How I Saved 4.5 Million Dollars with AI
What if we said that global emissions could be reduced to a level that sounds almost like science fiction simply by installing a few new components? Troy Aaron Harvey, an American startup founder and engineer, spoke about this, among other things, in his presentation at Proptech Hungary.
“In North America, the top three percent of commercial buildings are automated, while the remaining 97 percent are full of basic control systems. And this only gets worse when we move over to industrial buildings,” said Troy Aaron Harvey, the startup founder from Utah, as he opened his presentation on building automation.
After a 30-year career in building technology, the engineer, who graduated from the renowned private university Brigham Young University, decided with several partners to help the industry implement the latest automation and simulation technologies, primarily with an energy focus. PassiveLogic was founded in 2016 with this goal in mind, which means the company is celebrating its anniversary this year.
“A few people from the tech industry and the world of automation came together to solve the problem represented by the automation of traditional buildings,” Troy said in his presentation. Accordingly, his talk was highly engineering-focused, full of numbers, data, detailed diagrams and process metrics that approach building services engineering in a way that is completely different from the usual methods.
“Buildings account for 45 percent of total energy use, according to the figures of the US Department of Energy,” Troy explained. “If we say that 40 percent of that can be saved simply by using better control methods, then it is no longer like replacing the internal combustion engine with an electric one, but as if the entire emissions of transportation were eliminated from the planet in one fell swoop.”
Why is this the case? According to Troy, building automation solutions are often stuck in the past. Even when we talk about the latest available technologies, we are actually referring to methods that are 20–25 years old. Meanwhile, the pressure on the industry is increasing due to sustainability. As an example, Troy referred to new EU regulation, which treats the automation of new industrial buildings as a fundamental requirement and will even extend this to smaller building types in 2029. Decision-makers have therefore already committed themselves; now it is the industry’s turn.
Of course, industry players have also committed themselves — to AI. And this is connected to energy solutions, because it makes it possible to simulate building operation more deeply than ever before, even at the level of physics. It is clear to everyone why we cannot simply snap our fingers and replace all existing buildings with modern ones.
Well, that does not mean it is impossible to assess them, simulate them, and increase their efficiency on that basis. On the contrary. According to Troy, by equipping buildings with sensors, data can be extracted and then integrated into digital twin models. The entire system can be controlled by a language model, translating the client’s everyday language into the IT version of engineering terminology.
“There are three concepts in AI: abduction, deduction and induction. The question of abduction is whether I can take a model from the world I see and form a hypothesis. Can I infer from the hypothesis what will happen, meaning will there be deduction? Do my actions based on my predictions induce all the elements I cannot see?” Troy outlined the logic of AI-based simulation.
He then translated this somewhat abstract idea into the language of engineers who create buildings: if we imagine what the perfect, optimal operation of the building would look like, then we can reverse-engineer all the steps that make up this model. Instead of programming, we give people tools, and with digital twin models we can demonstrate all concepts, even if putting them into operation would be difficult — all of this supplemented by physics-based modeling.
The latter is very important because it can also monitor external factors and track energy-related changes in real time. If we know how quickly the water heats up, when and how much a component consumes, or what the pressure is in the pipes, then we can optimize the operation of an existing building without having to dismantle anything. Of course, to do this, we need to know the most important characteristics of the building.
“Where is your building? That is the question of the environment. Where are your sensors and what kind of IoT device ecosystem are you using? That is your input model. And what is your system, meaning what will the input model control? These make up the quantum world model.” The model of a fully autonomous building can also move to the cloud, becoming accessible and changeable anywhere and at any time, and capable of adapting in real time to all changes in the building. And if we want to know how a particular component is performing, it is enough to send an AI agent after it and wait for the results. Before the era of smart building technology, this would have involved human labor, a lot of time and possibly downtime.
The model outlined by Troy is forward-looking not only because it expands the possibilities of using AI and digital twin models. It is also forward-looking because it identifies the current shortcomings in the way we think about building energy performance. In many cases, the goal is cost reduction, and sometimes compliance with ESG targets. These short-term decisions may look good in Excel, but over time they only contribute to the snowballing difficulties caused by obsolescence.
It is important to point out, of course, that these methods are not cheap either. Troy also mentioned that the hardware ecosystem they use requires Nvidia GPUs, whose prices have recently been rising even faster than the rate at which logistics centers are multiplying across the country. So the high cost of the necessary components is still an obstacle to the widespread adoption of fully intelligent, AI-based building services engineering and automation. But just as digital twins were once considered something exceptional, today’s costly solutions may also become more accessible very quickly.