0:00 Everyone hear me good? My name is Noresh, and I’m an engineer by practice, and I lead the competitional design effort for enging groups at the Northeast region for stantech. I’m also a doctoral candidate at the building science and engineering group at drexo in Philadelphia. With me today is Augustine Salos, who’s a senior architect himself and practices out of the Miami office.

0:25 So, a little bit of plug: stantech is a global top rated design firm offering architectural, engineering and scientific evidence-based climate solutions for our clients. Some core business lines are listed here. Me and my colleagues here come from the building sector predominantly, and under buildings we have this digital practice entity, where, you know, our competitional design groups gets folded under that umbrella. Definitely a big presence in North America, but we, you know, collaborate actively across borders, and our computational teams are pretty dispersed all across the world.

0:56 Yet another plug about building science group at Drex, led by Dr Jinn. These are the people who are currently actively involved in the lab. Big focus on the grit interactive building projects, controls and fault detections, occupant behavior and well-being, which is my personal research focus, smart buildings, so on and so forth. Typical funding from our lab comes from like DOE, NSF, U, NIST and ashray, for the most part.

1:26 So, focus of the presentation, big question: why occupant centric design? You know, why should we consider personal preferences when we design spaces that we actually spend a lot of time at? What is the gap? Why are we already doing this at the moment, and how competitional solutions can actually be used to accelerate the adoption?

1:44 This is a picture of the current project we’re working on in Dubai, it’s Haman Bin Rashed Cancer Center Hospital. Not really going to talk about this during this presentation, I just wanted to include it for marketing purposes because it just looks cool. But disclaimer though: the content today is more of a sampler of different approach to occupant centric design in throughout your design processes, rather than a very very deep dive into a single aspect of application. So just, you, you’ll be getting bite-size of appetizers, going back to Sean’s reference on food yesterday.

2:17 So, going to fly through these numbers, but basically we spend a lot of time indoor. You may have come across some of these numbers already: we move from one man-made enclosed entity to another, car to building, building to car, in the back to car. You know, the one big thing I did want to highlight here is that almost 83% of executives and business owners recognize the need for improving IEQ, and they’re planning to invest heavily in this space moving forward. So this is a big thing, especially coming out of COVID, there, there’s a lot of lessons to learned from, from whatever, you know, hit us four years ago.

2:52 So, pretty much a Google search might lead you to a lot more publications or articles giving you these crude facts. The one thing I did want to highlight here is the recent study that came out of Berkeley lab, right here in the chart, that sort of compares perceived comfort between WELL certified buildings and lead certified buildings, right? For those who don’t know, lead predominantly deals with more of an environmental aspect, sustainability, energy, materials and such. WELL, on the other hand, gives a heavier focus on like comfort driven metric, so occupants themselves, you know, reported feeling much much better operating, or behav, walking around in these buildings. So this just, you know, proves to show that a well-designed building is a building that actually centers occupants and their needs in the design process itself.

3:43 So you probably already know this, some of the big factors that really impact the performance of buildings I’ve shown here. Aside from IEQ, I also wanted to partly talk a little bit about this occupant behavior, which might be a new concept to some of you here, strictly on the simulation perspective. Conventionally, some of the AC folks here might know, building designers consider occupants as a very very static entity, a space is occupied or it’s not, sort of a ones and zeros possibility there. You know, for mechanical engineering specifically, we consider occupants as a heat output, right? You know, when we’re doing our calculations, ranges from like 100 watts to 430 watts, depending on, you know, heat loads of the space or the activities that you’re actually performing in this space. But rarely do we actually consider these dynamic behavor, behavior, of people when we’re doing any building performance simulation.

4:32 So I’ll talk a little bit about a toolkit that’s developed at Drexel called the Habit toolkit, to model these dynamic conditions that are really tailored to different people. So computational modeling and simulation opportunities in this realm exist all the way from conceptual design, U, to post design in the operational phase. The SD, the early DD face, is probably where you would see all the cool geometry optioning, form finding, all the very architectural concept that you saw in KPS presentation yesterday, and and some of the facade presentations. But back end of it is a very very data heavy process, where we mine the building management system data, sort of infer trends from the meter or sensor data, and improve the performance, or, in some cases, come up with sort of a predictive algorithm that, you know, helps you operate your building efficiently, that can get integrated with controls, predicts the space when it will be used at specific time, and who’s going to use it, and how do we, you know, set the set point according to this occupant that’s using that space.

5:34 So there are a number of different KPIs that can be used for simulating some of these cases we’ll talk about. This publication really goes over each and every one of them in depth, you know, sort of really helps to hone in on what impacts, you know, thermal comfort, what impacts visual comfort, clity comfort, and what, so on and so forth, that can be used in your modeling perspectives. So with that, Augustine will be talking about thal comfort in a very early design stage pro process.

6:13 Gustin, thank you, Nores. And so this lectures is about thermal assign. So a lot of designers and architects always are looking to enhance their ideas based on several early studies. This project is blend computational design tools with AI and optimize indoor thermal comfort levels. Our goal is to predict and enhance thermal condition in spaces, ensuring they are not always visually appealing, but also tailor occupants comfort. This work was developed in the IAK, the architect, the institutional institution ofan architecture for Catal.

This methodology include, as you can see in the diagram, and some of the key aspect of simulation. And, for example, we took AI Plan Finder, which is a parametric design, parametric developing tool that create a lot of iration of floor plants in 2D. So from here we took the application and develop a script that can help us to bring 2D graphic solution into a 3D modeling information. And from there we color coding every single aspect of the modular condition of the apartment units, and from there we develop a screet, we develop a script into the Grasshopper, utilizing Lady Hop and analysis with MRT, and from there generate a a strong data set of images that can help us to organize a huge package of information based on this performance.

8:13 After that we use PTO PCK with conditional GUN, just to generate all the data set that we later need. Here you can see more or less the process, a little watch out, but in the first stage we use the Fli Finder, second, second step we use the the color coding system to visualize all the materials of the floor panss, and the third portion you can see the simulation aspect of more than 5,000 apartment units, so we can collect a lot of data for our later, later develop, developing app.

8:56 Here’s a sample of the color coding information, as you can see, we, we, we had a layering that explore and define exterior wall, floor plans, interior doors, windows and interior wall, two set of material and two different type of color coding, to understand the the simulation process. And here some of the data sets that we, we, that we, at the end we, we have in here some of the simulation resolve. And here the methodology that we use to bring this technique of computational design tools, organizing all this data into a flash web application, and then uploading an a simple image from the, just a sketch drawing, and from that sketch drawing you will have like a small anar image from your, for your study. And in here, of course, everyone have an outlook, and here we’re trying to just say that at the end we’re trying to develop an a plugin that will be interfacing with Rabit, or Rhino, or any other other type of software.

Thanks, Augustine. So that’s really using MRT, mean radiant temperature, as sort of a KPI, to sort of simulate a large data set that can actually be used to train a model for really really early design stage predictions, right, of floral plants, how it, it’s performing.

10:41 I want to talk a little bit more about thermal comfort analysis. U, CFD really allows us to really see how the fluid is reacting in the space, and, you know, around the occupants themselves in the space, so it’s really a good way of quantifying performance in a sensitive environment like a hospital. We did this for this 1 million square feet Cleveland Clinic project in Ohio, it’s a Neurological Institute, a lot of glass in the building, and the perimeter where the examinations rooms were located.

11:08 One thing about hospital design is, a, it’s a very code driven process. The architects already know the type of footages they need, the adjacency criteria, with room programming, so on and so forth. This is a IPD project, integrated project Delisa, was moving really really fast, particularly fast, and from yesterday’s presentation from KPF, you know that, you know, you have 3 weeks for submissions, and you’re trying to do CFD in 3 weeks, I don’t know how well that’s going to go. So as we’re preparing drawings and feeding them back for client review, contractors, we’re really getting ready with the shop drawings, which is the whole intent of the IPD projects.

11:43 So, you know, in this phase of the project, we didn’t really do CFD too much parametrically, we only had like few different options anyway, because the space was already planned for us. But what I did want to highlight and talk about is using CFD to computationally influ form, what window to- wall ratio and what room height would be appropriate to be, you would be a useful way of designing these, right? Like if you think about it, seiling the height window is going to bring so much heat in, and, and that’s going to be very very uncomfortable. So how do we inform this using CFD? That brings me to talking about this, you know, how, how do we manage computationally expensive models like CFD? Again, design is moving very fast, and no one can wait for one team to make a decision.

12:28 So, traditionally, building performance simulation, like an energy modeling, probably going to take you about minutes to do, like few iterations. CFD probably take what, hours, weeks, depending on the complexity of a building, or what Inlet, Outlet conditions are. But if you start doing BPS plus CFD plus optimization plus generative design, I don’t know how long that’s going to be taking, right? So it could really Skyrocket pretty quick, and you saw in Rick’s presentation, in in the Berlin session, he was talking about 11,000 few different options that they were doing, would have taken five years.

12:59 So one of the solution here really is, you know, Gabrielle has already talked about this earlier, using a Sate model. You could pre-train a lot of CFD simulation data, feed it into a Sate model, get your trained model, and then use that in your in Loop simulations for your work when you’re doing these predictions, right?

13:21 And so this sort of Technique, we was applied at a project in Drexel. This is a is a very complex project, it’s a heavy simulation, he simulation heavy project, where there are few different data exchanges that are happening between different models, but for the sake of this presentation I’m just going to be focusing on the building occupant data exchange model specifically. The goal of this project was to really model the impact of the stochastic behavior that I mentioned earlier from occupant behavior model. So, when a grid signal is sent to either load shift from the building, or shed the load from the building, the building is going to want to change itself set point accordingly, but that might actually impact how the occupants are feeling, right? And so they might actually want to go and change the set point back up, or open up the window, so on and so forth. So it’s a really really complex simulation, but the goal is to see how the reactions or occupants actions in this space impact the energy performance, and how do we optimize the controls of the buildings accounting for these type of really random signals.

14:22 So this is the study that you can pretty much scan and read through, if you’re interested, in Reading scientific Publications. So the airflow model, as I was talking about, was going to be really computationally expensive, especially for a Time series based in Loop control simulation, just as this project. So, like I mentioned, for each time step there are certain actions that are being evaluated. So the way they, the way to address this, was essentially training a sgate model, where, in this case, they produced a simulation data and trained Ann model that then helped with in Loop simulation, as they were exploring the design space of optimized operation conditions.

15:04 Next, on the behavior model, this is a agent based model developed at Drexel, basically have these actual occupants simulated in your room, in your building, and they have various varying different thermal preference profiles, as well as the constraints of behaviors, right? So it’s very very very stochastic, and, and, and they would produce different actions in the building that you’re simulating. Sequentially, a series of actions are taken depending on the discomfort range that they feel, but what it does in form is that, considering their behaviors, you see sort of really wild difference of uncertainty in performance of the building itself. Savings in energy can still be achieved, but really really huge impact, that ranges from like a person, one person, to all the way 58%, depending on different actions that the occupants can take. So it really makes a simulation and understanding how actually your building performs pretty rich, and inform your clients that you got to be careful with, you know, how much of a thermal thermostat control you give someone, where you place certain things, and really account for their behavior when you’re modeling and designing buildings.

16:14 So performance- driven IQ design, as such, takes almost the same path if we’re strictly talking now in the design space, from the SDS and DDS, you know, going from building model down to parametric model developed in Grass oper Python. We set like objectives in, in objectives, in the in in in, like honeybee and ladybug, for thermal comfort, visual comfort, so on and so forth. You generate a large design space using wallsa or octopus or any other plugins that you might be using, and, you know, go through a design Explorer, select the right design that actually works for your space. So this is a pretty typical process that, you know, computational designers are building, designing the buildings, would go through.

17:00 So some of the additional indoor Environmental Quality information that you can plug in, that right now exists natively in grasshopper, is this contemp IQ model as well. So, of, you can actually model voc’s, pm2.5, and and all the materials that actually gets spit out indoor environmentally, and then you can also do acoustic modeling using padm too. So I want to shout out to Arthur, I’ve been working with them, to sort of use this in loop simulations as well. Well, it’s pretty rich rate tracing simulation packet them that you can sort of use for your acoustic modeling as well.

25:00 So, with that, I want to talk a little bit about cross modality concept, where one ieq actually impacts how we actually perceive the state of the other ieq. For example, when you actually feel thermally satisfied in a space, you raise your rating of your comfort on Acoustics, even though it’s probably loud in that space. So this is called this one veto effect, and there are a lot of studies that are looking at these sort of interactions that happens good for us, because we can use these sort of informed empirical data from experiments, and feed that back into our simulation models, and sort of really understand how complex buildings can get, as you start simulating these situations.

Drawing inspiration from that concept, aside from my personal re research at stantech, we’re working on a modeling, on modeling a combined comfort situation, and understand how that would actually impact the design. So in this case study, with a behavioral hospital Northtown, we try to look at how thermally sensitive people or acoustically sensitive people would actually choose preferred location within the building itself, if we model certain criteria according to this equation. You know, obviously the grasshopper script is not shown here, it’s not visible here, but it’s messy, pretty messy, and still needs to be improved, but we use an agent based model to visually see where they would arrange themselves up there. So we’re sort of working backwards here, to try and understand the complexities of these different preference profiles of this agents. We can then further include more criterias and kpis, and make the process more robust, so it actually becomes more useful in our early space planning exercises, where you can influence more of the interior design and look into specifics of patient adjacencies.

So the additional kpis that I was talking about, that we’re looking into, are these sort of a multi-sensory environment aspects, material types, behavioral health condition is an interesting one how do you know if someone has this Behavioral Health condition, how, how close can you place one over the other. These are sort of constraints and kpis that you can still include in your simulation, and sort of inform how you plan your spaces, right? Clustering algor, algorithms, are probably going to be useful when we get to that stage.

So we’re constantly experimenting and working with experts in real case studies, such as this one, to really flush out these kpis that we can then integrate in a simulation study. So we’re doing both empirical studies, experimental studies, down in the field, collecting those kpis, going back to our comput, computational teams, integrating them, testing it out how it works, and bring it to fruition.

Similarly, at Drexel, we did a chamber experiment looking at the interaction effects of IQ on comfort and stress. We did a four-week study collecting both subjective surveys and objective biometric data. Generally, as you would expect, you know, occupants pretty, you know, felt pretty terrible in extremes of temperature and lighting conditions, but what really was interesting, though, is acoustic level did Mask the other ieq perception. So we’re almost seeing those one veto effect that I was talking about early, early in, in the few slides.

So the more we collect this data, it’s going to be much more useful for a machine learning training model. Stant, we basically have this Roomba like robot that just moves around your space collecting these data. What’s cool is, it actually builds a digital twin in a few days, and starts alerting the users about issues that persists. These data collections actually becomes more useful and informative when we further fine-tune our comfort model, right? So this essentially becomes a feedback loop into how we actually be considering these nuances within the design process, specific to even certain locations of building types.

Even so, you could ask me: what is Stopping Us in practice? Nash, what, why, what is the gap, where is this data coming from, what can we do move forward? And Gabrielle has talked about this too, and and Pastor Labs talked about this too. You know, we need clean labeled data, right? All buildings are different, so identifying these nuances and different data sets from buildings, and labeling them for machine train model, is going to be the much more useful way of prediction workflow in the future. Most building models are case study based, with limited generalizable labeling, whatever algorithm that’s developed for this Navy Yard building is not going to be applicable for a hospital building, right? So how do you pick out features that can be generalizable enough, or even transferred to another building? Probably you could use a transfer learning model, so on so forth, or you can even use active learning, as you train your Stargate model, you pick on new set points and retrain the model in the loop itself. And, you know, comfort is subjective, we need to actively work to objectively quantify it for modeling purposes.

So there’s a lot of experimental work, and and fuel studies going on, and the the need for design toolkits for occupant Centric design for designers is something that we need as well. And these annx 79 and 95 that you see, it’s an international agency organizations projects, they pretty much work with a lot of research institute all across the world to try and quantify this information for us as a design toolkit.

So the main takeaway, you know, integrated occupant Centric building design is the future, that’s how we’re going to be designing the buildings, and slowly that paradigm shift is happening. We’re collecting and labeling data accordingly, coming up with design toolkits, and this would, you know, help us with the modeling effort that we can use, especially in a simulation World. Obviously we’re working with different tool sets that natively already exist in grasshopper at this point, and, you know, we might turn to python to code something specific, but at this point I think we’re, we’re going in and out of different simulation platforms, and maybe even pasal apps might be something that, you know, you might consider what we working with.

I do want to close out saying, you know, we need to follow the lead of other Industries, and I want to talk sort of like put the slides from some of our participants here too. Some of these product designs, obviously, they’re very very much more user Centric design principle, integrated design process, right? But when it comes to building, if you think about it, the feedback loop is almost lost. You design a building that gets built in four years, then you do a post occupancy valuation, you understand how that building is actually performing, you collect that dat, data, but you can’t really go in repair at, at that point. Like these products actually gets continuously improved every time someone complains about it, but buildings don’t operate that way. So design toolkits, you know, however design toolkits they’re developing has to be much more robust into account for this, and we have to actually learn from buildings that’s been built too. So collecting this data from existing building, even age old building, is going to be useful for us moving forward.

Watch list, in case anyone’s interested, this is pretty much all the institutions, or, and more, there’s actually more that are actively working on publishing data sets, U, that can actually be useful for your model training. Information on how to do occupant Centric modeling in databases, labeled clean databases, that, if you are interested in doing machine learnings, you could do that, you know. So if you start looking into the annex, like I mentioned, the annex 9579, you’ll probably, you know, find your way into different different Publications and data sets.

25:00 So, with that, I also wanted to highlight this publication, came from that anx 79 group, shout outs to Dr William O’Brien and team. This book pretty much goes over, from A to Z, everything that you need to know about occupant Centric design, from simulation tools to optimization studies, to how to do an experiment, everything, U, from Soup To Nuts, you know. And so with that, I’m calling upon Augustine to sort of close it off, why we need to look at the past.

25:33 Thank you, everyone. I think we, as a designers and Architects and Engineering, we just trying to fill this gap between a complex engineering and aspect of the practice, and the early design process for Architects and designers. So I think, for a lot of Architects, it’s really important to see that approach into more integrated solutions that can go from the early very simple simulations aspects to a very complex engineering conditions, and and simulation projects.

25:51 So I think why look at the past, because the past can help us to understand some of the key components, on fundamental components of our practice. For example, we have been learning from bet that functional and form is something that we can really create a criteria that can develop a a whole holistic approach for our practice, but also, in our time, it’s very seen the complexity of our world, and we seen that it’s key for us, and for designers and Engineering, that include a human Centric Evolution for our vision, is, it’s a critical H into our practice. And the main reason is because what Nori was explaining us, in, in the building experience, we, we build the building, after the building gets built we start seeing some of the either issues or benefits of that. So I think coming and bringing human Centric aspect of design, it will definitely improve all the aspect of the the cities and the buildings. So thank you. Is that some of the references? Thanks everyone, appreciate it.