was from started to exploit our strange failure. (My plant supplied their plant with some of the product we produced. Their plant machine had customer complaints on the same product that we produced) this turned into a big mess. The director of the Engineering group thought that our plant just didn’t know how to process the machine. He came down himself and ran a trial based on what he thought the settings should be. (The trial was not a planed experiment – he just made adjustments from the hip and ran a production run order) So the head of the department ran an order outside of what the company stated was the SOP. The guy did some pre-inspection test, said it looked good and went back home. The entire production run was worse than what the operators had been running.  (The product has to be tested days after because of a chemical change during the first 12 hours) My boss let me run a DOE. I didn’t have a big budget, no software other than Excel and no buy-in from anyone else in the company. I got the operators to tell me what they thought were the main settings that could cause the failure. Then what were the max extremes I could go to and not just crash the machine or have the product not be in one piece. (Some of you DOE veterans already know what is going to be the outcome) The settings were not nearly extreme enough to see a change and because I was limited to doing an L8 because of cost the DOE, just about was a failure. My findings were that only line speed was the setting that had any effect on the process out of the setting in the DOE. As it was I showed we were running too fast to make good product. I truly felt like my DOE was a failure due to not being able to use more factors or a larger experiment. I went in to my presentation meeting thinking that because there was no belief in quality tools in my company I was just going to be a laughing stock. The Engineer that started all the trouble was ironically the one that gave me the most credibility. He said during my statements of what settings had the most effect on the process, “Hell I could have told you all that. If that’s all you came up with you wasted your time.” I said “Yes, but I came to the same conclusions as you without knowing anything about the process and I proved them mathematically.” That statement got me a new DOE with a larger budget and a copy of Minitab. My next DOE had every setting I could change in it. But I wanted to include changing back what the Engineer had done to the machine. That of course was not going to be possible. The DOE proved that only line speed could affect the product and we were running to fast. That gave power to our plant Engineer and Maintenance to call in a consultant to look at the changes made. The consultant said the changes would cause problems and we were lucky we did not blow up. During that time our operators kept doing things outside the system and a company decision was made to not run the product at our plant until we got our act together. I had another DOE lined up, the machine was now fixed back, I got new OPC software to log machine data to a database. The company chose not to let us run the product due to an increased capacity at the other plants. I blame the worthless operators that ran outside of the SOP and ran bad material because they have no accountability. (I guess I left out the part of the story where the machine operators run non-conforming product because they leave the machine unattended and run out of raw material or forget to do monitoring and needed maintenance. Or just decide to use non-conforming raw material because they don’t want to reject it. Also just change setting on the machine all the time.)
     
    So all the work I did building the charts was for nothing.
    One good thing will come of this, I now have an OPC server that is going to be connected to the machine’s PLC. After I write a DDE client to a database I will be able to log everything the machine does. Things like line speed, downtime, time it takes to respond to an alarm, machine setting, sensor outputs, basically I can pull up what the machine was doing when the operator made a change to the settings.  If I get approval for a vision monitoring system I can even relate defects to alarms and the response after. As part of the vision system proposal I stated that it will be made to shut the machine down after so many defects. So my revenge to an operator who steps out for an hour long smoke break will be a detailed report of when the machine shut down for defects and how long before the operator came back and restarted it. That way he will have to explain why he did not stop the machine for defects and why it took him x amount of time to start it back.   

    0
Viewing 30 posts - 1 through 30 (of 30 total)

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