Find the Optimum Seal Strength of Packaging Using DOE

Six Sigma – iSixSigma Forums General Forums Tools & Templates Find the Optimum Seal Strength of Packaging Using DOE

This topic contains 7 replies, has 6 voices, and was last updated by  Sergey 8 months, 3 weeks ago.

Viewing 8 posts - 1 through 8 (of 8 total)
  • Author
  • #55950

    Hi All,

    I need your advice,

    Currently in my company, to find out the optimum setting of setting is still using traditional method using matrix approach to assess the ability. Example:

    Temp. Setting 195 200 205 210 215
    Sealing Time
    1 X X X X X
    2 X X X X X
    3 X X X X X
    4 X X X X X
    5 X X X X X

    I will collect 30ea/setting and then measure the seal strength in four location on the pack. So, each sample will have 4 data of seal strength and the total is 120 data if I collect 30 sample (pack). After that, the result will be used to determine the window setting. Currently, we only use simple statistic e.g min, max, average, stdev. to determine the machine performance.

    Here are my questions.
    1. Is it enough to use above statistic tools to determine machine performance?
    2. How to use capability analysis where each sample have 4 data? my company has minitab. How to do it?
    3. Is there any methodology or tools to simplify the testing? I just find that we can use DOE to make lesser trial and now I still try to learn it.

    Thanks in Advance


    Robert Butler

    As I understand your post you are looking at a matrix of sealing temperature and time where each of the two variables are at 5 levels. Given what you are doing I’d recommend a simple three level 2 variable factorial (time and temp at their respective lowest, middle, and highest values). This would be a total of 9 experimental combinations. There are at least one and perhaps two major issues you will have to address.

    Issue #1: Setting Randomization. My experience with anything thermal is that it takes time to change a machine from one temperature setting to another and the time needed to insure a stable temperature level before running samples can be excessive. If this is the case then you will probably be forced to run your temperature setting in sequence from low to high – which means no randomization on temperature. If you have to do that then you will at least want to randomize the time settings within each temperature setting.

    Issue #2: Sample/measurement independence. If the samples are sequential within a given time/temp setting they will be repeated measures, not replicates, and the 4 measures within each sample will repeated measures within a sample. If you don’t have the ability run a repeated measures analysis then you will quite likely find significant effects where none exist.

    If the temperature settings must be run in sequence then after running the entire sequence you will need to reset the temperature to the lowest setting and run a replicate of as many of the three time settings at that temperature as you can.

    If it is easy to randomize the combinations of temperature and time you will still want to replicate at least the very first setting of time and temperature that you ran in the initial experimental run and, if possible, at least one other temperature setting.

    In both instances the output will still be repeated measures but you will have a second population which can be contrasted with the results of the first run(s) to give you a better estimate of the actual variability associated with the process. You can use this estimate of variability when examining the results of your analysis to make a judgement call with respect to the actual significance of the differences in results between the various time/temperature settings.


    Chris Seider

    Consider a basic 2×2 factorial with centerpoints.

    Are you sure you have good precision on seal strength? It would be a waste of time if you didn’t have good precision to detect changes in the output to get statistical confidence.


    Drew Peregrim

    I ran a seal testing DOE at a medical device company. The biggest factor is seal pressure. It turned out that the max pressure of the equipment was the best. I now use that DOE in my MBB lectures. The results showed that time and temperature settings could produce an optimum seal at max pressure, the caution was that just a little higher in either produced very low seal strength. (essentially a cliff with the strength falling off dramatically near the optimum settings. For robust design, the settings were backed off from optimum so that a 4 sigma process was created.

    I started out with a 3 level factorial design and created response surfaces. I then created additional levels based on the initial results to focus in on the critical area.



    Have you considered using the testing just as you said and calculating a process capability for each of the four locations? If you want to assign one capability number to the process, you could select the lowest of the four since you ultimately want the sealing process to produce seals within the tolerance limits at all four points.


    Thank you @rbutler for your advice, I think the issues that given to me is very useful.

    Hi @cseider, our measuring device is calibrated every 6 month. According to your statement, I think you have some advise for our validation activity that you can share.


    Chris Seider

    I’m saying if you look up gage R&R for variable data, you’ll see you can measure how much precision you have in the instruments.

    Seal strengths are often impacted by “setup by testers” so variability of the device may not give you good enough precision to know best settings. If you don’t have good precision, your statistical confidence may not be enough to know which combination of settings are good.

    Nice follow up, A.F.P.



    Hi @aufafadhli
    It might be you have an excessive reinsurance for seal strength testing. We’ve done some DOEs for that parameter to understand the whole system, based on my experience it should save time and effort. In case you have historical data, you may easily understand the possible variability in a system, take few samples from current process just to check whether it is not so far from historical data and create a simple design (2×2+cpt or 3×3) or consider so called split-plot designs (having hard-to-change factors). It is more rigor in dealing with issues @rbutler highlighted.

    If you use Minitab, some reference to thermal processes example provided here:

Viewing 8 posts - 1 through 8 (of 8 total)

You must be logged in to reply to this topic.