The control systems we use in commercial buildings today have evolved over the last 30 years to provide systems that are cost effective, and provide good performance. Beginning in the 1980s, building control technology transitioned from pneumatics to direct digital control. The transition was motivated by the low cost of digital control technology, better comfort control, ease of programming, increased accuracy and reliability, reduced energy costs, reduced maintenance requirements, and actionable building data. This has resulted in innovation, including a major shift to the use of open standard protocols, such as BACnet, web pages for user interface, and more network-based controllers. Today, we have a well-understood controls architecture that is supported by a small group of established, high-quality suppliers. But, at the same time in which we have seen steady progress in building controls, the world of computing and information technology has made quantum leaps. Think of some of the tools we rely on every day, such as Google Search (1998), smartphones (2007), smart thermostats and the Internet of Things (IoT) (2011), and digital assistants (2014). Today, this innovation continues at record speed with major innovations occurring in areas including artificial intelligence, quantum computing, and machine learning. As these new technologies continue to evolve and become easier and more economical to implement, it’s becoming clear that it’s time for the industry to move to the “next generation” of building controls. This article will cover the basics of why this change is needed and examine some of the key technologies anticipated to appear in the next few years.

Control Systems Today

Control or building automation systems (BAS) in common use today consist of a highly distributed architecture of small, economical controllers that operate equipment in a standalone manner or can be connected into an integrated system. Most control architectures start with dedicated controllers that are either included with a piece of equipment, such as a chiller or rooftop unit, or provided by the controls contractor and are field-installed on devices such as VAV terminals. These are often called “application specific” or “advanced application specific” controllers because they are intended primarily to provide a set of specific control functions. For example, a zone controller can be used to control equipment such as a fan coil or VAV terminal and will have a series of analog and digital inputs that connect to sensors for temperature, CO2, and airflow. It also will have outputs that can be used to control stages of heat, dampers, and valves. The controller operates using an onboard program that follows rule-based logic. Outputs are typically controlled based on a set point and an input variable using proportional, integral, and derivative (PID) logic. These small controllers are connected over either a serial (BACnet MSTP) or Ethernet connection. When connected to the rest of the control system, they are able to share data. Examples of system data may include schedules, optimized set points, occupancy information, etc. Larger controllers with more processing and memory are used to manage communications and run programs for system coordination, such as optimized sequences, schedules, alarm management, and trending. These controllers also provide a user interface by serving up web pages and may also have onboard inputs and outputs to control larger equipment, such as air handlers or central plants. Larger control systems may also contain dedicated servers that are used to store data and serve up user interface pages (See Figure 1).

There are several benefits to the current architecture. These include the fact that it’s relatively low cost (averaging $1-$2 per square foot), highly distributed so that it can work even in the event of a communication failure, and well understood by the industry. 

If it Works, Why Change?

There are many drawbacks to the current solution, including the fact that much of the communication is very slow; the controllers are very limited in their ability to process and store data; and there is significant effort needed to properly set up, check out, and program the controllers. For example, setting up a control loop requires properly selecting the type of input (voltage, current, thermistor, etc.), calibrating the sensor, selecting the type of output (voltage or current), and then configuring and calibrating the PID loop. All of this takes time and effort and with project schedules and budget challenges, it may not get done adequately. New technology, much of which has already been adopted by other industries, could help preserve the benefits found in today’s systems and eliminate many of the issues that are keeping systems from operating optimally. 

Drivers for New Controls:

• Improved Building Performance: Controls have been shown to be able to greatly improve the safety, comfort, and efficiency of commercial buildings, but they often fall short due to problems with proper design, configuration, programming, check out, and operations. The use of new technology that moves away from many of the currently complicated and hard-to-replicate processes, such as the use of analog inputs and outputs, manually configured PID loops, and rule-based programming, has the potential to provide improved performance. 
   
• Cost Reduction: The costs for controls today are relatively low; however, they often are not providing the value an owner is paying for. By moving to new technology, there is the potential to have control systems that have better performance for even lower costs. 
   
• Grid Services: There is a strong potential for buildings to utilize their demand flexibility to provide a series of services to help the electric grid. Doing this would allow for better integration of renewable energy and would be valuable in providing improved grid reliability and resilience. While these services can be used with existing control systems, they will be much more effective in new systems that are using advanced technology that has the ability to both control and also to adjust performance to meet grid needs, such as model predictive control.  
• Support of IT Standards and IOT: The avalanche of products that has started based on internet standards (often referenced as the IOT) shows the potential for innovation based on low-cost, readily supported technology. Years ago, we thought we would never see an internet-enabled light switch — much less internet-enabled light bulbs that are now readily available (See Figure 2). 
   
• Improved Security: Most building control systems are not well protected from a potential cyberattack. Fortunately, building controls are not the most intriguing target for attacks, but this still remains a risk. New control solutions can be designed to utilize security standards as a native function. 

New Technologies and How They Apply to Buildings

Over the past decade, we have observed the first indicators of another technology transformation. This is driven by the ever-increasing availability of new data sources from IOT-enabled devices; interoperability standards adopted and incorporated in new products; advanced data collection and management tools; advances in data-driven modeling; the broad availability of analytics; building simulation tools that take advantage of novel computational science methods; the low cost of computing technology, such as cloud computing; and specialized hardware, such as graphical processing units and field programmable gate arrays. In this context, the complexity of building systems faces order of magnitude increases, and this complexity must be adequately supported by new control technology that can process exponentially increasing and heterogenous data volumes and optimize the building operation for dynamically changing cost functions, such as energy cost, energy efficiency, grid service, renewable energy utilization, etc. This opportunity offered by these new technologies and advanced building applications creates a need for new (next-generation) building control technology that can address these challenges. 

Traditional control methods and control sequences that have been implemented broadly in the building stock across the U.S. face scalability challenges when integrating new data from other building systems and IOT systems, as the control sequences would have to provide a similar level of complexity. Such control sequences are becoming increasingly expensive to design, implement, verify, and maintain. While PID control will remain the cornerstone of servo actuation, the advanced performance-oriented needs in building systems can only be addressed by advanced computational optimization methods and off-the-shelf computing advanced. 

In the past two decades, motivated by advances in computer engineering and computational methods, several key technologies have been developed to advance the approach to engineer and operate new high-performance systems at lower costs. Two such technologies are digital twins and model-based predictive control. 

Digital twins are computer programs that capture the physical characteristics and operational behavior of real-world engineered systems. Such programs can be used to design systems and analyze their properties before real-world implementation, allowing for early detection and correction of design and operational flaws. This accelerates the design process for new products, enabling the industry to refresh product lines at a faster pace and with lower cost. Digital twins can also be informed by real-life measurements and streaming data from sensors. The ability to simulate the expected (nominal) behavior of a system under current conditions is used to analyze the health of a system, inform predictive maintenance decisions, and conduct what-if scenarios to identify failure points. Furthermore, digital twins are used to inform control designs and evaluate and verify their performance. In the building energy domain, open-source digital twin technology examples include solutions developed by the U.S. Department of Energy (DOE), including Energy Plus, Spawn-of-EnergyPlus (Spawn), and Open Studio. Commercial providers include Microsoft (Azure) and IES (ICL platform).

Model based predictive controllers (MPC) make use of models to evaluate the expected performance of a system for multiple set point combinations over a look-ahead time window and select the combination of set points that results in the best expected control performance. Essentially, MPC controllers repeatedly answer the question: What is the combination of set points that will maximize the performance of the building system in the near future while satisfying all the operational constraints? MPC can consider constraints, such as upper and lower bounds on set points and available actuators, system capacity, and can optimize for multiple performance objectives, such as minimizing energy use while maximizing comfort. Model-based controllers can also optimize control decisions when faults are present in the system and their effects are captured by these models. Some early experiments at building sites have shown that MPC can yield significant energy savings (up to 17%-65%) compared to the performance of installed control sequences. MPC algorithms can make good decisions even when the equipment models and load forecasts are inaccurate, and the predicted variables have uncertainty. For example, MPC could be used to select the optimal air-handler static pressure set point based on a model of the air-handling system, current box positions, and other variables. The data would go into the model, an algorithm would iterate various solutions, and the one that provided the best comfort and lowest energy use would be selected. This approach operates more effectively than current methods, such as “trim and response,” which select a value, implement it, and watch to see how the system reacts. MPC is especially effective in simulating future conditions, such as would be introduced, as we move to make buildings more grid responsive. 

Fueled by the big-data revolution, we have seen fast progress in the use of technology that allows computers and controls to make data-informed decisions. As the technology has evolved, it has gone by many different names. The overall area is often called “Artificial Intelligence” or “Machine Learning.” The technology that enabled most recent performance leaps in the artificial intelligence and machine learning areas is called “Deep Learning” (See Figure 3). Deep learning is a key technology that can augment capabilities of MPC methods, take advantage of digital twins, and improve performance and deployability of intelligent building control. Deep learning systems applied to building controls are already under development — the most prominent example being Google Data Center optimization system. Furthermore, a system that is collecting and analyzing data about building operations can use deep learning to diagnose faults and recommend (or even in many cases implement) solutions. These tools will be effective both in increasing building efficiency and assisting operators in assuring building comfort, continuous operation, and occupant safety.

Finally, technology advances are required to provide for improved cybersecurity. As we make buildings more connected and integrated, they become potential targets for cyberattacks. New technology for cyber protection that is being developed for applications, such as protection of financial information, defense, and the electrical grid, can also be readily applied to building systems. For example, some of these new technologies not only provide protection by restricting access to equipment but also evaluate network traffic and look for anomalies that may be indications of an impending attack. 

Costs, Benefits, and Disruption

With the advent of all of this new technology for controls, why has there been such limited adoption? The reason for this is largely due to the high cost of change. Developing controllers and their supporting software is a large investment, and the volumes of devices used for commercial building control is relatively small in comparison to those for consumer electronics. In addition, making changes requires retraining of technicians and installers, which is also a large and risky investment. 

At the end of the day, suppliers may be intrigued with the new technology but the return on investment may not be clear. So, what will drive these changes, and where should you expect to see them coming first? Here are a few potential scenarios:

1. Equipment Control: Many pieces of mechanical equipment come from the factory with specialized controls. Examples of this are chillers, rooftop units, boilers, and systems, such as mini splits and VRF units. Manufacturers use these controls for safe and efficient operation of the equipment and are able to invest in technology to differentiate their products and improve performance. Expect to see the first implementations of new technology embedded in the control of equipment and packaged systems. 
    
2. Disruptive Entrant: It’s quite possible we could see one or more companies introduce products based on IOT technologies for the commercial building space. This could come from a number of different directions, including from existing technology companies or companies already providing consumer IOT products, but the most likely scenario is that they would come from new startups looking to differentiate their offerings. 
    
3. Innovative Controls Company: Over the last decade, the controls industry has largely consolidated down to a small group of large global suppliers. These companies have large resources to develop and support new products. When these companies decide to introduce products based on new technologies or architecture, it will cause a disruption and be up to owners and designers how quickly these new technologies are adopted.  

https://deepmind.com/blog/article/deepmind-ai-reduces-google-data-centre-cooling-bill-40