Hello. I'm really excited to be presenting some of the work we've been doing at Jotun Paints, and I'm thrilled you've chosen to watch this recording of the presentation, which I'll be giving at the Discovery Summit. I'll be talking a little bit about who are Jotun to start with as a bit of a general introduction, what are intumescent coatings, and why is it important we understand the nature of the furnaces in the work that we do here in the Intumescent Department. Also, how can we do this using JMP and utilizing some of the more advanced analytic JMP Pro features?

My title is: Understanding the Positional Temperature Trends to Increase Testing Reliability Using JMP Pro. My name is Jay Richardson, and I'm an R&D chemist here at Jotun in the UK, and I've got a real interest in leveraging R&D data to make the most impact and value in the most efficient way possible in order to improve the processes in R&D.

Who are Jotun? Jotun are a global coatings and paint company which operates in four segments: decorative, marine, protective, and powder coatings, with each of their own solutions and innovations in these areas.

Jotun is headquartered in Sandefjord, which is a town just about an hour out of Oslo. It was established in 1926, mainly focusing on marine coatings and ship coatings, as this was a massive industry. It's then further globalized to over 100 countries and own 68 companies with 41 production facilities.

What is an intumescent coating? An intumescent coating is a coating that expands and is reactive to fire, and it forms an insulative barrier, like you see here, to coat the steel and protect it from the heat source within a fire.

In the Intumescent Department, we're in the business of creating market-leading intumescent products for the protection of steel to fire. As I showed in the video, and as you can see on screen here, the coatings themselves will swell up to 60 times the original thickness of the coating itself, and this allows critical time for the rescue services in order to arrive at the scene for the assets to be saved and personnel to be recovered from the building.

I think it's very important, and everybody knows the consequences fire can have, and our job is really crucial to protecting assets and lives on a day-to-day basis. It's a really crucial part of buildings.

One of the biggest challenges we have is trying to understand the nature of our testing, because it might not be obvious, but in order to test our products, we need to burn them. Our test follows standard certification type temperature profiles, which are monitored by temperature sensors called thermocouples. You can see one here in the furnace sticking down as a wire.

How these work is as the thermocouple heats up, the resistance within the wire changes, and so a certain current passage, which is proportional to the temperature within the furnace itself.

In 2023, our facility was expanded significantly, commissioning two massive new furnaces and essentially quadrupling our test capacity. You can see in 2022, we had two identical small cube furnaces, and with our expansion, we've added an extra two large furnaces. Our vision was in 2024 to be able to test at full capacity using these furnaces.

But in order to understand that we really needed to be able to understand and be confident in the reproducibility of each test, but also that the temperature profile was as expected for the test itself. We did this by temperature mapping the furnace, by looking at different positions within the furnace and seeing how the temperature trends differed. We were quite limited within our testing, given that the furnace itself had very specific preset drilled holes, because it's not possible to just add a new hole here, there, and everywhere, and set up a very conventional DOE type design. We were quite limited.

The other fact to consider here was that each test is very expensive. The amount of gas and the amount of product that we have to put in, to the fire tests themselves can cost up to £5,000 in order to run from start to finish. You can see that the testing itself is very expensive, but also we need to expand how much we know about it.

How it works within the testing is we had six thermocouples from the top of the roof, which monitored the temperature of the furnace and controlled the temperature as per the standard curve that we run. Then we had 15 surplus thermocouples, which monitored the temperature profile at specific positions within the furnace.

We need to really understand the reliability of these thermocouples, as I said, and the nature of the furnace itself in order to really gain confidence that our formulations we're comparing are reproducible, or the results are comparable. It's really important that we understand these trends when we implement new testing, we produce the most accurate product test we can from the work we do.

If I just... Sorry. Let's come back to that. The problem we have is that conventional analysis isn't really fit for the detail we're putting into the furnaces because our data is based on, firstly, a time-temperature type of profile, but also it has a positional variant. We have a 3D diagram of where all the thermocouples are here. It's not easy to track the reproducibility of each of these individual thermocouples.

We need to build more complicated models in order to map these trends. For this, we deployed the Functional Data Explorer as a dimension-reducing technique by fitting a model to describe these temperature-time curves in terms of their principal components. This enabled us to more easily interpret their positional component, respectively.

Functional Principal Component Analysis would then be used to understand the relationships between the temperature, time, and position, and really understand the reproducibility within that test.

Using this model, the temperature can then be simulated using FDOE to better visualize the positional temperature variations and use Scatterplot-3D to aid communication to other R&D chemists who potentially aren't as interested in the detail or maybe confused by the prediction profiler, but also manages to see clearly what is the impact of the work we're doing.

I know this is a lot of information, so I think a demo is due. The first step within this is to generate the JMP table with all of the different components, so the positional components and the thermocouples with the time and the temperature, with the rows colored by the thermocouple number, which is equivalent to the position of that thermocouple. The position of the thermocouple stays the same test to test.

We then can use data processing to look at what the general trends look like on first perspectives. If we look at the mean function, this sits very closely to how our standard temperature profile curve would run within the furnace. If we look, we can see that there is a differential relationship between the mean function and the standard deviation; which is what you'd expect, because if you think early in the test, we're producing a lot of heat to try and ramp up this temperature, so we can expect a lot of deviation within the furnace. Then as the temperature starts to plateau, we can see the mean deviation starts to reduce, which is what we'd expect it.

In this model, we utilized B-splines, and we can see that the mean function is best described by the combination of the shape functions, FPC 1 and 2, equating to 98%. Then using this PCA and the score plots… We can then use the score plots to assess the consistency of each run by the thermocouple position, which as I said before, is based on the color.

If we look at what the score blocks look like, it's quite obvious that we have a close relationship between the thermocouples in the same position between tests. Test 1, test 2, test 3 are all in the same color for the same thermocouple. This allows us to see that there's good reproducibility between the tests in the same position, and we can have confidence that if a test specimen is within that position, it can expect the same temperature profile. It shows good reliability and confidence within that.

We can take this one step further and look into the model from the FDOE profiler. We can see that early on in the test, we have the most variation, and then we can see later on this variation starts to plateau slightly. We can also see between tests, this same relationship that there is minimal movement within the X, Y, and Z components, and the temperature component of the two curves showing the same trend.

The other information we gained from this is that the positional variation position on the X-axis appears to be consistent. Left to right positionally in the furnace seems to have the same trend or there is not a massive difference. But when we look at front to back of the furnace and top to bottom of the furnace, we see that there is a trend. We see that at the back and at the bottom of the furnace, we have the hottest temperature, and at the front, at the top of the furnace, we see the coolest temperature.

This is all well and good, and people who are very aware of how JMP works and all of the different profilers might understand this detail, but it might be more difficult to people who maybe aren't as versed in JMP and all the analytics. The next thing we can do is we can use the FDOE Simulator in order to make this more clear and simulate the model itself to 10,000 points.

We can see here we have a simulation in the JMP table. What we can then do is use this simulated data within Scatterplot-3D with a local data filter set to time. We can visualize how this data looks in three dimensions and with a time component. You can see here we have an X and Y-axis, as you might expect, and a Z-axis going backward. What we can do by moving the time is we can see how the model predicts the furnace is running and how the temperature profile will look based on the model that we've run.

We can see that early on in the test, we have the most temperature variation between 400 degrees, but as we continue through the test, this plateaus, and we see less of a temperature variation between different time intervals. This gives us a good visual representation to show to managers and others that might not be as keen on looking into the analytics too deep. It's very visual with the temperature being colored.

Essentially it shows all the same things that we saw on the FDOE Profiler. We see this same trend from top to back bottom down here, and we see left to right that there doesn't seem to be a great deal of difference between the two if we look between here and here. We can say conclusively that we do see trends in the furnace, and these are things we need to be mindful of when it comes to development.

The reason we expect to see these trends is because we have the extraction system at the back bottom of the furnace, so that tends to pull the heat of the furnace down towards itself. This fits with our expectations and shows that where we have the most heat is potentially away from the specimens, which are at the top of the furnace.

FDE in JMP Pro can be used to efficiently understand the reproducibility of fire tests, giving confidence in the reliability of development with minimal test cost. As I said at the start, the tests are £5,000. If we were to look at this from a traditional standpoint and try to understand it, I think the amount of cost we would have to endure in order to understand these as deeply as building this JMP model would be a lot greater.

It also gives us great confidence and reliability that we've relied on a data-driven approach to make our conclusions. We've built a model in JMP. We understand all of the different parameters that are associated with it, and we're not relying on someone's belief or anything like that. It really gives good confidence.

It's clear that in JMP, visualizations are absolutely key to communication between people that might not be as interested in data, but also managers. Using the FDOE Simulator in conjunction with Scatterplot-3D has really given us a great way to visualize our data in a 3D positional way.

This analysis has allowed us to better understand our furnace for development and reliability, and we understand that the effects we make when we develop, and we formulate our coatings are based on the changes we make within the formulation rather than being based on any furnace variation or test variation. It gives us that confidence to say we can trust the work we do. Which is important because when we talk about fire protection, and we talk about building safety, this is critical that we're confident in our results, and we can put those in places where they're much needed.