Virtual Speaker Series Featuring Greg Pavlak - Smart Buildings:Opportunities and Challenges

Virtual Speaker Series Featuring Greg Pavlak - Smart Buildings:Opportunities and Challenges

Show Video

(Captioner on stand by) >> PAUL: It is 12 PM. Welcome to the Virtual Speaker Series. See the room is filling up. As always, we encourage you to let us know who you are in where you are from today. Put that information in the chat so we can welcome you in to today's Virtual Speaker Series. We have a reprogram lined up today and talking with Dr.

Greg Pavlak. He will talk about the work he's doing in the Penn State architectural engineering program. He will talk about smart buildings and all the good work he is doing there. I see Mark from PS FBI and Tim down in research Triangle Park, North Carolina. Steve tuning in from Bluebell, Pennsylvania.

Class of 66 tuning in, good to see you, Jim. Brian in Camp Hill. Thank you for tuning in today and we will get started in just one moment. I see Charlie from Shamokin. We were just talking off-camera about the region I grew up and you are there in the heart of the cold region.

I see Lauren from OPP. Good to see you. Welcome to the Virtual Speaker Series featuring Dr.

Greg Pavlak, assistant professor of architectural engineering here at Penn State. I see Mercedes from State College joining. Good to see you. We will get started here just one moment.

Where else would you rather be than a Zoom room full of Penn Staters? I am Paul Clifford, CEO of Penn State Alumni Association and I would like to welcome everybody to today's Virtual Speaker Series which is being recorded. Live close captions are available for the event and you can access them by clicking the closed caption button at the bottom of your Zoom video window and then click show subtitles. You can also customize your caption view by clicking the string text link posted in the chat. We are live streaming today's presentation and this has been made possible for the gracious support of a donor and fund for access ideas and audacious goals.

Today's presentation will be archived and available on our website after the event. Welcome Dr. Greg Pavlak; assistant professor of architecture engineering at Penn State. He has 10 years experience in predictive modeling and optimal control of the building heating, ventilation and air conditioning systems.

Buildings play enormous roles in our everyday lives and present important opportunities to promote the sustainability and well-being of society. This talk will highlight key areas where smart thing technology can help address pressing climate change related issues. Solutions being developed within Penn State architectural engineering will be discussed along with several challenges and barriers to adoption of such technologies. Dr. Greg Pavlak current work focuses on developing of intelligent control technologies for buildings to lower operational carbon emissions and increase utilization of renewable energy.

Prior to Penn State, he served as lead scientist for a startup company in model predictive building controls. He holds BS in engineering from Hope College in Michigan, and PhD in architectural engineering from the University of Colorado older. Please join me in welcoming Dr. Greg Pavlak.

>> DR. PAVLAK: Hello, everyone. Thank you for joining and thank you for the introduction, Paul.

I would like to thank the alumni Association for the invitation to share more about the work I am doing and going on here in the architectural engineering department. How do the slides look? >> PAUL: Looks great. >> DR.

PAVLAK: Wonderful. As Paul said, I did my bachelors degree at Hope College in Michigan and then went to the University of Colorado for my graduate degree in architectural engineering. I spent a couple years working for the coefficient trying to do these building technologies. It was 2017 when I was able to join Penn State architectural engineering.

Today, I would like to talk more about our work and smart buildings and with focus on sustainability. For those not familiar and don't work in this space very frequently, buildings are significant consumers of energy and more specifically, electricity. If we look at the global energy picture, buildings are fairly large piece of the pie. Heating turns out is the largest single at 36%. Cooling is the fastest growing in use. A lot of emerging economies taking advantage and using refrigeration and cooling technologies.

All this electricity consumption is contributing substantially to the global carbon emissions. 28% of global carbon emissions is associated with building operations. This is the day-to-day activity in energy use of buildings.

11% is the building materials itself and construction phase. If we want to address some of these pressing issues, we need to consider buildings as a prime sector for improvement. There is a lot of opportunities today and I hope to highlight some of those opportunities to improve the building sector today. With the carbon and environmental greenhouse gas emissions, taking center stage recently over the past several years, we see a lot of targets being established. Recently discussions at the federal level have been 50 to 52% reduction in greenhouse gas pollution by 2030. Carbon pollution sector by 2035 and net zero emission economy no later than 2050.

These are not just buildings, these are overall emission targets. If we looked at the state of Pennsylvania, the most recent climate action plan of 26% greenhouse gas reduction by 2025 with 80% by 2050. There are a lot of local implementations.

For example, if we look at buildings the city of Denver have set goals for net zero energy and all electric new homes for the 2024 building code. Same all electric net zero criteria for commercial buildings in the 2027 building code. In New York City, local 97 was established with limits of buildings over 25,000 ft.². These are limits in terms of metric tons of carbon equivalents per square foot of the building that will be enforced.

These all highlight the need to now identify tangible strategies and pathways to achieving these targets. In the first specific holdings and then communities as a whole. Before we move on and talk about the details, this is a common question that comes up, what are Smart Buildings? Smart Buildings can help us address these climate and energy issues, but what do we mean when we say Smart Buildings? I think the important thing here is it's hard to define because the definition is always changing. This is a moving target as technology has become widely adopted. Maybe we no longer consider them smart.

They are the state-of-the-art or baseline technology. The definition has always been changing. If you look back 40 years, smart what have been defined as buildings that are automated so they can manage their own system.

Minimal human intervention in terms of scheduling and turning systems on and off and making sure they are working as needed. I think we would have made smart synonymous with automated. Now I think one definition that is proposed by the ASHRAE handbook of applications, ASHRAE is firmly the American Society of heating refrigeration and air conditioning engineers, professional society published many standards and guidelines for buildings. This is a definition, this is more like introductory paragraph taken out of the handbook.

Some key criteria they note are that smart buildings exhibit characteristics analogous to human intelligence. May be drawing conclusions from data, making decisions, taking action autonomously without being programmed. You can see we have moved from things being automated, to things being more human-like or autonomous. There is the 2015 version and a new version but the 2015 version is where this chapter on smart building systems first appeared.

It has grown in popularity and become a whole topic of itself over the last five to 10 years. As often happens, when things are hard to define you don't have clear answers, what do you do? You might ask colleagues. I did the same thing here. I ask colleagues and the Penn State architectural engineering department, what do smart buildings mean to them? What is their idea of Smart Buildings? I want to spend the next handful of slides just sharing responses. What this will do is introduce some of the work they are doing in this area because there is a lot going on here in terms of smart buildings. First, my colleagues Rebecca, JP and Wes, primary response was smart buildings and cities will need to have smart materials.

What they mean by smart material is everything from materials that are self-healing and self repairing that can adapt to changing environmental conditions. They are talking about materials that actually are sensors or part of the sensing network themselves. We can design smart materials that can actually help provide us information about the environment.

Perhaps even going to the extent of having some of these new materials be part of the computing that's needed to process some of the data as well so we can think about having more computational power located at the edge of these networks, rather than centralized computing environments. There are very interesting materials underway in this topic related to Smart Buildings and cities. My colleague Javad Khazaei, he really talked in detail about the cybersecurity aspect of smart buildings and cities. This idea that as we develop more intelligence, or computing power distributed throughout the cities and more intelligent algorithms that they will actually be able to help and dissipate in making sure the networks are cybersecurity.

From the dive -- identifying cyber attacks to implementing countermeasures, both in the physical system themselves and the information networks. My colleague Donghyun Rim, he spoke about the need for Smart Buildings to be more responsive to indoor air quality and occupant needs. This is very timely topic.

This idea that buildings if they have more sensing and data processing capabilities, they can actively monitor indoor quality in more detail and unique occupant conditions and respond accordingly. Part of this work in both developing and more detailed understanding of how pollutants move through spaces so the dynamics of pollutants. Also, how some of these pollutants may lead to secondary chemical reactions within spaces.

My colleague Yuqing Hu, she said that smart cities will definitely need to leverage urban scale models in digital twins. This is big area of work in terms of building information modeling where we have computer representations that are very detailed in terms of what's inside of the building, how they are constructed, where things are located. This is used more and more throughout the design and construction process, these detailed models.

What has not been realized yet to the full extent is how we can leverage these four interior lifecycles for buildings and cities and improve operations. This is an area she's working in developing better models and processes. My colleague Alp Durmus mentioned lighting will be a very important in not just lighting quantities but the quality of lighting with L. E.D. You may have heard of light emitting diodes are lighting technology that recently have become more available and cost-effective in residential and commercial applications. Now that the primary technology has been developed, there are opportunities for optimization of the quality.

We can tune the spectrum of light that comes out of the L.E.D. For things like improved occupant conditions. The spectrum and quality of like and affect our circadian rhythms.

It also impacts energy efficiency. How much energy these devices consume. We have complex trade-offs and this is really the focus of the work.

My colleague Nathan the topic of sustainable smart cities mentioned the need for new construction materials and structural materials. When we think about improving the operation of buildings, I mentioned that was one piece of the carbon picture. At some point, if we make that piece small enough, what is left is the amount of carbon in the materials themselves and construction process. In Nathan's group, they are really focusing on reducing that embodied carbon or carbon part of the building materials itself. One way to do that is me more use of wood and timber construction. He's working on new computational methods for early design.

We have to think in different ways about how we designed these buildings and use timber. We have buildings that have adequate structural performance but can help us me some of these carbon goals. That's what my colleagues thought in the areas they are working in. What do I think and where does my work fit in? My work as Paul mentioned earlier, is really on the control side of the mechanical systems. Through improved controls and design of the mechanical systems, the heating, cooling, and ventilation systems, how do we make buildings more efficient and flexible? When I think of Smart Buildings, I think things that are efficient and flexible. What we mean by that? This figure on the slide here first shows typical energy profile in commercial buildings.

We have maybe higher usage during the day in the commercial building. People are at work and temperatures are higher so we need more cooling in the building. Energy efficient buildings may use less energy during all times of the day. We see shift from the original blue profile, down to the yellow profile. If we add solar PV electronic to that, the low-profile, the yellow, what you might see is big tips in the middle of the day when it is producing electricity. It turns out that these steep changes here, steep drop in consumption and going negative meaning we are producing excess electricity, steep ramp events as they call them, they are challenging for the electric grid to absorb.

If we had a smart building, one thing it might be able to do to help is shift energy consumption in ways that removes these steep ramping events. We might use more energy during the middle of the day and have that offset some energy use during midnight in early morning periods. That's what I mean by flexibility; the ability for the building to change when it consumes energy based on some information, some extra piece of information. That is really where my work focuses, is building control to ensure efficiency and flexibility. An example question or an example you might think of related to this is if you told a building when energy was clean and dirty, what would you do differently? How could it change the energy consumption pattern to use energy when it's clean and less energy when it's not claim? The way that we do that or achieve that is by developing new control approaches that can help the building intelligently change its energy consumption patterns. Next, I want to highlight one the projects, specific projects we have done in this area.

A lot of my past work involves something called model predictive control. This is where you take a model of a building that can describe how it uses energy and couple this to mathematical optimization all routines. You use the model to try out operational strategies.

What if I change the temperature set point for this device? What if I turn the system on this time of the day? By using the model, try out many different control actions and find the one that achieves the objective you are interested in. You can find the control strategy that lowers energy use. It could be fine control strategy that reduces cost. It could be find the strategy that allows me to maximize my use of solar energy. This is technology that has been developed, fairly studied over the past decade or more by a lot of researchers.

There are challenges in implementing this in practice. Some of those challenges are developing the detailed physics -based energy model can be time consuming, costly. If we have detailed model of one building, it may not be able to directly be used for another building.

If we try to use these models within the optimization framework, it turns out it can be very computationally intensive. We need a lot of computers or high-performance computing environments to solve the problems. Also, it can be challenging to explain output. Overall, there is need to develop methods more scalable, transferable and interpretable. That is something we have been working on here. Just to outline three approaches, the first one is the approach I just mentioned were used detailed model for optimization and arrive at the control strategies.

The second approach that is considered and study is use the detailed model to training machine models. This is simplified models and learns from the data produced by this detailed model. Once we end up with the simplified model, we can use that for optimization and generate new control strategies. The approach we are taking in this work, I call it role extraction. Here we use the detailed model to do optimization and we end up with best set of control strategies.

After we have developed a whole library of control strategies, the next step is to feed that to machine learning process to extract the rules. I will show you what I mean in a little bit. First, I would like to mention a lot of the results showed today are based off of work that I have done in collaboration with men, fourth PhD student in the architectural engineering.

The process we developed for extracting these roles come at the idea of role extraction is think about this process of developing detailed data set where we know what the best strategies are. Once I have the data set, the process of role extraction is look through the data set for patterns. I want to pull out patterns that repeat and may be able to use over again or reuse for good performance.

That is exactly what we have done. We start with the full state -- full data set from the full optimization. We go through the filtering process to select subset of solutions which meet preferences.

Once we have subset solutions, we group them. We use clustering machine learning for patterns and put them in groups. Once we have groups, we can train another classification model. Declassification model form -- the output are decision rules or decision trees. That is shown at the bottom.

I will talk through these in later slides. Just quick note about the building and test case we are looking at, it's an energy model for 1 million square-foot office building. It has fairly typical HVAC system. If we took all control strategies we developed and plotted them, this is what it would look like. We have energy consumption on the buy access and think of this as total cost.

Each of these great points represent one potential control strategy. When I say control strategy, think of it as temperatures in the space that control when the HVAC system turns on and off. Each gray dot is potential strategy and you can see lower energy consumption and some have higher energy consumption and higher cost.

The same points are plotted on the right figure but with comfort metric on the Y axis. Higher is bad or I should say worse. Points with higher penalty mean there is more discomfort in the building and zero means no discomfort. The great points are all solutions. If we apply preferences we end up with red dots. The most preferred strategy for each day is the blue dot.

You can see a lot of ranges of strategies we can select. This is what it would look like if you plotted in terms of time. We have these temperature set point on the Y access in time of day on the X axis.

We start with 400 possible control strategies. These are all good control strategies and produce by running the model predictive controller with the detailed model. These all minimize at least one of the objectives.

The next step is to apply clustering. We go from having 400 control strategies after we apply the clustering, we end up with about 30. We have gone from 400 down to 30. Once we have our smaller sets of common patterns, the next step is try to decide when should we apply those common patterns? This is the process of fitting the classification model. We took those 30 strategies and generated classification models and this is the output and what it looks like.

It is a decision tree restart at the top and follow the logic down the tree. When you get to a terminal, the number that is predicted here corresponds to set point strategy for the building. If you look at the rules that, we start at the top and included occupancy is one predictor variables.

Occupancy less than 385, remember the large 1 million square-foot office building? 385 is low occupancy number for this building. The first level found key distinguishing factor is whether or not the building is fully occupied or very minimally occupied? In essence, this separates the weekdays when people are at work versus the weekends when they are likely home. That is the first branch.

If we dig down into the tree, you can see split on future weather. If future weather is less than 31, we can follow this path. If the current weather is less than 26 Celsius, we follow different paths. Eventually, we end up with these lower notes here. If you follow the logic, these are hot days for the past temperature was hot and the previous day was warm and it's forecasted that the current day will be warm as well.

On these days, in order to reduce energy expense through peak demand or energy use reduction, we may need to start the building up earlier or may need to do pre-cooling or sub cooling. Temperatures need to be lower in the building and they normally otherwise would be. Overall, what this highlights is using this rule set, we map to these typical control strategies rather than running the full detailed optimization problem. How well does it work? After we generate the rules, we want to test those.

I have three figures. We have the comfort penalty that we talked about earlier. We have energy consumption in the middle panel and energy cost on the right panel. The first piece I want to bring your attention to is the Graybar.

This is the default building operation without any intelligent controls. Just standard schedule operations. If we compare the Graybar to the dark Lubar here, there is large reduction in the energy cost.

Implementing the intelligent controls are able to save quite a bit of energy cost. There is also similar reduction in use and you can see slight increase in this metric. On the next slide, we will put this in perspective. We will unpack that in a little bit. On the slide and want to bring your attention to the savings and cost and energy from the smart controls.

If we compare, each of these colors, green, yellow, light blue, purple and red color, these are different versions of the classification model. There is big jump in savings from running the fully detailed optimization problem. There is not much loss in savings when we moved to using the simplified rules.

For example, the yellow, blue, purple and red, these are all using different versions of the simplified rules that we just looked at. We are able to achieve almost the same energy savings and almost the same cost savings by using the simplified rules that. To look at the data more in terms of time, we have time on the X axis and the first panel showing temperature set points. It shows the control strategy implemented compared to the optimal case.

There are a lot of similarities between optimal case and simplified rules. What I want to highlight is the lower panel here, TMV is used in buildings between +0. 5 and -0. 5 is the acceptable range.

I have highlighted that in the gray shaded region. What you can see is most of the models are within the gray box during most of the periods. There are a few spikes, but those excursions tend to happen in a lot of models even the opt case.

I think what that means is it helps us interpret this comfort penalty result and even though there are some variations, all of these are fairly low in terms of thermal comfort degradation in the building. Just to summarize, we saw several simplified rule sets performed quite well and the rules are quite simple and interpretable. You can look at the rules and interpret the splits. The limitations, we need to do a lot more testing. We didn't exhaustively explore the feature set used to build these models.

Just to comment, we were hoping to go with this, now that we have simple set of rules, the thought is if the rules are simpler and more interpretable, it may be easier to apply these to other buildings. Rather than running the full optimization, full model predictive control for new buildings, can we take these near optimal simplified rule sets that we have developed for one building and transfer or slightly adapt them to another building? That's an open question and something we plan on testing. If that works, I think the consequence or result will be we can help overcome scalability and and limitation challenges that come with model predictive control. Just to sort of conclude overall, highlighted several technologies and opportunities and by no means is this exhaustive. I have seen this going on within department, there is more across the University departments that intersect with this area.

Some of the challenges I didn't touch on in detail is this idea of new construction versus retrofit. This is a big challenging -- challenge for technologies that work well when you have all of the design variables and opportunities at your disposal and that would be appropriate for retrofit. I think that's very challenging. Something we encounter more is what I call option process.

A lot of products being labeled as smart and being for smart buildings. With all of those options it's most impossible for any one person to know the details to distinguish between which ones work well or which ones serve the function needed? That leaves slow adoption of some of these technologies because it's hard to pick one. Sometimes the economic opportunities and evaluations are challenging or not clear with these technologies.

There can be diverse range of incentives city to city and location by location. It might be cost-effective for technology to be used in one place but not another. Often times, with these more complex situations, it does require coordination across multiple stakeholders and vendors.

I can say there are a coupled bright spots that I see. I think there is more demonstration projects happening in this space looking at making individual buildings more smart or smarter individual buildings both retrofit and new construction. Also, looking across communities.

Going beyond just these technologies improving single buildings by themselves? How can adopting these technologies across entire communities help promote the well-being, efficiency, sustainability of the entire community? As we see more demonstration projects being completed, we will start to learn more about the technologies that work well. We will learn about the challenges and how to address those. It's an exciting time and keep your eye out for these demonstrations. With that, I think I will end my formal presentation here and turn it back to Paul for Q&A. >> PAUL: Thank you.

I think we have a number of questions that have come in. It's get to some of those. If you have questions you would like to ask, use the Q&A tab at the bottom of your Zoom window and we will try to get to as many questions as we can. Let's start off with how common are smart buildings currently being applied? Of all the new construction going on across the United States, what is the percentage that these elements are incorporated into those projects? >> DR. PAVLAK: Great question.

I think it varies across different sectors. As we saw also, different levels of smartness. Almost all buildings are taking advantage of advanced data analytics and automation. Most systems have ability to collect more data into processing of the data even if it's at the basic level.

That is one thing fairly widespread. Other advanced technologies like controls I mentioned are not widely adopted yet and more up-and-coming. We have done some deployments in large commercial buildings and I mentioned the million square-foot office buildings in large cities. Still plenty of opportunity for these technologies to be adopted.

>> PAUL: You talked about retrofitting at the end. Are there different levels of maturity of smart building technology and approaches that can be applied to existing buildings? >> DR. PAVLAK: Absolutely. One of the exciting things about these control technologies is that they do tend or can work well as retrofit.

You don't have to change a lot of the physical features. You don't have to completely redo the mechanical systems or the envelope. They tend to be more may be some control upgrades or maybe programming that has to happen.

Comparatively, that process of connecting buildings and improving controls is often lower than some invasive measures. I think that is something that is promising for retrofit and new construction. In the residential space, things that are fairly mature and being adopted more widely are things like smart thermostats. A while ago they came out with the first program ago thermostat and they were not being used very much. One of the reasons was they are a lot like programming an old VCR. A lot of clicking of buttons and things like that.

These newer versions you might call smart thermostat you can program to your smart phone or web interface. They are more graphically oriented and intuitive. They find people are using these more and more. They are programming them which is leading to energy savings.

Things like L.E. D. lighting have reached price points where they make sense for residential and commercial products.

Those are all opportunities for technologies available now that are suitable for retrofit. On the other end, you have things like (Indiscernible) which is the next version of L.E. D.

That optimize lighting conditions based on the needs of the human occupants. Not just lighting levels for visual tasks but the biorhythms of people that improve our cognitive function, well-being and mental health. I think there is a spectrum. >> PAUL: My time on my VCR was still flashing before the technology became obsolete so I was never able to figure that out. It's good to hear some of these technologies are becoming more intuitive and simpler for the users to figure out.

You mentioned COVID-19 and the air quality piece of the work that you are doing. Steve is on the Zoom here and I know he thinks about this from a healthcare perspective and the air quality in hospitals and healthcare facilities, but any thoughts on the minimization of transmission of viruses that's going into the work you are doing? >> DR. PAVLAK: Great question. Just in general with existing technologies, there are opportunities to improve control of these airborne particles or infectious aerosols. Through filtration.

We can get the filters and systems that have recirculation of air. If we make sure we use high-quality in the filters, that can take out the small particles and can be effective means. Spaces that don't have recirculated air and don't have dedicated ventilation systems, we have portable local air filters. The key there is to make sure we pick products that have certified performance and don't have adverse effects or maybe remove one particle but admit another that can be harmful.

There are strategies that exist today that may be we wouldn't call smart. The next version or round of these types of products that may be would fit into the smart building category is smart ventilation products where we have more close monitoring of these conditions in the space. If we can actually detect when more these particles are present in specific sounds and we can pick from a set of intelligent ventilation strategies to reduce occupant exposure, or things like dynamically changed airflow patterns, not just change the amount of air we put into the space, but make conscious decision to change how the air is moving and influenced the dynamics of how the particles enter and leave the space. I think that is the next level. Making these more intelligent and responsive to the unique conditions that happen in each building. Buildings are very unique.

The transport phenomena that happens within each building is very unique and complex. Developing technologies and control systems that understand dynamics and can respond accordingly is the next level. >> PAUL: My wife and I have this discussion all the time around the house about the perfect temperature.

I think it is 72°. The house should be 72. She is more of 68 percent. Tom ask the questions maybe along these lines is the modeling account for the individual occupant variation? Specifically the outlier whose comfort preferences are some occupants who might install electric space heater under the desk because of the control is too cold. Talk about how the model takes into account the individual occupant or outliers? >> DR.

PAVLAK: Great question. In one of my slides I had the gray box for the acceptable comfort range? There is a range and people have different references. People have different preferences in the models can take that into account. What we see more and more now is the use of stochastic models. They capture the uncertainty of the variability. Instead of picking one of those regions or picking a single temperature, we can model different occupants and the likelihood that they will be comfortable in taking more probabilistic perspective on quantifying these effects.

That is another area we are doing work in is developing control strategies that can account for this uncertainty and variation. If we can model and variation, develop more controlled strategy instead of responding to a single number or based on a single strict preference, can actually respond to most of what the occupants would like or what occupants would find most comfortable. Then figuring out how to dynamically change those values over time is interesting. Another related piece is more on the design side.

There is work in personalized thermal comfort products. Either having more localized ventilation and air conditioning and smaller office spaces and things like that. You bring up a -- it's a well known challenge, but doesn't have the perfect solution because we cannot make everyone want the same thing. >> PAUL: Exactly. A number of questions coming in about the information being collected or generated in the systems. We start with Tom's question.

Is the idea here to use AI to observe the building over time to learn how the building performs and make automatic adjustments based on required parameters and cost versus comfort? Renewable versus fossil? Is that ultimately where you are working towards? >> DR. PAVLAK: Yes, that's an exciting direction and there is a lot of work in that area. It does overlap with my work. The approach that is described there is what I would call Shirley data-driven approach where you collect as much data as you can about the building and then develop machine learning models or artificial intelligent models that can improve the operations. Where that intersects with my work is what we are also trying to do is embed the knowledge that we do have about buildings and help support the process. Is building engineers, there's a lot we know about physics and how buildings behave in the preferences of the owners and occupants.

That can help reduce the amount of data we need. It can help these algorithms learn faster and help them find strategies that don't violate known constraints that might exist. That is exactly right, there is a lot of work in that area and trying to figure out how to do that efficiently without needing huge amounts of data and huge processing times is part of the challenge. >> PAUL: Absolutely. Question here from Stewart, what are your thoughts on using single occupancy network based on the lighting system to share data with other systems to feed into the decision tree? >> DR.

PAVLAK: Sorry, could you repeat that? Is it written -- >> PAUL: It's in the Q&A but single occupancy network based the lighting system to share data with other systems to feed into the decision tree? >> DR. PAVLAK: I think that could be a good strategy. I should say when we presented the decision tree is one version of what that could look like. You could imagine having different sets of rules for different zones. You could imagine customizing and tailoring these four different building sectors.

The information that these decision trees use to eventually decide on a strategy to implement, there are infinite amount of possibilities in terms of what pieces of information these rules can respond to. I would say there are a lot of ways these can be configured. We need to try more of these out. If you had some ideas on what might work well in the field you are working in, please follow up and I would be happy to talk more.

>> PAUL: I know nothing about this topic. Everything I know about this topic I have learned in the last 54 and its. I'm thinking of when you walk into an office and the lights are off, but then the lights flipped on. We know that's an energy-saving tactic, right? Does that also trigger something in the cooling system that kicks that on? Or in the water heater anticipating someone will need to use the water at some point and have that trigger other regulations? >> DR.

PAVLAK: Yes, exactly. A big recent topic of interest has been developing more accurate models for occupants. Not just the thermal comfort and preferences, but what is the most likely action or next action going to be? How do occupants move throughout the building in different spaces? Part of the reason those models have been hard to develop is because it's a hard data set to collect in the first place.

Tracking enough occupants and having them respond and noting preferences and what they are doing in the building, that takes effort and a lot of time. Those data sets haven't been available to develop models. As we improve those models, I think you are exactly right; we start to get to predictive-type of control approaches where they say the person has entered the building, just opened the refrigerator and most likely to do XYZ next. That opens up the issue of security and privacy.

>> PAUL: That is my next question. We have time for about two more questions. I apologize we are running up against time. I want to shift to the security issue. As buildings get smarter, what are some of the top two cyber threats from your smart building technology in your point of view? >> DR. PAVLAK: That's a good question.

Now more than ever we are seeing building separate the networks that they use for building automation and controlling the building and separate that information technology network from maybe general-purpose of the Wi-Fi network. That is good practice. In buildings where that is not the case, you just have a challenge with too many people using the same network. That is a easy one that can be solved for new construction and making sure we separate IT and general-purpose networks.

I would say the access control can be a challenge. Another major challenge are all of these devices that are coming from different manufacturers that are maybe wanting Internet connectivity. It's hard to know what cyber security standards they may be implementing, if any. It's hard to know which device may or may not be the next vulnerability in your network as people start buying and adding more devices to their network.

I would say those are two that come to mind. >> PAUL: We are studying this. We are developing these technologies here at University Park.

Has this modeling been applied to any projects at the University? >> PAUL: -- >> DR. PAVLAK: Good question. Are you referring to specific technology I presented on? >> PAUL: Yes, the modeling you talked about.

>> DR. PAVLAK: We haven't done it yet on the building here at University Park. I think the more detailed control technology is what I mentioned. We were commercializing in some of the large office buildings in Chicago and New York. Would say this new approach where we try to extract rules from the more detailed models, I would say that is fairly new and we have tested in simulation but not in the field yet and I think that's another opportunity. After we do more testing and simulation to get the models and rules right, field testing will be very interesting.

>> PAUL: Excellent. If people want more information about the work you are doing, where can they go to find information about that and be supportive? >> DR. PAVLAK: If they are interested, I would first say the architectural engineering department website has a faculty page where you can find all of the faces that you saw in my presentation today. If there is particular work that stood out, all the faculty I mentioned today would be more than happy to receive direct contact from you while. Feel free to reach out directly through email and the phone is listed there as well.

Email is usually the best way to get a hold of me. I would say that is probably the best place to go to start with. Most of the time there are links to personal websites where maybe there is more detailed information about specific projects. >> PAUL: I know Carrie and Lisa just put the link in the chat for people who might be interested.

Thank you for joining today and sharing all the great work you are doing here at Penn State. We greatly appreciate it. >> DR. PAVLAK: Thank you for the invitation. It was my pleasure.

>> PAUL: Reminder, we are hosting additional Virtual Speaker Series in the coming weeks and months in the program is in addition to wide array of career programs available throughout the year you can view the full listing

2021-05-03 13:34

Show Video

Other news