Your Building with Artificial Intelligence on the Edge

A bit scary? This is terrifying! But anything and everything is terrifying if you put it in the wrong hands.

Sinclair: Hi Nicolas! How have you been since Atlanta? I know you have two small kids, and that usually means sleepless nights. Maybe you have a need for some Artificial Intelligence?

Waern: Hey Ken, everything is great here in Sweden, can't complain! And even though it was great in Atlanta, it certainly feels like it happened a lifetime ago. Yes, there are the occasional sleepless nights with Isabeli and Maximus. But there's so much going on at the moment that I feel like my intelligence level is increasing every minute! Maybe, I, don't need artificial intelligence in the future, but I know for certain that our customer's buildings do! Because as you know, if you want to sleep better, you should invest in a smarter building. Not sure if it would help, yet, getting toddlers back to sleep, but combined with sensors and wearables, it would give me insights as to what might be the problem.

And that should be the starting point for any future AI/ML _Insert technology here_ initiative. Not directly see what AI could do, but think about it from a company perspective, "How ready are we for AI to begin with?"

Sinclair: That's what I wanted to talk about. AI - Artificial intelligence, ML - Machine learning, Big data, Data lakes, all of these things, are they for real? Or just buzzwords?

Waern: You forgot Brain-Computer interface, Quantum Computing, IoT, BioT, Edge computing, Fog computing, Distributed intelligence and probably 10 000s more examples like these. According to Gartner, we've got our hands full, and a lot of the hype hasn't gotten through widespread adoption just yet. But I would say that things are moving. A lot of buzzwords 2-3 years ago have materialized into real products and solutions in the market, and I thought I'll briefly go through some thoughts that I have had.

Running your building two hours from the future might be interesting to know more about, and it's actually not that difficult. I am often asked about the need for historical data in buildings. And as always, I answer "it depends." The data quality plays a major role in current AI/ML initiatives, and not all data is something you can work with straight off the bat. It's often vice versa that a lot of effort goes to cleaning historical data and where new targeted data can lead to faster times to value creation. And that's why having more logic on the edge (in the buildings), such as ML/AI running on a box, could offset that fact and at certain times even replace the need for historical data.

Let me give you a real example. We have a customer who owns a mall. They want to run this mall from a digital twin operating two hours into the future, which is an actual use case which we are in the process of developing for them.

Ideally, they want to optimize energy usage, improve tenant well-being, sell some services to their customers in the food court (restaurants) so that they can improve their customer experience. This, in turn, will lead to more people in the mall and gain additional insights on how to save energy, predict equipment failure and to get the whole 3/30/300 rule benefits that come with a smarter building.

One approach that is quite easily executed is to put one of our DINGO BMS Controllers/Microcomputers in the building. Connect HVAC-R, meeting booking system, outside temperature, CO2 sensors, occupancy, camera and we pool that data in the microcomputer in the building, connecting it easily via RESTful APIs, and BACnet Web Services. We start feeding the DINGO data, standardize it underneath a BACnet umbrella, and an algorithm created from a 3rd party small-footprint library will learn how the building behaves during any given day or week. It will create a DNA for the building from scratch. This algorithm can be distributed at sensor level as well as cloud level, making use of brain power where it needs to be.

The building (with some human help to get it going) will learn that on Thursdays, at lunch-time, there's a 30% guest increase to one of the restaurants in the food court because of the daily special "Pea-Soup and Pancakes." The camera, who's also got some logic through a 3rd party application, detects the long lines, as well as disgruntled faces, and combined with social media ratings, we can see the restaurants get 0,6 stars lower rating because of this fact exactly at this point in time.

So, how can AI/ML help? Well, it can then deduce that the next time, the building will mitigate the 30% increase in people by supplying more air before lunch, off-setting any negative impacts it might have. A third party app connected to the network translates machine-to-text-to voice, sends a WhatsApp message to the office manager, reminding the restaurant that they need to staff up and prepare, because "remember what happened the last time."

In summary, the building will learn when people are moving where, how they behave and become aware of what it needs to become aware of. Furthermore, if we have 200 buildings connected, these can also learn from each other, which is where the true value lies in that building to building communication will lead to exponential increase in value, insights, new business models and of course, ease of innovation.

And this is all possible today, quite easily as well if I say so. Getting things connected, forming a platform of data to draw conclusions from is a must for any AI/ML driven approach.

Sinclair: Okay, Nicolas, that sounded... very far off! But I like it! Having the building talk to you is definitely humanistic and inclusive. But isn't this a bit scary?

Waern: A bit scary? This is terrifying! But anything and everything is terrifying if you put it in the wrong hands. So yes, there are definitive challenges with data security, privacy, and technical challenges as well, and there are immediate concerns with hacking and not least the ethical perspective to think about.

But maybe the real question is if this approach is really needed? Maybe there's no room for traditional food courts anyway so this approach to innovation is just obsolete and we need to think about deletion, instead of evolution. Maybe the answer is just these self-checkout solutions instead?

That's the thing. Companies need to identify what is there right now, where do we want to go in terms of functionality, and possibly feelings as well? The existing status is to be annoyed, warm, irritated, because of the long lines and poor indoor climate. The solution is to get rid of those feelings with any means necessary. And that's what AI/ML and any parts of technological improvements are all about. To make things better, easier.

The whole ransomware movement for Personal Computers will slowly but surely find its way into buildings as well. And that will be very scary. Imagine if someone would hold your entire HR department (no offense to HR) hostage, locking doors, supplying too much air to the room, short circuit some stuff, and cause an explosion. Or just suck all the air from the room, or something else made possible having a "connected building." They will only release the hostages if headquarters wire $20M to some company on the other side of the World.

Without the correct skill-set in securing the IT and OT infrastructure, we will see this more and more.

Sinclair: That sounds more than scary, terrifying even! Is there anything AI/ML can do to help offset these risks? Any real value to the HVAC-R part perspective?

Waern: Going back to the HVAC/R parts of the equation I think that "the connected everything" has its pros and cons.

"Because Building Automation is simple," right? David Peters, General Manager at Elliot Controls Inc, started a very interesting discussion on Linkedin the other day, that has got a lot of attention worldwide. He posted this image below arguing that;

"All we have to do is control three variables (flow, temperature, and pressure) in two types of media (fluid or air) using four pieces of equipment (valves, pumps, fans, and dampers). The logic is very easy to arrange. The sequences of operation may not take long to write."

But it's the variables of all the dependencies and the physics around it which makes it extremely easy to wreck any setup. And also extremely difficult to get the full perspective.

Whereupon James Cheesewright, District Technical leader at Honeywell, made this interesting comment, highlighting on the importance of AI/ML from a BAS perspective.

"It also helps highlight why A.I. has such great potential in helping radically improve the way we commission and optimise the built environment."

Wherever there's complexity, there's room for technology to help make it easier.

Getting things connected also means that security must be a close first thing you think about; not a close second. Addressing these challenges beforehand how things should be connected is vital for everyone. But most of the time, there are already existing infrastructures in the building, and it's here you might run into challenges, where AI/ML can help. There's something called "Predictive - self-healing" which basically is what it sounds. If errors occur in the network, the network itself will try to fix these issues, as well as send alarms to the people who need to be notified. AI/ML algorithms constantly detect anomalies and networks can adapt to changes instantly modifying its response depending on what is happening. These robots or procedures can scan the network at all times and detect if something is wrong. And of course, we also see the emergence of hybrid clouds, private clouds, where servers are controlled by companies themselves, instead of having data sent to the other side of the world.

I haven't seen that many companies are offering in-depth security enterprise solutions for building automation and the OT-side of things (Operational Technology) yet. But this is definitely where companies like NanoHeal and Site1001 will have a huge impact in addressing these security concerns in a sophisticated way. I really want to find more companies like them.

And as discussed earlier, the building automation industry can only do so much, and it is here other companies with AI/ML powered solutions can come in and add value for system integrators, owners, as well as improving security for tenants and end-users of the buildings.

There was an article coming out just now that machine learning predictions are making all the wrong plays and this could lead to a negative value in the end. Because one of the most dangerous things when it comes to ML and AI, is the possibility to corrupt data at the source. AI and ML can't be super rigid. It's like you say to someone that they should walk 1000 steps in the x-direction, and only after the 1000 steps, think about where they are going. If they are just 1cm off to start with, they'll end up in a totally different place than you want, and definitely what they want. But if we have mechanisms for self-correction, improvement, or some kind of human control at set intervals, we build more robustness into models as well where we self adjust and validate after every 10 steps or 1 step for that matter. The amount of data can be bad, but equally great. It depends.

The dangers of getting everything connected could be mitigated through rigorous security, but maybe that the most important form of data is additional knowledge and extensive data sets. If you think that something is wrong in a building, but you are not sure, you (might) be much more comfortable seeing that all of your other 300 buildings have the same problem/or that they don't have this problem and that it might be an anomaly. However, this leads back to the question if there's an underlying problem with the model, or if it's with the data, which might lead to different models and approaches being applied as well because that is the value of Big data. That you have options, and multiple sources of information to choose from to decide what might be the best outcome.

Furthermore, regarding the tagging craze that is going on at the moment, AI and ML can also help to identify products and technologies by their unique DNA.

Sinclair: Now you lost me again, Nicolas. I know about tagging in the sense of the BACnet 223p standard, and that Project Haystack tagging, is doing wonders for this industry in terms of increased interoperability and faster time to value creation? Yes? So what do you actually mean?

Waern: I am not saying that tagging will be useless, obsolete and unnecessary in the future. In fact, it is absolutely vital today that a company has a clear understanding of the relationships between standards and data to enable a solid platform to stand on. It's important to get started, and for companies to realize that they are in control of the information, processes within their organization. And also, that they should be in control of the data, but allow others to make sense of the data and to create value from it.

But what I am saying is that there are more ways than one to increase time to faster value creation. For instance, let's say we have a portfolio of 1000 assets of commercial real estate. Hundreds of AHU's VAV Boxes, meters talking different standards, products from different vendors, and we want to connect all of these in an interoperable way as soon as humanly (?) possible. Even though we see a race to the IP level, raising digital maturity in buildings can be extremely painful. But it's getting a lot easier every day thanks to technological advances, open standards, service transparency and a more IT-driven approach to traditional BAS thinking.

Let's start with two buildings and get them 100% connected from an existing system- HVAC-R point of view. We've got Modbus meters; we've got controllers from different brands, we've got BACnet MS/TP, BACnet/Ethernet, we've got BACnet/IP, some LON, it's Siemens, Tridium, Schneider, Trend, Saia, etc. etc.

Let's also put in some IoT sensors from different manufacturers and standards into the mix, and then we'll say that we collect all of this in a data lake. No standardization in the building, no edge data strategy, basically no data strategy what so ever. Data lake = is a fancy term for a landfill of data. This is usually where the cleaning happens and without any meta tagging, of who's it from, how the data is structured, possibly also where in the buildings they are, etc. etc., this goes from data lake to landfill, to toxic waste dump pretty fast. Because API's might mean trouble if not done correctly.

But here is where AI/ML might come in handy and make things easier if done correctly. I wouldn't say a picnic, but it has the potential to revolutionize the speed of getting value from buildings on a large scale. This is a collaboration act if I ever saw one, and first, we get capable hands onboard from a Super MSI like Hepta to do a due diligence process on everything that exists and doesn't exist in the building. Once we have the information what is there, we can easily deploy sensors that talk to each other, will be absorbed by existing BMS systems, and that can scale up and down without ANY manual configuration. Boom, all the IoT gadgets will instantly become virtual BACnet devices, and off to the cloud, we go.