Carnegie Mellon University (CMU) uses Microsoft Azure and the PI System™ from Microsoft Global ISV partner OSIsoft to reduce building maintenance and energy costs. Now CMU has added Azure Machine Learning for better fault detection, diagnosis, and more efficient operations. With these capabilities, CMU personnel gain advanced analytics for improved operational insights and decisions. And CMU gains a way to cut energy use by 20 percent.

“We see Azure Machine Learning and the PI System ushering in an era of self-service predictive analytics for the masses. We can only imagine the possibilities.”

Bertrand Lasternas, Researcher, Center for Building Performance and Diagnostics, Carnegie Mellon University

Carnegie Mellon Sees a Way to Cut Energy Use by 20 Percent with Cloud Machine Learning Solution

CustomerSolutionStoryUnlock insightson any data

Photo: ©Carnegie Mellon University. All rights reserved.

Business NeedsCarnegie Mellon University faces the challenge of big data in fields as diverse as astrophysics and building management. In the latter category, the CMU Center for Building Performance and Diagnostics studies the operational efficiency of its own buildings as well as those of facilities worldwide.

The traditional way to monitor operational efficiency in buildings is to rely on collecting and analyzing disparate data captured from sensors and actuators that control everything from heating and cooling to lights, ventilation, plug load, and security systems. These systems lack the ability to meet challenges such as predicting failures and reducing energy use. The result is expensive system failures and wasted energy.

In 2011, CMU adopted the PI System as its infrastructure to connect sensors, data, and people to deliver real-time insights into facility performance through a real-time dashboard. In 2013, CMU extended the solution with the self-service business intelligence capabilities of Microsoft Power BI for Office 365 and migrated the PI System to a hybrid on-premises/cloud configuration using Microsoft Azure infrastructure as a service (IaaS).

Next, CMU wanted to add real-time predictive analytics so building managers could act proactively, for example, by repairing or replacing worn components before they could fail. CMU also wanted to enable automated building systems to act with greater cost-effectiveness and precision, for example, by anticipating when and by how much thermostats can be adjusted to anticipate heating and cooling requirements. CMU wanted the predictive analytics to be fast, easy to implement, and accessible to nontechnical personnel on a daily basis.

SolutionTo achieve these goals, CMU extended its solution with Azure Machine Learning, a Microsoft Azure platform-as-a-service (PaaS) offering. Azure Machine Learning uses a highly visual interface with prebuilt models and templates to reduce the time, cost, and complexity of creating and training predictive models for use with applications such as the PI System.

CMU used Azure Machine Learning along with historical data from the PI System to address the challenge of fault detection and diagnosis for environmental control-system components that may be hidden from visual inspection behind walls or under floors. CMU researchers used the

“The savings come both from reducing energy use and from being able to shift some energy use to hours of lower demand and cost.”

Bertrand Lasternas,Researcher, Center for Building Performance and Diagnostics, Carnegie Mellon University

OverviewCustomer: Carnegie Mellon UniversityCustomer Website: www.cmu.eduCustomer Size: 5,000 employeesCountry or Region: United StatesIndustry: Education—UniversitiesPartner: OSIsoft, LLCPartner Website: www. osisoft.com

Customer ProfileCarnegie Mellon University, based in Pittsburgh, Pennsylvania, is a research institution with seven globally recognized schools and colleges including the Carnegie Institute of Technology.

temperature of the water released by a valve as a proxy measure of its performance. They used the PI System and Azure Machine Learning to compare predicted and actual water temperatures, noting deviations between them to identify potential failures.

The researchers also addressed the building-automation issue. They constructed a use case in which the building temperature needs to be brought up to 72 degrees for the start of business at 9 A.M. The heating system is typically engaged at 6 A.M. or, on warmer days, at 6:30 A.M. But that likely wastes energy. Could the predictive analytics solution identify the ideal time to start heating the building?The researchers aimed to predict the internal temperature of the building at 9 A.M. using a model that included recent internal and external temperature, anticipated solar radiation levels, and other factors. But anticipated solar radiation data wasn’t available, so researchers had to first predict this variable. They trained a solar radiation model using a boosted decision tree algorithm available in Azure Machine Learning, tested the model to confirm its accuracy, and then used it in the internal temperature model to address the question of energy conservation.

Azure Machine Learning is one of four key components in the extended PI System and Microsoft Azure solution. The solution begins with an on-premises PI Server™ that collects sensor data from across the campus and forwards it via Azure-based PI Cloud Services™ to a PI Server running in Microsoft Azure

IaaS. An OSIsoft research tool cleanses, aggregates, shapes, and transmits the data in real time to a working repository of Microsoft Azure table storage, where it is accessed by Azure Machine Learning for analysis. The predictive insights can be accessed through Power BI, and the predictions are stored in the PI Server for use by the building systems applications.

BenefitsCMU has demonstrated that Azure Machine Learning and the PI system will reduce energy and operational costs and that it is fast, easy, and inexpensive to set up and use. CMU anticipates broad new uses for the solution.

Reduces Projected Energy Use by 20 PercentBased on the experimental results, CMU researchers estimate the solution could cut energy costs by 20 percent. Discussions are underway to implement it campus-wide, where it could save several hundred thousand dollars annually.

“The savings come both from reducing energy use and from being able to shift some energy use to hours of lower demand and cost,” says Bertrand Lasternas, Researcher, Center for Building Performance and Diagnostics, Carnegie Mellon University.

Cuts Setup Time from Weeks to DaysAzure Machine Learning simplified and accelerated the time-consuming process of creating and testing machine learning models. A typically weeks-long process was accomplished in a few days.“We immediately began using Azure

“We immediately began using Azure Machine Learning without having to prepare on-premises software; everything’s ready-to-use in the cloud. It’s significantly easier to use than other tools we’ve tried, and it fit seamlessly with the PI System and Microsoft cloud solution we already had.”

Bertrand Lasternas,Researcher, Center for Building Performance and Diagnostics, Carnegie Mellon University

Machine Learning without having to prepare on-premises software; everything’s ready-to-use in the cloud,” says Lasternas. “It’s significantly easier to use than other tools we’ve tried, and it fit seamlessly with the PI System and Microsoft cloud solution we already had.”The solution also fosters collaboration. “We can easily collaborate by sharing workspaces,” says Yogesh Venkata Gopalan, Graduate Student, Energy Science Technology and Policy, Carnegie Mellon University.

Supports Self-Service Predictive Analytics for “the Masses”The CMU researchers envision the PI System and Microsoft Azure supporting not just researchers but also managers, engineers, and technicians—the people who interact daily with building systems.

For example, tablet-toting field service technicians could access the predictive analytics to check and update remote equipment before it fails. Smartphone notifications could alert engineers to energy demand spikes. Because the solution is scalable and cost-effective, it could be used at building complexes and public-utility systems that can’t feasibly be served by traditional solutions.

“We see Azure Machine Learning and the PI System ushering in an era of self-service predictive analytics for the masses,” says Lasternas. “We can only imagine the possibilities.”

For more information about other Microsoft Customer Successes, please visit: www.microsoft.com/casestudies

Software and ServicesMicrosoft Azure platform• Microsoft Azure• Microsoft Azure Machine Learning• Microsoft Azure Storage

Microsoft Office 365• MicrosoftPowerBIforOffice365

Microsoft Server Product Portfolio

• Microsoft SQL Server 2014

OSIsoft PI Cloud Services™

OSIsoft PI Server™

CMU Center for Building Performance and Diagnostics©Carnegie Mellon University. All rights reserved.