167 - CFD for consequences and fire growth with Jonathan Hodges

Transcript

WEBVTT

00:00:00.040 --> 00:00:01.546
Hello everybody, welcome to the Fire Science Show.

00:00:01.546 --> 00:00:19.271
The mission of this podcast is to bring fire science to everyone, but it's not just direct fire science that we're talking about in here, and in this episode we're gonna talk about how fire safety engineering is being practiced and how fire science is actually used as a tool supporting fire safety engineering.

00:00:19.271 --> 00:00:30.030
I may say In the previous episodes I had some of my colleagues who are in the same business as I am at ITB, me and my team are doing CFD analysis for a lot of projects.

00:00:30.030 --> 00:00:47.301
We're using computational tools to design smoke control, assess the mobility in buildings, and that's the majority of my everyday work, and I like to meet with colleagues who do similar things at other companies, similar things at other companies.

00:00:47.301 --> 00:00:51.493
This time I have Jonathan Hodges from Jensen Hughes company in the podcast and Jonathan is a research leader at Jensen Hughes.

00:00:51.493 --> 00:00:56.328
He's a recipient of the SFPE 535 award Congratulations, jonathan, great job.

00:00:56.328 --> 00:01:07.490
And also he's well known for being a very skilled CFD engineer and someone who's basically helping all the colleagues around the global office of Jensen Hughes in applying CFD in their everyday job.

00:01:07.490 --> 00:01:17.012
So in this discussion we're not just going to talk about how CFD is used by, but about some clever ideas on how it can be used better.

00:01:17.012 --> 00:01:19.703
What are the restrictions for using it?

00:01:19.703 --> 00:01:27.155
How can we use CFD to iteratively generate our design fires to perhaps yield better simulations?

00:01:27.155 --> 00:01:34.909
And, if you stay with us till the end, we have some predictions, or our opinions, of how the future of fire safety modeling will look like.

00:01:34.909 --> 00:01:40.212
So, hopefully, a very interesting, insightful and very practical episode of the Fire Science Show.

00:01:40.212 --> 00:01:42.046
I'm sure you'll enjoy it.

00:01:42.046 --> 00:01:44.319
So let's spin the intro and jump into the episode show.

00:01:44.319 --> 00:01:45.381
I'm sure you'll enjoy it.

00:01:45.381 --> 00:01:49.343
So let's spin the intro and jump into the episode.

00:01:49.382 --> 00:01:51.004
Welcome to the Firesize Show.

00:01:51.004 --> 00:01:54.465
My name is Wojciech Wigrzyński and I will be your host.

00:01:54.465 --> 00:02:13.957
This podcast is brought to you in collaboration with OFR Consultants.

00:02:16.979 --> 00:02:17.924
Ofr is the UK's leading fire risk consultancy.

00:02:17.924 --> 00:02:27.772
Its globally established team has developed a reputation for preeminent fire engineering expertise, with colleagues working across the world to help protect people, property and environment.

00:02:27.772 --> 00:02:43.568
Established in the UK in 2016 as a startup business of two highly experienced fire engineering consultants, the business has grown phenomenally in just seven years, with offices across the country in seven locations, from Edinburgh to Bath, and now employing more than a 100 professionals.

00:02:43.568 --> 00:02:55.212
Colleagues are on a mission to continually explore the challenges that fire creates for clients and society, applying the best research experience and diligence for effective, tailored fire safety solutions.

00:02:55.212 --> 00:03:08.867
In 2024, ofr will grow its team once more and is always keen to hear from industry professionals who would like to collaborate on fire safety futures this year, get in touch at OFRConsultantscom.

00:03:08.867 --> 00:03:11.206
Hello everybody, welcome to the Fire Science Show.

00:03:11.206 --> 00:03:16.973
I am here today with Jonathan Hodges, the Director of Modeling at the Research Division of Jensen Hughes.

00:03:16.973 --> 00:03:18.567
Hey, jonathan, good to have you in the podcast.

00:03:18.979 --> 00:03:19.804
Thanks for hosting me.

00:03:19.804 --> 00:03:20.447
Great to be here.

00:03:21.000 --> 00:03:25.409
Yeah, congratulations on your 535 award, well deserved, mate.

00:03:25.730 --> 00:03:26.612
Thank you, I appreciate that.

00:03:27.159 --> 00:03:30.951
So, jonathan, we've talked briefly at SFP Copenhagen.

00:03:30.951 --> 00:03:33.139
I've read your recent papers on design fires.

00:03:33.139 --> 00:03:47.567
It seems we are both doing very similar things in our companies and that is using modeling in our fire safety engineering, our fire safety engineering, and that would be the theme of the podcast.

00:03:47.567 --> 00:03:56.149
While doing interviews with different people and talking about cfd and having conversations while traveling, I've noticed that cfd modeling means different things in different parts of the world.

00:03:56.149 --> 00:04:03.662
Like poland is the car park country, we do cfd modeling for car parks all of them and I know in some parts of the world it's not that common.

00:04:03.662 --> 00:04:07.508
I know some parts to uh atria and corridors.

00:04:07.508 --> 00:04:15.025
I wonder what kind of CFD person are you and what kind of CFD modeling in fire safety engineering you're dealing with?

00:04:15.445 --> 00:04:17.350
So Jensen Hughes is a really big company.

00:04:17.350 --> 00:04:19.601
We do a lot of different modeling efforts.

00:04:19.601 --> 00:04:21.644
My team does a lot of performance-based design.

00:04:21.644 --> 00:04:23.586
Smoke control my team does a lot of performance-based design.

00:04:23.586 --> 00:04:27.750
Smoke control, emission of fireproofing alternative means in that space.

00:04:27.750 --> 00:04:34.596
We do a lot in the transportation sector for subway ventilation design, some in the car park, like you had mentioned.

00:04:35.218 --> 00:04:50.930
We also do a lot in battery energy storage systems and looking at explosion prevention as well as looking at separation distance from adjacent energy storage enclosures and make sure the whole system design when you've got multiples is not going to spread when the system's going off.

00:04:51.620 --> 00:04:52.803
Having offices around the world.

00:04:52.803 --> 00:04:56.105
Do you also see those differences between countries how people apply CFD?

00:04:56.105 --> 00:04:58.646
Does it mean different things in different J&K offices?

00:04:59.369 --> 00:04:59.730
It does.

00:04:59.730 --> 00:05:05.165
We don't do a lot of high-rise timber, for example, in the US, but in our European offices.

00:05:05.165 --> 00:05:06.891
That is more commonly used.

00:05:06.891 --> 00:05:11.271
So we'll have teams in our European offices who are doing those types of analyses.

00:05:11.271 --> 00:05:16.672
But we actually, even though we are a global company, we are pretty well connected and organized.

00:05:16.672 --> 00:05:26.704
I host office hours for all the FDS modeling people in our team in the company, modeling people in our team in the company.

00:05:26.704 --> 00:05:30.980
So we've got about an hour a week set aside where we talk about FDS modeling and how we are using the tools to make sure we're improving consistency.

00:05:31.720 --> 00:05:33.788
And FDS is the main tool used by the company.

00:05:34.600 --> 00:05:36.264
It's the main CFD tool that we use.

00:05:36.264 --> 00:05:51.312
We also use FLAX for some of the deflagration stuff in the energy storage systems, as well as Fluent or some of the general CFD when we're looking at air conditioning performance or cooling in electrical spaces, that kind of thing.

00:05:51.312 --> 00:05:55.649
But then we also use a lot of zone models, like we'll use CFAST as well as.

00:05:55.649 --> 00:06:01.730
Ses the subway environment simulator that we use a lot for the transportation sector.

00:06:02.060 --> 00:06:10.589
That's interesting because in many places of the world, like in Poland, engineering design in fire safety would be largely synonymous with CFD simulations.

00:06:10.589 --> 00:06:18.829
Like, I wouldn't say I've seen any study with zone modeling in the past five years like a zone model focused study.

00:06:18.829 --> 00:06:21.000
I haven't seen any content in Poland.

00:06:21.000 --> 00:06:22.565
I haven't seen SES.

00:06:22.565 --> 00:06:40.805
Yeah, we had some coming with engineering teams from outside of Poland to Polish tunneling projects, but still they've learned a hard lesson that the Polish firefighter the one that has authority in here they want colorful images and CFD was necessary for those projects as well.

00:06:40.805 --> 00:06:42.992
So quite, quite interesting, interesting.

00:06:42.992 --> 00:06:46.221
And is it a part of performance-based design regime?

00:06:46.221 --> 00:06:49.250
How does it work for you, at least in the us side?

00:06:49.411 --> 00:06:54.750
so typically the zone fire models are not used as much in specific applications.

00:06:54.750 --> 00:07:15.074
I mentioned at the start those we do when you're looking at naval systems uh, okay, at ships where you have compartmentalization, and so we do a lot of zone fire modeling in that space as as well as in nuclear power plants You've got the detailed fire modeling to look at equipment failure from far targets, looking for separation of systems for safe shutdown.

00:07:15.074 --> 00:07:38.478
But a lot of those analyses as a first cut are done either with spreadsheet tools or zone fire models, and then those are very conservative models and then that are very conservative and then when you're seeing event that's very bad, you may then dive deeper into the cfd modeling to understand those, because just there's too many to evaluate with cfd so let's try to go deeper into how cfd is performed.

00:07:38.557 --> 00:07:39.920
It's also performed differently.

00:07:39.920 --> 00:07:43.425
Do you have a very specific routine like?

00:07:43.425 --> 00:07:54.228
Do you follow any specific guidance or you build up your internal guidance on how to put CFD as a part of performance-based design, because I assume that's also important for you?

00:07:54.939 --> 00:07:56.065
It depends on the application.

00:07:56.065 --> 00:07:58.148
We do have our general principles.

00:07:58.148 --> 00:08:17.809
We have engineering guidance on how to use FDS as a company, as well as general guidance on Contam and a couple others, and so we do have principles there of things that you should always be doing how to define boundary conditions and when you're reviewing someone's work, what is included in that review, things that you definitely need to be looking at the checklist.

00:08:17.809 --> 00:08:24.093
So we have a lot of that kind of guidance for specific projects, especially in the design space.

00:08:24.093 --> 00:08:34.611
We're usually working within some code framework, and so we start with understanding what is the requirement based on the code and then use that to design what we're going to be evaluating with the CFD model.

00:08:35.200 --> 00:08:41.727
With the CFD models and I know that this is important for you as well, because we've discussed this previously.

00:08:41.727 --> 00:08:44.548
Basically, what you put in is what you get out.

00:08:44.548 --> 00:08:44.870
Right.

00:08:44.870 --> 00:08:55.830
And I liked I think it was an interview with Mike Spearpoint when he introduced me to the concept of consistent level of crudeness, that the weakest part in your simulation.

00:08:55.830 --> 00:09:04.672
Like there's no point of running extremely complicated modeling if your input is extremely base and uncertain.

00:09:04.672 --> 00:09:16.712
Right, if you completely make up your fire, what's the point of having a really complicated fire model to solve that if this consistent level of crudeness is not maintained in your simulations?

00:09:16.712 --> 00:09:23.307
So how do you choose the input for your simulations, what steps do you take and how does it look like?

00:09:23.621 --> 00:09:48.914
So I think that goes to the difference between fire effects modeling and fire growth modeling those as an industry, where the cfd models that we use for where you've prescribed your heat release rate, your smoky so it yields and your toxic product yields, and then looking at where it's going within your space, are pretty robust, especially when you're looking in the kind of far field.

00:09:48.914 --> 00:09:53.927
You can look at the validation basis in FDS and see that does a pretty good job in those spaces.

00:09:53.927 --> 00:10:15.086
Where we've seen a lot of people really trying to push the envelope lately is in fire growth modeling, where you're trying to, instead of using a prescriptive design fire or using specific test data, trying to use the model to predict the design fire, to either scale it up in some way or to swap out materials, that kind of thing.

00:10:15.086 --> 00:10:30.892
And there you're stacking uncertainties on top of each other because now you're trusting that your uncertainty in the modeling parameters as well as the material properties and the model physics aren't running away with those interactions, and you see this a lot.

00:10:30.991 --> 00:10:34.649
I think your example of the consistent level of crudeness is a good one.

00:10:34.649 --> 00:10:52.988
Especially in something like the batteries world You'll see people who are trying to do detailed chemical modeling of the battery and thermal modeling of the insides of the battery, and then we're using that to try and come up with how it's progressing in its thermal runaway and when you're going to get off gassing.

00:10:52.988 --> 00:11:06.692
So you've got a lot of complex physics that we try to embed in that and then using a very coarse simulation of where it's going to go within the space and not looking at the detailed reaction kinetics of what's going on in the gas phase.

00:11:06.692 --> 00:11:16.970
So you're doing a lot of good work on the chemistry side, but it's not the level of fidelity there and then what they're using in the gas phase are inconsistent.

00:11:17.500 --> 00:11:22.011
The idea that designed fire is an output came out of Jose Toretto.

00:11:22.011 --> 00:11:24.148
I think he formulated that at some point.

00:11:24.148 --> 00:11:34.013
That designed fire is actually an outcome of a fire, and for me it's an interesting concept because I kind of get the interactions you would get in a fire scenario.

00:11:34.013 --> 00:11:49.828
Okay, if you're doing a very simple fire and a very simple fire for me would be a ventilation-controlled compartment fire, you know, fully flashoovered, only the uh heat transfer at the walls and the air coming in, flames coming out.

00:11:49.828 --> 00:11:53.583
That's the only things that that you know exchange the heat and mass in your model.

00:11:53.583 --> 00:11:58.644
For me that is a simple fire because I there's not that many thing you know that would drive it.

00:11:59.065 --> 00:12:05.475
If you think about the models or the fires that we would deal with in our engineering, it's not fully grown flash-overed fires that we would deal with in our engineering.

00:12:05.475 --> 00:12:06.278
It's not fully grown flash-over fires.

00:12:06.278 --> 00:12:15.605
Usually we would rather, like I always say, the fire engineering is for the first phase of fire, for the growth phase, because that's where you can take actions, that's where you need engineering.

00:12:15.605 --> 00:12:20.030
Once you get into fully developed fire, that's where you need firefighters and fire resistance.

00:12:20.030 --> 00:12:24.110
Right, I wonder, like is the fire spread modeling?

00:12:24.110 --> 00:12:27.981
Is it even possible at this point for the engineering purpose.

00:12:27.981 --> 00:12:29.265
What was your opinion on that?

00:12:29.687 --> 00:12:34.059
I think it's possible to be using these tools to inform design.

00:12:34.059 --> 00:12:36.886
It's not necessarily, I think, we need to do.

00:12:36.886 --> 00:12:39.643
When you're looking at fire spread, you need to have a grid convergence.

00:12:39.643 --> 00:12:42.793
We're running at least at a few grid resolutions.

00:12:42.793 --> 00:12:52.971
We'll typically do three, starting with one that's kind of an engineering scale where we think it should be based on our guidance, and then one that's a factor two larger and one that's a factor two lower.

00:12:53.539 --> 00:12:59.389
So you can look at the sensitivities and then use that to look at flame spread rates.

00:12:59.389 --> 00:13:06.279
Heat fluxes use that to inform how we're building the design fluxes.

00:13:06.279 --> 00:13:07.583
Use that to inform how we're building the design.

00:13:07.624 --> 00:13:27.712
One example that we had was we were doing some design in like an amusement park type area and they had plastic or polymer of some sort water slide type thing and they wanted to understand how big of a fire could this thing produce, since we're looking at some interior spaces and nobody's burned a full water slide without the water on it, and so you have to come up with some design profile to be using for that.

00:13:28.460 --> 00:13:38.365
You could do a surface area calculation for the whole thing and then do some linear flame spread rate and then do a heat release rate per unit area based on that, and that might be an OK starting point.

00:13:38.365 --> 00:13:42.553
But you also, when you actually burn these things, you can start getting dripping.

00:13:42.553 --> 00:13:45.570
It's not going to stay clean like it would in a model.

00:13:45.570 --> 00:13:48.389
So we came up with conservative assumptions.

00:13:48.389 --> 00:13:53.865
Use fire growth modeling to evaluate what's a reasonable flame spread rate.

00:13:53.865 --> 00:14:06.846
When you start looking at these complex shapes, where you're not necessarily just in a lift configuration or a horizontal configuration, use that to get a realistic idea of the flame spread rate and then prescribe that in the model.

00:14:06.846 --> 00:14:17.086
So we describe what the fire scenario is, and then we've defined our fire scenario now and then prescribe that in the model, where then we're looking at the fire effects from that scenario that we've defined.

00:14:17.840 --> 00:14:24.548
So, in other words, it would not be a spread modeling like you have one model, you just start the fire, it spreads.

00:14:24.548 --> 00:14:33.621
It's more like using simulation to inform decisions and refine the scenario until you reach something that you are comfortable with.

00:14:33.621 --> 00:14:36.008
Yeah, kind of in between of the worlds.

00:14:36.008 --> 00:14:38.013
That's an interesting approach.

00:14:38.013 --> 00:14:40.904
Do you see that applied elsewhere?

00:14:40.904 --> 00:14:44.931
Because, okay, the scenario scenario with the slide is very peculiar.

00:14:44.931 --> 00:15:00.909
But in spaces like car parks or For me, car parks are interesting because I understand the importance of the height of the ceiling, that we've done a lot of parametric research that's shown us that this is the most important variable, at least if we're considering the life safety in car park.

00:15:00.909 --> 00:15:12.503
And yet in a car park you would have very specific, very strongly prescribed fire curves, you know, up to a point where they are favorites of people, like I have my favorite curve for a car park, right.

00:15:13.264 --> 00:15:13.465
Yet.

00:15:13.865 --> 00:15:17.013
I imagined in every car park this would look differently.

00:15:17.013 --> 00:15:21.071
Do you think such a refined approach here would be something of use?

00:15:21.779 --> 00:15:24.826
Cars in particular are very hard to model.

00:15:24.888 --> 00:15:34.624
I'm sure you're aware of that because you've got a lot of non-combustibles surrounding your combustibles and getting the thermal model right on that is really complicated.

00:15:35.186 --> 00:15:39.606
So I don't necessarily think we're at a place where you can really be looking at flame spread in that way.

00:15:40.408 --> 00:15:49.472
Now you could be looking at you have a sign fire that you're really competent in for your first vehicle and be using that to be looking at heat fluxes to adjacent vehicles.

00:15:50.220 --> 00:16:15.041
But the approach that I would be using in that would be more related to okay, what does the literature say on critical heat flux for ignition of my adjacent vehicle, and then be doing that as something as a hand calc or doing it as an iterative CFD design where we're predicting the heat flux and then saying, okay, now the next one's going to ignite at this time, and then doing it that way rather than trying to actually predict the heat release rate of that individual vehicle.

00:16:15.041 --> 00:16:24.090
But that does open a can of worms if you start to look into it, because if you have just one car burning and you measure the heat release rate, that's fine, you can look at that.

00:16:24.090 --> 00:16:35.201
But then if you have two cars burning now, does the second car burn at the same rate as that first car, or do you have an additional accelerant due to the additional heat being added by the adjacent vehicle?

00:16:35.201 --> 00:16:39.251
But what about when your third vehicle does it burn at the same rate as the first one?

00:16:39.659 --> 00:16:41.167
Yeah, what about in the seventh right?

00:16:41.779 --> 00:16:59.586
Yeah, and so that gets to one of the things that we'll talk about today that scaling up of experimental data, where we need to understand how do you take data that's something that you can get in a lab from testing and scale that up to a realistic scenario where your conditions might not be the same as what you have in the lab.

00:17:00.081 --> 00:17:03.311
And what about the other end of the universe?

00:17:03.311 --> 00:17:10.334
You said you would separate the fire modeling into effects and fire growth modeling.

00:17:10.334 --> 00:17:16.451
So if you model fire effects, there are some design fires which are like.

00:17:16.451 --> 00:17:21.891
They're not even experimental, they're like just magic numbers that came out of thin air.

00:17:21.891 --> 00:17:28.059
In Poland our magic number for a long time, and still is in many cases would be 2.5 convective.

00:17:28.059 --> 00:17:41.090
So that's like 3.25 total heat release rate and just alpha T squared that fast until it reaches that and you're good For a lot of designs and commercial spaces.

00:17:41.090 --> 00:17:45.048
You have a magic number that would be a fan favorite in the US.

00:17:45.361 --> 00:18:11.314
So the historic guidance was 5 megawatts or 5,200 megawatts Sorry, 5,200 kilowatts was the prevailing guidance for many years and I think that has been shrinking over time as people are using test data for something that's sprinkler controlled, and if you're looking at a balcony spill plume or something and you're crediting sprinklers, it's probably something around like 1.5 megawatts is pretty typical for that.

00:18:11.779 --> 00:18:35.778
Whereas for an axisymmetric plume typeks if you're looking like in a building or you're looking at Christmas trees or like upholstered furniture, that kind of stuff, and you can always come up with some combination of those things being close to each other.

00:18:35.778 --> 00:18:37.301
That'll get you to that number.

00:18:37.301 --> 00:18:42.084
But it doesn't necessarily mean that that's the conservatively bounding scenario in all cases.

00:18:43.971 --> 00:18:49.002
But I don't think there's strong experimental evidence for any of these particular numbers.

00:18:49.002 --> 00:18:53.894
I think they're just at the scale of what you would see Like if you go NIST calorimetry.

00:18:53.894 --> 00:18:59.055
You can have a loft seat, you can have a couch, there is a kiosk.

00:18:59.055 --> 00:19:01.557
I think there was an office configuration.

00:19:01.557 --> 00:19:06.710
They used the office configuration for World Trade Center as well, so they had some office configurations.

00:19:06.710 --> 00:19:09.332
Of course, christmas trees go Maryland.

00:19:09.332 --> 00:19:17.194
They have endless collection of burned down Christmas trees, which actually shows you how varied it can be right.

00:19:17.194 --> 00:19:23.232
It's so varied to a point we have a competition to guess the number before we burn the Christmas tree.

00:19:23.232 --> 00:19:25.758
That's how variable it is.

00:19:25.758 --> 00:19:27.200
It's not a single number.

00:19:27.200 --> 00:19:30.085
That is going to be two and a half megawatts, right, or five.

00:19:30.085 --> 00:19:34.461
There's no strong experimental evidence that it is five.

00:19:34.461 --> 00:19:42.618
It's just a choice someone made a long time ago and it propagated everywhere and now, out of convenience, we're using that.

00:19:43.010 --> 00:19:46.740
It's also in some ways an artifact of the labs that we have.

00:19:46.740 --> 00:20:04.843
There's very few labs that can measure heat release rates higher than five megawatts, and so you know, when we're burning stuff in our lab, we kind of design what we're going to test so that it is not going to be larger than our hood can handle, and so if your hood can only handle five megawatts, you're not going to be testing things that are higher than five megawatts.

00:20:04.843 --> 00:20:27.519
And so there's some artifacting there that you're selecting things that you can measure, and when you start to get larger than that with like rail, car fires or vehicles or heavy good vehicles, and you start putting things in tunnels or other things, now you get a lot more geometric effects and ventilation effects that are affecting your measurements and a lot more uncertainties associated with it.

00:20:28.131 --> 00:20:36.217
Some time ago I've put forward in Poland an idea that I see some value in that, even that it's artificial number.

00:20:36.217 --> 00:20:44.978
If you do that on multiple buildings, if you do that a hundred times at least, it becomes some sort of a benchmark.

00:20:44.978 --> 00:20:51.801
You know, a benchmark test, like in a way the ISO curve we use for standard fire resistance testing.

00:20:51.801 --> 00:21:09.813
Like we know it's bullshit, like we know it's representative of a very small portion of fires, right, but nevertheless it became a benchmark in assessing the fire resistance and you at least can refer the fire resistance to this standard test and compare worldwide In a similar way.

00:21:09.813 --> 00:21:33.490
For me, a very specific two and a half five megawatt design fire like perhaps it's not going to tell me everything about the performance in my building, for sure it's not going to show me the behavior of the building as a response to real fire, but at least I can tell okay, this atrium is so much better than the one I worked at a year ago Because with the same design fire I see completely different outcomes.

00:21:34.171 --> 00:21:55.855
I would agree with that because it does let you compare apples to apples across different designs and structures, and the difficulty you run into with the US or really any municipality you're going to have lots of different designers and consultants who are doing these analyses and because the prescriptive design opens up and allows the designer to choose a reasonable design fire.

00:21:55.855 --> 00:21:58.500
Some people are going to choose 2,500.

00:21:58.500 --> 00:22:00.392
Some people are going to choose 5,000.

00:22:00.392 --> 00:22:02.336
Some people are going to choose 1,000.

00:22:02.336 --> 00:22:23.378
So you get into a position where it's up to the engineers, almost their comfort level, with how conservative they want to be with the design fire, and it's just a difficulty because no one really wants to prescribe what design fire you need to be using in a situation they want to put that on to the engineer to decide.

00:22:23.800 --> 00:22:56.040
But then because of that you get a lot of variability in what the engineers decide to use of design fires, boundary conditions, things that you put into your model and then becomes just a standard test that you could compare against a bunch of other results that you have.

00:22:56.040 --> 00:23:08.638
You're an AI expert, so you know that better than most of us that if you would have such a database, it would allow for quick comparison, a robust assessment across a large database of outcomes.

00:23:08.638 --> 00:23:20.300
Where does this outcome place in the ladder of or whatever, and performance based design CFD, where you would go back into engineering your design fire, and that being a part of the PBD task itself.

00:23:20.300 --> 00:23:22.693
I think that this would be interesting times.

00:23:22.693 --> 00:23:24.778
I think we'll see that in the future.

00:23:24.959 --> 00:23:25.359
I agree.

00:23:25.359 --> 00:23:36.701
And you look at round robin studies where they'll have, you know, send out the same prompt to different firms and ask them to all design the same thing or measure the same heat release rate per unit area in a cone.

00:23:36.701 --> 00:23:44.411
You see a lot of these kinds of things where you get to see that variability and it always comes back as highly variable in whatever case you look at.

00:23:44.411 --> 00:23:51.134
You can even look at the SFBE PBD conference that we were at earlier this year that all the PBD design examples.

00:23:51.655 --> 00:24:05.823
We saw a lot of diversity in the different design solutions, so we don't want to cut out the creativity of the industry and being able to come up with unique solutions, but having standard benchmarks, I think, is a reasonable way to do it.

00:24:06.269 --> 00:24:07.480
Actually for the round robins.

00:24:07.480 --> 00:24:16.135
I remember an exercise so I'm in the European Commission that writes the standard CN and we're working on part five, which is for smoke control.

00:24:16.135 --> 00:24:21.461
And there was this exercise that some people did about simulating balcony spill plume.

00:24:21.461 --> 00:24:23.215
I remember that case study.

00:24:23.215 --> 00:24:33.333
That was a decade ago and we've agreed on very specific design, very specific boundary conditions, like we've agreed on everything you would put into the model and we've run it as.

00:24:33.333 --> 00:24:47.518
Like our team did it in Ansys, some other guys did it in FDS, someone else did use, I think, jasmine and actually there was not that much scatter, like pretty much everyone got very similar results but everything was prescribed.

00:24:47.518 --> 00:24:52.555
There was like no discrepancy, no choice From that exercise.

00:24:52.555 --> 00:24:56.435
It was not published, it was just an exercise for the committee to see what's going to happen.

00:24:56.435 --> 00:25:04.079
But you could see that you could potentially get into the place where this becomes a test, where this becomes reputable.

00:25:05.451 --> 00:25:11.712
One more idea I also remember that from the discussions at the CEN, the robustness scenario.

00:25:11.712 --> 00:25:17.953
I know in some places of the world if you allow PPD choice of the design, fire would still have a robustness check.

00:25:17.953 --> 00:25:24.496
Let's say, one megawatt fire and assuming a failure of smoke control or something I think Swedish have it.

00:25:24.496 --> 00:25:31.474
I think in New Zealand it exists in the CVM2 method Like a one megawatt firewood also always simulate.

00:25:31.474 --> 00:25:32.358
What's your take on that?

00:25:32.559 --> 00:25:33.605
I think it's a good approach.

00:25:33.605 --> 00:25:54.278
You always want to make sure that if one system fails, you have some redundancies in place, and so in the US market we see this a lot in the transportation sector, where you're using jet fans to prevent back layering so that people can egress out of a stop rail car and we always assume that the fire.

00:25:54.278 --> 00:26:02.029
You know, we'll look at different scenarios, but if you have the fire located at one your most important jet fan, then that jet fan, it's failed.

00:26:02.510 --> 00:26:06.819
And then you we make sure that the system still works without that jet fan.

00:26:06.819 --> 00:26:09.461
And so we do that in all of our systems not designs in the transportation sector to make sure that the system still works without that jet fan.

00:26:09.461 --> 00:26:16.076
And so we do that in all of our systems not designs in the transportation sector to make sure that, even if you're losing one of your key systems, that it's still going to be okay.

00:26:16.851 --> 00:26:17.955
Using like a one megawatt.

00:26:17.955 --> 00:26:20.557
I've also got a good example for that.

00:26:20.557 --> 00:26:23.118
We use that design fire in metro systems.

00:26:23.118 --> 00:26:37.219
More so when we were designing the corridors pretty much the concrete corridors which people use to reach metro Like there's nothing you can burn in a concrete corridor right Luggage trash bags, something like that, yeah, yeah, but how often you see burning luggage?

00:26:37.239 --> 00:26:50.762
come on, only in a Swedish fire experiments you see burning luggage At that point and that was a heavy criticism to us Like why do you insist on having a fire in a place where there's no combustibles, like there's no fire?

00:26:50.762 --> 00:27:02.295
Why do we have to put smoke extraction in a place where there is literally nothing to be burnt and there's stuff on the metro that will not allow to bring in combustibles and fast forward six, seven years?

00:27:02.295 --> 00:27:04.821
We're living in the world of electric bikes, right?

00:27:04.821 --> 00:27:23.796
Everyone carries a one megawatt fire source conveniently with them across the metro station nowadays and I'm super happy that we've insisted on that, because if we have not now, perhaps we would have a challenge to retrofit the station to adapt, adjust for the new challenge that has emerged, which we absolutely have not seen six or seven years ago.

00:27:23.796 --> 00:27:28.862
So I'm in huge favor for those robustness scenarios.

00:27:28.862 --> 00:27:36.318
Another thing let's go back to some blazing between fire effects and fire growth modeling.

00:27:36.318 --> 00:27:39.692
What's the representative fire for the real world?

00:27:39.692 --> 00:27:43.405
How would you interpret the representativeness of the fire?

00:27:43.630 --> 00:27:51.223
Let me ask you this clarifying question Are you asking for a prescribed heat release rate curve versus, like an alpha T squared type curve?

00:27:52.130 --> 00:28:16.178
No, I would say, like you have an office and you would go okay, in this office, I would go with 3.5 megawatt fire because I find it representative that I see coming up a lot in CFD analysis Because honestly, the best would be to go multi-parametric and test 100 fires and see 100 outcomes right, but no one can afford that yet.

00:28:16.178 --> 00:28:21.558
Perhaps with GPU and AI-assisted CFD, this conversation will be redundant.

00:28:21.558 --> 00:28:27.257
And if you're listening to the podcast in 2030, just speed up 10 minutes because it's not going to be interesting to you.

00:28:27.257 --> 00:28:29.198
But in 2024, it is.

00:28:29.198 --> 00:28:42.722
So when you would say your fire is representative to a space, no matter if you just use an experiment, alpha, t-square or just best assumption of a megawatt, just say five megawatts is representative.

00:28:43.250 --> 00:29:00.134
So for something like an office space, the way I would typically handle that is I would assume that you've got enough fuel to reach flashover, and if it's not something that's going to have liquid hydrocarbons, then you don't typically see anything faster than a fast growth unless you've got liquid accelerants present in some way.

00:29:00.134 --> 00:29:06.576
So I would probably assume that it's a fast growing fire up to flashover conditions within the space.

00:29:06.576 --> 00:29:08.520
But this actually gets to.

00:29:08.520 --> 00:29:14.671
One of the things that I've talked about a lot with people in our company is that's not the peak heat release rate that you can get.

00:29:14.671 --> 00:29:31.869
That's the peak heat release rate that the oxygen coming into the door can support, but you can pyrolyze a lot more fuel than that in a space and so, depending on your local flame extinction and ventilation conditions in the hallway, you could get a lot more flaming.

00:29:32.330 --> 00:29:50.356
Or, if you've got exit signs or other potential sparking locations, if you had localized flame extinction you could still have ignition of the sunburned hydrocarbons that you're pyrolyzing and having come out, hydrocarbons that you're pyrolyzing and having come out, and so that's a.

00:29:50.356 --> 00:29:56.236
You know, if you're just looking at design within the room, I think those flashover correlations and the fast growth to that is a reasonable estimate to be using.

00:29:56.236 --> 00:30:19.105
But if you're using that to inform, like facade design or looking at smoke transport throughout the space, I think you also want to try and understand, you know, based on your fuel load density in that space, general heat of gasification of the materials, what would be the maximum pyrolysis rate you would expect to see and use that to come up with a design fire that includes these things.

00:30:20.211 --> 00:30:35.021
What you described is some sort of maximum scenario, like if we're at the ventilation limited fire, that's probably the max you can get in a room and the growth rate, yeah, okay, the boundaries of that would be characterized by the physics of the flame spread.

00:30:35.021 --> 00:30:38.019
David Morris's episode in the podcast, highly recommend it.

00:30:38.019 --> 00:30:40.939
Go listen to that if you want to learn more about that.

00:30:40.939 --> 00:30:43.529
But how probable is that scenario Like?

00:30:43.529 --> 00:30:49.123
That would not represent the most probable fire, which would most likely be something localized right.

00:30:49.123 --> 00:30:53.382
So it's always the maximum fire, the approach right.

00:30:54.051 --> 00:30:56.400
I have these troubles when I'm doing designs.

00:30:56.400 --> 00:31:07.842
Okay, in a confined space, I agree, because the likelihood that if it's not interrupted by a sprinkler or a person, it's most likely going to grow to a decent size.

00:31:07.842 --> 00:31:20.462
But if you design, for example, a shop in a mall or a car park, again it's very unlikely that my fire would grow to 100 vehicles in my car park.

00:31:20.462 --> 00:31:24.260
I'm not saying it's not possible because we've seen those fires.

00:31:24.260 --> 00:31:26.778
The last one was a week ago in Korea, I believe.

00:31:26.778 --> 00:31:28.365
Yet most of those fires would end up on.

00:31:28.365 --> 00:31:31.593
One was a week ago in Korea, I believe, yet most of those fires would end up on one, two, three vehicles burned down.

00:31:31.593 --> 00:31:36.758
So the likelihood of it growing out of control is also like a certain number.

00:31:36.758 --> 00:31:42.835
To what extent my design fire should be representative of very low probability events?

00:31:42.835 --> 00:31:44.458
How do you decide on that?

00:31:44.898 --> 00:31:48.032
That's a good question and it's a fire risk assessment.

00:31:48.032 --> 00:31:50.079
You have to decide likelihood and consequence.

00:31:50.079 --> 00:31:58.403
Do you design to the 95th percentile, the 99th percentile or the 99.99 percentile?

00:31:58.403 --> 00:32:01.210
And so that's a risk management decision.

00:32:01.210 --> 00:32:07.763
And that is when we go to probabilistic analyses, which is the direction that we're going with.

00:32:07.763 --> 00:32:40.074
When we go to probabilistic analyses, which is the direction that we're going with, like you had mentioned, once you have AI-driven CFD is prescribed at what risk they're allowed to take on, based on the Nuclear Regulatory Commission, and so there's an actual risk number.

00:32:40.134 --> 00:32:41.378
That's being regulated too.

00:32:41.378 --> 00:32:43.807
So it's a quantitative metric.

00:32:43.807 --> 00:32:51.577
You do your quantitative analysis, see what your risk is, and if your risk ever goes higher than that because of emergent events, then you need to be mitigating it in some way.

00:32:51.577 --> 00:33:11.736
But that's also a very regulated industry with very high controls on the spaces, whereas in your office space, if someone wanted to put 50 e-bikes in because they're trying to sell them on an online marketplace and they need to store them somewhere, they just have them all sitting in their office while they're waiting to sell them.

00:33:11.736 --> 00:33:14.310
There's nothing saying they can't do that.

00:33:14.310 --> 00:33:18.320
There might be city ordinances, but there's not rigorous controls to prevent that.

00:33:18.320 --> 00:33:26.650
That's a very low likelihood event, but it is something that could happen, and so we have to decide what is the worst case that we need to be able to withstand.

00:33:26.650 --> 00:33:27.030
What is the worst?

00:33:27.050 --> 00:33:27.872
case that we need to be able to withstand.

00:33:27.872 --> 00:33:40.420
I sometimes heard Wade describe it as the worst credible scenario, like the credibility here Okay, you brought an example of nuclear industry.

00:33:40.420 --> 00:33:46.854
That's the credibility there would be defined by the risk value that's regulated by the authority, I would argue.

00:33:46.854 --> 00:33:59.097
In tunnel space it's also fairly easy because in tunnel there's a high acceptance for risk-based engineering and basically that's the only way you can PPD a tunnel only through risk assessments.

00:33:59.097 --> 00:34:03.978
But for a shopping mall or a car park, I wonder where those boundaries are.

00:34:03.978 --> 00:34:07.311
Do you sometimes have those discussions with authorities Like?

00:34:07.311 --> 00:34:13.951
I feel it would be some sort of agreement between the authority having a jurisdiction and an engineer doing the project.

00:34:14.753 --> 00:34:34.916
We do have those conversations and it does end up being a discussion where we say this is what we think is the worst credible event, that, based on the fuel sources, what's around, what we think it needs to be designed to, we'll come up with a design brief where we describe that and the technical basis for it and how we're going to evaluate the system and then submit that to the authority.

00:34:35.217 --> 00:34:54.346
Typically we'll have a meeting with them to also talk about it and talk through it, and the difficulty you run into with that type of approach is a lot of times the authority isn't as knowledgeable on the topic as the people who are doing the analysis and so they're generally going to accept what the fire engineer is saying is reasonable.

00:34:54.346 --> 00:35:03.050
You'll get some who are very knowledgeable who may kick back on something, but that's where education of the different stakeholders is important.

00:35:03.050 --> 00:35:07.762
There's an SFBE guide on PBD for code officials.

00:35:07.762 --> 00:35:19.023
I think it was ICC and SFBE put out and I think there's a lot of good guidance in there on what to be asking for from a designer for you to be reviewing as a code official.

00:35:19.023 --> 00:35:28.748
I think educating AHJs on those types of resources before they're being consulted with on a specific project they're being consulted with on a specific project.

00:35:28.748 --> 00:35:34.159
But just educating the community on these resources so that everyone is more knowledgeable about it, I think, is how we can overcome that obstacle.

00:35:34.869 --> 00:35:36.916
And those credible FHIR scenarios.

00:35:36.916 --> 00:35:37.599
Where do you get them?

00:35:37.599 --> 00:35:40.177
I mean, there are resources that I know.

00:35:40.177 --> 00:35:48.085
There's NIST database that I had Matt Bundy in the podcast where we've talked about how it was created.

00:35:48.085 --> 00:35:53.195
There are other databases UL has or FSRI has their own database.

00:35:53.195 --> 00:36:01.596
There's NFPA 204 Annex with my favorite stacks of palettes for specific feeds which no one understands in here.

00:36:01.596 --> 00:36:12.061
A lot of sources, but you mentioned something like scaling experiments, so I'm really curious when do you get your design fires and and how you use experiments to get them?

00:36:12.371 --> 00:36:16.601
So I think the example I'll start with is let's talk about Christmas trees.

00:36:16.601 --> 00:36:20.179
Okay, so there's lots of data you can find.

00:36:20.179 --> 00:36:23.317
You can look at the SFB handbook that has.

00:36:23.317 --> 00:36:25.161
I'm looking at the chapter now.

00:36:25.161 --> 00:36:28.179
You've got a three megawatt peak for one of the trees here.

00:36:28.179 --> 00:36:31.032
You've got a 1.5 megawatt for another one.

00:36:31.032 --> 00:36:33.213
You can find lots of data on these.

00:36:33.213 --> 00:36:42.958
But most of these trees are about three meters tall, plus or minus maybe half a meter or something.

00:36:42.958 --> 00:36:46.942
You're in a mall or a hotel and they've got a five meter tall tree.

00:36:46.942 --> 00:36:50.003
How do you now scale up that measurement?

00:36:50.003 --> 00:37:01.594
Because it's not a linear scaling in terms of your fuel loading, because you've got generally a cone shape and so the fuel load density does not scale linearly with height.

00:37:01.614 --> 00:37:05.043
It scales with the volume of the cone or the density of your foliage or whatnot.

00:37:05.043 --> 00:37:11.900
So if you're going to scale that up, I would say at the very least you should be scaling it by combustible mass At least.

00:37:11.900 --> 00:37:14.692
Then you're doing some linear scaling there rather than by height.

00:37:14.692 --> 00:37:22.333
A six foot versus an eight foot tree I don't expect it to burn 20% faster, I expect it to burn more than that.

00:37:22.333 --> 00:37:32.864
So I think there's a lot of things you can talk about in terms of design fires, but I think the tree is a good example, because people can visualize a cone and say, okay, yeah, but the height doesn't really make sense.

00:37:32.864 --> 00:37:43.503
And so what we like to do is look at the data sources you mentioned the SFB handbook, fsri's database, fire calorimetry database that NIST puts out.

00:37:43.503 --> 00:37:46.199
All of those are good sources to find that starting point.

00:37:46.199 --> 00:37:51.940
But then we need to understand is that test data actually representative of what we have?

00:37:52.420 --> 00:37:53.302
You can have a couch.

00:37:53.302 --> 00:38:00.719
That's fine, and usually when people are burning couches, they're burning couches with kind of the worst type of cushion, the worst type of upholstery.

00:38:00.719 --> 00:38:03.139
So it's probably a conservatively bounding couch.

00:38:03.139 --> 00:38:05.336
But what if you have a sectional?

00:38:05.336 --> 00:38:12.498
You've got multiple, you've got two or three couches together and typically you've also got like a coffee table or something.

00:38:12.498 --> 00:38:13.981
Think about a hotel lobby.

00:38:13.981 --> 00:38:17.978
You might have two couches and a loveseat and a coffee table there.

00:38:17.978 --> 00:38:24.362
Do you need to consider all of those burning or do you just consider the kind of the one couch burning as a design scenario?

00:38:24.362 --> 00:38:43.780
And my argument would be that we need to consider those realistic configurations and then also, for each element within that realistic configuration, understand is that data appropriate or do we need to scale it up in some way, either by combustible mass or exposed surface area or different types of materials, etc.

00:38:44.402 --> 00:38:47.130
What if you put that in a different setting?

00:38:47.130 --> 00:38:52.847
That it was tested in the way how you would burn a Christmas tree under a hood is exactly what I've just said.

00:38:52.847 --> 00:38:55.434
You have a hood and you burn the Christmas tree underneath it.

00:38:55.434 --> 00:39:07.378
There's no sitting in between them, there's no walls that prevent the smoke flying sideways, there's no large opening five meters from the tree from which you might have a three meter per second flow on the tree.

00:39:07.378 --> 00:39:21.833
Do you have any approach to scale the design fire to the circumstances in which it will be used, or you would still just go with the measurements from the lab because you don't want to make your own source of uncertainty in that?

00:39:22.195 --> 00:39:25.914
So I think we would typically use the free burning results and scale them up.

00:39:26.255 --> 00:39:42.773
But I do see a place for fire growth modeling and understanding those other effects If you calibrate a model to where it's matching the experiments and you've scaled it up and then looking at some of these other geometric effects to understand if they make it worse.

00:39:43.373 --> 00:40:06.384
There's a few studies on heavy good vehicles in tunnels, for example, where they look at the cross-sectional area of the tunnel and there's kind of a sweet spot in that where you get this balancing of fresh air coming in, accelerating combustion but also retaining enough heat that you're getting enough pyrolysis to really support those huge fires that you see.

00:40:06.384 --> 00:40:12.050
And then if you make it too big, then you're not retaining as much heat even though you're making more oxygen available.

00:40:12.050 --> 00:40:22.525
And if you get it too small you don't have enough oxygen available or your flow rate's so high that you end up cooling your gases a lot and you're not really retaining that heat.

00:40:22.525 --> 00:40:34.362
So developing a model that can predict and experiment and then using that in these different scenarios to understand design implications, I think is a very important step of where we're going as a community.

00:40:35.170 --> 00:40:52.063
So the tunnels is an interesting one because there were also research like if you put a tarp on the pallets on the truck, it changes the heat release rate of the fire, which is exactly what I've asked about, meaning that you changed the environment in which the fire takes place.

00:40:52.063 --> 00:40:55.480
It already affects it significantly.

00:40:55.480 --> 00:41:01.583
What about trying to do reduced-scale experiments and scale up from those?

00:41:01.583 --> 00:41:03.231
Do you have any experiences with that?

00:41:03.231 --> 00:41:10.487
They're the only ones that we're ever using are again related to tunnel scenarios.

00:41:10.487 --> 00:41:17.943
There are some data on trains in 1 to 3 scale from RISE, but I'm not really fond of scale modeling.

00:41:17.943 --> 00:41:22.994
I wonder what's your opinion on that source of knowledge for design fire information.

00:41:23.516 --> 00:41:26.726
So we've done a lot of work on scaling rail car fires.

00:41:26.726 --> 00:41:39.065
So we've tested some quarter scale and three eight scale rail cars in our lab and looked at how well the scaling laws hold across scales.

00:41:39.065 --> 00:41:42.152
I think it's a good approach in some cases.

00:41:42.152 --> 00:41:48.547
But the difficulty is in fire you can't, you can never preserve all your dimensionless groups.

00:41:48.547 --> 00:41:57.525
In a typical environment you're going to be just looking at preserving Froude number and so that's going to be your heat release rate to the two-fifth power.

00:41:57.525 --> 00:42:02.255
But that's only preserving the fire plume because that's what it's designed around.

00:42:02.255 --> 00:42:30.597
But when you look at Froude modeling you're making the assumption that time scales I think it's timed scales with length ratio to the one half power or something and that leads to all sorts of difficulties when you start looking at flame spread, where your heat release rate really scales by length to the second power, because if you've got the same thermal environment then you would expect your heat release rate per unit area to be constant.

00:42:30.597 --> 00:42:41.608
And so you get into these difficulties where now you need to be scaling your burning rate because your time is scaling, but physically that doesn't happen with fluid modeling.

00:42:41.608 --> 00:42:44.371
So you get into these difficult places.

00:42:46.514 --> 00:42:57.451
I think it's a good starting point and with what we've looked at is really in post-flashover fires where you have mostly kind of a one zone environment, that scaling with Q to the second power.

00:42:57.451 --> 00:43:09.141
So just looking at maintaining the exposed surface area of combustibles and saying that that should be burning at the same rate, that seems to do a pretty good job in these cases.

00:43:09.141 --> 00:43:16.440
But the growth to flashover is not preserved in that and you have to the way you handle your ventilation.

00:43:16.440 --> 00:43:29.701
You can't linearly scale your ventilation size because your opening factor derives what that one layer temperature is and so you need to preserve that across scales.

00:43:29.701 --> 00:43:38.019
So you have to relax your geometric similarity if you're using that type of approach, but in a post-flashover tends to work pretty well.

00:43:39.063 --> 00:43:52.186
And what about approaches where you would take some very small scale material scale data imagine a cone colorimetry and try to scale that up into a design fire in your CFD real-world full-scale scenario?

00:43:52.186 --> 00:43:55.826
Any experiences, any opinions on that as a part of engineering?

00:43:56.315 --> 00:43:58.884
So that's something that we do quite frequently as a company.

00:43:59.536 --> 00:44:29.126
We'll test materials in our lab and this goes into the fire growth discussion that we had earlier, where you're taking that cone data, putting it into the model, looking at your flame thread rates, looking at your heat release rate, and we're using that really to inform what the design scenario should be, more so than just as its own parameter, because you do start to get your uncertainties stacked on top of each other, like some of the work we're doing right now looking at heat fluxes to surfaces.

00:44:29.126 --> 00:44:32.454
Heat fluxes are not grid independent.

00:44:32.454 --> 00:44:36.764
When you've got a fire impinging on the surface, lots of reasons for that.

00:44:36.764 --> 00:44:48.126
But that means that as you refine your grid more and more, your heat flux is changing more and more and more, and that's going to affect the answer you get out in fire growth, and it's not necessarily.

00:44:48.126 --> 00:44:53.027
You know you have to get down to about a one millimeter resolution before it really starts to converge.

00:44:53.027 --> 00:45:00.409
But even then the number that you're converging to might not be representative of reality, because it depends.

00:45:00.409 --> 00:45:05.206
If you're using infinitely fast chemistry you're at a one millimeter grid resolution.

00:45:05.206 --> 00:45:12.056
That might not really be how it would burn in reality, and so that's where we're working on improving those things.

00:45:12.317 --> 00:45:19.222
But when you're dealing with test data and trying to plug it into a model to use that to come up with a design scenario you need to.

00:45:19.222 --> 00:45:24.184
Really, this is where the fire science and the understanding of how things burn really comes in.

00:45:24.184 --> 00:45:52.096
I would rather have somebody who understands fire testing and has read through these intro and fundamentals fire dynamics books and really understands fire and have them work on this on a CFD project with me, Then someone who has never worked in fire but has done a lot of CFD modeling, who doesn't understand these things because I can work with them on the CFD side as well.

00:45:52.096 --> 00:45:55.983
But it takes a lot of experience to learn the fire science part.

00:45:56.445 --> 00:45:57.467
Absolutely so.

00:45:57.467 --> 00:46:11.257
In your case, the cone would be a part of this refinement loop, like informing your further refined design fire choices, the spread model that allows you to approximate and scale up the experimental results.

00:46:11.257 --> 00:46:14.681
If seeking for a design fire, you have a number for a three meter tree.

00:46:14.681 --> 00:46:16.465
You would use those to inform your decision.

00:46:16.465 --> 00:46:18.557
How a five meter tree?

00:46:18.878 --> 00:46:21.905
Okay, so we do cone calorimeter experiments.

00:46:21.905 --> 00:46:28.007
We'll typically do three heat fluxes and we have the scaling pyrolysis model that we've put in FDS now.

00:46:28.007 --> 00:46:37.563
It allows you to take the heat flux, cone data say 50 kilowatt per meter squared and dynamically scale it up or down based on the incident heat flux.

00:46:37.563 --> 00:46:44.922
So we'll use that pretty regularly when we're looking at design fires right now and trying to understand how things could burn.

00:46:45.382 --> 00:46:45.925
Fantastic.

00:46:45.925 --> 00:46:49.606
Okay, this went fast the interview.

00:46:49.606 --> 00:46:59.065
Perhaps for the closing remark, one thing that excites you in this world what do you see for the near future or maybe far future?

00:46:59.065 --> 00:47:05.500
How do you think the profession, from this point that we've discussed CFD, design fires, how we use those tools.

00:47:05.500 --> 00:47:10.545
Do you see any shifts coming that engineers should get excited for or scared about?

00:47:10.974 --> 00:47:14.945
So one thing that I'm excited about is just how far we've come in recent years.

00:47:14.945 --> 00:47:16.315
You think about it.

00:47:16.315 --> 00:47:23.916
The first public release of FDS was about 20 years ago, and now you see how broadly used it is in our industry.

00:47:23.916 --> 00:47:36.204
But in the scale of construction, design and engineering, that's not a long time, and so we're still refining the tool, learning more about the physics of what we're doing.

00:47:36.204 --> 00:47:42.887
But we've come a long way and I'm excited to see where we're going to be coming in the next 10, 20 years.

00:47:42.887 --> 00:48:01.644
You'd mentioned AI and leveraging AI-driven CFD surrogate modeling or AI-driven subphysics models in CFD, and I see a lot of potential for those going forward to improve the fidelity of the tools while not having to sacrifice on the computational time.

00:48:02.496 --> 00:48:11.405
I'm super excited about the growth in the computational space, like the GPU solvers and everything that's related to that.

00:48:11.405 --> 00:48:18.469
We're really almost at a point where you will be able to do your CFDs like we're doing zone model today.

00:48:18.469 --> 00:48:34.983
Some years ago I thought it's not going to ever scale up to this level, but today I see perhaps it's not yet this generation of GPU-based solvers, but the next one for sure, where we're going to beat the real-time CFD.

00:48:34.983 --> 00:48:43.385
We're almost at this edge right now, so we're definitely going to beat that and it's going to open a lot of new opportunities.

00:48:43.385 --> 00:48:49.563
Like even in this discussion, we talked about why you need to choose the credible scenario because you cannot run a hundred right.

00:48:49.563 --> 00:48:49.945
What if you could?

00:48:49.945 --> 00:48:50.710
Like that cannot run a hundred right.

00:48:50.710 --> 00:48:51.275
What if you could?

00:48:51.275 --> 00:48:53.681
That would be fun world right.

00:48:54.023 --> 00:48:59.206
Yeah, and NIST is doing a lot of really interesting work on the GPU computing with FDS.

00:48:59.206 --> 00:49:04.246
Now there's a research fork that was presented at the FM workshop.

00:49:04.246 --> 00:49:12.632
On it we're presenting and showing that you can get some GPU acceleration now with FDS, that you can get some GPU acceleration now with FDS.

00:49:12.632 --> 00:49:19.197
Eric also presented on a case for the wildland fire simulation with FDS that used 1.3 billion cells, Like it's.

00:49:19.197 --> 00:49:26.119
We're at the place where you've got if you've got enough computing horsepower, you really can throw a lot of computing power at these tools now.

00:49:26.119 --> 00:49:31.179
But on the flip side, if you give a scientist more computing power, they'll find a way to use it.

00:49:31.179 --> 00:49:41.590
There's always going to be more you can do in your chemistry in the gas phase or more radiation angles that you can do more detailed chemistry or thermal.

00:49:41.590 --> 00:49:45.425

167 - CFD for consequences and fire growth with Jonathan Hodges

Listen On

Recent Episodes