iSixSigma

DOE with a Variable Response

Six Sigma – iSixSigma Forums General Forums Tools & Templates DOE with a Variable Response

Viewing 15 posts - 1 through 15 (of 15 total)
  • Author
    Posts
  • #53796

    Euston
    Participant

    Hi
    I’m a deployment leader and MBB for a food manufacturer. While my primary role is supporting and mentoring the BB’s at our 4 plants, occasionally I work with our R&D group on various projects, helping with facilitation and use of various tools.
    I’ve run into an interesting DOE on a R&D project. This is a food product in an aerosol can, where the concern to the team/customer is the performance of the food throughout every dispense, particularly at the end, where consumers have expressed dissatisfaction. The team believes that changing the slope/drop in pressure will end up satisfying the performance issue over the last few uses.
    When I received the DOE results from the team, they had analyzed the experiment at each of the 8 response variables, ending up with changing significant factors.
    I’ve made the argument the one and only response should be the overall drop in pressure over the entire evaluation. The response curve is linear, which I believe supports use of the psi drop as the response.
    Any option of this approach or another way I can analyze the data?

    Jeff

    0
    #191464

    Robert Butler
    Participant

    As I understand your post a DOE was run and 8 Y’s related to the product performance were measured at each of the design points. If this is the case then the usual procedure would be to build a regression model for each Y and then take the ensemble of equations and predict an optimum performance across all 8 Y’s where each Y had some range of acceptable optimal values.

    However, your post gives the impression that only one Y is of interest – that being the overall pressure drop and there is no concern with respect to the values of the other measured responses. I find it odd that one would run a design this way but if this is the case and pressure drop is the only concern then one would need to know how the pressure drop was measured and assessed. Is the value for pressure drop a single number – i.e. the delta in pressure between initial and final use – or is it a series of measures of pressure? If it is the latter then you have a repeated measures problem and you will have to take this into account when analyzing the data.

    If you could provide more detail with respect to the problem perhaps I or someone else could offer some additional thoughts.

    0
    #191465

    MBBinWI
    Participant

    Jeff: Despite your inclination to use only one Y, unfortunately the customers pay attention to all the Y’s. This is one of the biggest benefits of conducting DOE’s, you can manipulate several inputs, measure several outputs, and make sense of it all.

    Hope this helps.

    btw, I hope I haven’t just helped out the competition as I’m currently working with a very large food company as well (wouldn’t it be ironic if it was the same one?)!

    If you append atsign wi.rr.com to my nom de plume, you can get an e-mail to me directly and we can chat.

    0
    #191466

    Euston
    Participant

    Thanks guys,
    Your responses made me realize I need to shed some additional clarity on the problem, then perhaps you can elaborate.
    To me, the DOE was set up with only one response variable, but that variable changes with each aerosol dispense. To explain how the data was gathered: From each run, they obtained an initial fill psi, waited 12 days (to represent S&H time), shook and measured the psi, dispensed 30 grams of food, waited x amount of time, shook and measured the psi, dispensed 30 grams, and so on through 8 dispenses. Hopefully you see how I ended up with 8+ values representing one Y.
    The objective of the group, which I learned after they ran the DOE, is changing the overall slope of the line, so that more gas is left to push out the food in the last few dispenses. This is what led me to look at the slope of the line, calling that our overall psi drop. When I ran an analysis on this, the only significant difference noted was the y-intercept, impacted by the quantity of gas (mass) used in the experiment.
    I’ve concluded that they focused on the wrong variables, and have since gone back and C&E’d what impacts the slope (total psi drop).
    I want to know if I should have analyzed this differently. Note that I did look at regression earlier in my analysis, finding that the slope of the lines for each run of the experiment were similar. From there, I created a well correlated linear regression to represent the slope. This is how I came to recognize that they are focusing on the wrong variables.

    0
    #191467

    Robert Butler
    Participant

    The measures described are repeat measures and, since it sounds like you do not have a package that can run repeat measures analysis, you did about the only thing you could do – take the difference between the psi measures of the first and last test of each experimental condition (the simple slope) and run an analysis looking for correlation between the design factors (mains, two ways, etc.) and these individual slope values.

    It sounds like you did this and it sounds like no significant correlations were observed. If they did all of the things they should have done in the D and M phases of the project and the choice of the variables in the DOE resulted from these efforts then it isn’t a matter of choosing the wrong variables, rather it is a matter of the best efforts identifying variables that, over the ranges examined, do not have an impact on the response of interest. While frustrating the results should not be forgotten because you now know, with a high degree of certainty, that these variables do not impact this aspect of product performance. This kind of information has the potential to save a lot of needless effort at a later date when the next issues concerning product performance have to be addressed.

    0
    #191468

    MBBinWI
    Participant

    Jeff: Where in cheeseland are you?

    0
    #191469

    Euston
    Participant

    lol. I’m in Denver. Did you have the impression I was in WI?

    0
    #191470

    Euston
    Participant

    Your exactly right about the destructive nature of the test and I’m glad to hear some reassurance about the analysis.
    Valid point on the variables….that is a solid way to articulate those results.
    Much appreciated.

    0
    #191471

    MBBinWI
    Participant

    Cheese (even spray cheese) and dabrewa (the brewer?) lead me to propose an hypothesis. Clearly didn’t meet the 0.05 threshhold.

    0
    #191472

    MBBinWI
    Participant

    You have reached the limits of the technology under the conditions used. Either you need to change the conditions (larger volume, higher starting pressure) or you need to change the technology. I can think of a couple of ways, but will leave that to you.

    0
    #191474

    Severino
    Participant

    Shouldn’t your response variable be the volume or mass of food dispensed at the last dispense rather than the pressure change? You’re performing a whole DoE on the teams “belief” instead of on what the customer actually sees. I would consider all dispenses up until the last dispense conditioning of the units. If that screening experiment determines that pressure is the key difference, then you can focus on the variables which impact that final pressure.

    How exactly do you define the “last” dispense?

    0
    #191477

    MBBinWI
    Participant

    Hey, Robert: How’s your knowledge of GRR?

    0
    #191478

    Robert Butler
    Participant

    MBBinWI….it looks like you’re asking me…if you mean Gauge R&R I’ve done some but it isn’t an everyday event. If you have a question go ahead and post it and I’ll see what I can do.

    0
    #191479

    Euston
    Participant

    Jsev607:
    Good questions. First and foremost, I wasn’t privy to the DOE setup or any of the work leading up to this experiment for that matter. That said, I do know they aren’t necessary concerned with mass of food as much as some other attributes of the food they measured post-dispense (proprietary information I can’t share). We were able to correlate the attribute measurements to the gas remaining in the headspace, enough to conclude we needed to determine what factors impact the pressure drop…this is what its tricky because of the ideal gas law. Most of us believe they have to find other ways around this through innovative packaging, gas mixes, etc.
    To answer your question….the last test did dispense the mass they wanted, but the ‘performance’ was poor enough to not give the effect they were looking for.

    This is all good stuff….

    0
    #191480

    Euston
    Participant

    MBBinWI…I’ve had a lot of practice with the tools encompassing the MSA, incl GRR. I’m with RB here…post it and we’ll all see if we can help.

    0
Viewing 15 posts - 1 through 15 (of 15 total)

You must be logged in to reply to this topic.