Controlling energy use and costs in commercial buildings
Johnson Controls came to exist when founder Warren Johnson invented the thermostat in 1885.
With that legacy it’s important to keep ahead of new developments in the field. Today the Milwaukee-based controls company is tackling one major shift in heating and cooling in collaboration with UW-Madison chemical engineers. As the company seeks to develop better HVAC control systems for its clients in large commercial buildings, it’s collaborating with Paul A. Elfers Professor and W. Harmon Ray Professor of Chemical and Biological Engineering Jim Rawlings and his research group.
The collaboration is essentially about giving the HVAC world the benefit of tools that chemical engineers have understood for decades. Rawlings, Chemical and Biological Engineering Professor Christos Maravelias, and PhD students Michael Risbeck and Nishith Patel are developing algorithms that will enable building managers to harness a broad range of data to run their HVAC systems more efficiently. Risbeck has also spent time interning with at Johnson Controls, working with staff in the company’s main office in Milwaukee. The UW-Madison researchers use a process control method called model predictive control, which involves forecasting the future behavior of a system and taking those forecasts into account to make real-time adjustments.
“In the chemical industry, optimization has been a much more important part in the operation of chemical plants because that's your profitability, that's the goal of the business,” Rawlings says. “So what the HVAC industry is able to do now is take advantage of all that development over the last 20 or 30 years in the chemical industry and bring it right over to buildings."
The reason this didn’t happen earlier, Rawlings says, is that building managers haven’t always had much data to draw on in making decisions about heating and cooling. But in recent years, it’s been easier to acquire good data for which chemical engineering-bred control processes are ideally suited: energy pricing, weather forecasts, the hours when workers tend to enter and leave a building, the impacts of other potential heat sources in a building. Software that harnesses that data can in turn help an operator get the most out of a given building’s system of boilers and chillers. And the impact goes far beyond a building owner’s power bills.
"Twenty percent of total U.S. energy consumption is commercial buildings,” Rawlings says. “I had no idea it would be that high. And half of that is education and office space.”
In his time working on implementing the algorithms as a Johnson Controls intern, Risbeck learned that the greatest challenge is creating something that suits the incredibly varied configurations of heating and cooling equipment in different commercial buildings. There’s no standard way to put together an HVAC system for a large building, and even within one building, that system might comprise many disparate pieces of equipment and several different software systems controlling them.
To analyze and control that system as a whole—or even optimize heating and cooling across an entire campus of buildings—the control framework has to be agnostic to the size and makeup of the system. Clients don’t hire Johnson Controls to tell them to buy more efficient heating and cooling equipment—but rather, to make the best of what they already have.
“You're going to be constrained by the choices people have made in the past,” Risbeck says. "It's a difficult optimization because you have to make not only continuous decisions but also discrete on-off decisions for these big pieces of equipment.”
The researchers and Johnson Controls see the current project—which is being tested at sites including the Johnson Controls building in Milwaukee and buildings on the Stanford University campus—as a foundation for advancing HVAC control systems in general. "It's an infusion of high-tech into a low-tech space,” says Dr. Robert Turney, a Johnson Controls engineer and technical lead on the project. "To date we don't really have optimization and control at the enterprise layer. They do control at the layer of the building, but not above that."
Turney says these efforts are of course driven by cost savings and by the new opportunities that arise as companies harness more data through the cloud. But he and other Johnson staff also find themselves speaking more often with newly appointed sustainability directors at client institutions, especially universities and airports, with lofty goals for improving energy efficiency. "Typically there's a customer-driven aspect where the director of sustainability is asking for better control,” Turney says. “There's really a vacuum in the solution space now.”
Rawlings points out that the opportunities and uses for model predictive control systems in commercial buildings are only expanding as engineers access better and better data. “You know how the building is running and you can look at how it ran a year ago, two years ago, 10 years ago, and you can detect things like equipment that is degrading, and you can find out what the use patterns are,” he says. “You can find out that over time, energy use in this building is rising or dropping, and you can start to ask what-if questions.”
As the optimization algorithms pass through different revisions and iterations, control researchers like Rawlings and his group gain an increasingly nuanced understanding of the real-world contexts in which their ideas are used. "From an academic standpoint, the examples you find in literature are all small, simplified systems,” Risbeck says. “When I was at Johnson Controls and actually trying to implement a version of this for a multimillion-dollar plant, there's just a lot more difficulty on the back end.”
While Johnson Controls provides that crucial dose of reality, Turney credits UW-Madison researchers with helping the controls industry seize on new possibilities.
“I count on the university to be the pioneer in solving HVAC problems in new and innovative ways,” Turney says.