Machine Learning 101: Is Predictive Analytics Possible in Your Facility?

However, the potential of machine learning in the built environment is neither HAL from 2001: A Space Odyssey nor an automated replacement for facilities managers. Instead, it can provide key insights into your facility and help you make more efficient and cost-effective decisions.

[Related: Internet of Things: How IoT Devices Enable Predictive Maintenance]

Are you ready to take on machine learning in your facility?

What Is Machine Learning?

Machine learning uses the data your facility’s systems and equipment produce to make your building run more efficiently.

“Machine learning uses data and statistical techniques to provide insights,” says Hugues Meyrath, Chief Product Officer of ServiceChannel, a facility management software provider. “The breakthrough idea here though is that machines (computers) can learn rather than be programmed.”

Humans are limited in our ability to create actionable insights with data. Applying machine learning to facilities allows an unlimited number of potential insights by virtue of algorithms and computing power that far exceed our own.

“Facilities management is a discipline with a huge amount of data: trades, jobs, rates, hours, assets, proposals, warranty, uptime, maintenance, performance, check in/out, budgets, etc.,” says Meyrath. “All of these can be optimized: How do we account for geographical differences? How do I reduce my budget yet provide better uptime? How often should I really maintain my assets (based on usage or performance rather than fixed schedule)? There are opportunities to reduce cost, automate, predict and provide better customer experience.”

The data that machine learning works with is based on past experiences. This is where it derives its value. Receiving actionable insight to make better decisions than before is key to using machine learning.

[On topic: A Case for Future-Proofed Indoor Location Services]

“Predictive data analytics looks at the past and extrapolates the future,” says Meyrath. “For example, based on historical data, the machine predicts/replicates the likely behavior or decision. Machine learning can learn and provide a better outcome.”

Is My Facility Ready for Machine Learning?

Data is everything with machine learning, so you want to have as much as possible. As buildings become more connected, the number of data points in facilities is growing quickly. But you need to know where to look to unlock this data.

“Visibility is the key ingredient – nothing is possible if you don’t even know what exists,” says Sri Chandrashekar, CEO and Co-Founder of Optio3, an IoT device management provider. “You first want to understand what types of devices and control functions you have in the facility that you are managing. Even highly sophisticated companies do not possess more than a rudimentary understanding of the various device types that exist in their facility.”

Identifying the devices in your facility can be difficult, since your building is likely filled with them. If you aren’t sure, you can enlist an outside company that works with machine learning to ensure that you’re able to use as much data as possible.

“A whole lot of existing buildings have a ton of connected devices, so how do you find out what connected devices exist in a building?” says Chandrashekar. “It could be anything from an HVAC system to a simple thermostat or governor.”

One of the properties that Optio3 is working with has over 100,000 control points and collects 40 million data points per day, explains Chandrashekar. “Obviously no human can process that amount of data, so this is a perfect area where machine learning can make a significant impact.”

How Do I Get Started?

To get started, you might want to enlist a consultant or expert in the field to make sure you’re able to collect and analyze as much data as possible.

“The first part is to get a system of record where you collect granular information over time from as many relevant sources as possible,” explains Mereyath. “All large hosted players have generic machine learning capabilities, such as Google, Microsoft, Amazon, IBM, etc. Consulting companies can help with specific project implementations. Domain experts explore particular problems that are often more targeted.”

[Machine Learning: 5 Steps to Optimize Your Facility with Data Analytics]

Applying a networked control system that connects your building systems for data harvesting is key. “You definitely need a networked control system. These solutions need to be connected to a network to be able to pull data out,” says Ash Awad, Chief Market Officer at McKinstry a design, build, operate and maintain firm.

Is Machine Learning Financially Feasible?

Of course, the biggest concern with something like machine learning is if it’s affordable. Fortunately for most facilities, the data is available so the real action to take is creating information what you already have.

The business case is relatively straightforward now that the process has become a low cost, effective way to pull data out, map it, organize it and use machine learning, explains Awad. “In short, you are buying into a solution that can do a great job effectively using computers to pull data out and use the first level of artificial intelligence – machine learning – and then organize it. You’re paying somewhere between 10 and 15 cents per square foot,” he adds.

Contributing to machine learning’s affordability is its limitless potential. “Machine learning can be applied to an unlimited number of problems: proposal approvals, assets, etc.,” explains Meyrath. “It makes sense to focus on bigger ticket items: large assets, large spend, biggest spend trade (e.g. HVAC), automation of repetitive tasks, etc.”

Because it requires some kind of networked software to analyze the data that you already have, machine learning won’t break the bank.

“The cost is so low and the return is so obvious and great, even those with really tight budgets have tools that they never had before,” says Awad.

What Is the Future of Machine Learning?

The main thing to consider about the future of machine learning is that it has almost limitless potential. The more facilities managers apply this technology, the better the yielded results.

“Machine learning will get applied to more use cases – going from more generic to specific problems, verticals, trades, etc.,” says Meyrath. “Learning gets accelerated over time, and thereby machine learning yields more and more insights.”

The main hurdle to jump to make this a reality is to share the data that you attain. Data points from a wide range of facilities in all geographic areas leads to more opportunities to improve efficiencies in all facilities. Sharing data widespread allows for more AI strategies to be employed.

One possibility for future machine learning technology is creating a digital twin of a building. When this level of analytics becomes available, facilities managers will be able to test the operational outcomes of any scenario.

 “You can organize the data and the way systems operate so you have a digital twin of a building or campus where you can test different strategies and scenarios over an entire year in exactly how the building would respond,” says Awad. “You can see how that will actually operate.”

This type of technology is still emerging, but it’s sure to be a game changer when this kind of intelligence becomes widely available.

Justin Feit was associate editor of BUILDINGS.