256 - Modelling turbulent combustion in fire CFD with Bart Merci

Conference Highlights And Big News

Wojciech 0:00

Hello, everybody. Welcome to the Fire Science Show. I've just came back from La Rochelle, where I have attended the 15th International Symposium on Fire Safety Science by the IAFSS. And oh boy, what a conference that was, uh, perhaps the best fire conference I've ever attended in my life. I'm really, really enthusiastic about what I've seen there. I've learned a lot, and from your perspective, the most important thing, I've identified a lot of really great speakers for the Fire Science Show podcast and even had a chance to interview some of them. And, uh, as this is a very nice round number of an episode, 256, let's celebrate this very nice number with a very, very nice speaker. And, uh, today I'm, I'm really proud to say I have Professor Bart Merci from Ghent University in the podcast. At the IAFSS, Bart delivered the Howard Emmons Invited Plenary Lecture. This is perhaps the most important lecture of the symposium and something that is universally considered as perhaps the highest academic distinction, recognition a person can have in the world of fire safety science. So this is not something that happens every day. This is a celebration of Bart's life as a researcher so far, and with a hope that, uh, his achievements continue. besides, you know, this distinction, Bar- Bart got, got them all, I think. the co-editor of Fire Safety Journal. I think we-- it's fair to call him the creator of IMFSC program, and that's just very, very few, the tip of the iceberg of his achievements. Uh, scientifically, Bart is looking at combustion in fires, and he's developing a new class of combustion models to be applied for fire problems, which are called the sub-grid combustion models that are used for turbulent combustion in fires. And why we need a new class of models, because there's a lot of fire phenomena, especially as you move towards extinction, especially as you move to-towards under-ventilated fires. Anything post-flashover, really. Uh, and if you want to apply this, this knowledge in, in very specific applications, perhaps microgravity, uh, is something. Then the models we have today are perhaps insufficient. Perhaps you can do it better. And, and Bart and his team are trying to find how to do it better. And that's what this interview really is about. But, while we venture to high-level technicalities of those sub-grid combustion models to give them justice, to explain how they work, the interview overall, I try to, you know, keep a touch to the life of an engineer, to the problems that we apply. And, and that was very easy because Bart is such a practical man. He also, uh, you know, teaches this. He teaches smoke control. So, so Bart also likes to, place this within the framework of practical fire safety engineering. So while the topic, the theme is turbulent sub-grid combustion models, the overall discussion is more about how do we model fires as fire engineers, and, uh, how deep are we supposed to go as an engineer, and how much deeper you can go as an academic, and how to link those two worlds. So we have the best scientifically sound, simple models that are practical to use in everyday fire safety engineering. That's what the interview is about. There's a paper also linked in the show notes, uh, if you would like to follow up with that. But so far, please enjoy the interview with Professor Bart Merci. Let's spin the intro and jump into the episode. The Fire Science Show podcast is brought to you in collaboration with OFR Consultants, a multi-award-winning independent consultancy dedicated to addressing fire 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 committed to delivering preeminent expertise to protect people, property, and the environment. With over 30 chartered engineers and a team of fire 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 tailored 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. And now back to the episode

Meet Bart Merci And His Path

Wojciech 5:16

Hello, everybody. I am joined today by Professor Bart Merci of Ghent University. Hey, Bart.

Bart Merci 5:22

Hello, Wojciech.

Wojciech 5:22

And, uh, yeah, we're here because, first of all, you are a scholar that, uh, we're looking up to, but, uh, now you've been also recognized with the Howard Damms Invited Plenary Lecture delivered yesterday. It was awesome, man.

Bart Merci 5:35

Thank you.

Wojciech 5:36

congratulations on, on the, on the award. That's probably the highest distinction one can get from the society, and I think it's very well deserved.

Bart Merci 5:43

Thank you very much. Appreciate that.

Wojciech 5:45

Uh, Bart, so, I think yesterday you mentioned you started as a combustion scientist and then got interest in fire. I don't think that's a natural pathway. I see a lot of people from fire then b- building interest in combustion, especially when the symposium is in Kyoto. But, the pathway from combustion to fire, I think it's not the necessarily the easiest one. So h- how did that happen?

Bart Merci 6:07

That's an interesting, uh, question, Wojciech. So actually my background is, uh, mechanical engineering. And, um, when I did my PhD, uh, I specialized in fluid mechanics- Mm-hmm and, uh, turbulent combustion in particular. Now we're talking end of the previous century, so computing power was not what it is today. And so this was the era where turbulence was, uh, mainly treated through RANS, Reynolds-averaged- Mm-hmm Navier-Stokes, so averaging out all, all the turbulence. And, uh, but I got interested in, um, in turbulence chemistry interaction, and how to deal with that in, in the context of jet flames, so high velocity-

Wojciech 6:42

Mm

Bart Merci 6:42

flames. Um, and I, I ran into the, the fire community almost by, by accident because there was a, an opportunity, uh, for an academic position at- Mm-hmm at Ghent University, and I figured flames are flames. Uh, and, and so I entered the field of fire from the flames perspective. Uh, little did I know at the time that buoyant flames are quite different from, uh, from jet flames. But nevertheless, when you zoom in sufficiently closely to where the reactions are taking place, there are still a lot of similarities. And, and so that's how I, um, uh, got into the turbulent combustion modeling in the context of fire.

Wojciech 7:18

Yeah. I, I, I love that zooming in. We'll, we'll touch on that in a minute for sure, many times. But, you said they're different. H- h- how they're different? Like, uh, in, in that, let's say end of '90s, at that stage, was there more knowledge about those, uh, high velocity flames or, or, you know, reactors, et cetera, versus buoyant flames?

Bart Merci 7:37

Not necessarily. So there, there, there was also a lot of interest in buoyant flows, but I'd, I'd argue that the context of the flames is significantly different. Mm-hmm. So in, in fires, gravity plays an important role, uh, uh, because of the high temperatures in flames. Mm-hmm. The, the flames themselves drive the flow field, whereas, uh, uh, in the jet flames that I was looking at back when, you would have an industrial device, a burner, whatever, where there was like a, a prescribed, uh, high velocity, uh, fuel mass flow rate. And, and the flame would sit where the mixing of the fuel and oxidizer- Mm tells it to sit, but the flame itself would not, say, determine the flow field. Okay. Yeah, yeah. The flow field was a given, and the flame was a, a result of the, of the mixing flow field imposed by, by the boundary conditions, say, of the problem. Whereas in fires, the flames determine the flow field to a large extent. So in that sense, the context is, is quite significantly different. Yeah, absolutely.

Wojciech 8:33

Yeah. S- someone used the, the, comparison that the fire is the pump of the hydraulic system of the, of the compartment, right?

Bart Merci 8:40

Yes. And this being said, uh, if, if I may, may jump in, uh, because the fire is the, say, the, the, the engine or the pump, whatever you wanna call it, of the flow, the entrainment of air into the reaction zone is a very important phenomenon to simulate. And this entrainment in the context of fires has a significant amount of large scale unsteadiness, large scale- Mm-hmm unsteady phenomena, which explains why in the context of fires, large eddy simulations is, is very important, um, because in RANS b- you're evening out, you're modeling- Mm you're leveling out all, uh, these, uh, even large scale unsteadiness phenomena Which you then have to model. But the issue is that these entrainment phenomena are very strongly problem-dependent. They depend on- Mm the configuration that you're looking at. So it's extremely difficult to find a universally valid model. Whereas in the large eddy simulation techniques where you simulate, and these large eddies, large vertical structures, you get that large-scale phenomenon for free. Well, for free, that means at a co- As a part of what you're doing at all at a cost, uh, indeed, as a cost of the CFD, but you don't have to model it necessarily. You only- Mm model the smallest structures, and these are much more universal in nature, and so much easier to find a model that you can apply for any- Mm situation. Yeah.

Wojciech 10:00

So, so, so the more deeper you go, the more you zoom in, the, the more uniform, less chaotic stuff looks like, right?

Bart Merci 10:08

I wouldn't say less chaotic necessarily- Yeah. Okay but certainly less problem-dependent. Less problem-dependent. Okay. Because as the large

Fire As The Flow Driver

Bart Merci 10:13

structures break up, they become more and more isotropic, and therefore- Mm-hmm. Mm-hmm less problem-dependent. And, and that's a big difference from the jet flames, for example, and where, uh, the, the flow field is driven by what you inject into the domain much more than by the buoyancy. It's, they're much more, say, justifiable to use an approach where you don't care about this large scale and steadiness all that much a- as in fires, so.

Wojciech 10:41

W- w- which bring us to, to an interesting problem of the scale at which we look into stuff, because you're also a very, uh, pragmatic researcher who's, uh, doing building fires and, teaches the new generation of, of engineers who e- eventually will venture to the world, do wildfires, do building fires at this scale. They're not gonna do a micro scale. No. They're not gonna do nuclear reactions.

Bart Merci 11:04

Mm-hmm.

Wojciech 11:04

Maybe they will if they want. Yeah. Yeah. But, uh, w- w- we're at very particular scale level. How does that influence the problem that you're looking at and the solutions that you have to- Yeah work for it?

Bart Merci 11:17

That's a very important, uh, comment that you're making, Wojciech, and because we're indeed looking at, developing tools, modeling tools for the practitioners to be used. Mm. So essentially what you want is that the fire engineer does not have to worry too much about the modeling that they are using. That's our responsibility as model developer, but we have to keep in mind that, practical problems, uh, for fire safety, uh, are indeed, say, fires that can last for a few hours in the- Mm built environment. If we're talking about wildfires, it could be days and even weeks. Mm. We're talking about real buildings, uh, so typically we're also interested in large buildings. So we're talking about a few hundreds of meters, and again, in wildfires, even more, kilometers. Mm-hmm. If you have to put your CFD effort, because we're talking here in the context of CFD- Mm-hmm in, in, uh, such durations and such large scales, it means you're limited in the level of detail that you can resolve. This is fundamentally different, for example, in combustion devices like an engine- Mm-hmm where, which is much smaller. You have very fast phenomena. We're talking there about milliseconds- Mm-hmm, mm-hmm rather than hours. So you can say orient, target your, um, CFD efforts. Uh, if you, if you're only interested in small scales and short, uh, timescales, you can put much more computational effort there in resolving that than you can do in fires. So that means that the tools that you're developing for turbulent combustion modeling in fires, you have to keep that in mind that you cannot go to the same level of detail. And, and much happens at scales that you don't resolve. They happen so-called sub-grid scale, so you have your computational mesh. Mm. Whatever happens inside the cell, you can't see, and so you have to deal with that.

Wojciech 13:02

And, and for the cells themselves, it's, we're also kind of, uh, budget limited or- Mm-hmm or pragmatically constrained by the projects that we're working with because, like, okay, I have a friend who's doing, uh, some cooling of nuclear rods and, uh, he has just a, just a rod. Well, okay. Mm-hmm. Sorry, sorry, but it's, it's very beautiful rod, but, uh- it's, it's, it's a cylinder. Uh, and he has like 100 million elements around it to, to really resolve the near wall flows and, and all the heat transfer that happens there on, at extremely, you know, narrow scale. I, I've never had 100 million models in my commercial, uh, CFD application of an airport even, right? Mm-hmm. So we're also constrained by the, what we pr- can practically deliver at the market. So y- y- you yesterday said that the, the problems are a million elements, 10 million elements- Yeah perhaps, right? Yeah. So that, that's the budget you work with, right?

Bart Merci 13:53

Absolutely, because this is also, now we're, we're not talking CFD, we're talking now about the reality and market, labor market and, and- Yes and commercial market. Clients have expectations. Uh- Deadlines clients have also deadlines. They have budgets that they want to spend. So if that means that you're also, as a partitioning fire safety engineer, you have to deal with that. Otherwise, if you cannot make an offer that is reasonable for the market- Mm-hmm nobody will accept your offer. Uh, and so indeed we are talking in the order of magnitude of a million cells for a realistic CFD simulation, uh, today in, in the

Practical CFD Limits For Engineers

Bart Merci 14:25

context of fire safety. That is correct. And,

Wojciech 14:27

and yet what's, what's brilliant, thanks to people like you, we now have models that actually work at this scale, given all of those constraints at this building context, and, and we apply them. There's a set of models used in FDS. There are a set of models that you are developing with your, uh, team at Kent. So, um, let's, let's perhaps, uh, move there- First, let's, let's start maybe with the chemistry of the flame, because that's where you also started yesterday. Mm-hmm. Going through the scales in time and space, from the smallest phenomena to the largest ones. you started with chemistry because it happens at the smallest scales, smallest time scales. in high school, you learn that you put oxygen to methane, you get the CO2 and water. Mm-hmm. That's a very simple equation. That doesn't look very complicated in it.

Bart Merci 15:17

True,

Wojciech 15:18

true. Yeah. Yeah. how much deeper is there-- W- when you look at this with, with detail, how much more complicated it becomes and how big influence that carries over to, to, to fire simulations at the scale we just defined.

Bart Merci 15:33

Yeah. So th- th- that's a very good observation again. So indeed, if the, the high school, say, equations, they look at your initial, uh, situation and then they look at the end result, which is, uh, before you have fuel and oxidizer, afterwards you have your combustion products. Now, in reality, this does not happen in a single step. You would have a lot of intermediate species, they're also called radicals, and these, um, you know, species are very reactive. And so that means that they do not live very long. Mm-hmm. The moment that they can connect to another molecule, another radical, they will do so. And so this is almost instantaneous if you look- Mm-hmm at the overall, uh, situation. So if you think about methane and oxidizer, that's a relatively simple, uh, hydrocarbon. It's what it is. That's, that's like as simple as you can get. As simple as you can get. Even there, you would have a few tens of these radicals and a few tens of intermediate reactions that are hidden, so to speak, by this overall global reaction. so some people are working on that- Mm-hmm uh, and really focusing on developing these detailed chemistry mechanisms for methane, but also for higher hydrocarbons, for more complex fuels. This is very useful, nevertheless useless for the practitioning engineer because it would take way too long to try and simulate that during a CFD of a fire simulation. So what we have to do is we have to simplify, uh, such things. And so one option is to go for what is called infinitely fast chemistry- Yeah. What does this mean? which is then mixed is burnt. Yeah. So that means is if, if you have methane and oxidizer and you find good conditions, so say close enough to stoichiometric conditions, uh, for, uh, fuel and oxidizer So if you optimize then you say, okay, they will need some time to mix, but once they mix, it's burnt. Okay. So it immediately goes to carbon dioxide and, and water vapor. And that's very reasonable because it's such f- a fast phenomenon anyway, but it's only reasonable if you have what is called well-ventilated conditions. Mm-hmm. So if you have reasonably complete combustion, um, then it's quite okay because it gives you also information on where you can expect the flame to be. Mm-hmm. And this is very important information to know where the flames are, and this, this is order one information. However, very often we're also interested in toxicity, like carbon monoxide. There's no carbon monoxide in this reaction from- No methane and oxidizer to CO2. That's two binding oxygens. And so th- so there's no, no CO. So if you want to do that, you cannot do that with a single step reaction. You need to do something more. Mm-hmm. And, and then again, there's choices to be made. Mm-hmm. So you could say, well-

Wojciech 18:10

I'll add an reaction for our CO

Bart Merci 18:12

For example, uh, and this is what is in FDS now, eh? You have intermediate steps where you can go s- to CO and soot, huh? So soot is also important in terms of visibility, in terms of, uh, fire dynamics

Wojciech 18:23

in general. But I want cyanide as well.

Bart Merci 18:24

Yeah, well, that's, then you need to do something more, huh? So if you want to have more species- It's

Wojciech 18:28

a battery. There's fluoride it,

Bart Merci 18:29

that's, that's a j- that's a very challenging one, huh? Yeah. So then you need to come up with additional equations. If you want to do that- Wow there's no other way around this. Yet you can, um, do that in, say, multiple step reactions in physical space, but beware, every reaction that you add comes at a price, and sometimes these chemical reactions are also very difficult to solve numerically. Mm-hmm. So they take a lot of time, so it comes with a big price. So actually what you typically do in, as a practitioning engineer today is you prescribe what is called a yield. Mm-hmm. And so you say, well- Mm-hmm if this is my heat release rate, I assume that a fraction of that heat release rate and of the gases that are emitted will be whatever you wanna call, this hydrogen fluoride, for example. Mm. You specify a yield. If you do that, then you make life easier in the CFD simulation because then basically what you do is you solve the transport equation for the yield, uh, of the species that you're interested in. The problem, of course, is shifted because then the question is, what is a good value for the yield? And that's- Mm and that's an unknown these days. This is very difficult to characterize.

Wojciech 19:33

I- i- is there any scenario in which this infinitely fast reaction assumption breaks down?

Bart Merci 19:40

Yes. It breaks down when you go to under-ventilated conditions. Okay. Then it's much more tricky- Makes- because all of a sudden the completeness of the combustion reaction in reality goes down very rapidly- Mm-hmm from close to 100% to a lower level, and that's something that you cannot easily describe with, infinitely fast, uh, chemistry. Carbon monoxide, if you want to go a bit more, say precise, it's also difficult to do that. If you go to flashover conditions- Mm-hmm difficult to do with these single-step reactions. If you have what is called ghosting flames or flames- Mm-hmm that start to w- travel around in the geometry, all these things are very difficult to do. And so for that, if you want to, go into that direction, and if this is important for the scenario also- Mm for the practitioning engineer, you need to do something better inside the CFD cells, and that's where this, say, novel sub-grid combustion modeling comes in- Mm-hmm because then you no longer assume, uh, infinitely fast, uh, chemistry. What we do there is we introduce finite rate chemistry. Mm-hmm. But not all the way because again, you cannot afford to do that. Not hundreds

Wojciech 20:49

or thousands

Bart Merci 20:50

of them. No, you cannot do that. So what we do there, and, uh, uh, I go a little bit more technical now- Please

Wojciech 20:55

go

Bart Merci 20:56

is you have the CFD cell.

Wojciech 20:58

Yes. One element.

Bart Merci 20:59

One element in the geometry Because of reality, we are limited in how small we can make those cells. And so again, if you take a real application, a real scenario, we talk about, say, cells of maybe 20 centimeter by 20 centimeter- Yeah even larger perhaps- Quite a brick depending on, on- Yeah on whatever configuration you have. So a lot can happen. Let, let's take- Mm let's take 20 centimeter cube just to set the mind. Yeah. You can have multiple flames. If you zoom in, you can have multiple flame structures inside such a cell.

Wojciech 21:31

Mm-hmm.

Bart Merci 21:31

Or not, depending on where you are in the field, depending on how turbulent the flow is, depending on the local flow structure. So we automatically detect that, and we then estimate how much, of reaction you can expect inside the cell. It could be- How

Wojciech 21:47

filled with flame it is.

Bart Merci 21:48

How filled with flame it is, exactly. Yeah. And then the flames themselves, we treat them as reacting structures.

Wojciech 21:55

Mm-hmm.

Bart Merci 21:56

uh, we assume, and it's called laminar flamelets. Mm-hmm. So it's all like small flames that are occupying this cell. Uh, the flames are not like a candle flame, and they are distorted by turbulence. They look differently, but you have to estimate how many you would have inside the cell. And then we do separate calculations in advance, so not during the CFD. Mm-hmm. Separate calculations in advance. If you know which fuel you will have, you can do very detailed, what is called autoignition calculation. So autoignition, what does it mean? If you create a mixture of fuel and oxidizer and you have the perfect mixture, so say stoichiometric conditions, you start from a certain temperature, and then you just check how long does it take for this mixture to ignite.

Wojciech 22:40

Mm-hmm.

Bart Merci 22:40

The higher the temperature, the faster it will ignite. The lower the temperature, the longer it will take. So you can make a plot of what is called the autoignition

Chemistry Simplifications And When They Fail

Bart Merci 22:49

delay time as a function of temperature.

Wojciech 22:51

You have a model to calculate that, or is it-

Bart Merci 22:53

Yes. This is very detailed chemistry, so there you go all the way. All the way. Okay. So you go the most- All

Wojciech 22:58

the radicals you

Bart Merci 22:59

know. All the radicals that are known. So you u- you take advantage of research in the combustion community, and you- Mm-hmm you spend time. Mm. This takes time, but because you don't need to s- fo- to solve any flow in these calculations- Okay it's affordable. Okay. So this takes a few minutes, let's say- Mm to, to calculate that for each temperature, and you tabulate this autoignition delay time. So you screen the temperatures that you would expect in a fire.

Wojciech 23:24

Mm.

Bart Merci 23:24

Say you go as low as ambient, and you go up to- K, I don't know, 1800, whatever you want. You, you screen- How much more, yeah whatever you can expect. And then- You go to simplified chemistry because you can only, say, afford to go to, say, two, three reactions, and you put whatever you want to put there. So take, for example, carbon monoxide is one of your species of interest. You have the reaction for carbon monoxide, and you do that with what is called Arrhenius rate type of expressions. Mm-hmm. So you n- you then have, it's a little bit more technical again, but you have an activation energy for these equations, and then you have what is called a pre-exponential factor. So let's say a few numbers- Mm-hmm that you can tune, and then you match your results with the detailed chemistry by tuning these numbers in your simplified chemistry such that you get the same autoignition delay time profile- Okay as a function of temperature. So you, in a nutshell, huh, you start from very detailed chemistry, you go to very simplified chemistry, and you use the results from the very detailed chemistry to tune the numbers- And

Wojciech 24:27

find something

Bart Merci 24:28

in between them and define, and find something in between.

Wojciech 24:30

Yeah. So something that's simple, yet gives more or less the results the accurate one gave.

Bart Merci 24:34

Yes. And not even more or less, almost precisely. Almost precisely. Okay. Almost precisely what, what they give. And then you u- and then you forget about the detailed chemistry from that point onward, and in the CFD, you only use the simplified ones.

Wojciech 24:45

But it works only for the specific fuel.

Bart Merci 24:46

It works for the specific fuel. So indeed, if you have a different fuel, you would have to redo the exercise to find- Sure your simplified chemistry. This is true. Yeah, so for every fuel you would have to redo that exercise, but it's a relatively fast exercise, and this is not an exercise I would say for practitioners to do. This would be something, uh, I mean, if it's a fuel that has not been studied yet, they would have to ask someone from academia, I guess, "Please tune those parameters for me now- Mm-hmm because I will have a different fuel that has not been studied before." Once you have that, and you have also the information of how many flamelets you can expect in your CFD cell, you can combine that and you can now do the transient effects inside the CFD cell. You can estimate that from the simplified calculations. With the

Wojciech 25:32

sub-grid combustion model.

Bart Merci 25:33

With the sub-grid combustion model. So you, I mean, this is a little bit of magic. I don't think the details are very important for the practitioners, but, but from that you can then get the reaction rate inside the cell-

Wojciech 25:45

Mm-hmm

Bart Merci 25:45

and the composition, and that you can then feed back into the CFD simulations for the transport of those species.

Wojciech 25:52

Does this happen Automatically within the CFD simulation without thinking. So every-

Bart Merci 25:58

Yes.

Wojciech 25:58

Okay.

Bart Merci 25:58

So, so, so this, uh, it's like, I mean, take FDS now and be- and because the FDS has a, I think, a very good combustion model as it is, it also has this, say, two-zone approach of thinking where you have reaction zones or reacting zones inside- Mm-hmm the CFD cell and surroundings. Um, the difference between the sub-grid combustion model that we've developed and the one in FDS is how we treat this finite rate chemistry. And so the- Mm what I've just, what I've just explained. the FDS model, also treats this mixing inside the CFD cell in a transient way. So this is also already very good. Yeah, so this concept of two-zone modeling is very good. The, the one thing where we really need to challenge and validate our models further is, again, these under-ventilated conditions, and this is, this is an area where we still need to, say, validate the models better. Wh- which

Wojciech 26:52

also, by the way, means, uh, being able to predict when flames extinct, right?

Bart Merci 26:57

Yes. Well, you can do that. And so a- again, take, take FDS as an example and, and, and, and- Mm but there are other models as well, of course, but take them as an example. What you can do as a user in your input file is you can specify an, autoignition temperature and an extinction temperature, a critical flame temperature. Mm-hmm. And so then once this is specified, uh, it is checked if you have reaction inside your CFD cell, do you reach the critical temperature or not? If not, then it's extinction. Mm-hmm. Uh, if yes, then it's ignition and the, and the combustion continues. So you can already model extinction phenomena with the models that we have. It's just that you need this extra step to define this critical temperature, and this is difficult to define in a universal way. With our approach, you don't need to specify that. It's this detailed chemistry calculations with the autoignition delay time that does the job for you. Mm. And then it's indeed checked, like does, does it ignite or not, but using, uh, these autoignition, uh, calculations rather than specifying an additional critical flame temperature.

Wojciech 28:02

If, if we may step back a little bit from the technical consideration because- Yeah you brought a very interesting practical space in here. It, it's the user of those tools, the engineer on the other side. Yeah. And, uh- You know, the, that's, that's also the reason why I brought you to the podcast because it's obviously very high level technical content. M- m- perhaps like people will never be able to, to, to apply that. But- Mm-hmm you know, if I have a fire in the middle of my shopping mall and I just do one megawatt heat release rate- Yes I, like, is it a pool fire? Is it a crib? Is it a burner? In the end, it's gonna be, it's one megawatt release- Yes in, in a, in a shopping mall. Yes. But if I put the same, uh, one megawatt in a closet with, uh, partially closed doors, it's a completely different fire. I- if- Yep it's a one megawatt in a vehicle with closed windows, it's a completely different fire mechanisms. And eventually, you know, something that was accurate enough for my one megawatt in the middle of a mall, it, it loses this, you know, accuracy or validity as soon as I move this into different geometrical context. And we do that all the time in- Mm-hmm in fire, you know?

Bart Merci 29:17

Yep.

Wojciech 29:18

So, uh, when you talk about universal, like I presume that for you, universal would be something that can capture all the physics. For me, universal is one that's gonna work in my car park and in my mall at the same time.

Bart Merci 29:30

Yeah. But the, these two are kind of correlated in a way. Yeah. Because indeed, if you have a model that automatically captures the reality of, of the boundary conditions, say, of the, uh, configuration that you're looking at, then it should be applicable in your shopping mall as well as in your car, as well as in your closet. And that's where the difficulty lies today, uh, that, practitioners, they-- You cannot expect from them that they figure it out to all the level- Mm of detail. I mean, the tool should be such, uh, that they can use it without thinking through all this level of detail. Today, this is not really the case, I would argue- Mm-hmm because then you would actually need to know quite well on how to deal with extinction phenomena and how to reasonably, uh, define critical flame temperatures. And, and, and this really is case-by-case dependent also to some extent because it's an artificial thing, this critical- Mm-hmm flame temperature. So, the question is also, like in a design, if y- if it's a shopping mall, probably you don't care all that much about possibly under-ventilated conditions because you have- Mm-hmm large geometries, you have a lot of oxygen anyway. So maybe you're not-- it's not that relevant in such- Mm-hmm configurations. If it's a car park, uh, for example- Mm-hmm yeah, then this may become different, particularly if the car park is not very large or, uh, if indeed the fire is inside the car, you have limited ventilation, you can have quite significantly different emissions. Th- th-

Wojciech 31:01

anything flashovered will already be touching that boundary more like- Yes most likely, right?

Bart Merci 31:05

Yes, absolutely, because in flashover, then you go very quickly from well-ventilated conditions to ventilation-controlled, uh- Yeah, yeah conditions very typically. It's a very rapid phenomenon also. Uh, so, so it- Mm-hmm there's a lot of transient phenomena taking place. That is difficult to do with the models as we have them right now.

Wojciech 31:22

Uh, I, I think the reasonable approach to this is to have models that are fine-tuned for these more complicated, cases, and they just work for an atrium case, right? Mm-hmm. Uh, rather than, you know, uh, it's probably not worth, uh, the, the 5% computational, uh, you know, optimization to have a model and burden the user to choose it. We just need good models- Yes in the tools that are appropriate. Yeah, yeah.

Bart Merci 31:47

Yep.

Wojciech 31:48

I mean, uh, yeah. L- okay, let's-- I, I appreciate that. Let, let's go back to the, to, to close the chemistry because you said also the final product are the yields.

Bart Merci 31:56

Mm-hmm.

Wojciech 31:56

Uh, so- I'm used to the world where I just specify my soot yield, which soot being my most critical thing I want from the model. Mm-hmm. Perhaps I specify my CO yield, perhaps I specify CO2 yield just to complete the model, et cetera. I need oxygen depletion in, in some context. But that, that, that's it. That's the granularity I'm, I'm, I'm looking at. If I, you know, take this made-up one megawatt design fire, I put my 0.1 soot yield on top of that- Mm-hmm and some, some yield for CO, how representative of reality that is and how big variation can be as the fire or the flame goes through those regimes that you've described?

Bart Merci 32:42

W- well, I mean, this, this is a neat reality. I mean, you will have to specify the soot yield, and in all honesty, with the, the sub-grid combustion model that we are- Yeah developing, we have not looked in detail at soot yet. And this- Yeah this is a, a separate story. And I think there we should also keep in mind that the fire engineers are not necessarily interested, I would say, in what happens inside the flame zone. You're much more interested in how much of that soot escapes from the flame, because that- How much goes

Wojciech 33:12

out, right?

Bart Merci 33:12

Because that, that is where you, you generate the smoke, uh- Okay,

Wojciech 33:16

because for you, soot is another reactive species that is created and destroyed at the same time.

Bart Merci 33:21

Yeah. Exactly. Exactly. And, and this is very difficult to, to describe- Yeah even difficult to measure, because, uh, in, in many of the techniques to, that try and quantify soot volume fractions, you need to have the emission of light. the moment that it escapes from the flame, it becomes dull, and it does not emit so much light, so it's very difficult to- Mm-hmm quantify even how much is escaping. Uh, so this, I think, is a, a, a problem that is, that goes beyond, say, the, the combustion model inside the CFD, because this is really like a, a combination of how much goes out, and that's where we are interested as fire engineers, right? Mm-hmm. And because this will determine- Yeah the visibility. It will determine also how quickly a, uh, a volume is filled with, um, uh, with smoke, which by the way is also, uh, with entrainment of air again. But that is the-- What I'm saying now, this entrainment of air into the smoke plume is relatively easy. Mm-hmm. If you have a sufficiently fine CFD mesh, this- Yeah the CFD does that for you. Uh, so there's not too much modeling required there because that, um, should be covered in the resolution. But how much gets out? That's, I would say, the $1 million question because you, you don't really know today how much that is. I mean,

Wojciech 34:32

it's, it's an interesting problem. I, I always was thought that i- it's soot yield is, is purely fuel property- Mm-hmm or kind of like- Yeah But I, I, I, I mean, I've seen a lot of fires- Yeah but I don't feel like that. Yeah.

Bart Merci 34:43

Exactly. It's not. Uh, I mean, it really depends a lot on, on the ventilation conditions- Yeah as well. I mean, in the SFP Handbook of Fire Protection Engineering, you will find tables, uh, where with different yields, and it's an order of magnitude difference between well-ventilated conditions and under-ventilated conditions. And obviously, it's also very strongly fuel-dependent. I mean, if you, if you have a mat- Yeah, yeah a mattress burning- Yeah or you have methanol burning, it's, uh, it's completely different, of course.

Wojciech 35:07

One of achievement of my career is to burn one, uh, liter of toluene in an open pool fire. That was not a great idea. Uh, uh, well, anyway- Yeah combustion was just one part at the bottom left side of your plot. Then you moved

Finite-Rate Sub-Grid Combustion Explained

Wojciech 35:23

into the flow and- Yes turbulence. Perhaps it's a- Yeah good point to, to start discussing turbulence. Yesterday, uh, well, I f- talked with some people, and, uh, one, one person said, "Oh, I finally learned that, uh, what LES is. That's a large eddy simulation." I'm like, "Okay, that's g- great." Mm-hmm. I mean, I, I mean, that's great, actually. Mm-hmm, mm-hmm. Yeah, that's a... So m- maybe, maybe let's, let's, uh, let's go there. You've already said LES simulations and- Mm-hmm run simulation. Let, let's try to do that starting with, uh- What's turbulence? That's a hell of a question. No, no. Uh, but yeah, w- how can we explain turbulence to people? Yeah.

Bart Merci 35:57

I'll try to explain it in a- Yeah, and, and, and- in a f- in a few minutes, um-

Wojciech 36:00

I know it's a hell of a question.

Bart Merci 36:01

No, no, but I think it's, it's a very important one because, because turbulence is... I'm not gonna say it's overlooked, but it's not something that, um, fire engineers talk about a lot. So what, what is it really? Turbulence is a very natural phenomenon.

Wojciech 36:13

Mm-hmm.

Bart Merci 36:14

So every flow has, in principle, the tendency to become turbulent. Uh, but there is always viscosity. Mm-hmm. And so viscosity, if you l- walk in air, we always have some friction, but that's very limited, for example, compared to the viscosity of water. Mm-hmm. And if you try to walk in a pool, then y- you know- It's, uh, yeah it's much more difficult, right? Yeah. So this, this is viscosity. Now, what viscosity does is it tries to dampen all the fluctuations that you have in a- Mm certain fluid, but in a fire, uh, of a reasonable size, typically, and the flow induced by the fire is sufficiently strong, uh, and it becomes turbulent by nature. Now, turbulence is chaotic. Mm. And that's why you... If you look at a campfire, for example, you see the flames dancing around- Mm-hmm on this campfire. That's not because of chemistry. That's because of turbulence, and that you see that, and they are, uh, dancing around. Now, there is structure in that chaos. It's not completely chaotic. Yeah. And, and these structures are called vortices, and in the literature they're called eddies. Mm-hmm. Yeah? So it's kind of a rotational motion on top of the average flow. Okay?

Wojciech 37:20

Y- y- I, I think, uh, for the listeners, uh- Yeah take any video of fire, any video of smoke- Yeah and you will... If you focus on that aspect, you'll see, just pick one. Like, if you h- if you see a flame puffing-

Bart Merci 37:32

Yes

Wojciech 37:33

focus on the puff when it- detaches from the flame, and it turns darker, it becomes smoke, and just follow the motion. Like, consider only that puff, and you will see- Yes it swirl, it grow in size.

Bart Merci 37:44

Absolutely.

Wojciech 37:45

And if you follow it, you'll, you'll, and you'll also see how it starts creating smaller swirls ar- around it in a way.

Bart Merci 37:51

Absolutely. Or next time you have a nice walk next to a river, and- Yeah and you see, like, an obstacle in the river, probably behind that obstacle you will see a wake. Yes, yes, yes. And if you focus there, you will also see some vertical motions.

Wojciech 38:03

And you can, you, you can some- sometimes also see it in shadows. Like-

Bart Merci 38:06

Absolutely.

Wojciech 38:06

Yeah that, that's, that's also very interesting. Yeah. Sometimes- Yeah uh, items cast a shadow in specific con- conditions, and you, in the shadow you already see. Yes. Like, that, that's, uh, actually a Schlieren method of, of imagining it- Yes where a lot of our knowledge came from, right?

Bart Merci 38:20

True, true. So, so we have these eddies. Yeah. Yeah, you have these vortices. Uh, but they are unstable by nature. Mm-hmm. So they kind of break up into smaller and smaller ones, and then the smallest ones that you can find in any- flow. These are called Kolmogorov scales. Mm-hmm. And Kolmogorov was a Russian, uh, researcher from the, the previous century in, in the, in the 1940s who, uh, developed theories on, on turbulence. But the smallest ones, yeah, so a- as the eddies become smaller, they also become slower as they, as they, um, uh, as they become smaller. They cannot fight against viscosity anymore.

Wojciech 38:56

But s- s- slower in terms of how much space they cover when they- Yeah rotate?

Bart Merci 39:00

Yes.

Wojciech 39:00

Because they can be a part of a s- subsonic, uh, flow at the

Bart Merci 39:05

same- Absolutely. Yeah. Okay. Absolutely. Absolutely. And so- So it's relative. Yeah, it's relative. It's relative. Yeah. And so, um, and they, they can't fight viscosity anymore. Mm-hmm. Like, they are too weak, so then they turn into heat. And so you have this turbulence that gets its kinetic energy because it's moving around. Mm-hmm. Its kinetic energy, it takes it from the flow field, and it's then dissipated into heat. But this is a continuous process because the flow field keeps going. It keeps feeding the turbulence, and some- and the turbulence then dies, let's say, at the smallest scales. And so th- there is an amount of kinetic energy, turbulent kinetic energy, and this is not at all uniformly distributed over these eddies. Mm-hmm. The largest ones, they also are the most energetic ones. Mm-hmm. The smallest ones, they are the weakest ones. They have- So kind of- least e- energy

Wojciech 39:49

every time an edgy, an eddy splits into two-

Bart Merci 39:53

Yeah, it loses- it splits

Wojciech 39:54

the energy.

Bart Merci 39:54

Yeah. Yes, indeed. Yeah, so the energy also is distributed, and each eddy then has less and less, uh, energy, right?

Wojciech 40:01

Yeah. Uh, wh- why do we care about how much energy is there? Like, what does this drive?

Bart Merci 40:05

So why is it important is because you- cannot afford to simulate all the detail of turbulence. Mm. This would again lead to unrealistic mesh resolution for the CFD. So you have to pick your battles.

Wojciech 40:18

Mm.

Bart Merci 40:18

And then it's best to follow the ones with most energy. Mm-hmm. And fortunately, these are the largest ones, because that means then that your CFD mesh, if it's fine enough, you can capture the energy at least of the largest eddies, which are the ones with most energy, which are the most influential

Wojciech 40:36

in the

Bart Merci 40:37

flow field. And

Wojciech 40:37

conveniently for us, they're also the architectural details of our buildings, right?

Bart Merci 40:41

Absolutely. Absolutely. Yeah. So, so that means that we are capable of resolving those-

Wojciech 40:47

Mm

Bart Merci 40:47

simulating those, and hence the name large eddy simulations. Yeah. Because you're simulating the large eddies.

Wojciech 40:53

Mm-hmm.

Bart Merci 40:53

How do you do that then, or at least how do you get rid of the smaller ones? Be- the ones that you cannot resolve is basically you put a filter on the turbulence. Mm-hmm. And you only look at the largest scales, and you filter away the smallest ones. Think about this as, say, taking a picture of a- Mm handsome man like yourself, huh? Thank you. And, and if you, if you take a very detailed picture, you see all the, the details of your, uh, your beard, so to speak, yeah? If you then start making the picture more blurry, yeah, you will still see how handsome you are, but you will not see all the details anymore. Uh, and, and so it's a little bit like this for, for now the, using the CFD mesh to resolve turbulence. You, you lose, say, the, uh, details of the smallest eddies. But you cannot pretend that they're not there-

Wojciech 41:40

Mm-hmm

Bart Merci 41:40

because that's the smallest eddies. That's where the turbulence get dissipated into heat. So if you simply ignore them, you would not dissipate the turbulence enough, and you would get an unrealistic flow. So you have to model that fact that these smallest eddies dissipate the turbulence, and that's where you hear names like Smagorinsky and Deardorff. I mean- Mm uh, if you look at the FDS documentation again, you will see that there the default model is the Deardorff model. Now, as a user, you don't need to worry about what this Deardorff model actually does. It's just good that it's there, because otherwise it would not give a realistic flow anymore. So this is again an example where the practitioner does not have to worry how the FDS developer did this. It's just good that it's there. And- But

Wojciech 42:25

it, but it arguably is a part where the practitioner has the biggest impact on all of this, because it's the practitioner who chooses the filter size.

Bart Merci 42:33

True. So the CFD mesh choice is a very important choice to make. Yeah. Because if that is not fine enough, then you might kill the turbulence too much, because the models that are in there- Mm they are trained to kill the turbulence. They are trained to dissipate the turbulence. So if you, if- But

Wojciech 42:50

basically we assume all the important stuff is above that.

Bart Merci 42:53

Exactly.

Wojciech 42:53

Everything beyond that we just average, and that- Exactly then that's it.

Bart Merci 42:57

And, and that's of course, I would say, an inconvenient truth in that in reality- People would not have the time nor the money nor the budget to do a mesh sensitivity analysis of the results. I mean, you would offer... You make an offer to the client. You do that knowing on your own capabilities in terms of computing power and in terms of, say, reasonable, uh, delivery times. You're not going to add simulations with a finer mesh because that would take much longer- Mm-hmm and it would be much more expensive. That's where academia plays a role again because there we do have time to invest in investigating this, and so we should give some guidance, uh, on how fine a mesh must be. One, one example is, uh, the D-star criterion in the- Yes FDS documentation. And that's, um, criterion has been developed, uh, looking at a fire plume and a smoke plume and what do you need to do to make sure that you get the large scale entrainment. However- Mm-hmm if you have, say, a different configuration where, for example, the flow through an opening is important, the flow through a window, the flow through a door, you should also make sure that you have enough resolution to have, say, sufficiently accurate flow through such openings. So you need to look at, say, the relevant length scales of your own configuration, and that needs to be resolved sufficiently fine. I, I

Wojciech 44:20

think practical example is jet fan systems in car parks. You also- Yes have done a lot of work on that, right?

Bart Merci 44:26

Mm-hmm. True.

Wojciech 44:26

Where jet, jet fan, like you can have a D* that sells, sells you like 25 centimeter mesh would be fine- Yeah for the fire problem, but then you have a 40 centimeter diameter jet fan. So you don't want to- No describe it with one cell, right?

Bart Merci 44:39

Exactly. And, and I know that you've also worked a lot on flows through vents. I mean- Yes natural ventilation. Also there- Same issue same issue, right? If you have, uh, if you say, uh, 50 centimeter mesh cell in a, in a large, a high atrium, uh- Mm could be, I mean, could be enough- Yeah depending on the, on the configuration. But if you then have a vent of one meter by one meter, yeah, you would have four cells, across that, that vent, and that's not enough often to have an accurate, uh, representation of the flow there. Now, it all depends on the, the details that you need because if this is about, say, smoke filling through an atrium, if the smoke layer is thick enough, then probably it doesn't matter all that much anymore- Mm-hmm how much is flowing out. However, if you're doing an A set, R set type of calculation- It's gonna matter a lot at the beginning, it's going to matter a lot, huh? So, so that's- I see that's where you need to

Turbulence Basics And Why LES Works

Bart Merci 45:28

pick your battles again, huh. I

Wojciech 45:29

mean, a- also, you know, uh, I know you use LES a lot. Uh, we tend to, to, to still use our RANS solutions for our smoke control problems in, in our, uh, in our practice. observing like, uh, my, my friend, Bert Blocken, also a Bert and also very good, uh, researcher, when he's doing a, a helmet analysis for a bicycle race, like he's, he's again getting to those hundreds of millions of, of cells, and he really cares about the, the boundary layers at the, the flows. I, I assume for him the, the, the viscosity at the surface to flow, flow interface must be critical. Mm-hmm. For us, it seems not because I, I've not, not seen those boundary layers- Mm-hmm in most of fire simulations. Like how, how important is that?

Bart Merci 46:14

Yeah. That's a, that's an excellent question. And- I'll, I'll first go back to RANS and LES because if you have, say, a large geometry and you know that you cannot afford to go sufficiently fine to resolve- Mm-hmm say, these large eddies in, in the turbulence, it may be a much wiser choice to go for a RANS model. Because in RANS, you model everything- Mm-hmm of turbulence. You know that in advance, so you don't need to worry also about tracking these large eddies. So if you do large eddy simulation on a mesh that is too coarse, you may get quite inaccurate results because as I said before, these models are trained to kill the turbulence, so it- Mm may become very unrealistic. So in that case, it really needs to be thought through if, if unsteady RANS could be much more accurate even than, than an, than a large eddy simulation.

Wojciech 47:02

And the, the, the way I viewed that, and I've considered this many times, like, you know, Because LES is very computationally expensive- Yes but also the time steps are much finer and, like, they have to- Mm-hmm match the, the, the flow, uh, velocities- Yes on your s- like it's, it's, it's more complicated. therefore more costly because of those comp- but, uh, if I end up with a half a meter mesh-

Bart Merci 47:22

Mm-hmm

Wojciech 47:22

you know, because of practicality- Mm-hmm. Yep of the problem solution. And then most of the eddies would be cut at that 50 centimeter mark, and then everything else would be resolved with some simple model, you know, that, that you use to, to, to account for that. And I could end up with a situation where most of my energy is actually in that end part. Yep. And, and in that case, if I just went k epsilon model and there's simply more equations that accounts for more things and gives a little bit more control on how it is resolved.

Bart Merci 47:55

Yeah.

Wojciech 47:55

for me it's perhaps a better sim- of course, if I went LES and did five centimeter mesh of the problem- Yeah, yeah. Yep it would be better.

Bart Merci 48:01

Yep.

Wojciech 48:01

But it's not reason- No feasible

Bart Merci 48:04

in that case. Exactly. And that brings us back to exactly what it is that you want to simulate, what it is that you want to resolve. And so this was, say, one s- end of the spectrum, huh, where you go for RANS rather than LES. If you go to the example that you've, you've given of, of colleague Berg Blokhin, where he is then very much interested in boundary layers, uh, around helmets. Well, if you go to an academic problem now again of fire spread along a surface- Mm-hmm or flame spread over a surface, then again these boundary layer phenomena become extremely important, and then you need to invest in resolution of- Or

Wojciech 48:40

forest fire

Bart Merci 48:41

or in forest fires. Again, so where you have very direct interaction between flames and surfaces.

Wojciech 48:46

Mm-hmm.

Bart Merci 48:47

And there I, I, uh, it's also something I, I mentioned yesterday in the lecture, I think non-intrusive measurements of flow fields and temperature fields, and even sometimes species concentrations, will become very, very helpful to improve the near wall treatment of turbulence- Mm of turbulent combustion. and I think there, there's a lot to be done still today- Mm-hmm because, because these techniques, these laser-based, uh, diagnostics techniques have not been applied all that much in the context of fire applications. They've been in, they've been around for decades in the combustion community- Mm-hmm but it's only fairly recently, and that this has now found its way into the fire community. And, and there I have very high expectations, and that also the CFD modeling for near wall phenomena will im- will make significant steps forward. Uh, uh, because today, for example, for convective heat transfer, yeah, we're still using correlations that have been developed for a heated plate, uh- 100 years

Wojciech 49:51

ago.

Bart Merci 49:52

Well, not hu- yeah, more or less 100 years ago, a heated plate, and then you have a cold airflow or a cold plate, and you have a hot airflow, and you have a free stream temperature. But what would be the free stream temperature if you have a surface, a flame, and then the surroundings? So, so-

Wojciech 50:06

It sucks that it's a podcast because now I could put a meme, you know, uh- you guys are calculating that. Like there's, there's a 35 in the Eurocode. Come on.

Bart Merci 50:16

Yeah, yeah, yeah. But- But, but th- this, yeah, true. Yeah, no, th- th- this, this is a, an, uh, a very valid point, yeah, that, that you're making, huh?

Wojciech 50:22

But, but it's, it's, it's actually the beauty of fire science because there's always a golden number for everything- Mm-hmm uh, like soot yield, like convective heat transfer coefficient. But if you start questioning it sufficiently, you end up with a PhD and a long career in academia. Yeah.

Bart Merci 50:35

That is true.

Wojciech 50:36

No. That's the risk if you- Yeah ask too many questions- Yeah, yeah. Yeah, yeah in, in, in, in that, uh, regard. Yeah. Uh, uh, we'll come back to the, experiments in a second. I was fascinated in your lecture- Mm-hmm that you as a modeler have given so much space to experiments, and I- Mm-hmm I love how you link them. But, I would just want to close this, uh, with how that knowledge of turbulence of the flow then translates into the sub, grid combustion models and the whole concept of, Treating the flame that, that you develop

Bart Merci 51:05

Yeah. So basically what, what we essentially need for our sub-grid combustion model is a notion of how turbulent the flow really is inside the cell. And we do that, uh, by using what is called a turbulent Reynolds number. Uh, so Reynolds number is a combination of velocity and a length scale and a viscosity. The viscosity is the viscosity of the gas that you have. Mm-hmm. And the length scale in this case would be the size of the grid. Uh, and then the, uh, velocity would be, say, an, an indication of the turbulent velocity of the structures that we have locally. And depending on that, uh, you can then determine what the, uh, reacting situation looks like inside the CFD cell.

Wojciech 51:48

Yeah. It, I had a funny thing, like ages ago, someone asked me like, well, in a, in some setting like, "What is the Reynolds number?" And I- Yeah I didn't have a good, a, a good answer in my head. And I-

Bart Merci 52:00

Mm-hmm

Wojciech 52:00

That's a really good question." Yeah. "What really is the Reynolds number?" Well- And then the person went for like 30 minutes and then talked Okay. Okay.

Bart Merci 52:07

About what is- So, so then I don't need to repeat that, uh, no.

Wojciech 52:10

No, but I, I always found the challenge in the length scale, actually.

Bart Merci 52:13

That is true. No, no.

Wojciech 52:14

That, yeah in, in this, in this case, like, okay, if I calculate a Reynolds number for a pipe flow for my hose- Mm-hmm uh, which is also a relevant fire problem, uh, the friction in the fire hoses, that's easy. Like, I- Yeah the length is I can take the hose diameter or- Yeah the hydraulic diameter or whatever- Yeah, yeah you know? But, uh, if you consider... And, and you said you're zooming in. Mm-hmm. So every time you're considering, you're smaller and smaller.

Bart Merci 52:36

No, no. So th- this, this actually links again to this, this breaking up of these eddies. Okay. If, if your mesh gets finer and finer, it means that you, see more and more of those eddies, and you need to model only the smaller ones. So actually from an asymptotic reasoning now, you naturally get into the right direction- Mm-hmm because now your, timescales, but also your Reyn- turbulent Reynolds number becomes smaller and smaller, so you need to model less and less. Mm-hmm. And so, and, and it becomes easier and easier to, uh, assume what is going on inside the cell. Again, take s- if you take 20 centimeter by 20 centimeter- Mm-hmm cell, you can have a lot of things happening inside- Mm-hmm the cell. If it's a two centimeter by two centimeter cell, uh- It's much less the uncertainty is much less of what happens- And then it's- because everything else has been resolved half mil

Wojciech 53:26

by half

Bart Merci 53:26

mil- Exactly

Wojciech 53:26

is gonna be- Yeah.

Bart Merci 53:28

Yeah. Yep.

Wojciech 53:28

And, uh, how does that knowledge of how turbulent the flow is translates into the chemistry

Mesh Choice RANS Versus LES

Wojciech 53:35

and, and the solution of, of what's inside the-

Bart Merci 53:38

Yeah. It does not go into the chemistry. So the chemistry is really like autoignition based. Okay. Yeah. No flow field, no turbulence there. You assume there the perfect mixing- Mm-hmm of, of fuel and oxidizer. It does get into the flamelet structures inside the cell. So the chemistry in the flamelets is not affected by- Mm-hmm the turbulence, but what the flamelets look like and how many you have inside the cell, that is determined by the turbulence. Well, can, can you

Wojciech 54:03

define a flamelet for the listeners?

Bart Merci 54:05

It's like a small, think of it like a small candle flame. Okay. Uh, so how many of these small candle flames would you have, and how distorted would the candle flames be inside your CFD cell? That's more or less.

Wojciech 54:16

So, so if you take a high detailed picture of a flame and just capture one snapshot of it- Yeah and you've actually put the exposure right so you see the flame, Yeah which is important for anyone shooting pictures of flame and- Yeah you get that by ex- you get there by experience. But anyway, if you capture that and you would like to trace the contour of the flame on that- Mm-hmm picture- Yeah it's impossible to track because it's not continuous. Like, there- Yeah you'll have lines- True, yeah everywhere. Partic- So those will be your flamelets.

Bart Merci 54:42

P- p- yeah. Indeed, particularly if you were to blow very gently into the candle flame, because then you would- Uh-huh see it's distorted. Yes, yes,

Wojciech 54:48

yes.

Bart Merci 54:49

Uh, if, if you blow too hard, you will extinguish it, but if you gently blow and then you see it's distorted, and it's that type of structure that you should imagine, uh, inside CFD cells.

Wojciech 54:58

And, and now the knowledge of how many the flamelets are in there or how complex the motion within the cell is, how does that translate to the final product, the heat release rates, the temperatures, the yields?

Bart Merci 55:10

Yeah. Because, uh, so now once you know what the situation looks like inside the CFD cell, you would solve, transient equations. And what do these transient equations look like? They have the chemical source terms in them- Mm-hmm and they also have the diffusion in them. Okay. So that is then the mixing of the combustion products and the fuel and oxidizer with the reacting zones, and that's what you then feed back into the CFD, the results.

Wojciech 55:34

And that, that gets a gas pedal. So because you assume- Yeah they burn immediately, so the diffusion will drive the whole process in that case.

Bart Merci 55:40

Yeah. Well, depends on what you call immediately now. Very quickly- Okay. Very quickly but it's fi- it's finite rate. Finite rate. It's finite rate. Yeah. Okay. Okay. Yes. Yes. Yes.

Wojciech 55:46

Okay. Good. Good. Good. Good. Okay, fantastic. So, so- where are you today with development of those models? And you are also now is a great moment to take, give credit to your amazing team, uh, at Ghent University. Like- Mm-hmm where, where are you now with, with creating those models and, where- what's your nearest field of application? Like

Bart Merci 56:07

w-

Wojciech 56:07

Yeah are you just developing them for fun or for something special?

Bart Merci 56:11

Well, we have fun and, uh, we find fun in our jobs, but, um, but jokes aside, uh, the, we are quite happy with the framework- Mm-hmm as, as we have it now. We, we think that the basic philosophy that we are following now with these turbulent structures combined with auto ignition philosophy to describe extinction and ignition, we think that's a solid framework. And why do we think that? Because we've already validated now this approach without tuning parameters to a range of open atmosphere flames, so flames out in the open. What we are doing right now is go into compartments.

Wojciech 56:47

Okay.

Bart Merci 56:47

That's interesting. And so now we are in the phase of, uh, validating, of challenging our own model for a range of conditions. First step is with well-ventilated conditions to see how that works. Uh, and then the next step will be under ventilated conditions. So this is where we are. So we are now in that stage. So far results have been very promising.

Wojciech 57:06

It, so it works for burners and now you're moving to, like, uh, real fires.

Bart Merci 57:09

Yes. Uh, so, so far so good. Even also in flame spread configurations- Okay so far so good. But well, we're not there yet. We also still want to challenge them for under ventilated conditions.

Wojciech 57:20

Does for, does orientation matter a lot in here?

Bart Merci 57:22

Like- Um, in principle for our model, it doesn't. Um, so but we've al- so far only tested it for downward flame spread. Okay. So we cannot make, firm statements yet on, um, on lateral or upward flame spread.

Wojciech 57:35

Oh, which is interesting because also we haven't covered that, but you also mentioned advection and, and that the fact that the CFD, for CFD it's natural that you solve the flow- Mm-hmm and the buoyancy and, you know, the upwards, flows that are created by fires, et cetera. So in essence, your model works at lower scale, so-

Bart Merci 57:51

Mm-hmm. It should work

Wojciech 57:53

it should work in lower scale. It should, yes.

Bart Merci 57:55

Yep.

Wojciech 57:55

So it naturally takes from this flow component-

Bart Merci 57:58

Yes

Wojciech 57:59

that's resolving in the, in the complexity of geometry, but- Yeah

Bart Merci 58:02

Yeah, it should work. One, I say, issue also with, um, uh, well, with downward flame spread but even more so with lateral flame spread is also what you do in the solid phase. Because there- Okay uh, most of the models assume 1D conduction, uh, so perpendicular to the surface. But if you have lateral flame spread and downward, maybe also 3D conduction, uh, inside the solid plays a role, and that's of course beyond the sub-grid combustion model in the gas phase. Yeah. So there is multiple phenomena, and that's a, a problem in fires in general. You know, we have so many interacting phenomena that, uh, sometimes if you improve a certain sub-grid model, maybe because of previous compensating errors, the result looks worse than before, yeah? So

Wojciech 58:47

Y- y- you may laugh at me, but you know, my very, very, very first CFD, uh, try, you know, I, I attend a lecture of, uh, late Professor Konieczny in, in, uh, in, uh, mine school fire service. He shows, uh, modeling, uh, of fires. I'm like, "Wow, you can model fires. That's awesome," you know?

Bart Merci 59:04

Mm-hmm.

Wojciech 59:05

There's a software, FDS. You can just download it and, uh, okay, that's, that's great. Like, I'm a student. I have a laptop. Mm-hmm. I have internet. I'm just one step away from it. I download FDS and I... Let's start with something simple. I don't know, a timber Oh, God. Yeah. My first block exploded, the second didn't ignite, and then I, uh, choose to become a modeler. Really, it was like, like that.

Bart Merci 59:28

Yeah, yeah. Yeah,

Wojciech 59:29

yeah. Yeah. So, and, and we, here, here we've not even touched the, the pyrolysis- No, no and every that, that, that, that- Yeah that, that's, that's beyond. Uh, w- what about applications in, like, microgravity? I know you've just started- Yeah the, the, the Synergy grant, uh-

Bart Merci 59:42

Yes

Wojciech 59:42

which is absolutely amazing. And, and is this also something that's gonna play a role in, in, in that?

Bart Merci 59:48

Yes, absolutely. And so the, the plan is for the, the sub-grid combustion model to also be applied in microgravity conditions. We, uh, Alex Negurev from the team has already done that, uh, by now. And, and again, very, uh, promising results. And so it, it does, uh, indicate that the approach is sufficiently universal to be trusted in, in different, uh, in different, uh, conditions and, and, and, and configurations. What is fundamentally different in microgravity is that you don't have buoyancy. Yeah. Uh, and, and so many of the models that we have, uh, implicitly assume that there is some buoyancy, or stated in a different way is-

Wojciech 1:00:26

And is there sufficient turbulence?

Bart Merci 1:00:27

Uh, well, there is less. There is much less turbulence, so this is- Okay this is one, thing indeed, uh, but also the timescales become different. Because of the lack of buoyancy, then radiation also be- has more time, uh, to, um, to take place. Uh, so then you have, uh, impact on temperatures, and if you have impact on temperatures, this is then also an input for the finite rate chemistry. And so there is a lot of, say, uh, effects that, um, start to build up and start to combine, uh, and have an impact on the outcome.

Wojciech 1:00:57

And now what's going through my head is that this could also be brilliantly applied to understand better small-scale tunnels- Mm-hmm because, uh, when we use Froude, well, we don't use that much. Mm-hmm. But the way if you, if one uses Froude-scale modeling- Yep for tunnel fires and they go into very small burners, what you do is you scale the, the Froude number of the problem, but the Reynolds number of the problem is, like, in a completely different, you know- Yep. Yeah range. Yes. Like, you can go from 100,000 Reynolds number fire that's extremely turbulent in a full-scale tunnel, and something that's almost laminar in, in a little model if you- Mm-hmm And then how much emit from the surface also. Like, uh, th-this, this could be a something we could look into. Like, like, uh- Yep I, I, I love how actually this, every-everything what you've said in this episode so far, how that actually is block that we need to understand this particular relation, which has- Mm-hmm a huge impact on, on the big field of science and engineering. So- Yep so awesome.

Bart Merci 1:01:58

But indeed, scaling is a very important thing. Yeah. And, and if you go, but if if you go too low, then the turbulence will be killed by the Reynolds number that becomes too small. Yeah. That's not going to change with the subgrid combustion model. Uh, so, so- Yeah the flow itself, that the, whether or how turbulent that is, that should not be affected by the subgrid combustion model. But- But you will indeed, in principle, have a tool that can still allow you to indeed look at details that you find at this, um, reduced scale. Combustion

Wojciech 1:02:26

rates,

Bart Merci 1:02:26

the- Yep. Yep

Wojciech 1:02:27

the,

Bart Merci 1:02:27

the

Wojciech 1:02:27

heat release rates- Yep et cetera, right?

Bart Merci 1:02:28

Absolutely. Yeah. Brilliant. But whereas if you have a combustion model that assumes fully developed turbulence, then that would break down- Mm if the flow is not, not so turbulent anymore.

Wojciech 1:02:38

You, you, you have mentioned, uh, the, the high level diagnostics, uh- Mm-hmm uh, and, and

Validation Plans Microgravity And Diagnostics

Wojciech 1:02:43

you've shown a lot of that in, in, in the, in the presentation, in the lecture. what are you looking in for those experiments like that, that we don't have yet? Or, or, or w-what is the ultimate, uh, experiment that gives you the most knowledge? You, you- Yeah you've shown some amazing things, and I will chase those people with the microphone you hold in your hand right now- to, to have them, uh, comment on that. I, I

Bart Merci 1:03:08

encourage you to do so. Yeah.

Wojciech 1:03:10

I, I will do so. But, uh, h-how does that translate to your work as a modeler?

Bart Merci 1:03:15

Yeah. So- Two things. Uh, so one is the heat transfer.

Wojciech 1:03:21

Mm-hmm.

Bart Merci 1:03:21

So if you, if you have detailed information on exactly how far the flame is away from the surface, and also how, what temperatures you find locally, and ideally also what composition you have inside the flame, because the composition will determine also the radiation, uh, from- Mm-hmm from the flame. If you have that information, then you get, uh, very detailed, uh, numbers on the effective heat transfer from the flame to the surface. In practice, what we do today in CFD is we have a correlation for convection.

Wojciech 1:03:55

Mm-hmm.

Bart Merci 1:03:55

and we calculate radiation. Calculating radiation is also done in a simplified way, but okay. But the convection, this is based on, uh, these old correlations that have been developed in non-reacting flows.

Wojciech 1:04:09

Mm-hmm.

Bart Merci 1:04:10

So this is a big unknown, uh, as to how accurate these really are, particularly if they are combined with pyrolysis, because then you have outflow of gases that will also disturb the, the local flow field. So there, I think it's, it's very useful to, uh, also have flow field information close to the surface, because this is something where we cannot really challenge our CFD directly because we simply don't have the information. You don't have

Wojciech 1:04:36

a- Yeah. Yeah you just see the out- outside of the flame- Yes not the inside of the flame.

Bart Merci 1:04:39

We're sailing blindly in that- Mm-hmm in that sense. Uh, so that's why I hope this will help. And, uh, I, I didn't touch it yesterday, um, but ob- obviously everybody's talking about artificial intelligence and machine learning. I think also there, these experiments can play a role because they can, create a database where you can train. say a machine learning-based model that maybe we can use to replace the existing near wall treatment correlations. I mean, we haven't dug into that at all, but this might be an- Mm-hmm a field where- Yeah, I was gonna say- where machine learning can, can play a role.

Wojciech 1:05:13

Uh, th- there's, uh, s- plenty of complexity. I'm look, I'm looking now at the, at the figure three of your paper. Mm-hmm. Yep. The, the one, uh, where you, you show, uh, reacting regions of, of a flame in a op- opposed flow. Is that right? Yeah. That is

Bart Merci 1:05:25

correct, yes.

Wojciech 1:05:26

Yeah, yeah, yeah. In microgravity, yep. Uh, in microgravity, and then you have, like, maximum heat release rate in one point of the- Mm-hmm of the plot and maximum temperature in another point of the plot. Yes. And, and that for me felt very non-intuitive. I love when- Absolutely. Mm-hmm I love when fires n- are not intuitive- Yep because that challenges curiosity. Yes. Why, why Bart?

Bart Merci 1:05:43

Well, uh, it's a, it's a very direct visualization of these finite rate chemistry effects, because if y- if you were to assume infinitely fast chemistry, then these two points would be very close to each other.

Wojciech 1:05:54

Okay.

Bart Merci 1:05:54

Uh, and now it's not the case because of the chemistry on the one hand, but also because of the flow field that is actually the larger the flow, the, the flow field velocity is in the, in this plot from the left to the right, the more the hot gases are pushed upward in this case, and because this is downward flame spread, and the, the further these points would be away. Now, this is something that is actually impossible to predict without doing detailed CFD, because intuitively you would put- Mm-hmm the two dots on, on, on the same plot. I don't- Yeah, and-

Wojciech 1:06:24

We don't have a tool to put into that flame and measure exact- Exactly.

Bart Merci 1:06:27

So that's where you would have these non-intrusive diagnostics again, where you can then have a direct one-on-one, say, uh, comparison- Mm-hmm to see whether or not this CFD makes sense. Um, but it's a big step forward also compared to, say, the, the research that was done in the '60s and the '70s of the previous century, where, uh, actually people had to guess what the temperature would be and, and a uniform flame temperature would be assumed, which is very reasonable.

Wojciech 1:06:55

Mm-hmm.

Bart Merci 1:06:56

If you have to do something, that's also what I would've done probably. But now we can do better than that also in terms of, um, experimental measurements.

Wojciech 1:07:04

looking at the picture, I, uh, the link to the paper is in the show notes- Mm-hmm uh, if any listener wants to follow, uh, up w-with us, it's figure three. I- That, that means that the reactions are also happening all, all the way up, right? They, they don't, they don't just happen in that star where there's max heat release rate, right?

Bart Merci 1:07:20

That is true. Uh, so they, they happen over a range, uh, in, in, in physical space, but it's also a combination of advection, uh, of hot gases upward, in this case, by the flow field. But it's true. Mm. And so the, the heat release rate does not happen in a single point, uh- Yeah it's sp- uh, which is also, again, uh, an outcome of these finite rate chemistry effects. Because if it's mixed, it's burnt, and you have stoichiometric conditions, you would say, "Okay, this is, this is- This is it this is where it is," huh? Yeah. So

Wojciech 1:07:47

if, if this was my CFD model and I had a cell that just covers your, your blobs, uh, using the old approach, I would release everything in the, in the, in the cell in here because they were sufficiently mixed at that point.

Bart Merci 1:07:59

Well, you would have, say, a, a, a flame, uh, uh, let's say a stoichiometric, uh, iso-contour, so it would not be in a single point, but it would be- It would be- mo- at, at, at this contour, uh, al- mo- almost, say, the same value everywhere.

Wojciech 1:08:12

Which is brilliant. I mean, uh, y- you then go into visualize using flames, and you have, uh, those iso-contours of temperature usually or some energy density iso-contours. Mm-hmm. Uh, we like to visualize flame by the iso-contour of, you know, uh, stoichiometric equivalence. That's also a nice way to-

Bart Merci 1:08:29

Mm-hmm

Wojciech 1:08:30

to visualize that. But, uh, only looking at that and, and trying to understand, you, you suddenly understand that the physics is so much richer, you know? Mm-hmm. Yeah. That perhaps the image that you see is just, you know, a nice i- illustration to show you, oh, the flame would be more or less there- Mm because the engine, the energy density is there. Yeah. But in reality, the chemistry already happened somewhere else in your model that you didn't account for. Yeah. In many problems, that's fine, but- Yep you may encounter that one where it's not.

Bart Merci 1:08:55

Mm-hmm. Mm-hmm.

Wojciech 1:08:56

And I'm really thankful, Bart, that you're there to stop me from doing that, developing the good models- that, that, uh- Yeah forgive me a lot of my ignorance and, and- Mm uh, lack of, of, of knowledge and, and, uh, are appropriate for the far problem that we're studying. Mm-hmm. Thank you so much for, for doing this, Bart.

Bart Merci 1:09:12

With a lot of pleasure, Wojciech. That's, uh, my task. It's my job, so.

Wojciech 1:09:17

Ah. And, and you're, and you're great at it. Yep. Thank, thanks, Bart.

Bart Merci 1:09:20

Thanks, Wojciech. With pleasure.

Wojciech 1:09:21

And that's it. Uh, I hope you enjoyed that. Oh boy, what a journey that was. Uh, he called

What This Means For Practice

Wojciech 1:09:27

me handsome on air. That's, that's, that's great achievement for myself. But more importantly, what Bart is doing is that he and his team, they're developing models. Models that eventually will find their way to mainstream fire safety engineering and will improve our capability of doing what we do. You know, we've talked in the podcast about the difference between modeling or predicting fires and, you know, just using the design fire assumptions for design purposes. And, uh, today those worlds are so far apart from each other in terms of what can be realistically done, from a perspective of a constrained fire safety engineer who has a budget, who has limits, who has deadline And, uh, with developments of those sub-grid models, we're getting more and more to the space where there's, you know, real considerations of fires, flames, chemistries, productions that respond to the conditions in your model, that they become achievable. And, and that's, that is an outstanding future for, for us, for fire safety engineers. What you've seen in this episode is probably the high end of academia. I mean, Bart just gave an Howard Emmons lecture. That's, that's the top of the top. That's, that's the biggest, uh, recognition a person can have in the, in the fire safety science community. So obviously, this is the absolute top of, uh, of what we have. Uh, at, uh, at, at the same time, he's so-- he's keeping touch with the reality of, of our profession and, and perhaps that's the part that I appreciate the most. Um, Bart's paper is out there if you would like to learn more about sub-grid combustion models, and there are more episodes of the Fire Science Show on turbulence, on other aspects of flows, on CFDs, on scaling, on practicality of CFD. So if, if you would like to listen more about those, uh, aspects, uh, feel free to do that. There, there's a lot of content we already have, and I think I'll just stop here. I'm just really, really thankful for Bart to taking some time at the conference to sit down with me and give me this interview and share this Emmons lecture, not only with, uh, the academics present there in the room, but also with the practitioners, uh, listening to the Fire Science Show. And, uh, yeah, IFSS was a fantastic place. It was great to hunt people down for interviews and I have so much more, to share with you in the upcoming weeks. So the next weeks of the Fire Science Show are going to be amazing, uh, and I would love you to enjoy them with me. Uh, next one coming next week, next Wednesday. See you there. Thank you. Bye.