Scaling DOE Collaboration with JMP Live (2025-US-30MP-2548)

Welcome and thank you for your interest in our talk today. My name is Wendy Zhang and I will be co-presenting today with my colleague Bill Worley. We both work with customers who are striving to increase the impact of the DOEs in their organizations. This topic is of great interest to us.

Many of you are scientists and engineers that are conducting designed experiments in your organizations to innovate. Or perhaps you work with colleagues who are doing this in your organization. These statistically designed experiments, commonly referred to as DOEs, are valuable sources of learning. That's why we do them to learn about a new system or process, and to use that knowledge to improve and innovate. How do we ensure these valuable DOEs are being shared with our collaborators and stakeholders so that as an organization, we are getting the most from them? Our collaborators need to be able to readily access, interact, provide input and make decisions using these experiments seamlessly. In today's session, the focus will be on how our impact as scientists engineers can be bigger when we can share, consume and collaborate on DOEs in JMP Live.

JMP Live is a complementary tool to JMP that enables enterprise wide collaboration and knowledge sharing. We are going to share two stories that are based on organizations that use JMP Live to scale DOE collaboration. These two stories will be described to you using examples from a fabricated company. Aquapure. Aquapure is a chemical company that makes raw materials for other companies in their industrial business unit and consumer products in their consumer business unit. This first example is a story of how Aquapure's product development and process engineers in the Industrial Business unit collaborate with JMP Live. Product development is responsible for developing processes to make new products and improve existing products. They use DOE to identify important process variables and understand how they work together to achieve specific product characteristics. When they have identified the process factors and ranges necessary to make a high quality product, they transfer this knowledge to process engineering who sets up the process in pilot to make the product at pilot scale. Let's take a closer look at how Aquapure's product development and process engineering groups collaborated to bring a new solvent, PureSolv solvent to market.

Product development leads for PureSolv conducted several DOEs in JMP in the development of PureSolv's final pilot process. If you're not familiar with these cube diagrams, they are often used as representations for types of DOEs. The cube represents the process base being evaluated, and the dots represent the experimental runs. When developing a new process, it's common to conduct several experiments. The first experiment, screening, is all about identifying the important factors. That experiment can then be augmented to study how factors interact with each other and then a final experiment may be focused on honing in on the process space that results in the best performance. Aquapure product development engineers final DOE focused on optimizing for process variables to achieve a yield greater than 90% consistently. Product development use this DOE to develop set points and ranges for volume, time, temperature, and catalyst to make PureSolv batches consistently above 90% yield. Let's take a closer look at how the product development engineer accomplished this.

Thanks, Wendy, and I'm going to share my screen now. Does everybody see my screen?

Yes.

As Wendy said, we did the optimization or the optimization DOE was done. From there we wanted to take the next step, and we developed a least squares model that looked promising for producing batches with greater than 90% yield. We opened a prediction profiler to show that the process may be what we expected, but we needed to do some what if scenarios to see what an optimal process could look like. The first thing we wanted to do, we needed to check a couple of things, and we wanted to first use something called the extrapolation control in the prediction profiler. We were able to quickly see that we may not have enough DOE runs at the lower combinations of volume and catalyst, to be sure of what would happen if the yield would happen to yield at those settings. That is something product development wanted to investigate as the process matured. We ran the optimization step and everything looked good, but the production engineer stated that it would be pretty much impossible to run the exact settings predicted in the optimization due to process variability.

They asked if there was any way to get some wiggle room on the settings to be on the safe side and have a more robust process. With that, we decided to try a newer tool in the prediction profiler called the Design Space Profiler to check out their concerns. The first thing we saw was that the input ranges from the original DOE, the process would only be InSpec around 6.3% of the time. We wanted to make sure that everyone knew about this, and we talked about that to the process engineers. We were going to share with them as best we could. The next step in the process for the Design Space Profiler was it allowed us to use a Monte Carlo simulation with 10,000 simulated points to adjust the ranges of the design of experiments, so that the process engineers would achieve an InSpec batch with up to 99% confidence.

Thus giving us process settings that would yield 90% or greater for every production run. You can see that over here where our lower spec limit was 90, and then by adjusting the design space profiler or playing with the settings, we were able to get that value up to about 99% certainty that we would have InSpec batches every time. Now I'm going to show you a quick demo. I'm going to step out of this. Escape. Minimize that and pull over my data table. This is the data from the design experiments. We were able to do virtually everything that we needed to with this data and then communicate that to the process engineers as needed. I wanted to build a model and that we're talking at standard least squares. We've got all our factors set up, and we're going to get run, and we get a pretty good setup. I'm going to clear some things out here real quickly just to get us to a better spot.

I get rid of that. Maybe a couple more, and then we should be good to go. We've got our model set up, we've got our prediction profiler, and we have our desirability setup as well. Let's pull this down a little bit, so we can see things a little better. The first thing we wanted to do was check out that extrapolation control. That's in the prediction profiler. This is a JMP Pro tool, but we felt we just wanted to let you see what it's all about. It's an important not interaction profilers. That's on me. There we go. Interaction extrapolation control. We're just going to turn the warning on just to see where things might be a little bit iffy. If we look at our volume, we drag that over to the lower side and drag temperature to around the lower side. Then we can start seeing things that it's looking a little bit iffy and then catalyst as well, if we drop that down to around five.

Of course, it's not working. There we go. It's showing us the extrapolation. We got a possible extrapolation at the lower levels. That's something we would want to investigate as we go forward. Let's turn that off. Let's go back to the prediction profiler and turn on Design Space Profiler. Now this could get a little bit long in the tooth if we do this manually or if we just let JMP do it for us. I'm going to show you a couple of different things you can turn on here, and we're going to make and connect a random table. We're going to use 10,000 points, and we're going to embed factor space scatterplot. Now we can see... We don't need this table in the way. Of course, it's going to be pain in my rear end. There we go. Sorry. We've got that. Now we are going to play with these tools over here. We've got the space 10,000 points.

We want to see what we can do about getting this InSpec Portion up to that point where we feel that we're very confident at least 99% of the time, we're going to get a yield of 90% or greater. Let's just play around with these a little bit. Dial this in. If we really wanted to we could actually just input the values that we know will work. We're going to play around with this just to show you. As you can see, the space gets smaller and smaller in the scatterplot and the factor space. As we dial these in a little bit more, you can see things get better and better. We're already up to 58%, almost 59%. We can play around just a little more to get some more understanding. Let's do this. Let's move the catalyst out a little bit. We're up to 86% and maybe a little bit more. What we did was we got this up to a 95% or higher, 99%.

Let's see if we can go a little bit higher and get the 99% value that I promised. Nope. Going the wrong way. There we go. We're almost there. Now we've got these values, and they're a lot tighter than they were before. But these are the values that we would shift to, or would share through JMP Live with the engineers. With that I'll turn it back over to Wendy.

Great. Let me steal the screen from you. Bill's demo showed the behind the scenes of how the product development engineer used JMP to derive the process ranges. Once this work was complete, the product development leads were able to share those results with process engineering in JMP Live directly from JMP. In addition to sharing the specific process factors and recommended ranges, they provided the results of the DOE to process engineers in interactive output. The same interactive output that you saw in JMP with the prediction profiler was shared with the process engineers, and they could access this interactive output and JMP Live using a browser.

When Aquapure did not have JMP Live, process engineers received limited information statically in email and PowerPoint. Let's go to JMP Live to take a closer look at how process engineering used the interactive DOE results to make data driven decisions. Using a web browser, we are accessing Aquapure's JMP Live portal. Let's take a look at the DOEs in the folder dedicated to PureSolv in their industrial business unit folder. This folder is permission to all the engineers working on PureSolv.

Let's take a look at the final yield optimization DOE as the process engineers did. The Design Space Profiler communicates how the process ranges were determined and helps build confidence in the ranges. For example, we can see why the ranges for volume are recommended to be between 2.5 and 5.2 by looking at the Design Space Profiler. Here's the graph specifically for volume, and we can see the vertical dotted lines represent that range of 2.5-5.2 that's recommended. We can see this is where the highest percent InSpec portion is accomplished. In this case, InSpec is really speaking to that goal of making product consistently above 90% yield. Scrolling up to the prediction profiler. The prediction profiler gives process engineering the ability to explore the process space that had been mapped out by the DOE. For those of you who are unfamiliar with the prediction profiler, it's a representation of a model. You had a chance to see Bill build this model using fit model and standard least squares. In this case, the model is representing the DOE results how volume, time, temperature, and catalyst connect to yield.

This enables process engineering to ask questions of the DOE. It allows them to understand trade-offs with making changes to the process variables. Let's take a look at the profiler and see what we can learn from it. For example, we can see by the slopes that temperature and catalyst are much steeper than volume and time. This means that yield is very sensitive to changes in temperature and catalyst as compared to volume and time. We can see this not only from the slopes, but by using our mouse to change in this particular case. The value for temperature, and we can see how that is reflected in the yield predicted value. Now let's use the profiler, as process engineering did, to solve two specific challenges that came up. The first challenge came during the initial setup of the process to make PureSolv in pilot. Process engineering was interested in increasing the starting volume from the recommended upper limit of 5.2 to make more product in a single run. Using the profiler, now I'm moving the slider that changes the value for volume.

They were able to see that they could increase volume up to around 6, again above the recommended setting of 5.5 and still achieve yield above 90%. The second challenge presented itself when process engineering was about to hand off the process to production. There was a significant price increase in the catalyst used in the process, so process engineering was asked by leadership to see if they could reduce the amount of catalyst used in the reaction without significantly impacting yield. Again, with the profiler, they could lower the amount of catalyst and make a prediction of what would happen with yield. They could see that they could drop catalyst lower than the recommended value of 8.5 and increase temperature and keep yield close to 90%.

Let's summarize what the process engineers could do with the DOE results in JMP Live. With the Design Space Profiler, they could see how the process input ranges were determined with the Design Space Profiler that connects the process ranges to the percent in specification. They could also understand the sensitivity of yield to incremental changes in volume and make changes independently with confidence.

They could also make a swift data driven decision to lower catalyst because they could see the impact of yield per unit change in catalyst, and also see catalyst dependency on temperature. Before the organization had JMP Live, the process of considering alternatives and making changes was slow because process engineers had no visibility into the DOE results beyond ranges that they were provided. Any adjustments they wanted to make had to be done by relying heavily on email communication with product development, and as a result, the process engineers did not feel empowered to make decisions. Let's go back to JMP Live to see how these two teams communicated within JMP Live. We're back in JMP Live looking at that final optimization DOE. If we click on to these comments, we can see the conversation that occurred between product development and process engineering during the setup of the pilot process.

You can add mention your collaborators, and they will receive email notifications indicating that there's a question that's being asked. Let's recap the first story.

In this first story, product development engineers shared the DOE results with process engineering interactively in JMP Live, process engineers explored the DOE and were able to make decisions efficiently and effectively because they could conduct what if analyses and understand the trade-offs independently. Product development and process engineers collaborated inside of JMP Live transparently to all team members. The second story comes from R&D formulators in Aquapure's consumer products business. R&D formulators use DOE to develop new products and reformulate existing products, and JMP Live is how these DOEs become a source of knowledge and a place for collaboration. To hear the second story, follow the QR code link to watch the on demand webinar. This concludes our talk.