Extracting More Information Out of Your Moldflow Insight Results

JAY SHOEMAKER: Good day, everybody. I'm Jay Shoemaker, and today, I would like to talk about extracting more information out of your Moldflow results. So if you who don't know me, I've been with Mozilla and Autodesk for about 36 years, I've done all sorts of things from project engineering, customer support, training certification. I have a book through Hanser called The Moldflow Design Guide, and I was honored with an award from SPE Mold Designer of the Year for 2011. I've since left Autodesk and I've been with iMFLUX for about a year. And I'm a senior simulation and development engineer. And concurrently with both positions, I've been an adjunct professor at Western Michigan University, where I helped teach plastics classes.

So the description of what we're going to talk about today. So we've all run hundreds if not thousands or even tens of thousands of analyses, and we usually run several analyses to optimize a part. And we found that often, we'll have subtle changes in results, sometimes making it a good design versus a great design. And we also know that minimum and maximum values don't necessarily tell the whole story. It's distribution of results that is important.

So case in point is volumetric shrinkage. Say you have a range on the part that's quite wide between 1% and 10%, but most of the part is between say 2.5% and 4%. Now graphically, you can see that on the screen by the way the variation in colors, but it's a little hard to quantify. So as you know, we have scripts in synergy where we can do anything we want through this API in scripting. And so we can export data. And I'm exporting it so I can pull it in to a spreadsheet and use histograms.

So in this class, we will be exploring the use of histograms for looking at multiple results. So the specific learning objectives is to really discover the power of using histograms. We'll look at the APIs that are involved for extracting the data. We'll look at how we take that extracted data and get it into an Excel spreadsheet with that. And then probably most importantly is we'll look at how to interpret that data.

So many of you may not be familiar with iMFLUX, so I wanted to give you a brief introduction. So iMFLUX, we really want to innovate and transform the plastics injection molding industry. We are a wholly owned subsidiary right now of Procter and Gamble, and as such, iMFLUX was started to solve some problems within P&G, but we've since broadened our horizons to more than just P&G, and we work with customers around the world.

So if we look at the history of iMFLUX, as I mentioned, we are from P&G, and probably everybody uses a P&G product whether they know it or not. And here you can see some examples of P&G products. And like many companies, time to market is everything. We need to get our product to market faster than our competitors, and of course, we want to do that with high quality and the least capital investment with that.

So that was the challenge that was formulated around 2011. And then after thinking about that challenge, we looked at, in about 2013, how to do that. And we really started with the concept of aluminum tooling. So we all know that it's easier to machine aluminum versus hard tool steels, and so that was the initial thought. And one of the first things we came across with the concept of aluminum tooling is that injection pressures were kind of high.

So we started looking at ways to reduce injection pressures, and that's really the starting point for the company that is now iMFLUX. So through the 2013 through 2016, we really started to perfect what is now the iMFLUX technology. In 2018, we really started to publicize it through various medias and papers and so forth. And since then, 2019 on, we've been growing quite rapidly.

So there is actually a number of parts of the iMFLUX business, but I really wanted to concentrate on the iMFLUX technology itself. Now this presentation is not about the technology, but I wanted to introduce it because that was in part, led to the motivation for this presentation. It is a low constant pressure molding process. So really, it turns 50 years of best practices of injection molding on its head. And I've been with the company about a year, and I'm still getting used to and wrapping my head around the technology, but it is just wicked cool from my perspective. I just have a lot of fun working with this technology.

But from a company who uses iMFLUX, there's lots of benefits. But the main thing is because injection pressure clamp tonnages are lower, we can reinvest that pressure reduction in a number of ways. As an example, we can lower the press size that we need to mold a given tool. Or we can up cavitate. Instead of a 16 cavity tool, we can make a 24 cavity tool and run it in the same size press. It's just one of many variations that we could do.

But that's not what we're here for-- to talk about iMFLUX-- I just wanted to give you that introduction. We're really here about extracting data. So my motivation for this wasn't iMFLUX directly and its technology. It was looking at volumetric shrinkage results as one of those key results related to iMFLUX. And for the decades that I've been teaching others how to use Moldflow and writing the training material, we've talked a lot about how to look at volumetric shrinkage, how to make it more uniform in the part.

And we know that we typically have a wider range of volumetric shrinkages than is desired. And we have techniques for volumetric shrinkage optimization, which wasn't necessarily my goal here. But I wanted to understand this range of volumetric shrinkage. So as I mentioned before, the values of volumetric shrinkage are often difficult to quantify. And I'm mostly interested in looking at one analysis to another. The variation within one study is important, but it's really between studies is where I think we have the biggest bang for the buck, and it's important to understand.

So the maximums and minimums, while important, don't really tell all the picture I want. It's the distribution that matters. And I mentioned that before, but it's worth repeating. I think if we sit back and think about our own journey with looking at Moldflow results, we sort of get it, even if we haven't really thought about it in those terms.

So I started thinking about a way to look at this data and histograms came to mind. So histograms, if you're not familiar with them, are a statistical tool that we basically group numbers of data points into bins or buckets of similar value. And so these histograms work for numerous results. I've mentioned volumetric shrinkage a number of times, but it works nicely for other things like time to reach ejection temperature, flow front temperature, shear rates, frozen layer fraction mold temperatures, and hold pressure.

These are all things that I have done, most of them quite regularly. I've only done shear rate maximum once, as it turns out, just a few weeks ago, but it was quite insightful for the problem I had it in hand. And I'm definitely going to keep that in mind for the future.

So what's the procedure involved? It starts out with an API script. It turns out I had drafted this script some number of years ago and just never did anything with it. And it was just a pretty straightforward, no bells and whistles script that took volumetric shrinkage at ejection from a dual domain or midplane model and dumped the data out in a tab delimited file. And since then, I've updated that script a little bit, but it's still a fairly barebones script, nothing all that fancy. But once I have it into a text file, I'll copy that information and put it into an Excel spreadsheet.

Now I could have this API directly within an Excel spreadsheet. I have not tinkered with that yet. I'm still tinkering around with the process and ironing out some minor issues that I have. It boils down to that I don't necessarily want all the data that's exported-- for instance, volumetric shrinkage and hot runners. There is no volumetric shrinkage and hot runners.

Turns out the data is stored in a very special key number. I don't think it's exactly the same, but the values are e to the 37-- I mean, a huge number. And kind of convenience it's always at the top of the file. So what I'm doing is I'll delete that information while it's in a text file before I pull it into Excel. And there's some other reasons why I do that. But as I mentioned, I'm interested in more than one study. So typically, I will do it for one study and then copy the Excel page, and do it again for another study.

So kind of fine tuning the procedure a little bit. Many of you may be familiar with the assigned macros panel. It's normally on the Tools ribbon. I've moved it or copied it, I should say, to the results ribbon so it's handier. And I've developed Excel templates for every different type of data. So volumetric shrinkage, frozen layer fraction, time to reach ejection temperature, mold temperature, and et cetera are all different templates because the ranges of the data is a little bit different. And I have other simple statistical analysis I do with it that's appropriate for that data. So I have a different template. But then I execute the script and copied the data in.

So here, I've got a little video that runs through the process. So I executed the script. You can see it just took a second. I'm copying the data, going to my template and pasting it in. And my template automatically has the histogram on there.

Now I've put in this text box where I have some interesting information, and I manually change those numbers. Well, I haven't updated that little video, but now those numbers are automatically populated. I just did that in the last couple of days.

So it's just really simple. And in this case, I copied the sheet, I went over, got another study, pasted it in. And now I'm going back and forth between the studies to make sure it's similar. And then once they're similar, I'm copying and pasting both histograms to the same page. And there were some areas on that video that I sped up, but the basic process for going from synergy, starting the API macro, to populating the histogram is typically under a minute. It's the subtle manipulations and so forth take a little longer.

Now I mentioned this basic script was for volumetric shrinkage and ejection. That is my original script, but to get the other results that I talked about, I needed to make special versions. And I've got two other special versions. One is any single data set result. And as you know, a single data set result is a result that we only have one value for the analysis such as time to reach ejection temperature. It could occur during the filling phase, the packing phase, or both, but it's one value.

And then we have an intermediate result as well, such as average volumetric shrinkage. That happens to be a 3D result, where we have multiple values at different times for that given result. The data is stored differently. And again, these are barebone macros. I haven't gone through the process to make them a little bit more sophisticated where it would extract the data in the appropriate method. I've just gone through and said, OK, I'm executing the correct macro for the type of data. Again, it's simple.

Now to execute these three, it's really the same way. Again, the first one is just a single button push. And I get the data, and that text file has the study name, a couple of blank lines, then the data. For single data set results, I need to tell the type of result, and I do it by a plot ID. Well, guess what, I don't remember the plot IDs, so I have a couple of different ways to give that number. I entered the plot ID and go. Same way with intermediate results. It's really the same input, the same list. I just have to pick the appropriate data.

So the downside is if I execute the single data set script and I give it the plot ID for an intermediate result, the script will run, but the text file will have nothing in it. No big deal-- you just run the correct script. Again, it's quick and dirty. Not all that sophisticated.

Now for the fun stuff. How do we interpret these results? So results interpretation for a histogram realistically is not much different than any other type of result. I don't know how many times I've looked at-- you too, probably-- you're looking at a result, but you don't really have any clear objective or reason for looking at it. So you go through the motions. You turn on the result, looks pretty, red spot here, blue spot there, who cares, go on. So you don't get much benefit from that result.

So like anything else, we need to have a clear problem or objective. Like other results, it kind of helps if you understand something about the model which you're pretty much going to anyway. But in the context of here and the reason I put it in is, for instance, a volumetric shrinkages. And we all know plastic part rules is we want a uniform wall thickness, and we pretty much know no plastic part has a uniform wall thickness. There's always thick areas, thin areas. The question is are they really thick or really thin?

But the context of volumetric shrinkage-- the greater the thickness likelihood, the higher the volumetric shrinkage. So if I know I have really thick areas, I'm going to assume that when I look at my volumetric shrinkage histograms. But if I really didn't understand that I had a thick area and I look at a histogram that has some outliers, what's going on. And then you, of course, would investigate why, but it's nice to have that knowledge ahead of time.

I am really picky when I look at any result from one study to another. I want to make sure that my scales are the same so my color ranges represent the same thing. Or do whatever is necessary to tell the story I want to tell. For histograms, that is the scale of the x and y-axis. So the x-axis is pretty easy. It goes from some value, and I've set up my API script in my Excel template to leave rooms where I can put in a value manually so that my x-axis for all my data is the same.

As an example, my lower right graph, where it says set pressure of 19 megapascals, I have a value of 1 for both the right bin and the left bin. So that histogram would be scaled the same as all the others. And that is important.

I do that for the y-axis as well. Quite frankly, for me, my brain cell gets taxed every once in a while, and I forget to go in and scale the y-axis the same. So I'll make this pretty picture and notice that the y-axis is different, and I can cuss at myself, go back, and set them up the same. So that I struggle with. It's my own issue.

And sometimes, I'll change the color of the bars. And I can do that two ways. I can do it inside Excel, where I can select a bar and change its color to whatever I want. So what I'm doing here in this particular example is I'm representing 2% of variation in volumetric shrinkage. So that's sort of my target, is 2%. And that graphically shows me what a 2% range is so I get a graphical representation of what's outside that range.

I can also do it in SnagIt or any other screen capture editing program where I can set the color I want, click, click, click, click. It's actually faster, but I'd still prefer it to do it in Excel so it's a permanent record.

Now how do I actually interpret? We sort of talked about the big picture, but how do we go through and interpret? I mentioned before, we need to have an objective. And here, it was comparing a conventional process versus an iMFLUX process for this part. And the interpretation is, geez, the results are nearly the same thing. Now the one thing that is interesting that we'll come back to and think about-- the conventional process, the part is slightly lighter. I mean, splitting hairs here-- it's hundreds of a gram lighter than the iMFLUX part.

Now usually, iMFLUX allows us to save 1% to 3% on weight. So one of the things this suggests to me is my iMFLUX process may be able to be modified to lower the part weight. And of course, I want to keep a reasonably uniform shrinkage. But guess what-- the volumetric shrinkage on average would, in fact, be a little bit higher.

So let's look at another example of volumetric shrinkage. This time, I'm looking at primarily changes in set pressure between different iMFLUX processes, going from 16 to 19 megapascals. Now in this particular case, I just batched them all up, made the changes, kicked them off, and ran it and I was curious of the differences. And I see the changes and the trends I would expect. I was wondering if I'd have more variation than I ended up doing. For me, that was sort of a little disappointing.

But anyway, it is what it is. So the left upper and left middle are really the same two studies we saw in the previous one, where the top left is a conventional process, sort of my benchmark. And the middle left is the one where it's an iMFLUX process where actually, they had a set pressure at 16, then jumped it up to 25 to pack out the part. And that's where my part weight was slightly higher.

So let's take a look at something other than volumetric shrinkage. And this is temperature at flow front. Now for those of you that know me and have maybe even been to one of my training classes, I really talked a lot about figuring out what your proper molding conditions-- mold temperature, melt temperature, injection time-- is for a part. And typically, injection time is by far the most important and the most sensitive.

So I really like molding window. I do it all the time. A lot of people want more detailed information, so they'll run a regular fill analysis with different injection times and look at the results. Well, that's sort of what I did here. So doesn't matter what set of analyses-- molding window or a regular fill analysis-- to judge injection time, I'm looking at temperatures for one. And with a full fill analysis, I'll look at temperature flow front. I want that to be fairly uniform. And next slide, we'll look at another result.

But looking at these three times that they have, 2 and 1/2, 1 and 3/4, and 1 second, they're all reasonable. But it's clear that the 1 second is far more uniform. And this is a perfect example of scaling my x-axis. So my interpretation would be a little bit different if I let the x-axis scale automatically. But by having the same x-axis scale, it's really, really clear how uniform one second is compared to the other two.

Now since I have a fill analysis, I can also use frozen layer fraction as a guide for determining an acceptable fill time. So at 2 and 1/2 seconds, it actually isn't too bad. Of course, with frozen layer fraction, we want things at the lower end of the scale closer to 0 rather than 1. 1 and 3/4 seconds is better, and 1 second is better yet with only about a tenth as a variation. But if we look up at the 2 and 1/2, it's still well within acceptable limits. It's way under 0.25. So it's acceptable.

This time, let's look at time to reach ejection temperature. So here, I have different geometries. The top one is the original geometry, which has a thick cross-section. So if we look at my little text box there, it kind of summarizes things nicely. Is our average is 12 and 1/2, our 95th percentile is kind of high at 35 seconds, and our maximum is at 41. And again, that summarizes the histogram, which, of course, summarizes the graphical results pretty nicely.

And again, using that same scale, I reduced the thick section, and show that I've gotten rid of those higher cooling times. So my maximum is less than 26 seconds, down from 41. And my average went from 12 and 1/2 to 9 and 1/2-- all good things.

And this is a scenario-- how many times have you done an analysis, and your customer says, you can't change, you can't change that, and you know making a change is going to really solve the problem, so you run it anyway. That's the rebel in me, that's what I do. And I think better than just time to reach ejection temperature graphically, this result supports hey, we really need to make that thickness change. It supports that argument, I think, better.

So now let's look at hold pressure. Now for those of you that don't use hold pressure very much, you can be excused, because it's not a default result. And so quite frankly, I sometimes forget it's there, and so don't look at it for a while. But it can be a handy result. And so here, we're looking at a conventional process on the top, an iMFLUX process on the bottom.

And the hold pressure result is the pressure on the part, its maximum value after VP switchover. Basically during packing or during holding, hence its name. And the conventional process has a really, really tight range. Just a couple of megapascals between min and max. And then you look at the iMFLUX, you'll say, jeez, this isn't very good. It's lower values, and it's not a tight distribution. I definitely agree with that interpretation in terms of a wider distribution and lower values.

Now what struck me on this is one of the benefits of iMFLUX, which is just wicked cool to me, I absolutely love it, is we know if we create short shots with the conventional process-- I've got a favorite part that I mold at Western, and when we mold it in short shots and propylene, we see tons and tons of orange peel, because propylene shrinks a lot. And it's really cool to see, and especially new students who had never seen it before, they kind of think it's cool. And then when we pack it out, we represent the mold surface really well, and the part looks a lot better.

Now what iMFLUX does-- it packs as it fills as part of the process. And so I've done it here a couple of times. It's just neat to be able to do it myself-- is I'll have a part. Conventionally, I'll get that orange peel very heavy shrinkage in that short shot. Do a short shot of about the same size with iMFLUX. It looks like a fully packed out part away from the flow front. It's absolutely amazing to me.

Well, this histogram is an analytical version of that. It represents that very well. Because I've taught in training classes, it's in the training documentation for Moldflow that the pressure on the part when it freezes off determines the volumetric shrinkage. Well, traditionally, I've thought about that like a dual domain or a midplane part, where you have a single value for a given element, or region on the part.

But as soon as we do that same part in 3D, we know the story is a little bit different. We have really, quite a wide variation of volumetric shrinkage through the thickness going from maybe even slightly negative at the mold wall to a high value in the center-- say, 6%, 7%. Where because we pack as we fill with iMFLUX, more of the part is freezing at a higher pressure, ultimately lowering the pressure when it freezes off. But it has a lower volumetric shrinkage. And I should have put it on this slide, but these are the same two studies where this bottom study has a higher part weight-- mind you, a hundredth of a gram-- higher than the upper study. And that was just sort of mind-blowing to me, is this histogram kind of quantifies that packing as we are filling concept.

Now looking at a different result mold temperature. Here, I've got a part where I made a simple water line's first pass, mold design is not done. So I just had simple water lines, same coolant temperature on the cavity core, 85 degrees C. And we get a distribution. It's not great. It's not horrible either. But I recognized that the core ran hotter than the cavity after looking at the results graphically.

So I went through, modified the coolant temperature on the core actually quite a bit. Dropped it 30 degrees C, ran the same cycle. And the results are a bit better. My average is lower and the amount of the part that's within 10 degrees C is better. But both of them have a spike in temperature of around 145, 155 degrees C. And so this histogram now is giving me a clue of the outliers. And this allows me then to take this information and scale my graphical results.

And as it turns out, for this part, this high temperature was in a slide that isn't directly cooled by the cavity or core, so we'd need to add cooling in a slide, as difficult as it would be in this particular case. If I hit control over the design, I'd use a high thermal conductivity alloy to take care of that.

So we've talked about a number of things. We've looked at the use of API in synergy to extract the results. I get it into a text file, which then I copy it into an Excel template. And this template automatically has the histogram chart there. And so we put the data in, it automatically populates and creates the graph. And then once I get one, I'll copy the sheet that this histogram is on, make a new copy, go to Synergy, copy in another one, and keep going.

And I've probably done this where it'll have 10 or 12 different histograms in the same file. That gets a little busy if I tried to look at all 12 together, but it allows me to really look at this data critically so I can take appropriate action, whether that's looking at some of my results graphically in a different way to understand the problem. Or it helps me understand the problem directly where I can change the process, change the PART whatever it needs to be.

So with that, I hope I have sparked some questions for you. For those of you that know me, I love questions. I love lots of teasing and everything. So please do. And again, thank you very much for this opportunity to talk with you. Please ask questions now. If you have questions in the future, please reach out. As you know, I love to talk to everybody and talk anything and everything Moldflow. So with that, again, for the third or fourth or 10,000th time, thank you, and have a phenomenal Autodesk University. Thank you.