| Distributed Analysis and Optimization in Building AutomationComplete system optimization will require the full complexity of existing building systems and will be best achieved by the BAS manufacturers themselves. We look forward to these exciting changes on the horizon. |
 Mike Donlon, Director of Research and Development, Computrols |
Recent advances in Machine Learning and Artificial Intelligence are steadily making their way across diverse industries and products. From speech recognition to financial analysis and weather prediction, these new computer technologies leverage big data to recognize patterns, build mathematical models and make predictions. Building automation systems are not exempt from this technological revolution. Large buildings and automation systems, rich with big data, controllers, networks and computers, are poised to take full advantage of these new technologies.Roadmap to Complete Building Optimization
Today, many companies are beginning to use these new technologies for analysis, but the analysis is only half of the picture. Our ultimate goal in building automation systems (BAS) is Complete Building Optimization. That is, we want to take all of the equipment and information that exists today, analyze it, then operate the building in a way that minimizes energy consumption and maximizes tenant comfort and equipment lifespans. This is not a matter of if it will happen, but when. If we look at the evolution of BAS over the past few decades, we see a continuous trend of more automation, more efficient operation and less labour. The simple illustration below helps us visualize this trend:
Although this illustration shows complete computer control as the future, this isn’t to suggest that the building staff won’t be involved. Like most technological advancements, job descriptions will shift as automation progresses. Management, planning, service, installation, etc. will still be performed by the building staff. What will be completely computer-driven will be the tedious calculations and analysis that are required in building optimization. Still, computer optimization will be overseen by people, and the end result will be a valuable tool for building managers and engineers.
However, the road to true building optimization is not so clear. As we know from experience, technological advancement in building automation can be slow and cumbersome. Standards and protocols can be debated, delayed or rejected. This article intends to highlight some of the potential challenges and pitfalls along the road to complete building optimization, as well as suggest some approaches and solutions that we hope will be useful.
Rise of the Energy Dashboards
HVAC, Lighting, Fire and other building systems continue to grow in point count and complexity as the cost of automation drops and technology advances. Even Elevator Systems are becoming more complex with the advent of technologies like Destination Dispatch. Unfortunately, all of this increased complexity leaves building owners, managers and engineers overwhelmed with information.
BAS vendors have generally failed to manage this complexity in a way that makes it easy for building engineers and managers. Systems are riddled with different software programs and tools, cryptic user interfaces, and workarounds. Building staff has to rely heavily on service while still not completely understanding their systems. When combined with the pressure for ever-increasing energy efficiency, measurement and reporting, many building managers are in a difficult situation. It’s clear that classic BAS systems are doing a poor job assisting building owners and managers with a simplified, big picture of their building, especially concerning overall energy usage and efficiency.
Seeing this opportunity, dozens of small Internet and mobile-savvy startups have sprung up to fill this void. Collecting data from disparate systems and integrating data into clean, modern user interfaces, this new wave of integration products assist building owners and managers in their quest for simplicity. Often referred to as Smart Building Platforms, these systems regularly use web and mobile friendly dashboards and graphs.
Computrols Real-Time Energy Dashboard
Ideally, the addition of these new systems is a valuable tool for building engineers and managers. In other cases, new systems further complicate existing problems. How does this new system integrate to existing systems? Who is responsible if the displayed data is incorrect? How much control is given to these systems and taken away from other systems? These are all issues that building managers need to consider when shopping for one of these systems.
Usable Analysis in Large Buildings is Difficult
Although there is some limited value in simply visualizing building data, the true intended use of Smart Building Platforms is to perform analysis and highlight problems in the building. The extent to which valuable analysis is presented varies from product to product. Certainly, a cursory scan of building data can reveal common problems, like running equipment or lights outside of scheduled hours. Realistically, buildings should not add to the complexity and cost of a separate system to identify simple problems.
A deeper analysis of building data might reveal more elusive problems. In HVAC systems, a building may be using suboptimal setpoints or suffer from a restricted flow of air or water. The problem with analysis performed outside of any complex system with limited detail is that perceived problems may be nonissues. Engineers and technicians with a more intimate knowledge of the system may already know about highlighted issues, and living with an issue may have already been chosen as the lesser of two evils.
Take for example supply air setback. Recommending changes in setpoints are a staple for analytics programs. Initial analysis often indicates that raising supply air and other setpoints would reduce plant energy while still meeting all temperature requirements. Setpoints are low hanging fruit for HVAC analytics. However, did the analysis consider how raising the setpoint would affect humidity?
Most building engineers know that humidity problems can be challenging. They not only cause tenant discomfort but potential property damage. In medical facilities, along with temperature and pressure, compliance requires that humidity is continually measured and reported on. Failure to comply can be very costly—often much more costly than potential energy savings. Recommendations—even those as simple as changes to HVAC setpoints—require a deep understanding of the mechanical system and operating requirements.
As shown in our illustration above, the current state-of-the-art for building automation is for automated analytics to make recommendations only. Recommendations should be followed by careful consideration by experts before action is taken. But even then, actions taken will often require service from individual system vendors themselves as control logic is reworked. Basically, real analysis that produces actionable intelligence is usually more difficult than it is described.
And Analysis is Only Half the Picture…
As stated above, many industries and systems are benefiting from the new technological advances in machine learning and AI. However, building automation is significantly different from most of these systems and therefore must be handled differently. As an example, let’s consider weather prediction.
Like in buildings, weather data includes temperature, pressure, and humidity, as well as quantities like wind speed. These quantities are collected at thousands of locations worldwide. Like in buildings, this data is continually stored over time, and meteorologists leverage these new technologies on their big data to model weather systems that ultimately provide weather predictions.
But what meteorologists don’t do is turn around and try to control the weather.
Weather data is input only. There are no outputs. It is only analysis—not automation and optimization.
In building automation, we also want to analyze, model and make predictions from building data. But unlike these other systems, our ultimate goal is to use this analysis to control our facilities automatically and optimally whenever possible.
The problem is that current analytics software aggregates complex building data into simpler, bite-sized chunks. For big-picture analytics, we want to remove “undesired” complexity. For optimization, we need to propagate a complete optimum strategy back into these individual systems. All of the data we removed during analysis will be missed if the intent is to use analysis for optimal control. This is illustrated below:
Lesson Learned: Distributed Control
Early building automation systems from the 1980’s used a Master/Slave approach to control. In these early systems, all of the “smarts” were centralized in a single large computer. The field controllers themselves were “dumb” and would rarely make decisions on their own. At the time, this was the only way to achieve automation. Electronics were not yet cheap and powerful enough to perform distributed control, and automation software had not yet evolved to the point where control problems could be broken up and distributed throughout the building.
This changed in the 1990’s when revolutionary Direct Digital Control (DDC)systems began to take over. As an industry, we learned that distributing the “smarts” as close to equipment, sensors and actuators was the only way to achieve high reliability and management. It’s not simply that DDC systems are only more tolerant of network failures, but more importantly, it allows equipment and controller manufacture to encapsulate complex details within a controller and expose only the data that is required by other systems. This way, networks of disparate systems can be more easily managed, and manufacturers develop faster and more freely. They don’t have to explain or export internal complexity to the larger system.
Future Solution: Distributed Analysis and Optimization
As we learned with automation and control in early BAS systems, a single Master computer collecting data from many sources is not the best way to perform analysis and optimization. So why is this the preferred method today? Simply, it is the only way that current technology allows us to do it.
But as technology progresses, it will become more and more feasible to run machine learning algorithms in a distributed way directly on equipment controllers. Tiny embedded computer systems are becoming commonplace, and open source algorithms are also becoming more common. This will allow individual equipment and controller manufacturers to not only embed analytics directly into their controllers, but also perform optimization continuously—even when the Internet or networks are unavailable.
Conclusion: Machine Learning at the Edge
Today, it’s hard to imagine a control system that is not DDC. As automation progressed in the 90’s, we expect the same progression for analytics and optimization today and tomorrow. Individual controllers and systems in buildings will perform their own analytics and optimization, encapsulating complex details while feeding results up to larger systems. This trend will not only prove to be the more reliable solution, but also allow more rapid development and easier management.
In order to accomplish this, BAS manufactures need to do a better job of incorporating this new technology into their products. It is no longer enough to simply control equipment in a suboptimal way and treat energy usage as an afterthought.
Building-wide add-on software packages will always have their place. Centralized graphics can be a great convenience, and a redundant audit of other systems can be useful. However, true optimization will not be easily achieved without the cooperation of large building systems like HVAC and Lighting. Today’s chiller plant optimization programs are a testament to the power specialized system optimization. Complete system optimization will require the full complexity of existing building systems and will be best achieved by the BAS manufacturers themselves. We look forward to these exciting changes on the horizon.
About the Author
Mike Donlon is the Director of Research and Development and one of the owners of Computrols. He received his Bachelor’s Degree in Electrical Engineering from the University of New Orleans in 1986. He returned to UNO to get his Master’s Degree in Engineering and focused his studies on computer vision and image processing. While in graduate school, he was employed as a research assistant at Kresge Hearing Institute of the LSU Medical Center. In 1989 he was hired by Computrols as the principal software developer for the original CBAS computer system. Soon afterwards he was invited by the company’s originators to become an equity partner. Since then his duties have been to oversee all software and hardware development for the company as well as researching and applying new technologies to Building Automation. His recent areas of interest include machine learning, building system modelling and open source technologies for building automation. He can be contacted at mike.donlon@computrols.com
http://www.automatedbuildings.com/news/sep16/articles/computrols/160828105505computrols.html