Choose Language Hide Translation Bar
Level I

Reducing the Fit Variation in Denim Jeans Manufacturing (2021-US-30MP-892)

Level: Intermediate


Anuja Khairnar, Quality Control Systems Manager, Surry Chemicals Inc
Helmut Hergeth, Associate Professor, NC State University

Lori Rothenberg, Associate Professor, NC State University


VF Jeanswear experienced a decline in sales due to the fit inconsistency in its jeans. Specifically, customers complained of inconsistent waist sizes and irregular inseam length in the same size jeans. The objectives of the research were to identify the sources of variation in the manufacturing process that led to poor fit, design a solution to reduce that variation and implement that solution. The tools used in JMP were Gage R&R and DOE. Several experiments were performed throughout the manufacturing process. Shrinkage variation was the outcome of interest. The factors were fabric type (cotton, bi-blend tri-blend), wash type (rinse, stone wash, hot bleach, cold bleach), washer type (high-speed extraction washer, no extraction washer), dryer type (dryer with humidity sensor mechanism, dryer with algorithm mechanism), type of laundry (product development, industrial), and mill (several mills). The results of the study indicated that by increasing the bi-blends and tri-blends the variation in fit was reduced by 23% in men’s and 12.5% in women’s jeans. To improve the fit further, VF needs to work with the mills closely to help everyone understand the variability in raw materials beforehand.



Auto-generated transcript...




anujakhairnar Hello everyone.
  This was a project that I did during my master's at NC State under the guidance of Dr Helgeth and Dr Rothenberg. And it is a pleasure for me and Dr Rothenberg today to present this to you and the topic for this project is reducing the fit variation in denim jeans manufacturing.
  So this is basically the agenda for today would be. I will give you the background of the company that I work for on this project.
  And the introduction will cover what this project entails, what's the manufacturing process, what methodology we used to reach our results of experiment or design of experiments that we did, and finally the recommendations and solutions that we had for this project.
  So a little bit about VF was that they wanted someone from
  the curriculum to come in and help them and use their technical know-how and our academic know-how to work together and to come up with a solution on this project. But just an overview about VF itself.
  It's a $12 billion apparel company...apparel, footwear, and outdoor company, and it was founded in October of 1899.
  It has nearly 70...70,000 associates in over 170 countries across the world, and this project was when VF had gone to brand which included in them, and this was a jeanswear project.
  kids, women, men,
  old people, everyone, just no matter who it is, everyone loves denim. And over the years, it has improved, changed its facets from being something that you use for dirty rugged work to something really fashionable to be walked on a ramp with. And because of the changing
  needs of denim there have been changes in processing, but this processing has effect...has resulted in an issue which is called the fit variation in denim and that's what we are going to look more into.
  So the purpose of this research was to investigate fit variation in the VF jeanswear denim.
  to assess that where in the internal manufacturing was the variation seen, and to come up with recommendations and solutions to reduce this variation, and finally implement it.
  So the manufacturing process for denim jeans starts from fiber. And this fiber goes into the spinning stage, from spinning it goes into the yarn dying stage. This yarn dye is then weaved with undyed
  fibers and formed into a denim fabric. This denim fabric is finished with either singeing(?) and
  basic stuff like that. Then it is go on to pattern cutting where the ??? cutting creates a pattern for the jeans to be made.
  This pattern is later sewn on and these finally sewn jeans go in the wet processing, which is the washing, drying and finishing process, and then, finally, it goes to the retail market. That...this is a just for denim manufacturing.
  And shrinkage is a big big portion in denim manufacturing, as it has cotton in it. Along with cotton, it has
  a yarn, ??? yarn, which which shrinks and that's why shrinkage is a big factor in denim.
  So the initial problem statement for this was that VF Jeanswear denim jeans show large deviations from the specification limits for the waist and inseam.
  The team believed that this contributed to the lost in sales. The objective was to study what was the reason for this variation and to reduce it in the waist and inseam.
  The baseline work for this, after we defined initial problem statement, was to see the waist variation distribution among
  the range of production...production denim and then comparing the range of fit between the sew...after sewing process and after finishing process...after finishing audit data on 100% cotton denim and stretch denim.
  Then we had to come up with the cause and effect diagram on JMP, which helped us to understand which factors to focus on. And JMP also help in making the process flow chart for the entire denim manufacturing.
  This was the data which I uploaded to JMP to understand what was the initial waist variation that we were seeing to define the problem
  realistically or factually. And here we have the means...the range of denim for male sizes 32 waist and 32 inseam. And as you can see, the range of fit is varying from
  up to 3.24 inches. That is the amount of variation that the denim has in the sizes 32 to 32.
  This means that a denim, which is supposed to be 32 inches in the waist can range anywhere between 29 to 35.
  That is what the range of fit means. And same with the female sizes. We just considered sizes 10 and 12 and when we measured the range of fit amongst the sizes, we do see
  a difference of three inches here as well. This led us to believe that definitely there is a fit problem which we can quantify and we defined our problem statement.
  So this is when we compared the data. We wanted to go more deeper and to find out where the actual variation occurs.
  And we collected audit data from after sewing of denim before any wet processing. And then there is a lot acceptance data, which is basically before the denim leaves the warehouse for retail.
  And as you can see here from the graph that the lot acceptance range is way more than the plant audit range.
  This lets us know that something is during the wet processing after the denim is sewn and before it is actually sent out in the retail market, something over there is causing this variation.
  And we see the similar results in stretch denim. The only thing is, we do see higher variation, a higher range in the stretch denim, which led us to take stretch denim as our basis, instead of 100% cotton.
  So the redefined problem and objective was VF Jeanswear denim jeans showed the range of fit for men's waist as 3.24 inches and showed the range of fit for women's waist as 3.12 inches.
  The team believed this contributed to a loss in sales. The objective for this study was to reduce the range in waist variation by .5 inches by April 2019.
  This was the cause and effect diagram which we made in JMP. We used a certain
  identifiers, like people, what procedures that they're performing that could be wrong, or there could be an issue in the measurement area, or just the plant, which causes...which leads to variation.
  This was the process flow chart of the entire manufacturing
  floor. And starting from the bale of fabric until packaging, every small detail is covered in this process flow chart.
  And this is what I used to make that cause and effect diagram. And I divided it based on human error, high variation zones, the selected zones for my thesis project, and the measurement errors zones.
  So the plan of attack to work on my selected area was to work in the red processing, which was washing and drying of the denim. And for this, we did a Gage R&R study and design of experiments.
  The Gage R&R study was performed to see that the people who were going to measure these fabrics that I was going to perform the study on, it was consistent. That these people could repeat it and reproduce similar results within a very small confidence interval.
  So here is what we chose. We chose 12 different fabric swatches and we chose three different operators who were the only ones who would measure this all throughout the course of the project.
  And we asked every operator to measure each fabrics square three times in random orders. And that's when we put this data in JMP to get the Gage R&R study data.
  So this is what JMP gave us back after we put the data in it, is that the part to part variation was about...
  part to part reproducibility and repeatability was about 99.98%. That is how accurate they were in measuring the data, which led us to believe that whatever procedure we...whatever experiments we were going to do here on would be calculated correctly to the 99.98% level.
  And this the other one was fulfill measurements, and this is for the warp measurements, but it was the same data.
  Now these were the experiments that we designed, where we took three different fabric levels, we used four different wash types, we used two different washer types, and two different dryer types. And this resulted in 96 runs on JMP.
  And this is just the division of what fabric codes I used for this experiment, in terms of
  what were the different fabrics, which were the recipes, what was the...what were the names of the washer that I used for this experiment, dryer, and the are total different measurements, which was after wash and after dry.
Lori Rothenberg I'm going to talk about the results of the experiments and I'm going to go to JMP to show you an example of one of the experiments we analyzed.
  Let me switch over to JMP and I'll show you where we get these results from.
  Here's what our data table looked like. We had designed a full factorial using JMP and we collected data on after wash fill, after warp...wash warp, after dry fill, after dry warp and I won't go through all of those.
  What I would like to do is show you an interesting problem that we came up with when we had to look at after wash fill. To go ahead and analyze this data, we go to analyze, fit model.
  I'm going to go ahead and take after wash fill, make that my Y.
  I'm going to take fabric type, wash type, washer type, and dryer type. Those are the four that were on the video that...or on the previous slide that Anuja mentioned. I'm going to go down to macros and do a factorial to degree.
  So this will include all our two-way interactions. I'm going to change the effect screening to effect leverage, so we can look at a few of the plots and now press run.
  Here we are. When we take a look at the effect summary, it's right below the first plot, what you can see, in the final column under P values is that we have some things that aren't significant.
  Here in JMP we started at the bottom with our two-way interactions, and we went ahead and started to remove those.
  We've got rid of that first, here was our next highest P values, so we selected that and removed it. We left dryer type in because it's a main effect. We looked at our next two-way, which was fabric type by dryer type and we removed that.
  And then, if you take a look up above, we no longer have dryer in anything, so I can go ahead and we clicked on that and removed it.
  At this point we checked our residuals, and we did have...the variance was fine, the data were normal, but if you look under lack of fit we had a lack of fit, and so we had to think of something else to do.
  I won't show this to you right now, but what happened was fit...worse fit has a higher negative number.
  And so we had to start making transformations and transformations aren't going to work on the negative numbers, so we flipped the scale, just for this one. And so a higher number means worse shrinkage.
  So we have that column in here, so I can go ahead and just show you then how we did a Box Cox transformation on it. I'm going to go back to analyze.
  And I'm going to select fit model.
  I'm going to go ahead and recall what I have, because that's so quick. I'm going to get rid of after wash, remove that one and, instead, I have the absolute values, actually, of the afterwash fill here. So I'll go ahead and put that in and now I'll run it.
  And now I'll do the same. I'll go ahead and drop those two-way interactions that aren't significant.
  And you'll see washer type, dryer type wasn't significant. Again I have to go to fabric type and dryer type.
  And then finally I'm back to dryer type and you can see dryer isn't occurring above so I can go ahead and select that and remove it.
  You'll notice in the lack of fit, I still have a problem.
  And so what we did was, we went ahead and went to up above to the red triangle next to response. So we did factor profiling and went to the Box Cox transformation.
  At the bottom, we saw the Box Cox transformation curve and, if you take a look at it, the best lambda it's showing is .329.
  Well, if you look at this, I know that's the best line, but we kind of have a little bit of a range to go ahead and work in.
  And so we did. I won't bore you with doing all of this, but what we found was that actually .25 did the best.
  And I'm going to go to the red triangle next to Box Cox transformation and I'm going to refit with transform. And now the lambda that I'll go ahead and use here is going to be 0.25 and I'll click OK.
  Now take a look at what happened. We do have our residuals in good shape, and if you take a look at the lack of fit, now we're in good shape.
  With that, we went ahead and looked at the interactions that were significant. And we found some interesting things.
  I'll start over with...let's start with fabric type by wash type.
  We went ahead and looked at the least squares means plots and so for...
  for VF they liked looking at the plots and they were pretty, you know, understandable. I'm going to go ahead first and create the interaction plot and I'll have fabric type.
  I'll click OK, and here you can see the usual least squares means plot when you have an interaction. Green down here is cotton. You can see that cotton
  had the least shrinkage and up here in blue, you can see...that's our tri-blend and it had the most shrinkage and that was across wash types and those were the ones that were
  rinse and stone and hot bleach and cold bleach. And so you can see there were occasions where we did have things that
  weren't significantly different, but sometimes we want to know what those are. Real quick, before I leave it, let me put up one more least squares means plot.
  I go to the red triangle next to fabric and I select least squares means plot. This time I'm just gonna leave that unchecked. This is an interesting way to look at the data also. And if you compare these two, you can start to see right here, if we take a look at our
  our bi-blend and then you look at our wash type there, you can start to see the patterns. So up here, we also had
  in our red, we had our bi-blend and, if you look here, it also takes the same shape. It's just another way to visualize the data and we use that too. Now, we also did the statistics on this. We went ahead and did the least squares means Tukey.
  And we get a nice big matrix, but we went down below and we could see from the nice chart that has all the letters
  on it, which ones really were significantly different from each other and which ones weren't. And what we see is here, in fact, that cotton still winds up having the least shrinkage, as we saw, but also we have this bi-blend...
  bi-blend that also was in a rinse that also wasn't significantly different from one of the cotton's that was also in our rinse.
  We repeated that. We did that also with our other interactions so we could go ahead and see what was going on with fabric type and washer type. We did the same steps.
  Im just going to show you the least squares means plots here. You'll see some more on the slides that will come up.
  So we could start to see here again that cotton, regardless of the washer type, had the lowest shrinkage. You can see up here the tri-blend had the most shrinkage and then, if I take a look over here at wash type and washer type,
  here we go. If I go to the red triangle and I select least squares mean plot.
  Here you will see there was an interaction also between the wash type and the washer type. And so, if you take a look here at the bottom, LB is rinse. It had the least amount of shrinkage.
  And then, if you start looking over here with LT and ST, LT was the cold bleach and then ST was our hot bleach, so you can kind of see our highest point is right there. And then likewise over here, if you take a look where at
  BT was hot bleach, sorry, LT was our cold bleach and ST was our stone wash.
  And I'm going to go ahead at this point, I'll stop demonstrating JMP.
  And I'll go ahead and bring you back to our results slides and turn that back over to Anuja, who can explain some more of this. I'll flip past...we included in here slides for everything that we did, just so that you would have a copy of it,
  in case you'd like to have that. So these were all the things that I showed you, and then we got into after dry and Anuja, I'll give this to you.
anujakhairnar Yeah so even here, we can see that fabric type, wash type and the interaction is significant.
  And these were the results, similar to what Dr Rothenberg just performed on JMP. We performed the same on JMP and we got this data, and here we can see that the bi-blends and 100% cotton and the tri-blends, all three did show variations, but moreso in the tri-blends and the bi-blends.
  So this was just another study we did, where we got batches from different mills, to see if that was a cause of the variation there.
  And yes, we surely did, because the PO number was basically different batches and we can see that it is significantly different.
  And this is what we got, the results. The bi-blends and tri-blends, you can see that it's not linear. We can see the variation in this, which gave us a good information that
  we needed whenever a different batch comes in, we need to needs to get tested for to see in which bucket it falls in, especially what shrinkage bucket it falls in.
  And then we also did different dryer temperatures, if that caused a difference, if that caused an effect, because there was no difference between the dryers as much.
  And yes, we do see again that there was a significant difference in the dryer temperature.
  So 180, if you see the numbers are inverted, but 180 caused the maximum shrinkage, and 140 and 160 caused almost similar shrinkage and also lesser.
  There was one more study we did, where we had product development washers and industrial washers and dryers.
  And we had two different fabric types, three different wash types and the different washers and dryers.
  And here we do see that the washer type, after wash, we saw a significant difference which told us that between different types of washers there is a variation.
  And, similarly, we put the data in JMP and we found the least square means plot...means plot for the wash type and washer type.
  And here we can see that in rinsing, the LB is the rinse part, we can see from whatever we have concluded from our older data that rinsing caused the least shrinkage and the cold bleach and hot bleach caused the most shrinkage.
  And this was the industrial after wash results. And we can see, even in this data that there was a difference between rinse and cold bleach and hot bleach, and the fabric type and washer type do to show significant difference.
  And similarly, we ran the test on the dryers.
  And these were the after dry results on the product development
  And we just put all the data that we need from earlier...
  from earlier.
  And we put it together in a chart. And we can see here the PW1, PW2 and the industrial washer 1 and industrial washer 2 with after wash and after dry, we saw up to 12% difference between these dryers and washers.
  And these were the improvements that we
  recommended. Before this trial was done, only 40% of the fabrics which were coming into the production were tested for shrinkage so that they could be put in different buckets.
  And after this improvement what we recommended was all of the bales that come in need to be tested, and it should give us a significant difference. And the other recommendation was that we needed to use PDL and industrial machines and...
  we needed to use PDL machines over industrial machines and the industrial machines need to be monitored and given more maintainance.
  So this is how we put it for them, the changes that were to be done and which were
  implemented. From 40% to 100% testing and switching from industrial to PDL.
  And these were our results. We can, if you see the graph, we started this study in August and we gave them results, we kept giving them improvements. And if we can see that Feb and March...Jan was where my project ended with my
  work project ended, and when I gave the recommendations. And you can see that Feb and March, we do see a significant
  drop in the range. And my project ended end of March, so that's why I don't have any data for April.
  And this was the conclusion and future scope. So because JMP helped to get so much details that is that we had a range of 3.24 inches and
  when we recalculated it after the improvements on the later months, we reduced it to 2.49 inches, which is about 23% decrease in the range.
  Our idea was to get at least .5 inches so we succeeded in that. And similarly for women's waist, we reduced from 3.12 inches to 2.73 inches, which is about 12.5% reduction and the
  reduction is lesser because they have...there's more stretch in women denim, like that's just why it still has slightly higher variation.
  And we also...because we saw a significant difference between the batches that came from the mills,
  the other improvement led them to contacting the mills and asking them to improve their quality assurance and quality control, which, in turn, resulted in better results for production of jeans.
  Then, as the PDL and industrial machines showed about 12% difference in shrinkage on an average, it gave us more
  details, I think that was. And dryer temperatures did not give a significant data in all the JMP trials that we saw, dryer temperature never had a significant P value and that's why there was no need for us to switch.
  And tri-blends definitely needed to be studied further, as we saw, it always gave us a varied data, so these were our future recommendations.
  And there were a lot of limitations in this experiment, because of the time we did not have as much time to run it. And we did not have as much data because
  the factory was located in Mexico, so every time you had to run something, you had to travel there to get the data, come back, read it. So the location and lack of data and also no constant access to the information, which hopefully those are the limitations for this project.
  And this is my acknowledgement of all my advisors and everyone who helped me there. Any questions?
Article Tags