BrainBox AI brings machine learning to building controls

Having spent more than a decade working on smart buildings and energy efficiency for commercial property, BrainBox AI cofounder and chief technology officer, Jean-Simon Venne, sees AI-powered autonomy as the necessary next step in the battle against entropy — and against climate change.

The problem with engineering buildings any other way is that efficiency tends not to last, Venne says. “I’ve worked on energy efficiency in large buildings pretty much all over the planet in places like Dubai and Las Vegas. When you’re done with the project, and the building is pretty efficient, you’re happy with the results, and you go away. That’s when things slowly start to degrade.”

Improving the energy efficiency of buildings is a global concern, given that, by some estimates, commercial buildings account for about 40% of greenhouse gas emissions. But what happens over time, according to Venne, is that subsystems malfunction, or else they get turned off or misconfigured. Conditions change, and problems not accounted for in the initial design emerge.

What’s needed instead is something more like the engineering that goes into a race car, where the initial design is as perfect as the engineers know how to make it upfront, but every few laps during a race, they fine-tune it further for the specific conditions on the track that day, Venne said. His inspiration for a solution that is less labor-intensive than car racing also comes from the world of automobiles — specifically self-driving cars.

In addition to knowing the basic rules of the road, a self-driving car needs to be able to adapt to the unexpected, such as swerving to avoid hitting the squirrel crossing the road ahead, Venne said. “It occurred to me that if we’re doing this with cars, we should be able to do the same with the technology that drives the mechanical side of the building.”

BrainBox AI focuses primarily on controlling the heating, ventilation, and air conditioning (HVAC) systems within a building, which accounts for the majority of the energy consumption in most buildings, Venne said. A next-level goal is to get multiple neighboring buildings in a city working in tandem to produce better results, like helping utilities balance the consumption of electricity during periods of peak demand. A pilot project based on that concept won a Tech for Our Planet challenge at the recently concluded COP26 United Nations conference on controlling climate change.

Better efficiency for buildings and the planet

The Montréal, Canada-based company is also winning financial recognition, last month announcing a $24 million series A fundraising round led by ABB, the industrial technology conglomerate, with participation by Esplanade Ventures and Desjardins Capital.

Commercial buildings have been getting “smarter” for years to the extent they contain increasingly rich sensors, controls, and analytic dashboards that can be used to optimize air conditioning and environmental controls. However, it’s typically up to a building engineer sitting in a control system somewhere to make those adjustments, Venne said. Putting more of that control in the hands of a machine-learning algorithm makes it possible to take many more factors into account — for example, weather forecasts or the position and intensity of the sun — to achieve more precise control over HVAC systems, window shades, and other environmental controls, he said.

Where a traditional thermostat is by nature reactive — waiting until the temperature rises above or below certain boundaries before making adjustments — a smart building control system can be more efficient by anticipating changes and compensating for them, Venne said. It can also make use of shades or other passive climate controls, rather than just cranking up the heat or air conditioning.

While the laws of thermodynamics haven’t changed, the possibilities for manipulating them have. “You’d probably need 45 engineers working 25 hours a day, making all these calculations, to come close to what AI is giving you for a few pennies on the dollar,” Venne said. “You should not attempt to do this manually for the same reason you should not play chess with a computer, because it’s just too good.”

Building for the future with machine learning

Since about 2015, the computing power and the programming frameworks required for machine learning have become much more accessible, and BrainBox has been able to capitalize on building blocks readily available within Amazon Web Services, Venne said. The bigger challenge has been getting the machine learning specialists and the building engineers to speak each other’s languages. For example, computer scientists need to understand practicalities like the need for and function of water pumps in a tall building. In that sense, it took “a good two years” for the product team to come together, Venne said.

Once created, a building control model can continually improve using the same deep learning technologies employed by autonomous car engineers. The idea of networking together control systems from multiple buildings is still just a concept for now — albeit one that intrigues Toronto Hydro, which collaborated with BrainBox on its COP26 demo. “We’re prototyping now, but it’s not ready to go to market this year,” Venne explained, noting that the concept would be more powerful if half the buildings in downtown Toronto incorporated BrainBox technology, rather than the handful that does so today.

Beyond controlling heating and cooling within a building, BrainBox could also control the air quality within the building — for example, by adjusting ventilation for smog conditions or pollen levels. “The amount of C02 is often not monitored, or even considered,” Venne said. “So those would seem to be the next logical steps, focused on the wellbeing of the people in the building.”