Advice for SS Project Concerning Cutting Product in Squares

Six Sigma – iSixSigma Forums General Forums Implementation Advice for SS Project Concerning Cutting Product in Squares

Viewing 10 posts - 1 through 10 (of 10 total)
  • Author
  • #240390

    Aaron Olson

    Hello all,

    Our facility has 2 main functions:

    1. Cutting Material (custom extrusions, plate, regular stock, etc.)
    2. Fabrication (CNC, punch, drill, bend, weld, etc.)

    Cut material can be sold directly to customer or sent to fabrication for further processing.

    We have had an issue with ensuring that product is cut in square according to the customers’ specifications (or specifications required for proper machining internally).  Upon implementing 5S at the saws, it was discovered that at least 4 different methods are being used to measure how square the part is.   When discussing what gage/instrument should be at each station, some differing opinions surfaced.

    1. A single 90 degree angle block and feeler gages
    2. A simple square and feeler gages
    3. A custom instrument that uses a flat plate, a straight edge, and a depth gage that can move in 2 axes
    4. A surface plate, 2 x 90 degree angle block and feeler gages (to make essentially 3 datums)
    5. (also available but rarely used due to capacity/complexity is a CMM and a “Romer” arm CMM)

    We are in the process of determining how often and for how much $ we end up with defects concerning how square the product is.  When i have that information, I will update this thread.

    My current steps are as follows:

    1. Determine the impact (including $) of the defect; create a measure (and collect data) for the number of defects due to cutting material out of square
    2. Perform Gage R&R (or other MSA if advised differently) on the different methods currently used to measure the square-ness (including the ways that seem absurd from a quality/inspection point of view – hoping that proof will aid in the conversation of why it is not effective); will likely do separate tests for different “classes” of material [standard stock, complex geometry extrusions, different height/depth etc.]
    3. Use above information to aid in the creation of standard work (work instructions, training, etc.); also determine a standard for how/when to measure for square-ness at the saws
    4. Use above information to identify potential “X”s that cause material to be out of square; develop hypotheses and perform tests
    5. Use above hypothesis tests to guide the ideation, selection, and implementation of improvement countermeasures to decrease the amount of measured defects

    This is a broad-stroke idea but I wanted to see if it made sense.  Any thoughts or suggestions for first steps?

    Quick question while we’re at it: While performing a gage r&r, should one of the inspectors be the “expert inspector/method”?  Or is that going to mess up the results?  I remember hearing something about this during training but that has faded a bit.

    Thanks everyone!


    Mike Carnell

    @aaronolson I assume that these are large squares? It looks like options 1-5 are going to have capabilities that are going to differ based on the tolerance requirement of the square. You might consider that you qualify the measurement system based on tolerance requirements i.e. not use a CMM if there is a tolerance of +/- 1 inch.

    when I look at measurement system 1 and 2 it looks like they will tell you in a go/no go way if the corner is 90 degrees. That does not mean you have a square. Just a thought.


    Aaron Olson


    Thanks for the response.  To clarify, the parts are generally smaller (always less than 10″ x 10″on the profile and varying in length); the difficulty of squareness is that it has a tendency to interfere with our machining tolerances (typically +/-0.010″); and perhaps I worded it wrong but this is concerning cutting product (usually custom extrusions) in square (axes are all perpendicular) and not cutting into actual squares/blocks.

    Qualifying the device based on the tolerance requirement is a good idea, thanks!

    Regarding measurement systems 1 and 2, I would be okay if we just stopped using them in general for checking squareness.  1 is a measure of straightness and not squareness (only 2 axes vs 3); 2 is a difficult one as it is largely reference and more difficult to check all areas of the profile.

    Did you have any thoughts on my last question above regarding gage r&r methodology? (“While performing a gage r&r, should one of the inspectors be the “expert inspector/method”?  Or is that going to mess up the results?”)



    Mike Carnell

    @aaronolson I apologize on missing the part about the expert. Your measurement systems may make a straight answer a little more complicated. It looks like 1 & 2 are going to produce attribute data. That is going to be an attribute agreement type study and you will need the expert included by definition. If you go to a measurement system such as a CMM that will produce variable data I would include the expert and run the analysis with and without the experts input. Not with anything particular in mind but just to see if there is a difference.

    If I am checking 3 axes doesn’t that mean measurement systems 1 & 2 are not adequate?

    I would think part of the answer is going to be what is the volume. I would think if you are doing a high volume then 3 and 5 make the most sense. You would need to fixture #5. With a tolerance of +/-0.010 you have a lot of room and I would think a CMM is overkill but if it can be fixtured and programed then it is a fast solution.

    Maybe it is just me but option #4 seems very prone to operator technique issues. I am not a big fan of opening the door on measurement error/variation to becoming a function of how much partying took place the night before.

    In terms of your approach I like #2. Without the data on the current systems then most of your discussions are going to be opinion. As long as you have a person considered to be the “expert” that person is probably going to have an inordinate amount of influence on the group. If you do include the expert in the MSA and they do poorly I would not put that data in front of the group. You should have that discussion on the side or else risk having an enemy for life.

    Who knows if you do step #2 and some of your systems fail that makes decisions on how to measure and what data to use for steps 4 & 5 a much easier decision. That is what problem solving is. It is a subtractive process. You begin with all possibilities and then decide what to throw out and what to keep. You don’t make any progress on this until you eliminate the poor measurement systems and the data they produced or at least understand how inadequate they are. That is almost as valuable as knowing what works.

    The first 3 steps are really just to enable you to do steps 4 & 5 which is really the long term solution to the problem.

    This is just a thought but no matter where you go with this I would think sooner or later you are going to have to answer the question about how good is the die for extruding. I would put that thing on a CMM as soon as I could just so you don’t have to sit through some agonizing pontification about the effects of the die.

    Just my opinion,


    Aaron Olson

    @Mike-Carnell No worries! I appreciate your advice!  You’ve been gracious enough to give your opinions to someone who is still fumbling their way through this.

    I agree that measurement options 1 and 2 should be disqualified since they don’t seem to be able to control for all 3 axes.  However, there is resistance to this because, you guessed it: “This is the way it’s always been done”.

    Measurement Option 4 is actually the one that is thought to be the most consistent behind the CMM.

    Measuring dies would be quite difficult: they are at our suppliers and there are 100s (or 1000s) of them.  At this moment, the issue of squareness is suspected to be in the saw process and not the die.  Still worth considering as a course of action?




    Mike Carnell

    @aaronolson I would think that to a bunch of mechanical people #4 would appeal because it is traditional. Old time mechanical measurement stuff. Also technique dependent.

    This whole 1,2 & 4 thing seems to be pretty opinion based. Probably healthiest to stay out of the fight and let the data talk.

    That is a lot of dies. Probably best to wait.

    The sooner you get your team onto data I would think the better you will be. There are a fair number of people who, once they take a position, will not move regardless of facts. Probably best if you get them looking at data before they dig a really deep hole.


    Aaron Olson

    I fear I may have made a mistake and not collected enough samples for an attribute gage r&r for our squareness gage.  It still looks grim but we’ll see if I screwed up the method first.

    2 separate gage r&rs (different parts/type of profile, same operators)

    Each Gage R&R:

    3 operators

    2 trials

    10 parts


    Gage r&r for part 1: 60% with and between themselves against the standard ; 80% with and between themselves only

    Gage r&r for part 2: 70% with and between themselves against the standard ; 70% with and between themselves only


    Is this enough to gain anything useful?  I’m assuming despite the small sample size that this tells me the gage/operator training is incapable and I’m going to have to dig in further?


    Thanks for the advice!



    Mike Carnell

    @Aaronolson Most of the literature is going to tell you 20 or 30. We had a huge debate on this in 1999. 20 or 30 is actually insufficient as well. The real issue is nobody generally will pass an attribute agreement study first time through so why do we want to do a couple hundred samples which is where one of out consultants had determined we should be?

    Tell you what. Run your numbers again and make it look like you had 20 samples and the additional 10 all were correct and see what your results are. Your inspectors are not agreeing with the standard i.e. the true state of nature. That is a problem. It gets even more complex when they don’t agree with each other.

    I ran a study in June with sample size 16. It was because of the way it was packaged. I could have done 32 but when you watched them on the job you realize they won’t make it at 16 why do 32.

    Once you fix the inspection you may want to run a larger sample just to be sure you fixed it.

    Remember this is inspection so this is a behavior based issue. Your control plan needs to cover training at probably around a 3 month interval because this just never stays fixed. You pay someone to find defects that is what they do. You improve the process and they get more critical. Take a different approach talk to them about their job being to verify everything is good. It sounds stupid but their seems to be a difference in looking for bad versus verifying things are good.

    Just my opinion.


    Aaron Olson

    @Mike-Carnell  Thanks for the response!

    After fixing the measurement system (re-training, evaluating design flaws, etc.), how do you recommend doing the study?

    3 Operators

    100-ish samples?

    2 trials

    What should the goal/standard be? >95%?


    Mike Carnell

    @aaronolson In all probability you are not going to make much difference by the number of samples you take because it takes a very large confidence sample to mean anything. I still stick with 30 and 95 – 100%. I do prefer 3 trials and randomize between trials.

Viewing 10 posts - 1 through 10 (of 10 total)

You must be logged in to reply to this topic.