267 - Modelling Fire Growth with Stephen Welch

Wojciech Wegrzynski: 0:02

Hello, everybody. Welcome to the Fire Science Show. If you've listened to the last week's episode, know compartment fires are super Uh, it's the essence of fire science for me. And if you've followed me for a longer time, you I'm also very much into modeling fires. That's, uh, that's a part of my career that, also a big chunk of me. And if you combine those two things, you get compartment fires, and here we somehow get into space where we are thriving with our design fire and have been doing that for ages, and the whole is able to, to do that pretty much. And we have sort of predictive fire modeling, spread modeling, fire growth modeling, really know how to do. I've talked about that with Jonathan Hodges a, a time ago about the differences between fires using the design fire concept and modeling a true fire growth in, in a compartment, and completely vastly different things. And even though there are tricks that allow you to kind of make a model that seems like the fire is in it, The physical mechanisms that, that growth, they're not really easily done in CFD, and they're, they're extremely difficult to, to, validated. They are extremely difficult to make sure you get out of that model is, is something remotely close to a real fire. And there's a very, very, very few academics in world who actually can claim success with, uh, fire growth. And as you can imagine, Dr. Stephen Welch from University of Edinburgh, who uh, the guest today, i- is one of those people. He and his team have shown tremendous, uh, successes with modeling, uh, traveling fires, recently timberline fires. a lot of them using CRIPS are mostly related to CRIPS as a continuous, uh, fire source. We'll-- more on that in the episode. But, uh, still, that's, that's a huge, huge step So today I invited Stephen to figure out how, how are they doing that? How difficult is that? What steps are they doing? What's important to them? And, uh, you'll learn a lot of interesting things about validation of CFD and, and trust in, in this episode. I loved the conversation. I hope you will enjoy it as well. So let's just spin the intro and jump into the The Fire Science Show podcast is brought to collaboration with OFR Consultants, a independent consultancy dedicated to addressing safety challenges. OFR is the UK's leading fire risk consultancy that this year celebrates its 10th anniversary. As experts in fire engineering, they are fully to delivering preeminent expertise to protect property, and the environment. With over 30 chartered engineers and a team of researchers at their core, they continually explore the challenges that fire creates for their clients and society so that the best research, experience, and diligence can be applied for effective solutions. In 2026, OFR will grow its team once again and is keen to hear from industry professionals who want to collaborate on fire safety features this year. Get in touch at ofrconsultants.com. Hello, everybody. I'm joined today by Stephen Welch from of Edinburgh. Hey, Stephen.

Stephen Welch: 4:08

Hello, Wojciech.

Wojciech Wegrzynski: 4:10

Yeah, you're very welcome in the podcast, and, especially that you're touching a subject that's close to my heart, that's modeling. And actually, uh, what I've seen from your group a very unique type of modeling because you guys to do fire spread modeling, and wow. I know a lot of people claim that they do that. I don't think a lot of people do that. So are you modeling fire spread or just Steven? Tell me straight on.

Stephen Welch: 4:37

I mean, we, we need to be honest about the, the here, and it's a tough, it's a tough problem. but I, I think, um, you, you have to be self-critical and carefully examine the capabilities of your models. and in that respect, it's not just looking at colorful dynamics. so yeah, I mean, there, there's, there's two major of a-areas of, of, of study here, I guess. There's the, um, uniform, um, fuel beds in, in large This is all driven by the interest in traveling fires and understanding what controls the dynamics in these but also recently, less published from our group so far is, uh, an interest in timberline compartments.

Wojciech Wegrzynski: 5:19

Right.

Stephen Welch: 5:19

again, it's quite easy to, to run FDS and, and make that looks impressive. It... By default, it's gonna look impressive. You'll have lots of... Well, pro-provided you've chosen the right parameters, you'll, you'll have a, a big fire. but is it, is it a fire that re-relates to reality is, is the question, uh, that we need to explore. So yes, a lot of this is, validated against experiments. I think that's where you need to start. and in doing that, you need to say, "Well, what, what, experiments, you know, do you start at TGA? Do you, do you do cone calorimeter modeling?" We've we've done all of this. Um, do you go to crib scale or do you, do you, do you, do you go direct to compartment scale? Uh, so there's lots of things to discuss and explore,

Wojciech Wegrzynski: 6:02

Yeah, you know, for, for me there, there was a uh, that was in, paper by Professor Blokhin, 50 of Computational Wind Engineering." cite it directly from the head, but it went the line that the worst CFD is not one that's not converged or, has slightly wrong assumptions, but one that looks good enough to look believable but is wrong, you know? And, uh, I, I, like along the lines what you said there, i- if you put some settings to FDS, you we, we have the pyrolysis models out there. There's like settings you can tweak, temperatures of ignition. You can get that stuff to ignite and burn and in the image and, you know, you can drop in heat rate curve on them and it's gonna get you Sometimes it's gonna get you pretty damn far. Yeah, but, it's a different story if you have one case you want to represent and you have an data and you want to match that one single point. And another thing is to have a model of which are confident that you are putting in an unknown fuel that has never been burned in that and the results will be more or less real and,

Stephen Welch: 7:15

It's a very, uh, ambitious challenge and I think, one, one of the first, papers I ask new students, had many students studying this problem, uh, over the uh, who've done all the hard work in running the but I, I always ask them to read, uh, Jose's, scaling up paper. It's not actually explicitly a, a CFD paper, but it you know, put- putting together fundamental them together, at a, a small scale and then building physics in that you can then get the thing to work when you move up to the, the, the scale of the large and, are conditioned by the environment that they exist in, so just, just the, the wood crib, so much has been with the wood crib, burning rates in wood cribs and, uh, fire spread in the open. But put it in a compartment and you, you, you have a complex problem, is coupling the, all the heat transfer, the, the fluid, uh, mechanics, the ventilation and your combustion, your soot, your under-ventilation. Um, so it's a, it's a fascinating, environment to explore with modeling tools. Uh, but yeah, you, you have to be critical and, and, zoom in on, you know, what, what can the model do? Can we systematically validate and explore, what it, what it's good at and, uh, and where it has limitations? So I'm always more interested in what, you know, what discrepancies, if, if you run the model and you find that doesn't match, you know, what, what's causing that? can we understand better how, how to, to push things

Wojciech Wegrzynski: 8:46

l- let's now move to the opposite end of Let's assume we have this solved, like there's an easy model people just put my fuel is wood this many sticks. Let, let, let's, let's assume we have it as, as possible. W- what would the world look like? What would that give us and a- allow us to do that we cannot today?

Stephen Welch: 9:08

Yeah, I guess this is back to, um, how we use traveling fires as a boundary condition for structural fire Yes. And I mean, the other element of, the scaling up paper is the emphasis of scenario uncertainty. So, you know, if you run the same fire, or even, you a, a slight variation on a, on a fire, you repeat it you'll get different results. Uh, and, and of course, the same comes out from the You, you make a small change and, and you get a very uh, result. Um, but yeah, I mean, the purpose of, of that modeling is to look at the boundary, the fire structure what kind of timescales and severities are we the structure. Of course, there's a whole, raft of empirically derived or, analytically based or phenomenologically based, models that can do part of that problem. And yeah, I mean, uh, they have their, their use and their role. I guess your question is if the CFD was perfect, what does that enable us to do? I guess this, it's not just the, the structural, uh, boundary condition that we're, we're interested in of the, massive interest in, timber-lined, uh, and the prediction of burnout. That's a-an-another kind of, arena where we would like test model capabilities. Of course, FDS might give you a prediction of burnout, it what, what, what does it mean and in, in relation to what assumed fire? So yeah, I mean, I'm, I'm talking around a, a complex Uh, you, you also need probabilistic, methods. There's no point designing something for, for one fire. So even if your results are perfect, uh, and want to that parametric space of, okay, so if I change the fire load density, if I change the fire load distribution, I change my opening, uh, locations, uh, what kind of might I get? And then, feed that then into the, the, the structural

Wojciech Wegrzynski: 11:06

if you go to the early first traveling fire like Jamie's model, then Egle's variant, et I understood it since from the beginning as a way to generate a lot of different plausible scenarios for the structure. So you have a lot of, you know, variants to go and see and, and by that means explore very exposures to structures and perhaps some of them critical, which Egle in her PhD shown a lot of

Stephen Welch: 11:38

Yes.

Wojciech Wegrzynski: 11:39

Yeah.

Stephen Welch: 11:43

No, that's, that is not, not realistic to, to be using

Wojciech Wegrzynski: 11:51

but, but for example, if I think about now, like add window breakage probabilistically to that how many experiments you would have to run? And you've run large experiments, I've run large Even if we join all the fire avengers of the planet, we don't have resources, people, and equipment, capabilities to run all the experiments we would to explore the full parametric spectrum. Uh, movable ceilings. Uh, have you ever found a building suitable for built in your lab where it was not movable?

Stephen Welch: 12:25

significant, uh, sensitivities as well, the ceiling obviously, and the ceiling thermal properties. but, um, yeah, I mean, you're absolutely right. We can't, we can't study all of, all, all of the So, um, I think the CFD is being used to provide into the burning, uh, regimes, and the mechanisms that fire spread. Of course, there have been, as you're well aware, a series of experiments, full-scale experiments in traveling There have been theoretical developments, uh, to, to what extent is, is the fire plume, the fluxes the plume, uh, versus the hot layer controlling the spread under different conditions. And tho- those are great because they get you a, a kind of overview of the kind of expected sensitivities. But the, the CFD, uh, does enable you to do more than you can do experimentally.

Wojciech Wegrzynski: 13:16

Mm-hmm. Especially in the parametric space and just

Stephen Welch: 13:21

that's, that's what it's perfect, uh, to do. So yeah, I mean, I, I mentioned, in the scaling up paper, um, there's a strong emphasis on, on scenario which we need obviously to bear in mind as a small could produce, drastic differences in the real, But, for the CFD, the comparative, studies are probably the most interesting. So you make a small change, you move the opening you add a- another layer of wood sticks, you change the soot, the rate of fraction, the soot concentration, ceiling thermal properties, and you observe what And the beauty of the, the CFD is, uh, n- not only do have the full detail of every, you know, burning surface, every heating piece of structure and piece of timber, you can extract the macroscopic parameters. So you c- you can look at the, the typical preheating, look at the typical, flame, flame heating. Um, it's not always easy to decouple the overlapping, but in principle they're all there. And this, I think, gives fascinating insights, into really causes the fire to, develop in different ways.

Wojciech Wegrzynski: 14:24

and one thing that you will never be able to do an experiment, at least not easily, is to decouple the co-contribution of the ceiling versus if you, if you consider t-timber lined compartments. Like, good luck in a real experiment to decouple, you know, the contribution of the structure from contribution of your removable fuel load. I mean, we've done it once with Danny, uh, when we weighted the ceiling as it was burning out, but was a, that was a hell of an effort, and it was, it compartment, but small from the perspective of fire, you know?

Stephen Welch: 14:55

Yeah.

Wojciech Wegrzynski: 14:56

very, very challenging. anyway, I think the rationale is, is now well set, so, uh, we know that it would be amazing to have and, actually the, where you got already is uh, it, it's like giving us huge confidence that can be done. where did you start when you were starting that? I, I assume the first one was the uniform fuel beds, then you moved to the structure. may-... Yeah.

Stephen Welch: 15:27

Yeah.

Wojciech Wegrzynski: 15:27

with the professionalization of firefighting UK.

Stephen Welch: 15:31

I mean, in terms of, uh, modeling flame spreads, yeah, I, got started getting involved, uh, in that, back in the century, you know, when we, we were using critical flux models, in SOFI, and exploring fire spread in, in scenarios. And, and in retrospect, I'm thinking that, that crazy. I mean, we were, were, we were too ambitious. but the, the current work, the origins of the current are pretty much in a project at Edinburgh called Re-real Fires for Safe, Safe Design of Tall Buildings, uh, when we ran some full-scale fire tests. they're called the, the Edinburgh Tall Building Fire ETFT.

Wojciech Wegrzynski: 16:06

Is, is that the one that ha- that's on YouTube, the story of building the rig, et cetera?

Stephen Welch: 16:10

yeah. That, that was wonderfully documented with, with Yes. Uh, and, and they were fantastically designed Uh, and part of my role in that project was the a- as you can appreciate, this is back in, uh, we had the, the first student being confronted by modeling this enormous compartment full of wood sticks. You know, where, where, where do you start? Should we s- do some cone tests? Well, actually we did, uh, Juan did, uh, the, the cone on, on those sticks, so we, we know how an individual of timber burns that was used in, in the test. But so one of the fascinating things in all this work how, how differently does a stick burn toward, you know, a flat piece of, of timber in a cone or on a, on a, on on a, on a boundary and they, they actually burn in ways. So the early students, uh, I give I wouldn't have to mention all, all the, all the names of people who on this project, but was Hongbo Zheng and, Penny Yang, they were running, these simulations of wood crib so the wood, the wood crib variants on the test series. There were two, two wood crib tests. they were running them on our local clusters on Eddie and Archer using quite a, a decent number of meshes processes. Um, and I insisted we must have more than one cell per

Wojciech Wegrzynski: 17:26

Oh.

Stephen Welch: 17:28

with two, two cells across the, the stick and tried to some grid sensitivity studies, but I think, they were, were quite constrained, uh, by computational in that era. But that was the pioneering, kind of in- initial, this topic. And, you know, the results kind of convinced us that, okay, uh, it's predicting the fire spreading. and it was a well-instrumented test. We had compartment temperatures. we had some kind of, indication of mass loss, but really in, in terms of proving that your CFD was, good

Wojciech Wegrzynski: 18:02

Uh, that was already FDS6, I think. What did you use back then as, as the model? So just ignition temperature and fixed heat rates per stick, or you went deeper into like

Stephen Welch: 18:14

I think from memory all of that work was based on what call simple pyrolysis. And I must, give credit really to the current PhD uh, Chang Liu. when he started, he was keen to try out the Arrhenius the detailed kinetics of complex pyrolysis. And I, I was a bit skeptical because on my, my own work with, complex, soot prediction, for example, the complex the model, the harder it is to control and get to give useful results. So I, I was, um, uh, thinking it, it might be safer just to, to stick with simple. Obviously, with, with simple you're not able to capture all of the physics because you ignite the stick based the ignition temperature, and then it, it follows burning rate. And if you're interested in, time-varying, your heat fluxes is changing, then you realize it will be able to capture all the physics. so in retrospect, it was, it was great that, Chang off in this direction. It was hard work. took a long time to get these models calibrated, and so the, the next stage of, this work was calibrating at the level of a wood crib. So we were very fortunate, um, that the, the second input to this work was a TRAFIRE project, so we were to design, an isolated wood crib. yeah, there was the Liège test series, so we, we about, uh, 15 different kind of wood cribs, which to be kind of nominally, consistent with the, fuel load density from the Eurocode. And, we then took that crib and put it in a large and, and ran a series of tests. This was the, the Ulster TRAFIRE test series, uh, run uh, care of, um, University of Ulster in Northern, in North- Northern Ireland. so that, I mean, that, that was fantastic because rare to be able to study a problem at different scales. So to have the detailed data from the crib temperatures, fluxes, mass loss, uh, spread rates, burnout rates, and then to go and, and look at what happens to the same crib in a full-scale compartment.

Wojciech Wegrzynski: 20:22

Yeah, you're touching on the early work by, uh, the, the stick versus flat piece of timber versus a structure fire. I guess a, a crib is an intermediate level in pie as well. So, uh, if you consider a flat surface, so you have a flat surface exposed in a cone calorimeter, you actually have a single isolated stick burning in surroundings it is in, and you have a whole crib. What's the difference? Like, I know it, it's different, but it's all

Stephen Welch: 20:55

So it did, did-- The crib differs in two respects. One is, yes, been used as the historic vehicle for of burning rates and compartments because essentially your fire spread is dominated by what's happening the crib, so it's, uh, it's isolated from your Yes, of course, the top layer is the burning rate at the top layer will be strongly coupled. if you're just looking at the burning of a crib, an ideal, mechanism for studying or for simplifying the problem at a compartment scale. you then want to study the fire spread, then of coupling to the top surface uh, becomes critical and, important. But, the other aspect of it is the sticks burn, much more easily than, than solid, uh, surfaces because they've edges and corners. So actually in the limit, the corner is kind of a thinned bit. it, it's really interesting to interrogate the, the of, of the models to see, well, which bit ignites first? Is it the vertical surface, the horizontal surface? if you've got enough resolution across the stick, which, which bit of the stick ignites? and in general, we're not claiming necessarily that the kinetics that are tuned for these models are They're not. They're not the same as the TGA kinetics.

Wojciech Wegrzynski: 22:06

Mm-hmm.

Stephen Welch: 22:07

they're engineering models that make the burning rate of the stick match the experiment. I, I think we need to be honest about that from the start. Uh, you can't just do a, a TGA experiment or a cone and simply extract the kinetics and put them into a stick model and expect it to reflect the reality, of experiment.

Wojciech Wegrzynski: 22:26

W-which is the, which is the dream, which everyone would like to, to, to, to be able to do.

Stephen Welch: 22:30

it would be better if, that was possible, but of there's good reasons that it, that it isn't. Part, partly numerical, so y-your, your stick in, CFD model of the compartment is, is still, a finite of cells.

Wojciech Wegrzynski: 22:42

And while your pyrolysis kinetics or, or just heat transfer in principle, one-dimensional heat to the depth of the timber is not gonna change the modes. You're kind of moving from this one-dimensional in, in a flat surface to a three-dimensional geometry and various modes of heat transfer at you know, edges and everything in a stick. Then you add the porosity of the crib, the of the crib, yeah, re-radiation inside of it, and then you put it in the compartment, and you now have a whole lot of external effects and non-uniformity of the heat flux field, which was constant in cone at the start, right?

Stephen Welch: 23:24

yes. So, we've talked briefly about simple pyrolysis. This is where you need an, uh, an ignition temperature then prescribe the burning rate. And, we believe the most useful way ahead, uh, in most our modeling has been using detailed pyrolysis. As you put our Enos kinetics, of course. there's another strand of, of research we've little bit with the S pyro, the scale pyrolysis models in, in FDS. So that kind of concept. You, you test it in the cone, you, you bring it into fire model, and then you, you do some adjustments upon the predicted flux. and, you know, that's obviously, that's, that's an uh, that's a, a halfway house, uh, as it were. but yeah, I think one of the reasons, uh, it was a good choice to go to, um, pyrolysis is because our, boundary conditions in large compartments, particularly when you approach the transition to rapidly growing and over fires, uh, the boundary conditions are extremely, changing. Uh, and it's, the sphere is ramping up. possibly at that point it doesn't matter anymore it's, it's going to flash over anyway. Uh, but, but, you know, uh, it, it is, uh, potentially more of the physics. Uh, and, and I think it's probably the best, uh, tool we have available for studying the problem.

Wojciech Wegrzynski: 24:37

I mean, that, that's actually kind of interesting if you, think about it. When the fire is really, really small and it's progressing, like you probably don't care because you could just, you know, wrap it into a linear spread and you're probably gonna get there. Like if the flame spread is consistent over the time because it's a small fire and it's slowly that's easy. If it's flashover, that's like you can as put that down heat release rate per square meter your crib and you're gonna be there. But all in between is unknown, and that's

Stephen Welch: 25:13

that's absolutely... That, that's, that's where the interest, uh, lies and in the post-processing. For example, looking at the, the sensitivities of the rates of the sticks to the imposed flux or the, the net, the net flux, the ridge flux, or the total flux. Um, you're absolutely right that it would be foolish to, pretend that the initial phase of the fire, the, the phase, that that's actually capturing the physics your, your sticks and your cells are, you know, if you to study that problem, you'd, you'd zoom in just part of the problem and you'd, you'd run a much finer and, and, and you'd model the accelerant and, and all kind. So we kind of, we, we treat that bit as, uh, sacrificial.

Wojciech Wegrzynski: 25:54

Mm-hmm.

Stephen Welch: 25:54

sometimes we have to do a little adjustment at the start of the predictions to, to match up the, you know, the, the start of the fire spread. Uh, but all of the interest lies in the bit in between. And in, in many of these large compartment tests, that's a, a decent space to, to operate in. You've got maybe a 15-meter in, in the Ulster test, a uh, uniform, fuel bed, uh, the fire spreading over to an hour in, in the most open case.

Wojciech Wegrzynski: 26:23

Okay, so trying to get those, crib models to a point where you actually believe the results that, that they show something, how did the journey look like? Uh, uh, you said that you started applying those a-advanced models. Uh, what were you trying to do, and what were you achieving with, with the next steps of it? Like, how did the path look like?

Stephen Welch: 26:44

I mean, it was quite a complex path. I must also give credit here to, uh, Dr. Shudai, who did his p- he, he, he, he took up this uh, towards the end of his PhD on, on traveling fires, he devoted immense effort, on developing the simple uh, representation for cribs. And then, we were s- uh, we, we, we were amazed when he then took the same calibrated model from the crib and it in the large compartment, the, the same that he was, uh, involved with. You know, we, we were on the site at Ulster where, where we ran these full-scale experiments. We saw how the fire spread. It's always massively valuable for, for fire modelers to go and actually look at the experiment. So, uh, he put the, the in-into the compartment and when we ran the predictions and he came back and it's amazing. It kind of actually matches really well." It matched so well that I was kind of afraid that, when we presented of a fudge. You know, you must have, you must have tweaked this." actually that's been one of the biggest challenges, with the c- conveying the results of, the work based on simple paralysis. it's surprisingly good. You know, we, we were able to recover the spread rate. It was a steady, uh, in test one, it was, a, a steady fire. The fire spread rate was well predicted, the we had a load cell in the middle of that compartment, so we were able to get a ballpark on, on the heat release, and that matched within, you know, maybe 10%, 20% k- of the best you could expect. so that, that was shockingly good. But we did know that actually if you pushed it, if you began to, reduce the ventilation, then eventually it was, it was, it was gonna unravel. So that, so, that was the f- the, the first, paper we, we published. Uh, we also called it scaling up. Um, I mean, scaling up here in this context has, has meanings. Uh, it, it's about assembling the physical processes putting them into more and more complexity, but, but it then means, okay, so can you then use the models that capture those things to, reproduce conditions, the of conditions in, in more and more complex scenarios or larger and larger scenarios? um, yeah, I mean, we were really pleased with that work. Uh, but when Chang came along, we, we went back to the again and, redid all of the calibration. He, he did all of the recalibration of, of the model um, detailed paralysis, uh, again, back to the and then scaling up to the large compartment and, and then running not just So having, having achieved what we think is a sufficient level of validation, so very pleased with the results against the test, uh, then beginning really explore what the, the model can tell us by uh, the boundary condition that, that we, you know, we, we discussed earlier. So running paramet- parametric variants.

Wojciech Wegrzynski: 29:33

Of course, all the papers are linked, uh, in the notes. so the listeners are very welcome to read as, as we talk. There's a lot of beautiful figures and, and images from, from the experiments as, as well. Um, was the goal always to create like an model that's gonna handle any wood crib in any of spec- any kind of setting of stakes, whatever? Or you were focused on a single crib that you is gonna be used in the experiments like you've in the TRAFIRE?

Stephen Welch: 30:03

Yeah, I think, I think the question as to, you know, uses of the model for non-validated cases is, is, is less interesting. y- y- you always have doubts there as to how credible the, the predictions are. So I, I think it's critical that your, your, your are first validated at different scales, at crib scale compartment scale. Uh, but yeah, there's a very interesting question as to, having calibrated those kinetics, can you then take, that model and, for example, you've, you've got a, a timber crib design, you've got different stick sizes. I mean, most of them are softwoods, so you, you might different, uh, density of your material, different, thermal properties. Can you just put the updated thermal properties in, in model and expect it to work? Um, we've k- we've kind of, uh, explored that to some So this is, uh, so far unpublished work, but we've quite a lot of effort also to modeling the, the Edinburgh Traveling Fire test. You know, as, as I said, we started with that about 15 years ago now, but, um, with student projects. but they were v- very focused on the kind of demonstration of the capability, the, the, the generic capabilities the Uh, does the model produce something that looks reasonable? Um, we've gone back to that, to do much more systematic studies, much, uh, higher level of, uh, interrogation the model against the experiment. And it's tricky, uh, Wojciech. You can't just take something calibrated for one case it doesn't translate. I think the stick size is one of the big things. So in the, the Ulster compartment, the sticks are small. I, I know they've been smaller in some other tests, they're, um, by three, so it's a fairly small of a stick in, Edinburgh Traveling Fire and, uh, I back to things like the Gordon Coop tests, the, the long compartments at Cardington. They, they, they used bigger sticks, 50 by 50, mils. and as I said, it's an engineering model, so it's, um, needs retuned. But yes, it, it's, it's a fascinating area. Uh, it's an area we still want to, to understand better,

Wojciech Wegrzynski: 32:10

W-w-what about like density of the steaks, uh, content in the different?

Stephen Welch: 32:15

Um, moi- moisture content, is not really explicitly in the model. It's, certainly it's in the thermal properties of the We use specific heat, um, the latent heat methods for possibly is in, uh, the stick heating model as well, but I think by the time the fire reaches the, the later of, of the fire, I think possibly a lot of your is already gone. So it's, I think it's probably important for phases of a fire where you've just lit it. But, yeah, it, uh, it is, it is massively uncertain. Some of the, um, full-scale tests, um, we, we had stored under tarpaulins, and it was raining, and some wind, and environmental conditions clearly important. and that, that obviously I think that's probably of an impact on the variability of experimental results. theoretically, yes, we, we could explore a bit more of in, in the modeling.

Wojciech Wegrzynski: 33:10

I mean, I, I agree and disagree. Uh, like okay, for late stages of, of the fire, yes, uh, it probably dried out. But, uh, I've also done fire experiments on a day after a rainy day, and I know they l- and I know look different.

Stephen Welch: 33:25

I, I, I f- I fully agree. They, they, they, they should have a big impact, on experiments. And if, if that's true, then we ought to be able to

Wojciech Wegrzynski: 33:34

But it's, it's like, that's another thing in, in to study that I, I wouldn't say that at this it would be an invalidating factor or anything. And, and the density of timber, like how, how, much do you have to know about the timber before go into the model and where you get that from? Is it cone TGA or, or you do something specific

Stephen Welch: 33:54

Yeah, I mean, obviously each stick is, uh, a, a standard deviation of variability on the for both and the Edinburgh, tall building fire tests, did cone tests. So, so we ha-- we, and, and we measured stick density, Uh, so we have the, we have the ballpark numbers. But yes, um, fuel is delivered to the experiment in, uh, I think it was six tons for each, um, Ulster traveling test. so within that fuel bed, there will be some variation. And as I said, while watching the experiments, you, you get some insights into, kind of causes of some of the I, I do certainly recall seeing the fire jumping when it, a discarded plastic tie on the timber was, was still there in the fuel bed.

Wojciech Wegrzynski: 34:41

Mm-hmm.

Stephen Welch: 34:42

that, that kind of thing obviously can happen when you're running an experiment. But, but each stick is different. So also when we were looking carefully, uh, in through windows of the, uh, Ulster compartments, we could see sticks bending as the fire approached. They, they bent towards the stick. So all, all the the models are based on your 120 mil gap and pitch between your sticks. is actually burning that gap, that gap is changing. So, know, there's so many of the things there are um, and we have to choose something. And yeah, if we think it's an important parameter, we do a sensitivity study on, on, on that parameter.

Wojciech Wegrzynski: 35:16

There's also like the late stage of the where the crypt starts collapsing and it doesn't the geometry anymore.

Stephen Welch: 35:24

absolutely.

Wojciech Wegrzynski: 35:25

like if, if your timber model is perfect but

Stephen Welch: 35:30

Yeah.

Wojciech Wegrzynski: 35:30

how can, how can you get there?

Stephen Welch: 35:33

that, that's, that's a good example and, and again, it, it makes it surprising the success of the model. I mean, I think it probably the fire spread along the surface of the crib is, a big part of, of what you're to capture, and then of course it's burning in depth and eventually towards the end of the test you're getting and crib collapse. but maybe it doesn't matter too much because that's at back and the contribution to the, the thing that you're most interested in with the, the spread front is possibly less. but yeah, back in the, initial, um, the, the work done Xu on the Liege crib, we did a very extensive sensitivity study on all those parameters, the heat of combustion, density, and, you know, ev- ev- everything that might the, the burning rate of a wood crib to get a feeling for whi- for which ones are important.

Wojciech Wegrzynski: 36:18

Mm.

Stephen Welch: 36:18

and I think, I think moisture content was in there as actually, Wojciech.

Wojciech Wegrzynski: 36:21

in terms of what's happening within the crib so what's happening in those open spaces sticks, how, how important is, is that in the model? Like, w- what, what's happening in there? I guess pyrolysis, some kind of combustion, sensitive it is for, for example, for porosity of the crib? How well the sealing software gets the air

Stephen Welch: 36:42

is a good question. It's, it's quite instructive to, to take an overview the, crib designs in, in the, the many... There's 10-ish large, traveling fire tests that have been done, and available in literature. And the, and the crib porosities vary quite a lot. I think there's only, one in what we call the porosity. There's, there's many, ways of describing this. You can call it porosity control versus open or, controlled. but the-there's only one of these tests, um, which has in, uh, the porosity control regime, and that's the, test, that's the Gordon Cook test, the long back in, in the early'90s. So most of the wood cribs that we're, that we've, we've been using, I say we, the royal we, uh, in, in recent um, have, been fairly open, um, some more open than So, for example, our crib, that we used in the Asian and the Ulster test has, um, we call it a porosity factor.

Wojciech Wegrzynski: 37:41

Mm-hmm.

Stephen Welch: 37:41

uh, literature, and, and it's measured in, in centimeters of 0.35 centimeters. Sounds a, very small number, but the threshold, the from porosity control to open is about, 0.1, centimeters. So, we're well in... So, so we've, we've got enough, potentially e-e-enough, uh, space inside our, uh, inside our fuel bed to support, uh, burning, with, ho-hopefully not having-- Well, we, can use the model to explore and look in and see what's going on inside the crib and see if we've got areas of underventilation. I, I think we're, we're tending to still have a lot of going on in, in the depth of, of the fuel bed. It's not just the fuel is, is generated and, and out at the top to burn. But interesting, interesting questions, yes.

Wojciech Wegrzynski: 38:29

Okay, um, now, th-this was the calibration work, you, you went further in the last IFSS, uh, the were presented on, uh, using that for different of compartments with different geometries. Uh, I, I guess you, you were playing with o- speaking, opening factors of the rooms because that's really, really interesting variable. Uh, as you said, ceiling is an important one, opening factor is probably the, the one that's most often controlled, say. So how did it look when you took that crib into, those settings? And did it again magically work, uh, and, uh, lived happy ever after? and, uh, pain?

Stephen Welch: 39:10

Yeah. ga- again, giving credit, this was a student project by Shiyu Ming. Um, it was her final year thesis project. And having validated the, the full-scale model against two tests, we said,"Well, okay, so what would happen if you, took, those openings in, in the full-scale and you moved them around?" So for example, instead of having a downstand and a long, horizontal opening, if con- concentrated that opening just to the middle. Or for example, if you've got a downstand, let's make an up, an upstand or a sill. You move it from the top of the compartment to the uh, now of course, your traditional, models, uh, par-- like the parametric temperature time curve, uh, by the opening factor i-i, sensitivities. and you're not really changing that, so you're, you're the area the same. And but, but never-- So the fascinating thing coming out of those, predictions was just the extent of the So we're seeing massively different fire spread So when the fire is confined at the start, and you're the heat, and you're, you're not letting enough, uh, air in, the, the fire tends to move rapidly towards and then it stabilizes in openings. but looking across all of those results is, is amazing to see the level of complexity. What, what you thought would be f-fairly simple as, change the confinement, you, you would get more more rapidly traveling fires is, is, is a massive It depends very much where that opening is located. And, obviously then you're able to then begin to the mechanism. So you can, you can look at the heat fluxes from the hot layer, and the hot layer position is changing. a big one that possibly we had overlooked the entrainment height is very important. So the, the bit of the fire plume between the fuel bed disappearing into the hot layer, uh, dictates your, your o-oxygen uptake, uh, and that dictates ultimately how energy you can release. so that, that's vitally important as well, and often it can be, it can catch you out. It, it can do things that you, you didn't expect. and then, yes, just the, the cooling effects at the edge of the crib. So we're often seeing complex, uh, evolutions of the shape. So the, edge of the wood crib, you've got more heat loss to the external environment. You've got possibly sometimes a lack of, of air in the of the, the fuel bed because you're blocking it by, a fire at the sides. So you get a, a kind of M profile developing. Um, but that's not necessarily always the case. Quite often that profile, changes for various of course the CFD, the detailed, uh, information in the CFD solution en-enables you to kind of explore what, what's causing those changes.

Wojciech Wegrzynski: 41:54

I think a big, challenge in here lies with the of the flow field itself, because it's kind of in those modeling approaches that the, fire or less uniform. You define it, but you define it like with the that it's gonna spread linearly along your or more or less linearly. While I think if you, if you have especially small openings that can, uh, introduce air at quite a velocity, so the momentum, uh, yeah, the momentum So it kind of depends if you're before the door or after the door opening, the flow is either you or pushing you, you know, and that's a massive, massive difference. And, I base this on, again, uh, large scale not traveling size experiments, but 70, 80 square meters where we, you have this large opening, but the swirl in the room kind of establishes in one can see what's happening on the right-hand side very different from what's happening on the side.

Stephen Welch: 42:56

Yeah. So, I mean, this is the whole interest really. When, when you've put the thing into a compartment, it's not just that you've got the coupling from the hot layer. Um, I mean, there's a beautiful illustration in Drysdale about, you know, you can, you can reduce the fire load if you distribute your fuel, uh, uh, over the surface the compartment. You, you get a flashover, and if it's a- all, all in a wood crib on the floor, you, you don't. but yeah, that, that coupling to the, the ventilation is the other aspect of that. It's, quite influential. I think, um, there's some really nice, contributions in Gupta.

Wojciech Wegrzynski: 43:35

Vinnie here next week.

Stephen Welch: 43:36

fantastic. Yeah, okay. That's, that's great. That, that, that, that is lovely work because, you they, they were the first to kind of explore this, um, or the severity of, regime two fires, momentum fires, and kind of demonstrate that actually no, it's just the, the hot layer temperature that matters. It is everything else. It's, it's how it, how is that, um, how is the flame by the ventilation and, potential for, for higher i- in those kind of fires as well. that's what we're seeing in our predictions. It's extremely complex. As soon as you put it in a box and you move the openings around, uh, the level of complexity in, in what's produced

Wojciech Wegrzynski: 44:17

for me, the most striking, like, difference really I, I was surprised with, which I shouldn't I was. We were doing this large, uh, scale CLT experiment, and it had columns, and they were in the opening of the room. The opening was not full, the, the whole front wall was just a giant window, you know. And the flow established in one direction. W- and the difference, between the right-hand diagonal column and the left-hand side, which were very narrow to the opening wall, was, like, You could see, like, on first glimpse, okay, this one, the smoldering in that column progressed further than on the left-hand side. And I was like,"Wow, that's a flow." And, and wow, that, that, that's profound.

Stephen Welch: 45:03

Yes.

Wojciech Wegrzynski: 45:04

And also, sorry, in, similar experiments, the of the columns after the experiment. We had giant glulam columns, and they were very narrowing down to the, to the floor, where we had most likely more oxygen, which I would connect and that, that is, of course, like, pro- most smoldering damage or just oxidation, but it's a phenomenon. Flaming combustion also will react to those

Stephen Welch: 45:29

Yeah. Yes. So, I mean, unfortunately, wind effects are very and, uh, we, study, and I know this has been studied as well, even back to the time of uh, on, on wind-driven fires in, in, in wood cribs. you can study that at crib level, but then translating to the design context for real buildings where you've got random winds that are changing during experiments, it's, um, you have to walk before you run. You have to s- to start with a simplified version of problem. Uh, but acknowledging that, of course, the things that studying might actually, in the big scheme of the end, be, overwhelmed by other uncertainties. So, you know, back, back to a s- uh, scenario everything that we're studying is highly idealized. We've got our uniform fuel bed. We've got often symmetric openings. They're fixed openings. We're, we're not looking at glazing failure problems. We could put that into the model if we wanted, but the level of complexity and, and, and what you of those results, you know, you, you'd need to validate them. So yeah, it's an, uh, an environment where there's a mass of complexity, but you have to choose your battles and, and progressively work forward in, in parts that you

Wojciech Wegrzynski: 46:40

uh, at least we're not getting out of jobs, uh, soon. There are still a few problems to be tackled, Uh, how did you move from the cribs to the compartments then? Because now, that now is a big step.

Stephen Welch: 46:53

Yeah, we actually had, two kind of, things that we had some collaborations with, with RISE on the FRIC test. So they, kindly shared their experimental data and some effort, or Chang devoted a lot of effort to, to those tests. Very interesting overlaps with the Edinburgh tall fire tests, a, a similar, crib structure, uh, but a uh, wood crib and, and timber linings. but again, fiendishly complex. Um, so those results are not published, but we were able to decent, of the fire spread. Uh, but the post, uh, spread burning behavior with h- high levels of inter-ventilation and, and lots of going on in, in the ceiling was, was very challenging the CFD. So that was one kind of arena, uh, where we started a few years ago. Uh, we were very lucky as well to have, a instrumented test done in our fire lab by, uh, Ali Awadalla for his PhD. so he had, timber linings on his ceiling and his They were glulam, uh, spruce, elements, well instrumented i- in terms of, you know, we, we're measuring mass loss calorimetry, temperatures along the floor and the heat flux at the opening. Uh, so that, that looked ripe for investigation with modeling tools and, and having the collaboration input actually being, being able to talk to Ali about the and so on. Um, so we've done a lot of, uh, work on modeling those I say we again, this is, uh, Chang. Chang has, has calibrated the, detailed process match these experiments and absolutely the, the res- the insights that the, the CFD is giving those punctual measurements that you, you can make the experiments. So we've got the, the whole temperature environments, got the predictions of the burning rates of the timber the flux boundary condition. Uh, and you, you can look at the correlations, for of, of the burning rates against fluxes, uh, and, and loss and so on.

Wojciech Wegrzynski: 48:53

I, I didn't ask that before, but how, how do you ignition in that, in that modeling? Like, because also th- in this case, it's mu- must be different for the crib, for the glulam, for ceiling, th- three different materials, different

Stephen Welch: 49:06

Uh, yeah, I mean, it's embedded in the kinetics, so it's you're, you're tuning the kinetics to do a lot of Um, they're multi-component kinetics. So we've got, uh, if, if you're familiar with the, the detailed process model, we've got cellulose, lignin, and you're producing char. Um, but yeah, you're, you're, you're really asking a lot of that model. So you're asking it to give you both the fire spread, is less important in, in, in the, the small, uh, so Ali, Ali's, uh, box is 0.3 meters high. We have some reduced openings. we have different wall materials to produce different loss and different, uh, burning, uh, regimes. B-b-different, different, uh, histories of burning, necessarily burning regimes. They're, they tend to be towards the, the underventilated side. but yeah, you're, you're expecting that, uh, the you put, put in to, to give you your spread from the, prediction of the pyrolysis, uh, the start, of the at low temperature, and, and then to give you a burning rate when the, the fluxes ramp up later on. So it is, it's quite a tough challenge, and especially you consider the complexity of the material. if you look at the experiment, you can see the, the and potentially then, Ali's, work eventually is, trying to feed into prediction of, of burnout, and then got delamination, uh, all these other complexities that might effectively, uh, you know, a-arise later in the

Wojciech Wegrzynski: 50:28

So, so in, in those experiments we're doing with with Danny, we were exposing, uh, the ceiling, CLT ceiling, it was approximately, I don't know, 60 meters of timber out there, uh, to, to gas uh, and, and actually like a flame impingement a, a, a thing that was quite typical to those was that the... Eventually the timber ignited on the ceiling. It, it spread very rapidly across the, the whole surface. Uh, as soon as it developed some sort of simple layer, perhaps a minute of time, may- maybe less a minute of time, the, the burning rate rapidly. And it- it's, it's a common thing for timber to do that. in a crib it's kind of different because due to the complexity of a tr- a crib, it creates this, you internal conditions that allow you to sustain the re-irradiation inside the crib. But, uh, in a, in a, in a ceiling, that's W- are you able to capture those effects with those models or that's, that's a, a job for now?

Stephen Welch: 51:35

Yeah, I think, I think we are. I mean, I think probably we can go back to look at for the cone calorimeter, and I think that, these things in the cone have been able to capture both the initial spike and then you get this pl- this long when the, the char layer is building up. Uh, and then at the end you get completely different when the, you've only got char left that's, that's I think, yeah, we-- I think the detailed kinetics is, successful in capturing this. So the real test of this model, uh, I should just go one step and explain the beauty of this experimental So it, it's not, only a timberline compartment, it's a compartment, this is Ali's box, uh, with a burner. So you got your gas burner on, it a good blast so you, you've got all your timber burning for the first, 10 uh, is it 12 minutes? Uh, for the first, uh, a, a period of the test, you, blast it with, with heat, you get the timber cooking and burning, and then you turn your burner off, and then you study, what is the residual burning just, driven by the, the pyrolysis of the, burning timber and, and how interacting with the compartment, heat loss through um, you know, some, some of them with, uh, with, with boundaries and, in the other extreme with, with steel. So in, in the steel case, it, it's burnt out. Uh, the, the fi- the fire goes out in a couple of Uh, but the ones with vermiculite and stone wool, they, they con-continue cooking and actually some of them a second peak maybe half an hour later. and, and, you know, the, the, the timber's quite deep, there's plenty to carry on burning, um, during the stages. experiments, really, really interesting from, from a perspective.

Wojciech Wegrzynski: 53:17

Also later the lamination and the mythical stack is refreshed over.

Stephen Welch: 53:21

Yeah, potentially delamination too. Yeah.

Wojciech Wegrzynski: 53:23

Yeah. that, that, there's still, l- a lot to, to go What's the next thing on, on your list, Steven? of this?

Stephen Welch: 53:32

Okay. So there's loads of, uh, really having established these modeling tools or modeling approaches, there's a box of sort of, oh, well, why don't we try that? So for example, we've just been talking about the compartment effectively it's at, um, one-tenth Uh, so the same model, um, at full scale and see you get. Now, of course, the emissivities of your, your hot layer are gonna be different because you've physically the thing up, so you expect it to be quite different. now it might, it might be a bit overreaching to kind branch out based on that, that, validation at, at one to, to something at a completely different scale where you know you've got other things going on like, currently s- we're studying the effects of the radiative fraction and the soot, uh, levels. Uh, they will be different at the different scales you've got different residence times. Um, but yeah, um, the mo- the model provides you with the, the ability to do that. So that's, that's one arena that we're, we're working in. on the full scale compartments, I think there's We talked, uh, earlier about wind. Um, another, fruitful area of work that is not yet is we've been exploring the fuel bed effects. So, know, having validated against a wood crib with layers of sticks, you can then say, "All right, so if it was 11 sticks or 13 sticks." Uh, so Chang, Chang has a, fantastic systematic study of, what we call load density, uh, sensitivity, uh, effects. And I think this one's turned out even more than, than the previous one. So, the, the level of unpredictability, it seems to be You, you, you make a small change, you, you add two, two, two more sticks on the top and suddenly your fire is, is smaller. Um, whereas of course all the traditional theories and are you put more fuel in, you, you'll have a bigger fire. But there's complex things going on inside the fuel bed in terms of the burning in the depth of the crib and coupling to the environment, the fact that the layer is a slightly different height. so that, that's, that's one area that we're, uh, uh, maybe not for the future. That's-- Well, we still need to publish those papers, that's one we're currently exploring. Another one I'd, I'd like to look at there is, uh, to do sensitivity studies on, on ceiling height. I mean, we did that for the wood crib in, in the LB7, Liège test, uh, studies by, by Xu. but we haven't really had the resource, to do that at the full scale compartments, and we know it's gonna parameter So, you know, I think one of the constraints is these simulations are expensive. We're running for the full compartment, it is 9 million cells. it takes probably a month to run, the complete using maybe 70 or 120 processes on, on the HPC machines. Uh, so we need to keep, uh, bidding for resource to be to do one more sensitivity study. back to the discussion about, you know, the You know, we, we, we go one step at a time. You have to choose your, your, your battle and whi- parameter shall we, shall we explore? Can we, and, and can we gain understanding then of, uh, from in- inter- in- interrogating the, the detailed

Wojciech Wegrzynski: 56:41

And there's endless richness of, uh, geometry

Stephen Welch: 56:48

Absolutely.

Wojciech Wegrzynski: 56:48

architect's dreams, you know, moving. and eventually, I assume moving from, uh, uniform

Stephen Welch: 56:58

another area. Uh, obviously there've been, there have been uh, with discrete cribs. in some senses they're easier because the fire blocks.

Wojciech Wegrzynski: 57:09

Oh, if you get the jump wrong, you, the,

Stephen Welch: 57:13

yes, everything else, uh, unravels after that. But, you don't necessarily have to get, the the stick level correct to, to predict ignition of block. So it's an interesting question as to whether or easier problem. But, but it's fair to say that it also doesn't correspond to real world, uh, fuels in, in offices where we have funny geometries, all sorts of things. You know, um, we, we have to go, uh, step by step. These simulations also produce enormous volumes of as I'm sure you know very f- f- very well aware Wojciech. We, we're talking about maybe 200 gigs from, from one, one of these full scale, um, scenarios. so it's not really necessarily also the, the, the is not running more scenarios. the, the volume of the, the treasure trove of it produces and finding the time to interrogate it look at correlations and, and so on. So that's, that's a c- a practical constraint.

Wojciech Wegrzynski: 58:13

I'm, very, very well aware of, of the risks of just brute forcing your way, uh, through, uh, a study and then facing consequences of that choice beforehand. That's probably story of my 80% of my, of my th-thus far on, on parametric studies with CFD. Uh, so, so I, I, I get that. Well, a-anyway, Steven, uh, I would just say note to the listeners, there is so much more to elements we've discussed. There's on technical details level, which we space to, uh, really, uh, give them, uh, time in this, in this one-hour interview, but it's all in the papers. I've done it on purpose. If you're interested in the how it's done, from the scaling up papers to the papers, all of them, they cover beautiful technical details, and I highly recommend anyone who would to see how that works to, to go there. And something tells me if people send you an you'll be keen to respond.

Stephen Welch: 59:14

yes, absolutely. I try to keep up, yes. But you're absolutely right. There's, um, there's a treasure trove of, of information out there in the public domain. There's, there's, there's several more papers that we will be appearing in the, in the near future. So, um, yes, um, more than happy to discuss, and th- so much for the opportunity to, to talk through

Wojciech Wegrzynski: 59:37

my pleasure, my pleasure. Thanks for, for delivering brilliant content, fine science, and I cross my fingers as you get further because, like, you really, uh, in not, not in the peloton. You're, you're escaping the peloton. We're chasing you, and then you're, you're far I appreciate you give us, you know, a, a, a of, of, of where to see now. Gr-great stuff, Stephen. Thank you so much.

Stephen Welch: 59:59

Yeah. Okay.

Wojciech Wegrzynski: 1:00:01

that's it. Thank you for listening I've enjoyed this a lot as I said in the opening, uh, it's my world, compartment fires. Uh, that's what I like, and I know it's, it's difficult, and I have very little trust to people claiming that they are able to model fire growth a compartment. Yet, uh, I've seen the papers of Stephen, I've the presentations of him and his team on and I think they've gone very, very far, maybe furthest from all of the fire modelers out there, in being able to actually recreate the mechanisms that drive the, the fire spread itself. I think it's, it's quite brilliant the progress uh, they've gotten and, And I think it's very you know. It's been demonstrated that we can do it, we, do fire spread, so now we, we have to move the of our capability. In doing that, we need to validate the hell out of the models and, uh, as Stephen said, uh, like possibilities open when you get the model that, you trust. Sensitivity studies of all sorts, so we're, waiting for them. It's, it's gonna be a source of tremendous in the world of fire. Therefore, I'm super, super excited for the near and seeing, uh, where they can push, uh, those ahead. Uh, anyway, I think that would be it for today's so, um, thank you very much for being here with me and I hope to see you here, uh, next Wednesday for another dose of your fire science. Thanks very much. Cheers. Bye.