Hello, everybody. Welcome to the fire science show. Nice to have you here is going to be hard to beat the last week's spacecraft fire safety episode by. Let's give it a try. Today, we're starting Let's say a mini series in the podcast. I'm going to make a. Group of, themed episodes. They will be themed around the experiments that have changed the fire science. Which means I'm going to discuss here the most impactful experimental programs we had in the fire science.
With the scientists who have performed them or ones that are very well-informed about the curves of the experiments. We're going to talk about their legacy. We're going to talk about what. Happened with let's do these experiments. Why were there needed, what were the ideas when they were conducted? What questions they try to ask? And. In the end. What did they change overall in the fire science? And today we're starting with the first one. It's on the round Robin pod of Dalmarnock fire program.
For me, it's one of the most inflation pieces of research that I've learned during my career. And they And they have really laid a new perspective on using of numerical tools, models. The difference between modeling and model. something that, that, from my own perspective from my own career has really changed. The landscape of, of use of modern tools.
The Dalmarnock fire program was carried on the supervision of professor Jose Torrero by the university of Edinburgh and it was obviously much bigger than the part that we're going to touch about today. It was first and foremost, motivated by creating more visibility for fire. Professional fire engineering profession. Bye. Making a documentary, a Skyscraper Firefighters that aired on BBC two. And, I was a major thing to, to produce a large documentary about compartment fires.
And that that goal was certainly achieved. It was also part of a Fire Grid program, a much larger study. It had massive instrumentation, all of this you can find and read, but here today I would like to focus on one. Let's say little aspect of this. Experiment. And that was the round Uh, Round robin was a, an attempt to send the data to some fire modelers. Who. To try and predict what would be the consequences of the fire in the building. Then they've built, burned the fire.
And then they've compared them. I know, boy, it was interesting too, to look at the data and that is exactly what we are going to talk about in today. I've invited professor Guillermo Rein, who was the lead scientist behind, a priori the before fire experiment, uh, round Robin. And Dr Wolfram Jahn at that point was leading the post-fire modeling, where they tried to improve the quality of, of modeling and predictions. Given the knowledge gained from the experiments.
So I hope this will be interesting to you. I've linked the Dalmarnock papers in the show notes. So if you want to learn more about the experiments themselves, the motivation and the other aspects of it, you're very welcome to do that. There are the round Robin.
Apriori posteriori papers as well a lot of a lot of knowledge have very fundamental research and now let's hear from the scientists why was it important and why you actually should read them so yeah that's been the intro let's go Hello everybody. Welcome to the Fire Science Show. Today we're gonna try something different. Well, the same thing, but a little different. I thought that, it could be nice to learn about what made our fire science the thing. Have today.
And for that, I would love to discuss some of the most famous, most impactful experiments and papers that that happened over the course of the decades as Fire Safety Engineering was developing. And to start it off, I have one that was suddenly impactful for my own career as a CFD engineer, and that is the round robin part of the Dalmarnock Fire Experiment. And today I have with me two lead researchers of that task in that big project. first Professor Guillermo Rein from Imperial College London.
Hey Guillermo nice to see you here for the fourth time.
Thank you, but it's great to be back.
I'll have to buy you like a t-shirt or something for the fifth one, I guess
I would love a t-shirt and a, a pen, or a book, or
I don't know. I'll, I'll figure out something. But it's great to have you and gui, you are leading the, a part of CFD experiment there, and I also have the head of the
a
posteriori
hey, how are you? Tich nice to be back.
Nice to be back. Yeah. You also have been a guest on the show talking about cfd, which perfectly ties to this
this in the previous e episode as well.
you, , if you have not, uh, heard the Wolframs episode, it's very impactful and you should just pause now. Go. I think it was episode 14 or 15. You were a very early adopter of the show. Thank you very much for that. Both of you are okay. Let's, let's jump into the Dalmarnock because there's lot to talk about that. So, fir first question, Why the hell such a project happened?
how did we end up having this massive set of fire experiments in the residential compartments, carried out in, what was the, the grand idea? I know that from the talks we, we had before that Ron Robin came to the whole program at some point of, of its development. Let's first talk about why we needed Dalmarnock in the first place. Guillermo, you were there observing it firsthand.
So Dalmarnock experiments of 2006 is the brainchild of Professor Jose, Jose Torero. He was at the University of Edinburgh at that time, and he convened the housing authority in Glasgow, the fire Brigades and the TV programmers of Horizon Show, which, signs the commentaries. He convened the three of them to have these experiments in a real building, in a real apartment, not in a, in a real building, but it was only two apartments that got two fire.
and he, made this opportunity an incredible and super exciting opportunity because they remain the, most densely sensed experiments conducted to date in fire science. And this was done in 2006 it's not been beaten yet because the amount of sensors of all kinds that were inside those two flats is incredible. many things happen in Delmar.
No. And afterwards, but one, one of them, one of the things that happened, the one that you have invited us to talk about is the modeling exercise around, Dalmarnock. And at that time, Jose and I were in very active discussions, about what is the difference between a model modeling, right?
So both are scientists and engineers, and we were concerned that if one says I've validated the model, that doesn't mean that the modeling apply by another person in another context means that the same level of or, position is, is applied. And we didn't know how to make this point. And I remember we had a meeting in London with our top and top FDS developers, and out of that meeting, I told Jose, Look, there is a way that we can address this. can do round robining.
And at that point Jose said, What is that? And, and I explained to him the idea, which I, I can, I guess your wants to know as well, what's around Robining and around round is when you have a task, for a group of people and everybody does the task independently, and then you put all the results and you compare what the group has done. Right. So the key is that this, they're independent.
It cannot be that I'm looking into who's at the left or who's in front use that results to guide my, So they are independent and you put them all together and you compare them. You have do a benchmarking of one to the other. So we said Dalmarnock is gonna happen. We tell everybody, the modelists around the world what Dalmarnock is going to do. And before we do Dalmarnock or before they see the results, actually it's more correct to say before they see the results.
Hmm.
we tell them to simulate this, fire, the development of the fire way they want it to do. They should simulate this before they see the results. And that's what we call the a priori round robin, priori for the Latin of ahead of the event. So before the fire happen, many people around the wall try to simulate with different software tools would have happened. Then they send us the results So they, they send us their predictions and then we send the results.
then the second part is what Wolfram led, which was a posteriori for the latin of after the event, which is,
Robin. Actually, it was just
which was not around Robin. Indeed. Good point.
Why is
was just one, one team, Well, me and Guillermo basically. Um, so it was just one team. It wasn't, it wasn't many teams kind of competing to do it. I think there's, or there have been more attendance of modeling that, but I don't think any of that has been published. think ours is that was actually published and
this is important Wolfram this is really important why they didn't publish.
because, um, it wasn't, it wasn't easy even knowing what happened. It wasn't, it wasn't easy to match the results. It was very, very hard actually to come up.
So this is, this is the key This is why we did the round robin. This is literally why we did it because, you go into a literature and the scientific literature and this, this massive filter in the scientific literature that only if you get amazing results, publish, you submit to journal. mean, you have to be out of your mind really to submit to a. An exercise where you prove you are useless at simulating the experiment. However, editor, please, everybody should know how crap I am at modeling.
that's why in the literature you only see beautiful examples cfd or it doesn't have to be a cfd cuz this happens to every single discipline. Non true is you publish only when the results are amazing in your modern You don't publish research. Negative have good press, doesn't create careers, doesn't take the time of editors. So with round Robin what we wanted to say is like, well, a group of people known to the field are gonna try to do this. the experiments are amazing and well done.
some of them will not do too well. It will be negative them. The other ones hopefully will do well. That will be positive altogether. think this is something that everybody will want to know about and that's, and that's what we did. We wanted to, to see the part of the filter that never gets to us. in the literature, which is the tens that failed to simulate it was attempted to do at the beginning,
we have to come a little bit back to, to what's from the beginning that why were these experiments done, in the first place? It wasn't, it wasn't about modeling in the beginning. It was about trying to. Being able to predict the fire based on, cuz it was all in the context of this Fire Grid project. the big idea able to predict fire development based on sensor data and trying. Then once you have the prediction, trying to manage the fire, that was the whole paradigm I think.
at some point we'd realized that we need modeling, but we can't do it. And that's basically the, ongoing problem it's been years and there has been lots of development many aspects, but some of the main conclusions remain. We still can't blind predictions.
if I ask you to move back in time 15 years ago, what, how did the modeling look back then? I mean, what version, what, what version of FDS we even had? Was there this four
I, I did the posterior modeling with four and I was criticized, I'm not using five. And so it was, was about that time that five appeared.
Okay. And app was, was definitely four then.
One thing because we just did something that I did it as well and we should not do the round. Robin is not about FDS
okay. Yeah, of course.
CD and, and the round Robin at least two zone.
Okay.
and it's important. When we invited people to participate, we did not hint it nor impose any model
Okay.
said the model that you are comfortable with, what happened the end is that the brave, I call it the brave cuz they're very brave. Only the brave, of them chose fds. one came with the guts doing Ansys. fluent,
That's
Sophie, None of these modelers have the guts to, to come with us, to the round robin only users of and this on the own model, came with us actually the two models that were use, a hundred percent of the models used in the round robin. were by NIST free
Nice. That's, that's powerful. And, what was the environment back then? I, I guess this was quite soon after the World Trade Center papers were published, which also were, This investigation is also something that I need to cover in this series because it was also groundbreaking, on its way. And I know it paved the way to Multiprocessor, fds and other, things that we take for granted today, like is obviously are here, but then it, it were problems.
but from, from your experience, how did it look back then compared to today?
you mean in, in terms of modeling or generally?
yeah, yeah. Of tools, of, of availability, of, of solutions, of the best practices.
that's a good question. think there was a little bit of, So, so there, there were these papers , to, to Convincingly, um, to reproduce the, fireball of the, of the double of the world trades and I think there was a bit of, of feeling that as we, discussed before, I think that kind of modeling could model fire. It was possible. We had the model that was fds. It worked and actually it did work for some, some applications. It worked very well.
you had, there was quite, quite a bit of validation, of the model. So it had been validated and it worked. So, um, I think there was, it was a very positive like feeling that we basically, there was a problem in a way. and I think that the big problem is, is actually a difference between, I always say that and Fire modeling and fire consequence modeling. And I think that's the key here. So you could back in the days and with FDS four, you could very well model the smoke movement if you
mm-hmm.
information about the fire, the hit release rate.
Okay.
and that's, I think that's the key cuz that is what they didn't have for the round robin. They didn't have the all the, all the information the fire we gave them, well they, the amount of testers, I wasn't there at the time and I missed them by week. they gave them all the information.
Yeah.
the fire development couldn't be done even with all the information, not blindly.
Yeah, but Wolfram you, just to highlight that this is a, we know this now,
yeah. what I'm saying.
this, this is, this is what easy with we publish the paper. This is what kept us busy with reviewers, the reviews were either saying, This is the best thing that I've read in 10 years, and the other one was saying, This is the worst thing I've read in. Our editor was Professor Dougal Drysdale, who is known to be tremendously fair and dedicate time to the process. So just to give you anm, we publish the paper, right? So we get comments and we up rebutting twice the length of the paper.
have the paper, which is a big paper, and we, the rebuttals, with all the reviewer comments, only half of the reviewers one, it rebuttal lend that to twice the paper in thickness. So actually we end up writing more toBut the reviewers than the paper itself.
That must have been a
um, you know, this, it's been, it's I wouldn't say a lifelong struggle, but a little like a career long struggle is to convince people the problem is not the model as, as Guillermo said at the beginning. it's about the modeler or the modeling process. CFAs for that matter. They are, , perfectly well, suited certain purposes, but you have to know which are the, you know, what they are designed for, then you can use them.
But, so when, when you criticize the, the modeling process, it shouldn't be understood as a critic. Or criticizing FDS or whatever of a simulation to what you have. It's about the modeling process. that's actually what, what came out as a result of all this, is that we have to be, you know, conscious about that process and where the limitations are. And I think people weren't conscious about it. They didn't know.
Okay. So to give, um, the, probably all of people in the audience may hear the word for the first time in the live, so, so let's give them a short introduction to what the fire did look like. Like whats actually the, the, the burning in the experiment? Of course, the experiment. It has, you have to give the credit. It was a much bigger thing. Ron, Robin and the CFD part is, Part of it. And, uh, there's a website of the project. There's many papers to be read.
There is the documentary tv, which you've mentioned GUI and I now need to find to link to the show notes. If I find it, it's gonna be brilliant. If I cannot find it legally, I'll steal it and then make a copy for anyone for the glory of fire science.
I have a copy in in CD
Okay. Yeah.
if it's the last resort I'll have a copy of.
Fantastic. I, I have no remorse. I, I live in Poland. No worries. I I can steal that. put it in the internet for the, for the benefits of fire science. I think it's actually important to do that. Anyway, it was a huge project, but, uh, let's talk about the, The part that touched myself, which is the Ron Robin and the a CFD part, because this, we talked about it for, for years, about the reasons of the project, and there's a reason why inviting you here. I, I had the CFD part in my mind.
I, I think this was something that if, if you ask a random fire scientist about Daher, they would probably refer to this, this part of the experiment because it, it aged the best of, of all of them for some reason. now let's go back to, I've asked you a question. What, what was burning? So what was the fire load in the Dalmarnock no.
the, the objective of Darlmarnock was to have a real. Not an experiment in a laboratory with a fake fuel or up fuel, no crib. This was an attempt to do, I mean, there were already a few experiments of NIST and other people, we wanted to do one, British style, so to speak. And it was a flat, a small flat with two bedrooms, one kitchen and one living room. remember that two experiments and just focus on one is the same. most of the action by far was in the living room.
The living room had a sofa, had a two bookshelves, had a table, had a a desk with a computer, with a chair, it had a little bit of decoration. was a blanket on the sofa. So it just tried to imitate a normal living room of a modern, flat. every single element in the computers were from ikea, and we actually even have the names of Ikea and the year of manufacturing. So it allows some reputability, so to speak.
Okay.
And if you see the photos, it looks modern, it looks very Ikea. The only thing is this absolutely pack of sensors. there are, there were sensors for the smoke that stand layer with lasers. There were, string gauges in a replica of a stress steel stress like in the Wall Street Center. There were a, the couple wires in the walls the center, uh, hit flex meters everywhere on the should face. were cameras, CCTV cameras. There were small detectors. it was packed. It was incredible.
It was very exciting. You can imagine the amount of data that was produced out of those two experiments of 15 minutes. we still have not processed all the data.
It will never be guess
that's beautiful.
some they would've done some experiments previously to the,
yes.
basically the sofa was burned, I think with the, um, under the hood basically to.
We bought three sofas of Ikea. One. One. Each of the flats receive a float. Four. Okay. Yeah, it's true. What happened to the fourth one? Wall frame is not official,
Yeah.
I don't know what happened. That one. Yeah. So the three I know, one of the sofa was burned in the lab and the laboratory conditions. walls, no environments under the hood. And the idea is to measure the heat release rate we measure the heat release rate and we provide the heat release rate the round robin participants. So it's not that they have zero knowledge of what was going to be the far development is that we said in the lap. is how it burns. Free burning, right?
No smoke layer built up, uh, in the room. No walls, no radi. And, and the question is, with this information, how do you think it would behave if it were to be in a room?
Yeah, but there was a difference though, because at the, during the tests, um, there was
Yes.
so there was a waste basket next to the, so that was the ignition source. and the fire should spread to the, to the sofa, which it did under the hood. It worked, but it took quite a time. during the experience, they make it a bit faster. I think they put a blanket over the, , like hanging into the basket and kind of over the sofa. So that was
And Wolfram spend a significant amount of time simulating these blankets.
out, because it, that was the end, it was quite important. And so we just, again, one of these things that you wouldn't think about beforehand, that was during the pre-flash over, one of the most
Hmm.
Bits of, of information, and you say, Well, you didn't, So round Robin us just said, Well, you didn't give us this information. Well, yeah, but that's part of the,
No one knew that. Like when you design a building, you don't ask where the rug will be lying, right?
No, but what this is important. The thing is we can tell the story from as he was happening afterwards, because remember the Wolfram has spent half of his PhD doing the
Yeah.
So Wolfram is telling you a hundred percent the truth. It took him three years to figure out.
Yeah,
years with trying to, trying to get this published, it was just
it's true because we met reviewers, opinionated reviewers, also with the posterior Well,
part of a year trying to, to figure out why was different from, why did, why it just didn't match. So it took me a year to match this and at the end it was as simple as I was just a blanket.
No, but wait, I, I want to frame this because I know how history can cast this. So it is, it was the blankets, okay? The blankets happened to be absolutely asension for predicting better, the fire. thing is, this was not known to absolutely anyone, so it's not that we did it on purpose. Oh, the secret blanket. No, we have absolutely no idea that we were doing this by putting a blanket.
the second one is if the blanket changes the results from, I have it here from 500% wrong to 20% wrong, then we are doomed in modeling.
that's.
I happen to believe that that's not the case. I think that there are details that matter and there are details that do not matter. And obviously in an immaterial science, like fire science, we have a lot of work. What we should not be doing is, we should not be saying the blanket doesn't matter. That's not true, and we should not say the blanket is the only thing that matters because that's not true either, right?
So there is a context, and in far science, we are dealing with such a complex phenomena that we cannot just focus on one element. Unfortunately, we have to focus on all of them at the same time and decide little by little, which one's count and which ones countless.
I'm looking at the picture of the room. Uh, right now in front of my eyes, there's actually a 33 pages long paper about the instrumentation alone. To give an impression about , uh, what scale of, of, of instrumentation we're talking about. I'm looking at the room. There's, uh, there's this, uh, couch with a blanket. There's.
The blanket
There's the three, three wardrobes with some papers on them. There's a computer desk to actually, there are some chairs, computers, like random stuff you would find in the, in office, a plant, an Ike lamp. I had an lamp like this. I confirmed that Ike . So in indeed, a very normal, uh, space, an office space that you, you would have in in a building. Uh, so Let, let's go round, drop in now. What exactly did you give to like, Okay, so far we, it hit release rate, but what else?
Now we give them photos every single one of the items independently and
Okay.
gave them the, the catalog description of ikea. Cuz they actually even give you the materials in ikea. They could actually go to their local IKEA if they want and bite the same version of the element if they want. No one did this, but they could, We gave them, we weighed some of the materials. Uh, for example, the sofa we waited for. Sure. Uh, we gave them the heat release rate of the sofa burning the lab. we actually, when we did the experiments,
Yeah.
we never thought about this, that there was press coverage. We, we didn't think about this problem, Robin, and we thought, oops, some modelers might be, good with social media to speak and might actually have seen the, the coverage, at that time we, we, to tell you the truth, we didn't know if it was going to ignite or not. Obviously, I can tell you that no had the many fire, so it was probably going to ignite, but we just didn't know if it was going to ignite.
I mean, it was going to, it was going, is the fire going to spread across the sofa? So anyway, so because it was press coverage, so we knew there was a fire, so we had to take our copies of the press coverage with some photos of the plume from the outside, and we give it to everyone and we say, Oh, by the way, and the, and the window, there was two windows. second window was broken on purpose. From the outside was Jose throwing a stone.
And we gave them the time at which Jose broke the window from the outside. We didn't say what happened to any other window because that was not an intervention, right? So the model is new that the fire happened and that the fire developed and there was some plume. That someone from the outside broke a second window but they knew absolutely everything that was inside the room. And then we asked them do farm modeling of this event and make predictions of the following variables.
And we said, Well, average temperature, local temperatures in the couple smoke the stand, flexes, obviously heat release rate. That was the first one. Time to flash over. I think that's it, right? Wellfront. Anything else I'm missing?
were the, the main.
Because what, these were the things that we could measure. We didn't ask them to predict anything that we could
Mm-hmm.
and we said, just send this file. And when you send the file, only when you send the file, we'll send you the data.
Okay. So now, uh, going into, this, this predictions you've already tasted a little bit with 500% scatter, but we'll unravel that in a second.
actually 800.
It's on . Just come on. Okay. Nice. Um, so first of all, the flashover, it grown. The flashover has happened in this compartment, and it was like a fully developed fire that, self extinguished. Was it quenched at some point?
No, it was, it was far good. That can
Okay.
I didn't
Yeah, we, we've been there, done that. Uh, Fire Bri case in large fire ex, they, they're always becoming nervous at the interesting part for some reason, Anyway. so, so let's talk Flashover, the predictions of the time of, of Flashover. I, I think that it's, for me, myself, having the knowledge I have today, in terms of doing full scale experiments, like full scale experiments are unpredictable, like absolutely unpredictable, uh, especially when you model, uh, 382 square meter abar. But,
Yeah, but Wojciech, Wojciech I have to
yeah.
we got this comment very often. Tell me in the literature, scientifically not opinion where that is set, but you just, I agree with what you said. Tell me where in the scientific literature that is a statement that can be quantified as saying, This person is saying that large scale experiments cannot be.
No, I, there there's none. And I, if you ask me the question five years ago, I would say we have perfect models to predict that. And now after doing like 20 of these, I have no idea. it's so hard to predict Now what, what I want to say is the moment from the moment of Flashover to some certain extent of time, it becomes more predictable. You know, because then you just have all, all the stuff burning.
And I think there's less variability because you have, it's essentially energy balance from that point onwards. And to some extent the intensity of radiation inside the compartment, which probably is affected by many things. But, but from the flashover, it, it becomes to me more manageable. but then again, If I model my shopping mall and I model a Flashover mall, I did not a great job as a fire engineer. So do not really commercially model these.
I don't care about a post flashover fire when designing my small control system, because that means I failed as an engineer.
Yeah.
So I care about this earlier phase and now this earlier phase, as you said, a dish rack can change the, fire behavior and actually build up of the conditions inside the compartment that did lead to flashover. That's an interesting, So this is why my first question is what was the scattering time to flashover? Because it, in a way, is an umbrella of all the things that happen as the fire grows, to reach a certain, certain size.
So, so this is why, like, uh, my immediate first, question, not, not temperature deviation, not, uh, smoke layer high,
a hundred. It went to a hundred to 1,200
Okay. That's a scatter.
and, and I want to emphasize this, but obviously you can go back to the paper and see the names of the team members. And the great majority of them are really, really well known. These were not a amateurs these, these were people that at the time and now were very famous and people will rely on them to have professional views on fire modeling. we didn't get the best modelers or the most famous modelers. because also I think these people, first, they didn't know us.
I mean, they said, Guillermo Who?, who is this guy? Right? This guy that just graduated. and also they had an invested into this because we were going to show the results no matter what.
hmm.
We, you, I think you mentioned this, we, we are not the first round robin in fire science project. We are the first ones to go to publication. There were at least two attempts before us. The results were never shared with anyone. Because when the room got together, they were so unhappy with the results that they themselves said they didn't sell, thought this will damage the field. We cannot go public with this message.
And when we did the round Robin organizations in the beginning, we said, We are not going to back up. we be accepted in journal, but this will go to our repository. This will go to conference. Maybe we are going to share the results of this no matter what. And I think that might have made some modelers uncomfortable that they would not be able to deviate if things go bad.
The best round robin we, we have right now is a yearly contest to, to find the hit release of the Christmas tree. And, uh, and it's fun and there's a lot of scatter. I wonder actually if there would be a value of gamification of, of this round robins and maybe just do them more often without, without hurting anyone because it, helps us understand our profession better. So, Wal firm, you said what was discussed from 80 seconds to 12 to.
some didn't. never flashed over. but I think,
One of them
um, I think the latest one, I'm just watching, seeing the graph here about 800 seconds. So it was between 80 and 800 seconds.
that's a massive scatter. and the flashover is sensitive to the heat fluxes to ground the, the small layer high to some extent, and let's say flame spread over solid surfaces. But I, I guess the whole SFA would be on in fire at that point. So, uh, how it looked for the predictions of the heat fluxes and, and layer
disaster. um, I mean, if one goes to the paper, we start with the heat release rate. The predictions of the heat release rate cover a very wide range of predictions. It's wild. typically, when I present this, I always say any dynamics to the planet Earth here predicted you will have to go to Mars to get a different curve. It's very broad. And then in the middle of it, towards one side is towards low. That is the experimental data with massive error barss.
wait, Wolf friend arrived one week late, so his predictions are not there. My predictions, the predictions of my team are there. We didn't do well. just, you just, you just cannot tell exactly who. You just know that the authors, all of them had models. We just, on purpose, we didn't link authors and
Yeah. Yeah. That's not point
a different game. So what I want to say is that when I say that the models didn't do well, just letting you know that we were there, I was involved personally in this and I didn't do well. Okay? So I feel, I feel that I'm fine and since then I continue using modeling as an engineer, as a scientist. So it's not that I did this to end a tool, of science, quite the opposite. I did this to bring balance to the tool, right? Bring balance to the force.
If the force was on balance, I thought, so I lost the,
Yeah, the, the layer. So, so let's talk layer, layer heights and
the heat release rate we didn't do well and everybody knows that if you're not doing very well with the hi rate, it's not going to get better. And, and it is true. It didn't get better. It just started to get worse and worse. The average temperature of a compartment, average of a smoke layer do well, but it did as bad as the hi rate when you look into the smoke, the scent layer. But yeah, I know you're very interested in the, descent layer is for how be inside. It was a disaster
Hmm.
pro probably. It is the worst all was an absolute random spaghetti plot of throw something at it. It's incredible. that was probably the, the most damaging result that we discover is. an important variable for life safety, the extent of the smoke glacier. And we cannot get it right or at least we for multiple reasons. It's not our
Uh, so sorry. the colors on the plots are consistent. Like the green group is the green group. So actually on the plot of hit release rate, there's the green group who did fairly well on predicting hit through release rate
yeah. Can I tell you a story of that one? This, this ends. So the green which I think I remember who they are, but I'm not even convinced and I lost the email, so I would not be able to tell you exactly. But I remember that the green team should submit this simulation a few days before we close the folder and we saying, you cannot submit anything. They send me an email and say, Guillermo peace please. We take it back. We want to remove the one that I send you and we want to add this new one.
And I thought, well, some groups like actually us, are not submitting one simulation. We are, we're submitting an assemble saying that anything between these two is possible. And they thought, Oh, that's really nice. Yes. And then please add these two. And they wanted to remove one that they nailed the he release rate with.
The, the green one. Okay. Okay.
my summary for this is not even when we get it right. We know
Yeah, that makes it even better. What I wanted to point is that they nailed the heat release rate perfectly, but they were among the worst to predict the layer height, you know?
the lady height is not the
but that, that's, that, that's so fascinating. Like, uh, it's not even about nailing this one parameter, you know, and, and you are sure your simulation's gonna be perfect. It's such a conundrum of different, physical things happening, flow rates, velocities, inside pressure, so many things that affect your flow. Even the, like, minuscule change in an external wind could turn the simulation around most likely, and you would not never know in inside of your simulation.
So that, that's fascinating. And if you, obviously, if you didn't nail the layer height, you're never gonna get the, temperatures in, in the vertical plots of your
no, it was the other way around is they produced the temperatures and then from
From that. The they, the picture. The, Okay. Okay.
some people might have done. The thing is we didn't say how they have to define this mo the layer height. We said, this is how we measure lasers, and then it was each of the teams to decide, well, I want to do this. We CFA this way with fds, this other way. So we, we, we were not constrained. We were not in given tiny little details. We were not micromanaging the simulations. We said, This is how we measure, do your best.
And they didn't do well, some of them, because of the definition of the layer height. But that's part of the problem. I cannot go back to people and say, Don't worry, I, your is super safe. I designed myself with fds and I'm super famous and, and, and don't worry because, um, I'm the best at knowing how to define the layer height. Like that's not the point. The point is, no, it, it's not how I define it. It has to be how the community thinks. It's the best way to do it.
a collective mind of the fire engineering. And, the oth the last variable that I wanted to talk, which also for me is, Not that, that huge variable, like the time to flashover would be important. And from flashover it, it goes on in, in a very certain way. Then after flashover your ventilation limited, the size of fire is what it is. The smoke production size is what it is.
So in essence, at that point, you should pretty well nailed the, the, the hot layer temperature because that's basically the hit balance equation of your compartment again. and there's a massive scatter on the hot layer temperature after flash over as well, because I see groups who, uh, uh, never went above to 420 degrees, and there's a group we had 1200.
So that's a massive scatter in like the final temperature, maximum peak temperature, which in essence will be highly related to the damage to the structure of the building. No, no longer humans, but, but the structure right.
No, no, absolutely right, And, and We didn't do well, predicting average temperatures, but when we went into the local ones, it's very important. Dalmarnock had the ability to look into several vertical,
Mm-hmm.
with the couples. We did, we did even worse. So the scatter is, is grows. of the conclusions of the round Rogan is we did robin average predictions, but we did even worse with the local predictions.
I hope.
flexes also fall into the realm of the local ones and in the world. Now, all this is really sad, but then Wolfram comes in and, and he was able to improve this tremendously. I mean, he literally turn around everything,
Okay. Save the day.
he did it because Wolfram did not submit his papers to publication until he minimized the errors between his predictions and the experiments because he had access to the experiments. He knew what.
So I had, I had all the results and I tried to find the heat release rate that would give me the match between the type.
Okay.
what we did, and after that actually, and that's the, fourth sofa, um, is that we, there was another one burned afterwards and I had, I had two sets of experiments for the heat release rate, to choose from for my poster modeling. And none of them worked Um, and they had quite, quite a big scatter as well. I mean, the difference between those two tests, in terms of heat release rate was over 200% as well.
So you mean Well fine. You mean the sofa burn outside the flat?
In the color measure, in,
So you, you, so are saying that, and just to recasting your question, you're saying we did not measure the HIEs rate in the experiment itself we didn't have the ability to measure. We could not measure hi release rate before Flashover. We could measure hi rate after Flashover by assuming that all the oxygen was consumed and we had meters the vents. But until Flashover, we didn't know the HI release rate and Guam is saying he needed it obviously.
For the initial phase. Yeah.
interface for the growth phase. So he went back the only two experiments, that existed. One I want to highlight. One was official, the other one was
other one
unofficial.
we, we did the, the second one for this posterior modeling.
perfect. Okay,
we, went,
it's not, Okay, perfect. So it's not part of the round robin.
we did this again. But what I'm, what I'm trying to say is that the heat release rate of those two experiments, only the sofa burning under the hood were tremendously different as well. and that's, that was when we concluded it was the, the blanket that made the difference because it, it kind of accelerated the growth. So, And it wasn't just the, load of the added, was the, increased growth rate of the, of the,
I, I have to stop you for one second. Wolfram, when you say that, the blanket influenced it, you meant that the, the growth, the speed to time to reach the flashover, but the blanket had no, I, the blanket had no impact. That post flashover temperatures, the blanket had no impact. That spoke layer heights. So, so is important to, to.
yeah. This was.
Let's not give an impression that the, the blanket broke. the modeling. a tiny variable. Yeah.
And that was a tiny blanket. I mean, it wasn't, it wasn't.
And the case, this happens to all of us, right? What takes us the most time is the one that we want to talk about most. The blanket was a big thing for Wolfram and actually was a breakthrough when, and I want to say this is important, he confirmed this scientifically. It's not that he had an opinion, which he obviously has. It is he scientifically can prove to anyone that wants to listen to him that it was the blanket that offset growth face.
And the growth face means that flashover happens later. It means that the time of burning is different. I mean, if you don't get the, the, the
The answer. The flasher. Yeah.
changes.
it's very sensitive to that. yeah, so we had these two sets of experiments as an input for our model. So now we have the, and none of them were, so we, as I said, we, we started to, figure out how to, you know, how to model this by comparing temperatures and came up with a blanket at, at some point we found a heat release rate curve.
They included this blanket, which we didn't have the data for, was, was a kind of an, we just knew that there was a blanket and we, recreated this and then we suddenly were able to, to match temperatures to a very good degree.
And by matching means you, you like this is the 20% point, the one where you almost got it right.
we
20%
um, average temperature
50.
time temperature curves that matched, Um, and then we had at certain times where the, you know, basically what, what we, we, we selected certain points in time where
Mm-hmm.
temperature in the compartment that we simulated was very close to the one that was measured
Mm-hmm.
times we the local temperature distribution and in h and, space basically. and it was consistent. So we had good results and obviously within, certain ranges, let's say. Um, but we had good consistent results with that. So again, I, I think this is, it's important to say that model that we use in that case and that this scale was FDS four, once we had the heat release rate and we had masked at that bit, could actually make good predictions and which says okay, the, the model is not the problem.
So we can kind of predict the of, of, of hot air and you know, of gases once we knew drives them.
So I guess that, that, that was a relief relief moment for a model is that, uh,
what
all be happy. Um, but they weren't, uh, it took us quite, quite a bit convincing of getting this published
now looking at the results of, of your posterior study, uh, which is interesting. I, I see that the heat rate is a, is fairly close match the temperature, I assume that's the average layer. Temperature is, is pretty close match, at least for the most of the, of the simulated time. individual profiles are still like very different and different in a particular way.
That in some cases, yeah, in some, some of them, especially the ones nearing openings, doors and the windows, where in simulation you would have, uh, actually observed two layer behavior and in measurements you didn't like. It seems like a layer and, and a strong flow on the, on the floor.
Right?
Yeah.
I mean, yeah.
But he reduced this cutter tremendously. I mean, if if you read the two papers together, the improvement is tremendous. So, I, it's something that I want to highlight because I know how history is written one of the reasons what we wanted to do round robin. main objective actually is, something that is modern now, you intelligence, everybody's talking about not in far science everywhere, and that they say they have a replication crisis, right?
That no one, most people cannot replicate the studies of the other. Well, I had a replication crisis of my own when I was studying CFD fire modeling. thought I could not replicate most of the results that I was seeing from colleagues in their, in papers, in papers. This is very important. No, I, I was in these beautiful papers with these beautiful predictions of the experiment and I was trying myself and I was crap.
So when I was, I was obviously telling to myself, Guillermo that's because you're a crap modeler. and then when I was talking to some people, some people said, I confirmed this. It's because you're a crap model of thought, I thought, Are you sure? Are you sure? Because how come all these beautiful papers are published and so few people can replicate the results?
So I do think at that time that we had a replication crisis as well, I wanted right, the round robin of farm modeling to, to show to everybody fire we do have a crisis. it's not that we as independently are crab modelers, it's that farm modeling is even harder fire science because you, you need to master fire science in order to master fire modeling. And fire science is in mature field that is just developing a few decades ago.
And farm modeling, an absolutely essential tool that everybody should be embracing, is just not the pan affair. not just a, a magic ball that we have. We touch a bottom and they give you results. The results look amazing. But that doesn't mean always that the results are
And, and Dalmarnock papers were probably the first ones who shown that on a plate, Look at this scatter and tell me it's otherwise like, it's so obvious looking at the spaghetti plots that, that, uh, you are just right. And I, I have this experience from the other side from trying to make my experiments as high fidelity as well controlled as I can and then replicate them with cfd. And I absolutely unable to do that. Like with, uh, with colleagues from Juelich Forszungzentrum.
They were trying to match visibility in smoke with a very highly controlled experiment of n-Heptan in a very specific 10 by 10 by four meter toll room. And I, I have good experience. It's, it's one of the most repeatable experiment. Like I can match the temperatures to like five degrees cel. Experimentally doing the test here and in Germany, this is how repetitive the test is. I've done it 10 of times and they are unable to model it.
They, they get the, the, the error invisibility, 500% modest error of the most important variable we use for the design.
there, there are people who are trying to model, simple heat transfer even, and they, they're getting errors of hundred percent because when you really, really want to solve a heat transfer on a vertical boundary and you stop using a number of whatever, 25, or sorry, the, the heat transfer coefficient based on whatever you, you have, but really that in mind, what is the value of heat transfer coefficient? Here you go. Scatter of 300%.
The more precise you go with modeling, the, the harder it is to really nail it. Right.
yeah, but I think that that's,
which makes.
the good bit cuz that that's actually, model and that has been improving. So I think there's
yeah, true.
towards
That's true. Yeah.
there's some intrinsic problem with modeling because you. I mean, there's some bits you can't predict, basically. And you have to, I think you have to, to separate that and clearly, so this is, you know, the, the fire models are very, very good these days. but there's still lots of room for
But the Okay, but the input is still, The input is still bullshit.
it better. But, but there's another bit doesn't even depend on that. It's just that we don't know, what the heat release rate will be, blindly beforehand. and, and we have to somehow come up with a way to model that and, and that will influence the results. And I think you have to be very, conscious about that, that you
And it's not even heat rate only. There's many parameters. Suit yields. Yeah.
Yeah. but this is, this is becoming set for me. So the comment about the hi release rate is a hundred percent true, right? If we don't know the hi release rate, we are in trouble. And this was one of the main arguments of some of the reviewers that we have to go through over months and months of time.
And at the end we said, Of course we, I mean actually that margin of ground robin shows this, that if you don't get the heat release rate right, then you are losing tremendous amount of ground to do predictions of anything else. But is absolutely no study in the whole literature where this is shown or set.
So this is one of the things where we uh, when we have a coffee, we are meeting in the lunch of the conference, we all are in agreement, but when it is time to go public to humanity, then absolute silence, right? It becomes like this deep, the. You don't have the hit rate, you cannot predict. It's like tell them they need to know. You know that it is a built environment where people live, where the buildings are done with people, right? So we tell to we talk is revere.
At some point we, I will publish the rebuttals. I said, Of course it's true, and you can now cite this paper show scientifically that is true. In the meantime, no one could cite anything about this.
Yep.
happened to me? Um,
Go.
the round robin paper, So I, we were hired as, as an expert in a, in a trial. So we did some fire model. It was about, showing a timeline in a fire that happened. and it was, there was smoke scene at some point in a camera, we had to basically come up with the, time that the fire started or, or, or reached a certain, um, so it was, it was basic. We had a very good idea of what happened. It was, you know, it was basically smoke So we, we were quite confident we could do that very well.
so in the trial, I presented my, my results and, and you know, very confidently so that the lawyer, no idea about fire, you know, and she used the round robin paper against me. She said, You said you can't.
How dare you use my own spells against me.
amazing. I, I, I wanted to congratulate her because, you know, she, she used my paper to, to, to, to invalidate my, my, my expert, which I mean from, it was very, very well played. obviously my answer was, you know, in this case I knew what happened. I knew the heat, his weight, and I just was modeling the, result so with stiff. you know, the fact that she, um, had that and, and had, you know, had found me saying that,
A small victory, right?
I think it.
So, we are nearing the, the end of the episode, so probably the, the most important question. What, what was the lessons for you? Like how did it change you as a scientist? You go first.
it had massive effect on me, in a way, very, extremely positive. it make me, made my profile visible because so many people had strong opinions about the of them hated it, half of them love it. It was one of these topics that you go into a room and the room is not the same Afterwards, I had to discuss one-on-one, these results with many known people, and that helped me know them and know their views, which is that otherwise people will have to spend years.
Uh, it gave me confidence of the things that I can predict that, that I can expect, and that the group that my, my group can expect from modeling, not just fire modeling, but modeling in, in general. For example, uncertainty became very important for us, for the, my whole career. Not, not only experimental uncertainty, but the modeling uncertainty. The fact that there are some parameters that we don't know that, and that it creates uncertainty, your predictions.
And since then, in most papers that I can remember, I hope, we have always had uncertainty in the modeling and uncertainty in the experiments especially. And we, we compare the two of them, right? Because we aim for the intersection of the two. we just don't have a model one simulation. also another thing that we embrace with Wal firm actually as well, is the value of ensemble.
It's not just as we have one simulation, it's that we have a group of simulations and we say in between things will be valid because we are just not sure of what, what would happen. Right? And this is something that the war en symbol, the first time that is used is in, in Wolframs thesis, I believe. before we had the concept, but we don't have the term, so then we call it symbol. Since then,
And your view on engineering through the, through the glass of, of Ook, through the filter of ook. How, how did you view the fire engineering afterwards in which the modeling is such a profound part of it?
I, I don't think it changes. what it, it created created a even discussion. You could see discussions where people were saying authorities and other experts and forensic trials like Wolfram, where suddenly the value of modeling was, discussed as opposed to be taken as, uh, no, top of the science half spoken as, Look, we have this paper here. What, what do you think of this? So it created more discussions. I don't think it changed modeling, continue developing tremendously.
Uh, maybe, maybe we help let it a little bit with a grain of salt to make it more mature, more. More professional? No, what I said before is we thought we wanted to bring balance to the force. CFD modeling is a force, it's one of the three forces of science. So maybe it was a little bit on balance. I think in, in the, in the beginning of the century, maybe Dalmarnock helped a tiny little bit to bring balance to the force.
And for your all firm, how did it change you? Is your PhD? Does it, So change you from Master to Doctor
yeah. much the same as, I think, you know, the round rowing per se was a bit of a, curse and a blessing at the same time. So, um, it brought a lot of problems, um, with people Yeah, because you were, was always perceived as an attack to the software,
Yeah.
Uhhuh,
And
which wasn't okay.
me, it's, it's still taking me, quite an effort to convince people. I'm not criticizing the tool, I'm criticizing the use of it. So that's, Yeah, that I think, And then I think that's, that's the main, takeaway for me is that you have to, to separate these two things. Um, these are tremendously valuable and they can be used if used, properly if used by people who know what they're doing.
and, but people who know their limitations, and I think that's very important, to, to know the limitations and, you know, know where to apply them and where to go a different path.
And, and now talking about this. Over decade after it happened, I think 15 years, more than 15 years since the experiments. It must be quite fun. It still discussed, it still leaves on it still. Like I saw you gear. I think you have the like presentation this year about Dalmarnock somewhere. So it's not a legacy part on the shelf. it's, it's still alive in the community.
now, lemme tell you this, it's alive. So last week I gave talk about Dalmarnock of Mechanical Engineers here in, in London. Now it's true, it's not that. They came to me and they said, super famous Dalmarnock speak about it. They said, Do you, do you have anything to say? And I said, And they say, You have what? say, But let me give the talk. And then, you know what, it create a fantastic discussion. So this is, this is 15 years old it create a fantastic discussion in the room.
Only like three people knew about that Madoc. The rest, the 20 others were like, Wow, amazing. Really? And you know, Wolf friend is absolutely right. We made a lot of enemies actually. Uh, unfortunately in the moment. I think some, most of them did peace with us at the end or we made peace with them. in that room, there was no controversy. In that room. when I presented this last month, everybody was Of course. Oh, finally someone says this. Oh, great. I'm gonna take this to my team. Right.
It was all very positive, very constructive. We didn't get in any of the aggressive respond that we got. Well, it is important, right? Because I was a very young academic at that time, Wolfram doing PhD. So you don't want, you don't want to be the center of a storm when you are in, in such a earliest stages of your Wolframs paper, poster was rejected from your or twice, right? Wolffer?
And I had, I had this previous, and we had this one paper we used to, the parameters of the, of,
yeah.
I presented that as at Ifss and the FDS developers were in the, in the audience. And I, it was my first, it was the first half year of my PhD. didn't know much about And yeah, I had to confront them and
So, so,
experience. But yeah.
so as, as we are fire engineers and our job is to predict, uh, uncertain future from very uncertain data point. I, I hereby predict that as we refresh this, piece of, of literature on the, on the shelf, I wonder who's gonna be the first to publish AI predictions on That could, that, that could be interesting. I have my hands if I had to bet. There's one group I would,
me.
That's, No, you can't. You have to do it ahead of time. So what I would like is artificial intelligence people predict and experiment before we do it.
Put
Blind
you can find me.
actual blind prediction. Artificial intelligence are trained on purpose against the result.
you find me five groups of AI who would like to do that, or 10 groups of CFD engineers who would be willing to submit their results. I have a place I'll find funds and I'll do it. I, I, I'll do the experiment.
don't review that
Okay.
Now you'll be
you are a professor now. You can send us your PhD students. That's, that's the whole point. And guys, for, for the very, very last thing, in some years ago, 2019 or 2020, I think that was, um, in the midst of pandemic, that was a SFPE Europe conference in which we've played the Triva game. And there was a question in the trivia. in 2006, University of Edinburgh has conducted a series of fire experiments that forever changed our understanding and trust in safety analysis of fire safety.
The form, the profound conclusions were related to the scatter of numerical predictions of multiple engineering teams. The question now, what is the name of the Glasgow district that gave name to this program? And there were answers, which are s Drum Chapel, Woodside, and Dalmarnock and, 60% of the responders got it correct. So I hope after this episode, uh, that would go to 90 something
and they could spell it and pronounce it properly, you think?
If, If, you want more funny things about the trivia, we also ask how much the SFP handbook weighs, which I saw behind Wolfram just seconds ago. And just, and that, that actually 7.57 kilogram and only two people got it correct. So
have to do this trivia again. That sounds like
that was a lot of fun.
Sorry, can I add one comment that you might, you can edit it in and out
please go on.
that, Okay. So I told you the difference between models and
Yeah.
really, really important. this has had been mentioned before, this was mentioned as an engineer, I don't like that as user effects, right? So you have the model and anything that is screw up is always the fault of the user, right? So if the mathematicians say, Oh, that's not my fault, that's the fault of the user, but an engineer, engineering cannot happen without the user.
Mm-hmm.
So I, I, that's why I prefer, instead of talking about user effects as the people who come to scrub the beauty of the model, I prefer to say, well, there is models which are mathematically mostly, and then there is modeling, which is engineering, which is not only science, it's also art, and actually protects people from fire. this is something important for me because we, I discussed that tremendously after the round row when be public. People were saying, just publish a paper on user effects.
I was like, Tell me that again. What do you mean? What do you mean user effects? So is that
Yeah. Beautiful. That, that's a great, uh, final thought in in this episode. Uh, I, I love this series. I'm gonna do more of that, okay, guys, thank you so much for, uh, spending an hour an hour with me. was, it's a huge pleasure. Thanks. Thanks again. And then see you around guys.
Thank you, bge. Thank you. Great. You're welcome.
Yeahs. Thank you.
And that's it for the first episode in the miniseries, The Experiments That Changed The Fire Science. I hope you've enjoyed that. Tell me if you like the format, tell me if you would like to hear more about this experiments and these research. Maybe you have ideas which to cover next. Actually, some of them are already in production. So, I guess if you will guess up what's the next step for this series?
I hope that it'll get some nice recommendations from you and I'm more than eager to continue the series because I absolutely love learning about. The experiments and, how the landscape of fire science has changed. Over the years. This is probably the coolest part of the podcast. Well, Second to spacecraft fire safety. That is for sure. So that that will be either I won't prolong this. Much because it was a lengthy episode anyway.
Just one announcement . As mentioned before, I'm trying to make an questions and answers episodes. Q and A's. And there we'll air. Monthly on at the end of the month. Uh, you can submit your questions either through emails or through the SpeakPipe add-on on the website. Please use them at ask questions about the podcast episodes about fire science, about meaning of life, whatever you like. And I'll try to answer them as well as I can.
In the monthly Q and a episode, I guess that there's going to be fun. I already had three or four really interesting questions. So the first one is looking, uh, Promising. Uh, please don't make me speak about meaningless stuff. ask questions. I can answer So that's it for today. See here next Wednesday bye