139 - Wind and Fire Interactions for Safer Open Car Park Design

Hello everybody, after a short technical break, welcome back to the Fire Science Show this week. I'm happy to be on the air again and I have something to tell you, something that kept me away from doing the podcast in the last week, and that is our successful finish of my first research grant that I have led as a principal investigator. Kind of a big thing for me. First time I was giving the responsibility and I really hope I did not fail. The grant was on the win and fire and in this podcast episode I will tell you about the case study we've done for this research project. It is on the performance of smoke control, natural smoke control of an open car park. So a system which is kind of a passive, active system, a system that is used to remove smoke and heat from a car park in an event of a fire, but through its open facades. And obviously it's a dream case study to investigate it under wind conditions, because wind conditions will obviously break or make the system. So in this podcast episode I'm going to tell you what I think about the performance of natural smoke control in car parks. That's obvious. But we're also going to go through all the hassle it takes to actually make a good case study on this subject, starting with the design fire on which we had a separate podcast episode going through the wind design scenarios, performing multi parametric CFD and, in the end perhaps the interesting part for those who are not big fans of wind engineering actually, how do we take a bunch of CFDs I mean hundreds of CFDs and how do we summarize them all together to get like one sound conclusion out of the project, instead of 1000 pictures that tell you nothing? That was a challenge and that's something I'm really proud of because I think we've done quite a good job. We've introduced the concept we call the operational uptime. We've also tried to calculate some risks. We've tried to evaluate life safety, firefighters access a lot of interesting things, and in this podcast I promise I'm going to speak mostly about how we have done things, not why wind is so bad, so it hopefully is more attractive to the general audience. Last week as you may notice, there has been no podcast episode we had a webinar. It was transmitted worldwide directly from ITB headquarters and we've also captured the entire thing on video. Hopefully the webinar will go live later this week and I will link it in the show notes. So if you want to check the final outcomes. There's much more to learn from the videos, and today we talk about how to do it. So, yeah, that's it. That's the intro. Let's roll the music credits and then let's jump into wind and fire. Welcome to the fireside show. My name is Vojci Wigziński and I will be your host. This podcast is brought to you in collaboration with all of our consultants. All far is the UK's leading fire risk consultancy. 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. 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 100 professionals. 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. 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. So that's it. That's the end of my journey as a principal investigator. In my first real research grant funded through a contest, I won it from a national body called the National Science Centre in Poland, which I am very grateful that they have allowed me to participate and actually awarded me with this grant In a scheme that's in Poland called OPUS. That's the, let's say, the second stop scheme in the program, so it is for senior researchers. I mean, really I feel like a grown up researcher now I've done my first research grant. When I've submitted it that was year 2020. The world looked a little weird in that year because we were just struck by the pandemic. Everyone had a lot of time to sit at home and write, so I'm not sure how many great grants there were submitted. Mine was evaluated fairly well. We have proposed a multi parametric risk assessment on the wind and fire, something that kind of has never been done in the world of fire science. I mean, if you follow me a bit, you probably know that I've been dealing with fire on wind for quite a long time. There has been a podcast, episode number 50, with Guillemorein interviewing myself on the role wind has in fire engineering. That was really fun and that was actually around when we started the grant for real in 2020. We knew that wind and fire interaction is definitely profound. We knew that wind can completely change the behavior of safety systems inside your building. We knew how to model that kind of. In 2018, we've published two papers on that in fire technology the wind and fire couple modeling reviews, which I highly recommend to you if you're interested in the technicalities of modeling wind and fire together. We've done an interesting case study when we were writing a book chapter for brand Mitcham's market magnum is handbook of environmental effects of fires, where we've investigated the release of heat and smoke into urban surroundings, including wind and fire simulations. So, yeah, we've been there, we've done quite a lot of wind and fire engineering. But think about it a case study in which someone would take a well-defined building and just submitted that building a fire in that building into all possible winds. That's something that has never been done and that was something that I kind of dreamt of because I did not knew what will be the outcome. What will we find if we actually study everything not a sample, not the worst wind direction that we think is the worst, not the worst wind velocity that we think is the worst If we really truly study every possible outcome that can happen in a wind and fire scenario. Thinking about magnitude of this research. It's not a commercial project. You cannot do this within the framework of a commercial project unless you work for a billionaire I'm happy if you know one, I'm happy to do that. But it was clearly a research project, and the research project driven by curiosity. What will we find there? And yeah, yeah, we found something. We found that it's not necessarily the strongest wind velocity, that's the worst. We found that dependency on city infrastructure is kind of significant. It's not just submitting your building object to a set of velocity fields. That's not necessarily the entire wind thing, but we'll get that. We'll get there in this podcast episode. So first let me briefly introduce the case study. As I mentioned, it was an open car park. So what we've chosen to do after seeing all those huge car park fires around at that time it was Liverpool, eco, cork in Ireland, stavanger. We had a big fire in Warsaw shortly before the Grand was our. That that was actually crazy. And recently we had Luton fire. We had a fire in Netherlands just literally three weeks ago. So clearly open car parks are something that kind of burns in our modern world. We thought this is a great case study. As I mentioned in the intro, it's not equipped with the system, it's kind of passive. It's there, the openings are there. The openings in facade are always there and they're supposed to be working and extracting your smoke and heat from your car park. In a case of fire, it obviously works on the principle of natural buoyancy. So the smoke is produced by your fire. It's hot, it's warmer than surrounding air, so it has buoyant forces. It will spread through the car park, obviously because of this buoyant forces and forming ceiling jet, and then it will eventually reach the openings in your facade. We kind of expected that the openings of this facade, the smoke will exchange itself with the surrounding air, meaning the smoke will escape the car park in the upper parts of the openings and the fresh air will come in the bottom. And funnily enough, when we were trying to study the background of that like did anyone actually ever check that? We didn't find real proof of operational success of this way of smoke extraction. It seems that it's mostly system designed for sanitary mode and then simply accepted as some sort of fire system as well as a supportive system in fire, with no real expectations towards what it should do. I think in minds of many investors, firefighters, authorities, actually, they would expect the system actually does something. So, yeah, that was first very odd finding when we were studying the subject that there's no real proof that natural ventilation can even work in a car park. In our car park it was not huge, it was something like 60 by 30 meters, so not your biggest car park out there. We've specifically chosen a smaller one, a smaller footprint, so we can study more scenarios and it fit better. The location that we've chose and location is downtown Warsaw is a plot of land just next to my office which we know very well where we have done previous wind and fire engineering studies in. So we were quite happy with how this quite varied urban infrastructure gives us very rich set of wind boundary conditions, because the way we have tall buildings around, we have small buildings around, we have a bit of flat ground, we have an urban canyon, it's very nice and to variate infrastructure around. So we expected a lot of different outcomes. We have built a massive domain for that, really going all the way through our wind and fire coupled guidelines. We really applied everything that's written in that. We done the model exactly as it is written in the papers that we've published. By exactly a really minute, I mean the most annoying parts of it, even the 15 heights of the tallest building in the domain in all directions from my city footprint, which kind of means that domain was like two by two kilometers and 300 meters tall. A lot of people were asking us like why such a huge domain? Isn't it an exceptional cost? And the answer is if you do it well, if you prepare your case very well, the domain, the size of the domain, is actually not a big limiter to your model scale. You can go out with quite a coarse mesh in the remote parts of your domain, which means that these kilometers of terrain around the part that is of my interest is perhaps 10, maybe 20% of all the mesh elements I have in my model. It doesn't really add a huge cost but adds a huge versatility to my model. It makes sure that I capture the things that happen in the wind From my city. I capture all the wakes behind the buildings. I really make sure that my wind model is not dependent on how I build my city around it. So yeah, going with big domains is making a big difference we have. Also, while I am at the model, we have also faced a significant issue with the wind profile in there. So, as we wrote in the Wind and Fire Coupled Modeling Papers, you need to put the logarithmic wind profile to your domain, which means you want your profile to represent atmospheric phenomena that are happening inside the atmosphere. Wind at different heights is very different. You may have no wind at the ground and still have like 20 meters per second at 100 meters into your troposphere. So yeah, it is a phenomenon. I sometimes joke that describing wind as a simple velocity inlet condition with constant velocity flowing in is as stupid as defining fires through a temperature profile. And yeah, that's a bad joke because it gives you a bad impression of what fire engineers often do. But anyway, wind is a complex phenomenon. You need to model wind profile and now imagine we have these two kilometers of the numerical domain. We put a specific wind profile at the edge of this domain and then the wind is released onto my model. Now, before the wind reaches my model, it has to go above one kilometer of the terrain and as the wind is flowing, it's interacting with my terrain, meaning that there is some shear on the surface of my terrain which kind of influences the shape of the wind profile. So one kilometer later, when the wind meets the building in the middle of my domain, it's actually a different wind profile than the one that I defined and it can be profoundly different. It's actually quite a big issue because you cannot state that the wind profile is the same and for that that was a big task for my postdoc, paulina Jemiska, who spent a good few months writing new atmospheric boundary layer ABL model for us. She took some literature, recommendations, literature data and wrote scripts that modified this shear layer at the ground that make sure that the wind profile is conserved along the length of the domain until it reaches the point where we want to test the object. So we have went quite far to actually program a new wind boundary layer model that helps us preserve the profile and make sure that once the wind reaches the building that we're studying, in this case the car park, this is the exact wind that we have prescribed at the edge of the domain. This is something that's used in ANSYS Fluent. Ansys Fluent is the software of our choice. You've noticed that from the FireSensual podcast. I'm not sure how you could do it in FDS. I'm sure there are ways. Fds has terrain roughness models, fds has wind boundary conditions, so I'm sure you can work out a way to introduce a good wind profile into FDS. Of course FDS is LES model. So then you also entered the issue of transient definition of the wind profile or introduction of turbulence at the boundaries. That's perhaps a little more complicated than the Reynolds average models that we use in ANSYS, but nevertheless I am sure you can do it. So we had our numerical model of the car park, we had atmospheric boundary layer model. Now we've chosen two approaches to modeling this, In which we would put the car park literally in an open domain and we would modify the wind through terrain roughness coefficients. So we would define roughness as for city, as for city with tall buildings, etc. This would be the way that we would influence the wind profile and turbulence in it, simulating kind of the mixed flow that you would have in large city. And we also done an explicit model of a city, explicit model of city surroundings, in which the wind would naturally develop itself, would flow between the buildings and approach the building that we wanted to study and, as you can expect, you get totally different outcomes of that. So for a second, let me stop here and comment on the consequences of these approaches. If you go with implicit wind as a model ABL model, you put your wind profile. You can pretty much evaluate impact of specific wind velocity and wind direction on your system. When wind is a very uniform field, and this is great if you're studying the performance of smoke control systems. To some extent, this is great when you're studying a building, because it gives you a good impression of what the behavior can be like, in what range the performance of your systems will be found of the thing that you are studying. However, when you introduce the city into your model, you not only get this turbulent mixing, you don't only get this additional vortices that form, you not only get a more random velocity profile at your building, you also get those very, very specific effects of wind flowing through, you know, parts of your city, through streets of your city. We literally observed that in some wind angles the wind would fall into a street that was leading into an edge. It would bounce from a building and then hit the evacuation doors of the building that we were studying, creating a completely different outcome in our simulation than what we have expected. And yeah, that was a lesson. We've learned that even though it's cheaper and easier to study wind only with an empty domain and atmospheric boundary layer ABL model, actually studying the entire cityscape sometimes can lead to a completely different set of outcomes. And I'm not saying you always have to study that. I'm saying you have to be aware where the differences may be and perhaps if you notice in your domain that there is some very odd wind patterns in there, you perhaps would like to investigate it with the city domain. Luckily, today we have some great tools to study wind without fire, some great solvers based on Lagrangian particles that allow for quite cheap and fast assessment of the wind profile. So you can perhaps discover those unique flow fields on your own and decide with this knowledge if you need to go the entire cityscape modeling or just go with a simple domain. Okay, let's move on. Oh boy, it's already like 20 minutes of the recording and I'm just on the beginning of the list of things I wanted to share with you. This is crazy. As I told you, there's a two hour long seminar that goes very in deep in every single aspect of stuff that I mentioned in this podcast episode. So if you want to learn more from my teammates in my project. I'll link to the seminars once they're uploaded, which should be later this week, and you can learn more from there. Anyway, I promised design fires, but I guess I'll skip that because there was an entire podcast episode on how do you choose a design fire for your car park, and this is largely based on what we found in this project. So we went through a lot of studies that describe fires in vehicles done in laboratories. We investigated how those experiments were performed and we got some conclusions on how you could choose a specific design fire for your car park. In this project we've chosen six or seven different design fire scenarios ranging in the heat release rate from 1.4 to 8.8 megawatts, so a big scatter of fires, and we actually tested all of them. I've also told you that we've tested all winds. So what do I mean by testing all winds? I mean you probably can come up with an infinite number of wind scenarios for any kind of analysis. Like wind is a continuous problem, so you can break it down into infinite scenarios. In here we've performed a specific historical wind expertise for Warsaw, in which we have investigated the historical wind patterns. In here we have gathered 50 years of meteorological data from six different meteor stations around Warsaw and we've investigated, with a resolution of 10 degrees and 1 meter per second, how often we find the specific combinations of wind velocity and wind attack angle in here. Then we decided to decrease the number of scenarios by merging them together. So instead of having 10 degree resolution, we've ended up with 30 degree resolution, which conveniently narrowed the number of angles to be investigated to 12. And for velocities, instead of investigating every single possible velocity, we've also grouped them together into four groups of velocities one, three, five and seven meters per second to which we've assigned velocities of similar magnitudes that we found in the historical analysis, ending up in 48 very specific wind direction and velocity scenarios. And for each of them, based on the 50 years of data, we were capable of putting a probability number. We knew how many percent of recordings from this period refer to this particular wind velocity and directional angle combinations. So we knew how often which wind is present. And once we got all the fires as I mentioned seven different fires we got 48 different wind cases, so 12 directions and four velocities. What's else to do? We just merged them all together and brute force 336 simulations in our model to get to the final outcomes of the project. I mean, that was quite the task because, as I said, we're talking about quite large modeling, quite large domain. In fact it was 172,000 CPU hours. That's how long it took to calculate everything. To calculate it we've run, for every case, we've run a steady state simulation for simply the wind to obtain the initial flow field in the area. Then we would check the wind profile Every single time we run the simulation. We were verifying if the wind profile is as we defined, at the boundary conditions of the car park, then we would introduce the first fire, run the fire, then read the base flow field, again run with a larger fire and repeat seven times. And when we were done with all the fires, we would rotate the domain by 30 degrees and repeat the entire thing, checking out if the results and everything we collected so far makes sense. So yeah, it's a really silly and simple description of few steps that actually took us more than a half year to complete. But with all this knowledge, with all those 336 cases, we finally were able to jump into data processing and I think this is the most interesting part of this research project. So, first start for a second try to contemplate what are we trying to get here? We have results of 336 CFD simulations temperatures, velocities, visibilities, everything you would get normally from a CFD simulation of a fire and we want to understand what is the overall impact of winds on the outcomes of the fire. I'm not thinking about the individual impact per scenario, saying that at the angle of 45 degrees it is significantly worse than at angle of 90. Of course, to some extent I would like to form such detailed conclusions, but I really want to know what is the general impact of having wind acting on my car park that is on fire. How often it is bad? How often it is good? Is it a problem? Is it not a problem? Should I generally include this in my analysis? Should I not? I mean, if you think about it going individually, one by one cases, it only gets you that far. Really, at some point you lose the ability to investigate differences between them because there is simply too many scenarios. You have to investigate them at bulk, and that was the challenge we were facing. How are we going to do that? So the first initial idea was we're going to go statistics. I take my, let's say, plot of the visibility of smoke, visibility in smoke in the car park at the height of two meters, and I analyze in how many cases less than 30% of car park is filled with smoke. If there's one third of the car park filled with smoke, I think there's a proper behavior of the smoke control system. I mean, most of the car park is free of smoke. In this case, that's great. If there's between 30 and 60% of the car park filled with smoke, my initial thought will be, yeah, it's probably not the greatest smoke control system, because more than a half could be filled with smoke. But then again, a large part of the car park is free of smoke entirely, which means there is some performance left in there. So I thought, yeah, that's an average outcome, and if more than 60% of my car park is filled with smoke, well, I cannot say my smoke control system is having a sufficient performance at this point. And this is the way how we investigated it, case by case, and we were plotting the results, the outcomes of the visibility in smoke, of temperature where we've put the temperature threshold. Of velocity where we've put the velocity thresholds, and we were investigating in how big chunk of the car park at particular threshold is met or not. Based on that, we would give a simple score. You can call it an index In this case it's good, in this case it's average, in this case it's bad. And then, going one by one, simulation by simulation, we would assign a score to this particular system, telling us that, in, let's say, 35% of cases the performance was good. In this percentage of cases it was average. In this percentage of cases this was bad. This was already very helpful to indicate the overall behavior of the systems, because we knew we started to see patterns. We started to see that which particular wind angles we see worse outcomes in larger fires. We saw differences between the fires. We saw systems that were performing well at low heat release rates that lost their performance at high heat release rate For very large fires. We saw that at low wind velocities the systems were completely overwhelmed and the smoke was everywhere and the outcomes were not very favorable, whereas in high wind velocities we had really good performance in some cases and really horrible performance in some. We started to see patterns, exactly what we were looking for in this project. But we did not stop there. We don't one thing that perhaps is, in a way, innovative. I really care about firefighters and I believe that our systems should allow firefighters to do their job. Their job is not easy and we shouldn't make it worse with our systems. Quite contrary, in Poland we designed systems to help assist firefighters, so we started tracing to the seat of the fire. How many pathways exist to reach that position in every single CFD? So we would take an outcomes of a CFD simulation, the plot of visibility in smoke in the car park. We knew where the entrance points are, which were the entrances to the staircase and the ramp through which the cars entered the car park. So we had three entrance points to the car park and then we would basically draw a line from that entrance points to the seat of the fire. If you could draw a line and there was no smoke across the line, basically you could find a pathway where there would be no smoke and one could reach from the entrance to the seat of the fire distance, let's say, 10 meters from the seat of the fire, and not come across smoke. It means there's a free pathway for the firefighters. Now we followed with a simple thought process. If all entrances are available for firefighters, that's kind of great. Whichever they pick, they can enter and they're good. If two of them are available, that's for me still okay ish, because they have a lot of options. Now, if there is just a single pathway available for them out of three that we initially had, it means that if they pick the wrong one they have to go back and change the strategy, change the direction. Perhaps their access to the third one is not so easy, not so fast. It means it will add time to their operations. So in the end they're still capable of reaching the seat of the fire, but you could expect this will be much delayed. Versus a case where they had entrance from every single point they would access the car park from, and in a case where they had no entrance, well, that's really bad outcome because they cannot do actions from the inside of the car park. They can only resort to extinguishing from the outside, and the large car park fires from the recent years show us that this is insufficient to actually contain the hazard and it usually leads to significant damage to the structure and the vehicle is inside. So that's an unacceptable outcome. So here again we can have a simple measure how many pathways exist for the firefighters for which we can give a good performance? Two or three entrances average performance. One entrance and bad performance, no entrances to the car park. And that's what we've done. So we have this indexing approach, this qualitative measure approach for different physical phenomena, and then we thought, ok, how about we just merge them all together into one index, and we did so. We've made the mattresses for each individual physical quantity or the firefighters entrance that we would be measuring, and then we just summarized them into one final metric. If a car park scenario was good in all of the outcomes that we've measured, it's clearly a good performance. If it was average in one of them so let's say, visibility was average and everything else was good we found that, yeah, that's an average performance. If two or more were average, or at least one was unacceptable, that meant that the performance was insufficient and we considered the car park not working as intended. So we've agglomerated all those physical quantities that we were evaluating in our car park into one final measure of how good the performance was. Now, as I told you, for the 48-win scenarios that we have envisioned for the case study, we could tell the probability of every single of those wind scenarios. We could assign a probability value based on the historical data from Warsaw. So we knew that an angle of zero degrees and zero is our west, and wind velocity below one meter per second. That's 4.11% of all the winds that we have in Warsaw. We had such a number for every single scenario. Now imagine we have the number of probability of each wind and fire scenario and we know the outcomes for every single wind and fire scenario that we analyzed. From this point we came with a new measure, with a new way of Assessing the performance of natural smoke control in the car park. We've called it the operational uptime. What is operational uptime? So if you think about the performance of natural smoke control in a car park, let's say it's Thursday, it's a nice sunny day, there is absolutely no wind in the city and there's a fire in this car park. What does it mean? In this particular scenario? The wind is not acting on a facade. Depending on the size of the fire, the performance is typically the outcome of the buoyancy that happens in the car park. If it's small fire, it's okay. If it's large fire, it's probably pretty bad. Now the same fire happens on Friday when we have strong western wind. It means that in this wind the car park is fairly well ventilated. It's blown all the way through by the wind, which means if the fire happens in these conditions, the smoke will be very effectively removed from the car park. If it's Saturday and the wind is kind of northern, it means that's the worst wind scenario for my car park. The wind is hitting a side that has no openings. It's creating wind conditions that entrapped my smoke inside the car park and I have the worst outcomes of this fire. Now it's Sunday. We have again western wind and again favorable conditions. We know that for, let's say, 70 days in the year the wind will be favorable. For this amount of days the wind will be average. For this amount of days the wind will be bad. How do we know it? Because we did the analysis of the historical data. Now, if good, bad and average comes from the indexes that we have established, the Visibility in the temperature, the velocity field, the firefighters axis and the probability comes from the historical data, we can assign a number of how much time within a year, for how many percent of days or hours of the year, the conditions, if the fire happened in those conditions would let to correct operation of the system. For how many hours, days, percentage of the year, the system would simply not work and that's what we did. That's the operational uptime. The part of the year, the part of the overall time in which the system Is ready to act, is ready to operate and if fire happens within those conditions, it will operate correctly. And this allowed us to basically take 336 simulations and cram them up into a single number, a single number that defines the performance of the system. The reasons are a little bit shocking. To be honest. So if you take the average and good performance and a low heat-release rate of 1.4 megawatt, which is a bottom 5 percentile of all the experiments done on car parks and usually 1 for 1.4 megawatt Would be something I would associate with an early growth of the fire For those fires, 86 percent of the cases were okay. So average are good and 11 percent of cases were unacceptable. Two and six percent are is we. We are not able to make an assessment because the wind was variable. So For 11 percent of cases my system would not be operating correctly. If we up the heat-release rate to four megawatts, which would be a common design scenario for car parks and something around the median size of the fire of an average passenger vehicle, the system works correctly, operates correctly for 38.8 percent of cases and its operation could be considered incorrect for 58.5 percent. Now that's quite a shocking. More than a half of the time the system would not be working. Now if we up the fire a little more, up to 8.8 megawatt, then it becomes quite challenging in such a large fires and 8.8 megawatt would represent upper 90 50 percentile of fires in our database For this large heat-release rate. Only 14.2 percent of times the system performance would be acceptable and In 83 percent of times the operational uptime is not. That is not. It means that if there was a fire, there's chance one to six that it's gonna work and five to six that it will not work. Quite staggering numbers. But it's not just about the number, of course. If you, if I rotated my car park Against the wind so the probabilities change a little bit because maybe I have a more favorable wind in my car park I can up those numbers a little bit and get maybe 30 percent, 25, 30 percent of correct operation instead of 15. Yes, you can do that, but still it's, it's quite shocking. But really I really don't want to focus on the number. I read this project was not to say that open car parks are really bad places in their challenging and we see that From the tragic fires that we see and acting on how we can improve. That is perhaps another step after this Project. I wanted to show you that you can make a very complicated, very large, vast Engineering analysis and get a single digit outcome, get a single number that Represents what you are seeking for, easy to discuss, easy to present, easy to act on, and I kind of love it. Another approach, a more classical, would be a risk approach. We tried risk engineering for this project. So we established the probability of a fire occurring in our car park based on Warsaw fire brigade statistics and the amount of vehicles that are in park and ride car parks in Warsaw which was available to us. So, based on those statistics, we came up with a number that represents a probability of a fire, daily fire probability for a vehicle parked in waso. Then, knowing the outcomes of the experiments, we have established some probabilities of the size of the fire of vehicles. Again, it comes a little bit back down to the design fire discussion that I had a few episodes ago. I highly recommend you listen to that one. Then you know the outcomes for every single fire we've analyzed through the recent analysis we have just performed, so you can come up with scenarios scenarios in which there could be fatalities in your car park, scenarios in which there will be no way for fire brigade to act on the car park fire and stop the growth, so potentially you're losing your car park. You could establish how many vehicles could be damaged in the fire, to how many vehicles the fire could spread, establishing the monetary cost of the fire for the goods within the car park and, based on those numbers, multiply that by the probability of your occurrence of the fire and probability of the wind and you can build a matrix that revolves around the risk. That's also an ultimate goal of this project. It was called the multi parametric risk analysis for a reason. So here we are. We got some final conclusions of the project. I'm really proud of how we managed to actually get so many inputs, so many data points, so many assumptions all together into one coherent set of simulations and then finally get them all together into getting one outcome of the project, one conclusion for this particular data set that we have investigated. When we were discussing the project, the next steps came up and I think there's one great thing we can do next. Actually, there are two great things we could do next and one thing that we could do pretty differently if we started today. So let's start with what we could do differently Today. We're at the brink of GPU revolution in computer simulations. I told you more than a half a year to finish the C of D simulations for this project, 172,000 CPU hours for the project. That was based on CPUs. I think with GPUs that we have at hand today, we could cram the project in two weeks, maybe a month, clearly less than a month. So we could perhaps do 10 folds of the analysis, or all of the analysis that we have performed, in one tenth of the time. This means that something that has been a research project, something that had been available only because of the funding I have gathered from my external funding source it was quite large grant it perhaps joins the regime of something I could do as a commercial project. This opens new pathways, new possibilities. So, yeah, if this started today, we could do so much more with the same resources that we got and for the things we can do in the future. So, if we could, because here we were doing the design fire approach, here we were doing just the assumed fire size. If for some way we could account for the growth of the fire, we could simulate how the fire spreads in the wind. We could simulate how fire in vehicle grows within the wind we have reactive design fire, not prescribed design fire we could make a much, much better analysis. For me it would be a simple switch from a prescribed fire to a reactive fire or growing fire. In reality that's a, I guess, entire research project to get there. But I think we would get then the ultimate answer on how fires grow and how different properties of the car parks ventilation, wind whatsoever define the fire performance of a vehicle car park. And you could also add a structural analysis to that to get an ultimate wind, fire and structure case study. If we could call the structure Earth, we would get Earth, wind and Fire. That's probably worth it. We'll see. For me that's a research direction that we will probably be pushing for. And the second future direction, perhaps even more exciting AI. I mean, come on, this is a great space for AI. We already see that you can most likely tell the outcomes of the fire analysis looking just at the wind field, like seriously, I don't have a proof. But from the amount of data that I've seen, I can really well tell, looking at the wind field, where the smoke would be and how bad the fire would be in this particular car park, just at the flow field. If I can do that, if I can see that for some reason, based on my experience, it means it must be trainable for an AI. And if we could train an algorithm to go through all those results and tell me for a new flow of field of wind what would be the outcomes of fire, perhaps in a different location, perhaps of a different size, we could generate not hundreds but thousands, tens of thousands of scenarios for a much better informed analysis of the outcomes, of the risk, of the indexes. It would improve significantly our understanding of the interaction between wind and fire by increasing the amount of scenarios that we have at hand. I think that's a very, very tempting evolutionary scenario for us and I'm sure we will be investing for that. And remember when I told you the differences between the city and an open field. With ABL atmospheric boundary layer model, you get the different outcomes when you model a car park in an empty building and just apply the field of wind through atmospheric boundary layer model, abl model, and a different outcome when you do the same but you introduce the city in the model explicitly. It's very costly to introduce the city, but you can do it here. You could probably simulate your car park in the ABL model, in the cheaper version of the model, get your baseline outcomes and then simulate the wind in the city with some simpler tool, not necessarily complex Eulerian CFD approach, perhaps Lagrangian approach. That is really working well fine in city aerodynamics. It's cheap and quick and reliable. You could get outcomes of the windfield from those models and feed them into a machine learning algorithm that was trained on your Eulerian CFD fire wind analysis and expand your simulations. Expand your results to new wind scenarios that are highly dependent on your city architecture, for a fraction of cost of the complex Eulerian couple wind and fire CFD analysis. It's really a mind blowing concept for me. Mind blowing gets swell that we have infrastructure and software to do that in future and that perhaps is my next research direction for me and my team. So, guys, this is a brief summary of the multi parametric wind and fire risk with numerical modeling project that we have just completed, funded by National Research Center in Poland. I am highly grateful for NCN, the Polish National Science Center, for funding me this grant. It's my first real research grant done as a principal investigator. I am a real grown up scientist now and I'm very happy about it, very proud about it For myself. I'm sure that the project has accomplished it goals and we got almost everything that we have planned. It was well executed despite the pandemic, despite the remote work, despite the challenges with the wind tunnel, despite the challenges with the CFD that we came across. A lot of work, but gladly a success in the end. This is obviously not everything that we found. This is not the. This is just the tip of an iceberg. As I mentioned, we had two hour seminar. Links to that will be up in the show notes as soon as the webinars go up online for replay, so you are very welcome to watch it. There's a long Q&A session in the end of the English seminar, so I hope that one is great and that one is useful for everyone and if you have any questions on wind and fire modeling, as always, this was the first topic that opened the world for me, the first topic where I could share my findings with the world, the first topic where scientists from different parts of the world would ask me questions and treat me as some sort of authority. I loved that I can help people with their wind and fire projects, and this has not changed. 10 years after, now that we are much more in the subject, I still love to help you. If you have any questions on wind and fire modeling, please let me know. Shout out to me and I will try to guide and help you. So thank you for listening to this today's episode. I'm very sorry for the mishap last week. We've missed the podcast episode. What happened was these seminars. They took the entire Wednesday last week, so my normal podcast release day was taken by a different event. But okay, one event would be not enough to take me down. But we also had to release two big tunneling projects as our commercial work and to add to that, I was feeling a little sick. You perhaps can still hear this in my voice. It's probably not my normal voice. I hope the voice situation improves in the future episodes. Anyway, this was a huge pleasure to share the findings of our project with you. For the next weeks I have a bunch of great interviews approaching your way. We're having a charring of timber in mass timber buildings. We are having a history of single burning item test method and pathway towards intermediate scale facade testing method. We are having the role of insurers in fire. A lot of interesting episodes coming your way interviews this time, so you'll have a little rest from my rambling. Thank you for listening and see you again next Wednesday in the fire science show. Thank you, bye.