Data Analytics and Smart Buildings Increase Comfort and Energy Efficiency

Page 1 | Data analytics and smart buildings increase comfort and energy efficiency

Data analytics and smart buildings increase comfort and energy efficiency

It can be easy for people to take workplace comfort and climate control for granted—until an air conditioner goes out and a room or building becomes too hot or even uninhabitable, which affects productivity.For workplace comfort, it’s important to ensure that equipment in cooling and heating facilities is working properly. And because this equipment is expensive and hard to replace, being able to predict failure helps lower costs and increase efficiency.

Along with providing a workplace that isn’t too hot or too cold, what about energy efficiency? According to the U.S. Energy Information Administration, “In 2015, about 40% of total U.S. energy consumption was consumed in residential and commercial buildings, or about 39 quadrillion British thermal units.”To reduce thecarbon footprint and reduce operational costs, it’s equally important to optimize energy consumption. Today’s smart buildings are designed to do just that.So, a real estate/facilities organization in Microsoft partnered with Core Services Engineering (formerly Microsoft IT)to work on a smart-energy building solution that helps address these needs. There are two components to our ongoing solution:

Chiller plant. We want to optimize energy use of cooling/heating components and heating, ventilating, and air conditioning (HVAC) systems. We’re working on three chiller plants that support eight buildings. But our roadmap includes all Puget Sound buildings—about 100 chiller plants and more than 100 buildings.

Climate control.We look at comfort levels in buildings, and aim to understand equipment conditions that precede worker complaints/tickets about a room being too hot, for example. We’re working on one building right now, but we’ll expand to several buildings in the next quarter of fiscal year 2017.

To reach our goals, we’re:

Creating predictive analytics models based on Azure Machine Learning to identify and predict issues and equipment failure.

Using theInternet of Things (IoT)to monitor our HVAC systems, do anomaly detection, and lower our costs.

Our work on the smart-energy building solution is in its initial phase.Down the road, we’ll analyze the results of our work thus far, and expand our initial effort to these other plants.

Even in this initial phase, we’re already getting lots of value.We’ve seen a 15percent improvement in efficiency over a 12-month period from just one of our chiller plants.And we’re excited about the future of this solution—likeenhancing our IoT capabilities, using automation to feed machine-learning intelligence back into the building management system, incorporating artificial intelligence, and scaling out globally. We’re also looking into possibly using Azure IoT Suite/IoT Hub for its preconfigured solutions, like predictive maintenance and remote monitoring.

Predictive analytics and IoT optimize energy use and climate control

Whether from a comfort or energy-efficiency standpoint, predictive analytics and IoT shed insight that helps us:

Identify energy-saving opportunities, and inject intelligence to maintain energy-related components.
Monitor the environment in buildings, improve comfort, and reduce operational costs.
Proactively identify conditions that mightcausea climate-related ticket, and correct the conditions before they become severe.
Understand the conditions that cause equipment to trigger a fault.
Do preventative maintenance and optimization—before issues arise or escalate. We get data about when an energy-related component is about to fail and needs to be fixed immediately to avoid emergency maintenance.
Prioritize maintenance work based on criticality.

Improve budgeting estimates of how many components are failing or are near the end of their lifecycle, and getting valuable data to help in planning and forecasting maintenance costs.

For even more IoT benefits and taking digital transformation to the next level, we’re exploring the possibly of using Azure IoT Suite/IoT Hub. Azure IoT Suite has preconfigured solutions that are quick to implement for scenarios similar to ours. For example, improving efficiency of building equipment by collecting millions of records each day, doing predictive maintenance, anticipating problems, and preventing failure. There’s a lot of potential value in IoT Suite for us—like ease of use, scalability, predictive analytics, and data visualization—so we’re eager to explore it further.

Defining our strategy for smart-energy buildings

Our work ties into an existing smart-energy building program that began five or six years ago. We putpreventative maintenance and intelligent-climate operations into place to reduce operational costs of HVAC maintenance and climate conditions that trigger worker complaints.We defined our strategy accordingly, looking attwo key areas:

Chiller plant.We decided to start by collecting historic data from one chiller plant that supports three buildings, and are now working on three chiller plants that support eight buildings. The intelligence that we get helps us forecast energy consumption and find ways to save energy—which lowers costs.

Without adding sensors everywhere, data scientists extract intelligence from existing building management data to examine a complex mechanical system where water cools buildings. We look at the amount of energy consumed to pump hundreds of tons of water. Metrics related to the water temperature, differential pressure, cooling tower fan, and water flow rate are ingested into machine learning algorithms. The building’s HVAC system carries out the cooling task, and air at the proper temperature is distributed to floors and offices.

Climate control.With machine learning, we can identify what contributes to equipment fault signals and worker complaints. Analytics help us predict anomalies in equipment operating conditions. We can apply that learning to reduce extreme-climate conditions in buildings and relatedworker complaints. Analytics also reveal, for example, whether air-handling units are functioning correctly, and whether they’re within the right temperature range. And if these units fail, to what degree will ittrigger a customer complaintabout a room that’s too hot or cold?

What’s included as part of our ongoing strategy? We’re working on tasks like the following:

Develop predictive analytics models based on Azure Machine Learning.
Use IoT to monitor our HVAC systems. Right now, we have a monitoring system that helps the operational center identify issues and send out technicians under certain climate conditions.
Optimize tuning of the chiller plant.
Use our energy smart building software to capture datain buildings about how often the cooling system kicks in.
Look at energy consumption under hot and cold conditions.

Detect faults early, and do diagnostics in our system that could save energy. Wewrote code to help us with early fault detection. Also, we’ve designed cost-modeling to determine the cost in one year if we don’t fix these faults.

Deploying our solution

We’ve completed work on phase 1, and are looking forward to iterating, scaling out, and delivering phase 2:

Phase 1. In fiscal year 2016, we finished chiller optimization for one chiller plant that supports three Puget Sound buildings.(Again, there are about 100 chiller plants that support other buildings.)
Phase 2.After we prove our concept and validate performance, we’ll deploy phase 2. We also plan to collaborate with the IoT teamat Microsoft on additional opportunities to enhance our solution.

For the chiller plant, we’ll take findings from phase 1 and apply them to two other plants. We’ll useAzure Machine Learning on data that we collect in 15-minute intervals to fine-tune chillerplant operations.

For climate control, we’ll expand from one building to several buildings. We’ll collect and analyze the data, make sure that the predictive model’s stable, and scale out. We’ll also do more preventative maintenance, predict conditions that make equipment fail, and keep finding ways toimprove the working environment of buildings.

Getting descriptive, predictive, prescriptive analytics

We use dashboards to provide a description of issues, and we’re working on the predictive and prescriptive pieces.

Descriptive element

Around 60 percent to 70 percent of the Microsoft campus has unified, integrated dashboardsand reporting that are related to temperature settings and climate. We use this dashboard information to monitor our systems. We can see if we need to send out a technician, based on whether something’s too hot or too cold in a building. We can tell which building has a problem,and drill down for details. Figure 1 shows a sample integrated dashboard with information about top performing buildings, energy demand, and load segmentation for a specified month.

Figure 1. Unified dashboard showing demand, temperature, and load segmentation for several buildings

Predictive element

The predictive part, which we’re working on now, is behind the scenes. Our energy smart building software tells us the 500 most expensive faults, in a descending order of dollar value. And it predicts potential cost-savings—for example, if we were to fix a certain issue, it would save us $5,000 a year. These faults auto-generate a work order for issues to be fixed remotely or dispatched to an onsite technician.

Prescriptive element

The fault rules and fault detection/preventative work are prescriptive—for example, under these specific conditions, we need to fix the fault. We’ve already had success with anomaly detection, where if a condition happens with a specified frequency, we fix the issue. We’re now testing the anomaly detection algorithm to be sure it’s robust and that we can fix what we detect.

Also, we’re using IoT to monitor the chiller plant. Sometimes we get an indicator that the chiller plant needs more pumping or chilling. We might have to turn on the chiller based on demand. To optimize energy use, we look for opportunities where we don’t have to start the chiller, and without having a negative impact on comfort level. To reach our goal, we’re looking at what/if scenarios like, “If we don’t start the chiller plant, how much do we save?”

Setting up the components, architecture, and data flow for the solution

One of theenergy smart building tools that sits on top of our building management systems collects data, which has its own, unique format. We’re in the process of bringing that data into a big data platform.

We use a third-party, work-order system. Based on faults that are detected, a work order is created, and the orderis dispatched. A technician works on the order, and closes it in the system.

Our overall workflow for this solution is basically as follows:

We collect data via energy smart building tools, IoT, sensors, and devices.
The information is managed in Azure Data Factory for data integration and processing, with eventual storage in Azure Data Lake.
For machine learning and analytics, we use Azure Machine Learning. Eventually, we’ll use Azure HDInsight and Apache Spark to scale out from an analytics and data processing standpoint.
In Power BI, we’ll showinsights that we get from predictive and prescriptive analytics, and integrate them into our existing unified dashboard. And we’ll use these insights to make prescriptive recommendations.
Transforming our data into intelligent action benefits workers, helps us improve our operations, and gives us an idea of what intelligence to automate, such as automating what temperature a system should run at.

Figure 2 shows a high-level view of how we’re transforming data into intelligent action.

Figure 2. High-level flow used in thisenergy smart building solution

Figure 3 shows a detailed viewof thearchitecture and data flow for this solution. This architecture allows us to integrate components like client systems, power monitors, devices, apps, and servers into a single network.

Figure 3. Detailed architecture and data flow for this solution

Here’s a closer look at the elements in Figure 3:

In the first layer of integration to the left of the figure, power monitors, control devices, and mechanical devices in our buildings communicate with device interfaces and with control panels that control heating, cooling, and ventilation. We use the control panels to manage and monitor the devices.

In the next layer of integration, we have our graphic user interface and server systems that run the devices. Building-management system products—like WebHMI—connect with a front-end server. These server systems also include a historian server, PMCS server (power monitoring control system for monitoring power of devices), Siemens Apogee server, and I/O servers. The I/O servers use BACnet IP and universal datagram packet (UDP) as protocols and methods for systems to communicate.

Thehistorian server stores information at frequent intervals (such as every five minutes) for up to five years. The advantage of this is that we can look at factors like outside temperature and energy consumption on a specific day last year and then compare energy sets from last year with those from this year.
Most of the server systems were already in place before we had energy smart buildings, but we added the application server and a software system that’s hosted in Azure. This software app and server suite join information from the different systems and bring it into the energy smart building system across the Puget Sound, and in various locations in North America and around the world.

The app server is a nexus that brings systems and data together from both ends.For example, it pulls information from the historian server to get data over time, bringsin sources from our feed storage (a facilities server called FAC03), and from external data that’s third-party managed.The change of value (COV) rate is communicated to the app server. This rate indicates whether we want systems to update us at a particular time interval even if nothing has changed. Or we can request updates only if there’s a change, such as a temperature going from 72 to 73 degrees.
FAC03 contains feed information and is a repository of building data like square footage, headcount, and building use. We populate columns in SQL Server based on this information.
The external, third-party data includes data from a computerized maintenance management system (CMMS) andfrom Facility 360 (our maintenance management system).

Benefits of this architecture

In the energy smart building industry, there are lots of hardware solutions for installing controllers and other hardware to get data from buildings. We used what’s mostly a software solution, even though we might need hardware integration in some places.With this solution, we can pull information from different systems, vendors, and environments within one overall, energy smart building architecture. As a result, we’re saving millions of kilowatt hours and we’re helping create a sustainable environment.

Technical limitations of this architecture

When we began, it was easy to overload our systems. Some of our equipment and controllers were old. We wanted to collect data each second from thousands of devices and two million control points, and our systems weren’t able to handle this. So weinitially lost data, controllers would go offline, we lost pressure in buildings, and temperature and lab cooling were affected.We overcame this challenge by really defining exactly what we wanted to accomplish.

If you’re incorporating older systems, be very mindful of the actual volume and content that you need.

Business benefits we’ve gained

Although some of our work is in progress and we don’t have a lot of historic data yet, we’ve already seen big value:

15 percent improvement in energy efficiency. We’re pleased with the 15percent improvement in efficiency (how much energy consumed for every ton of water used to cool a building) that we saw in a 12-month period for one plant. Now we want to replicate (or improve) and validate these numbers for other plants and buildings.
Gives us visibility.For example, we detect faults and get climate data that we were previously unaware of, and build on this learning to improve our models over time. Because Microsoft is a large campus, it can be difficult to know if a particular fault exists. After we scale out to other buildings, this solution will help uncover issues.
Saves us money, and helps us proactively identify problems.By being able to predict when we need to fix or replace a component, we save money by maintaining it up front, rather than having a costly replacement after it breaks or stops working.

Saves us time, and increases our productivity. Before we used energysmart building software, every five years we’d have to go back and check whether all equipment was calibrated to perform optimally. We no longer have to do this because we have intelligence on what’s happening end to end.

Challenges and lessons learned

Our energysmart building system was generating so many faults that we lacked the resources to fix all of them. We had to create a strong business case for adding resources and budget, so that we could fix the faults, save energy, and prove our return on investment (ROI). Generally, the larger the campus, however, the more the total expected savings. In our case, we’ve already gotten back what we put in because we’re a large company with lots of buildings. If you’re a smaller business, you might not see the ROI as quickly. But it’s all about the longer-term, holistic view and the ultimate gains.

Looking ahead

We’re just starting on this road, but the energy smart building solution offers big potential. Building on our current momentum, we plan to:

Use what we’ve learned as a foundation, and automate. Build on the work we’ve started of using predictive modeling and machine learning to understand when a device will fail or need preventative maintenance. We’re in the process of putting machine learning intelligence back into the building management system for scenarios like telling a system to run at 39 or 41 degrees instead of 40 degrees. So far, we’ve provided this information manually, but we want to automate this. The algorithm will tell us if taking a certain action is recommended. And we’re working on making sure that machines talk to machines.
Enhance our IoT capabilities.Build on our ability to do predictive maintenance and fault detection, reduce energy, and lower our costs. Azure IoT Suiteoffers very relevant scenarios thatalign nicely with our goals.
Move the solution entirely to Azure. This is where all the data will eventually reside.
Scale out. In terms of climate-control data, we want to ensure that our solution works well in the Puget Sound before we scale out to other locations like Hyderabad, Shanghai, Beijing, and Dublin.
Incorporate artificial intelligence.For example, workers currently create a ticket to report a room being too cold or too hot. With artificial intelligence, we’ll be able to do preventative maintenance to detect, anticipate, and modify equipment conditions, so that a ticket isn’t needed.
Use Power BI.We’ll take the output that we get from predictive and prescriptive analytics, and integrate it with our existing unified dashboard.

Roll out this work to other teams in Microsoft. Other teams are also excited to learn from IoT data, so we’re building a platform that others in the company can use. We want to clone the data, so that we can do offline analytics and prescriptive and predictive work to help others in the company.Microsoft needs to develop its own intellectual properties around handling, storing, and using Microsoft data.

Harnessing the power of insights

Energy smart buildings are revolutionizing operations and maintenance related to air conditioning, heating, and HVAC systems, while reducing energy usage. In our energy smart building solution, the data that we collect and analyze from using predictive modeling, machine learning, and IoTpays us back in actionable insights. We get huge benefits related to comfort and energy efficiency, and our ROI will continue to increase.At the end of the day, it’s very gratifying to know we’re helping lower the carbon footprint and improving comfort and productivity for workers. Our analytics and technology help make this possible, and we’rereally looking forward to what the future brings.