Artificial intelligence and its machine learning subset are all over the media these days. “Doom and gloom” commentary about deepfakes, misinformation, bias, privacy, and ethical risks to society abound—yet AI-related R&D is skyrocketing among tech giants Google, Microsoft, and IBM.
But tech companies aren’t the only ones investing. Siemens, Honeywell, and Johnson Controls are integrating AI algorithms into their cloud-based smart building management system (BMS) platforms to make actionable and predictive decisions that reduce operational costs and achieve energy and occupant-experience goals based on historical and real-time data. These algorithms provide step-by-step instructions for calculating optimal settings and for ML systems to identify patterns in a building’s life cycle.
Some BMS platform providers have internal AI teams developing algorithms, while others are partnering with AI consulting and development service companies like Intel, IBM, Accenture, and Infosys. Honeywell’s connected security platform, for example, incorporates Intel’s Vision products with advanced analytics, deep learning, and facial recognition capabilities.
AI and smart environments
Deploying AI for smart building automation requires large data sets, such as a project’s energy load, system performance, occupancy and space utilization, user satisfaction, and environmental stats, such as temperature, humidity, air quality, lighting, and noise level. AI might consider external factors such as weather patterns, climate trends, peak energy periods, and security threats. The amount of data to be verified, filtered, and analyzed are too complex for any human to effectively interpret and act upon. That’s where AI comes in.
To reduce energy usage, for example, AI would analyze data such as occupancy, space usage patterns, and peak demand before initiating adjustments to heat and air conditioning. ML algorithms would sift within historical data to detect and diagnose anomalies within HVAC systems and prompt owners to schedule maintenance and adjust valves, fans, filters, or controllers as needed.
Gary Chance, vice president of marketing and partnerships at New York–based Prescriptive Data, says its AI/ML cloud-based platform Nantum OS has already generated significant benefits for users. “We have saved thousands of dollars by catching water leaks before they happen, automating system startups based on when people arrive at the office, and automating demand management using AI,” he says. The platform has even reduced New York’s steam expenditures by automating the preheating and filling of steam tanks before peak hours.
Last year, the company discovered that 44% of the data reporting requirements for WELL certification can be automated using AI. Not surprisingly, automated reporting for WELL, Fitwel, LEED, and other building certification programs is now one of its service offerings.
Cloud-based platform BrainBox AI integrates weather forecasts in its energy optimization algorithms. “The system takes in-building sensor data and weather data from the nearest weather station to train systems based on what’s happening inside and outside in real time,” says chief product officer Omar Tabba. “Our AI system can predict the future state with more than 95% accuracy based on a certain forecast. If the data shows that it is going to be 78 degrees F outside in two hours, but we want the indoor environment to stay at 72 degrees F, the system knows [when] to turn on the fan and at what level to avoid extra energy consumption and expenditure.”
In work environments, AI can condition areas based on real-time data in addition to notifying occupants of busy cafeteria times and forecasting space needs. “Most smart buildings operate on set points with schedules and overrides based on planned events and occupancy, as well as temperature and humidity levels,” says Dani Stern, senior director of product management for commercial smart buildings at Honeywell. “With hybrid working, occupancy becomes very difficult to predict and schedule. AI can analyze historical and actual occupancy information from multiple sensors to help optimize the space. We even have a partnership with Cisco using its Wi-Fi system to triangulate the flow of people based on smartphone connectivity.”
Still, assigning ROI to space optimization can be hard, and without that guaranteed value, AI adoption in some use cases might be tempered. “AI is starting to determine worker satisfaction based on indicators such as air, lighting, and space quality in relation to the number of people coming back to the office,” explains Lucy Casacia, vice president of smart solutions for global engineering firm WSP, “but we will see more AI investment on the user experience forefront in entertainment, hospitality, and public spaces.”
AI is gaining momentum in security systems. The technology is initiating lockdowns, emergency alerts, and first responder notifications based on proactive intrusion and weapon and gunshot detection, analysis of surveillance footage, and incident reporting. Casacia is also seeing “systems that use heat mapping to monitor patterns and movement to detect theft in retail environments, such as someone moving faster than normal.” In transportation, AI and ML can “determine if travelers are disoriented based on measuring sentiment and happiness via facial recognition.”
Old Oak Common, a super-hub interchange constructed as part of Britain’s new HS2 high-speed rail line in London, recently leveraged AI to optimize station signage during its planning stage. The project team invited dozens of members of the public and designers to navigate through a virtual station and used eye-tracking ML, AI, and advanced signal processing from Swedish company Tobii to analyze traveler emotions, interests, and distractions. Here, AI has the potential to improve safety and security by reporting overt emotions, such as panic or irritation.
Adoption reservations
Despite the growing acceptance of AI, many consumers remain wary due to misgivings about data privacy, bias, and ethics. These concerns have even led to the downfall of AI-based smart city projects, such as Sidewalk Toronto, also known as Quayside, which Alphabet subsidiary Sidewalk Labs canceled in 2020. (Earlier this year, Waterfront Toronto announced new redevelopment efforts for Quayside that conspicuously shifted its emphasis on smart city features and data collection to affordable housing, sustainability, and diversity, equity, and inclusion.)
“AI technology is moving fast, and bad press has made many people nervous,” Casacia says. “As we saw with Toronto, these concerns have more weight in residential facilities.” She believes privacy should be less a concern in commercial buildings and public spaces, where personal data such as ID cards are regularly deployed. “As long as we ensure that data governance and standards are in place and that stakeholders and occupants are aware of the data being analyzed, AI will continue to evolve and grow in smart buildings.”
Cybersecurity is another concern restricting the pace of adoption in building operations, Tabba adds. “It’s imperative to have tight controls on access to data, ample redundancy, disaster recovery, and ongoing third-party testing to identify vulnerabilities—all of which are best practices in a modern cybersecurity strategy for AI.” BrainBox AI, for example, encrypts data in transit and at rest in compliance to infosec standards like SOC 2. Staff, located in Ireland, Australia, and Montreal to cover different time zones, monitor customer buildings 24/7 and can disengage AI at any time due to planned maintenance, threats, or other events.
Many technology experts leveraging AI in the smart building space believe that trust is the key to increasing adoption. “We see early adopters embarking on the journey,” Stern says, “but there are still fears and doubt—especially around cloud-based data and automation. As trust grows, the technology will evolve. In the meantime, we can use AI to make recommendations and allow the customer to initiate action. They can also take a private-cloud approach or host systems on site—whatever they’re comfortable with.”
AI and job replacement in operations
Though AI will overwrite some job aspects, parameter definition and algorithm programming will always require human input, WSP’s Casacia says: “The technology needs to be governed and managed with human checks and accountability. A plane can function on autopilot, but we still have human pilots at the helm.”
In other words, building operators shouldn’t grow complacent.
Even as Honeywell continues to develop AI in its cloud-based Forge platform and participate in open discussions with other tech providers, Stern believes people will remain integral. “[T]here is a lot of room for innovation, and we’re working with digital twin and smart building technology companies to make sure our systems are open and accessible to enable integration,” he says. “Still, there is no better sensor than a human. That’s why we’re also building systems that allow occupants to provide feedback about cleanliness, comfort, and other factors via rating systems that become part of the data sets that AI analyzes.”
With data-reliant building performance reporting mandates on the horizon, Chance sees opportunities for both AI and people to thrive in a more energy-efficient and productive environment. “Regulations like Local Law 97 in New York, where most buildings over 25,000 square feet [have] to meet new energy efficiency and carbon emissions [mandates], will put both building data and tenant data on the roadmap,” he says. “Real estate is typically two years behind in technology adoption, so we will likely see an uptick in AI for building operations over the next few years. But at the end of the day, we still need humans.”
Ready or not, here AI comes
All AI and ML algorithms require large data sets to learn and discover patterns or problems; the deeper the algorithms, the more data is needed. “Having ample quality data is the biggest hurdle right now,” Casacia says. “It starts with determining stakeholder requirements and the data type needed to meet them, followed by identifying what data needs protection, how much to store, and for how long. Then it’s about ensuring data quality with the right inputs and outputs.”
Initial data mapping, which matches data fields from distinct sources, must determine validation rules and data flow for continuous mapping to occur. BrainBox AI’s website says its AI engine can take two to four months to learn and map building data, depending on the facility’s size.
“Mapping building data is a challenge for an existing building with no historical data and thousands of inputs from different systems, some old-school OT-based and others IT-based,” Honeywell’s Stern says. “To leverage the technology, we must get all that data into one stream.”
Smart building owners and operators can work with solution providers to assess existing systems and ensure the equipment and devices are in place based on the data and actions required to achieve outcomes. “[We] then review existing systems to recommend a scope of work with the hardware, software, sensors, actuators, and other devices needed to acquire, analyze, and act upon the data,” Chance says. While cloud-based BMS platforms integrate AI algorithms, he says, solutions like Prescriptive Data’s Nantum OS act as an optimization layer that can integrate with existing systems via the cloud or on-site edge gateways that can read and write various protocols, like BACnet, to get data into the correct format.
BrainBox AI is also an AI-based optimization layer. “We can connect directly to existing cloud-based BMS platforms or local systems via software drivers or by deploying gateways that allow us to bring data into our cloud-based software,” Tabba says. But because many existing buildings cannot connect to its systems, BrainBox AI had to innovate. “We work with customers to replace existing thermostats with Wi-Fi-enabled thermostats … or we may need to add supply air temperature and CO2 sensors and ensure the ability to read and write to outputs that control fans, dampers, economizers, and other equipment.”
As with any emerging technology, smart building owners and operators should proceed cautiously to achieve value and meet their objectives. “Sometimes smart building owners want to differentiate themselves and launch into technologies that don’t meet expectations,” WSP’s Casacia says. “AI and ML need to be part of the smart building operations strategy, but we must be careful and understand what provides value. Sometimes keeping it simple is beautiful.”
This article has been updated since first publication to include a more recent link to the progress of Toronto's Quayside project.
For more news, projects, and profiles in the smart buildings ecosystem, subscribe to the SBT newsletter and follow us on LinkedIn, Twitter, and Facebook.
