UVA Engineering Earns Early Success in its Cyber-Physical Systems Initiativegreenwood@cstone.net
In fall of 2015, UVA Engineering announced the establishment of the Link Lab, an interdisciplinary initiative designed to create a world-class center of excellence in cyber-physical systems. The lab itself is under construction, consuming the top floor of the Engineering's School's Olsson Hall. Meanwhile, Link Lab faculty members have continued to demonstrate their national leadership in cyber-physical systems.
Computer Science Professor Jack Stankovic, an author of one of the earliest national reports on the then-emerging field, has co-authored a report on behalf of the National Academies of Science, Engineering and Medicine. The report, entitled "A 21st Century Cyber-Physical Systems Education," is expected to shape the future of teaching and learning in the field. Also, the University of Virginia made one of the first awards from its $2 billion Strategic Investment Fund to help the Engineering School establish the Link Lab, with more than 20 faculty members working collaboratively on cyber-physical systems projects. UVA Engineering has hired five new Link Lab faculty members this year, with plans to hire three more.
In addition, UVA Engineering this year secured three major grants from the National Science Foundation’s Cyber-Physical Systems Program - the most awards for any single university in that program. The awards, which total almost $1.4 million, represent nearly 10 percent of the awards the program granted during that funding cycle.
The distinguishing feature of all three Link Lab proposals is their emphasis on tackling fundamental challenges to real-world cyber-physical systems. Stankovic, the principal investigator of a project designed to standardize the way cyber-physical systems designers characterize human behavior, aims to clear the way so that smart wearables produce a revolution in health care. Electrical and Computer Engineering professors John Lach and Benton Calhoun are co-designing devices, circuits and systems for the dramatically more efficient, flexible components needed for the self-powered Internet of Things (IoT). Meanwhile, a team led by Kamin Whitehouse, associate professor of computer science and director of the Link Lab, faced with the tremendous cost of deploying cyber-physical systems at the scale required to reduce energy use in buildings, is using data science techniques to reduce the need for physical instrumentation.
Energizing the Internet of Things
Power is a stumbling block to the Internet of Things. Each of the connected devices, as it senses and responds to its environment, requires a source of energy, but batteries — even better batteries — are not a workable solution. Electrical and Computer Engineering Professor Calhoun has calculated that with a trillion connected devices in the IoT, each with a generous battery life of five years, it would be necessary to replace more than 500 million batteries each day.
“Harvesting energy from the environment is an obvious alternative,” said Lach, who is also chair of UVA’s Charles L. Brown Department of Electrical and Computer Engineering. "But when you depend on energy harvesting, you have to deal with real-world dynamics that affect the amount of energy you can harvest at any one time.”
Lach, Calhoun and their colleague, Materials Science Professor Susan Trolier-McKinstry at Pennsylvania State University, are developing a system that would adjust automatically to changes in the environment that affect energy harvesting. As a starting point, they are building sensor systems they will deploy in various IoT applications to help them characterize the dynamics of the environment and help them understand how these dynamics might affect specific energy harvesting techniques.
“One of the main things we are focusing on in this project is converting mechanical kinetic energy to electrical energy,” Lach said. “We will look at vibrations in the ductwork of buildings and use accelerometers to see how and when they move.”
With this baseline information in hand, they will tackle the problem of building dependable energy harvesting systems for each test application. To maximize energy supply, Trolier-McKinstry will create ultra-high-efficiency mechanical harvesting structures optimized for specific applications. To minimize energy use, Calhoun will design electronics with power management circuitry that adjusts to fluctuating energy supply. Finally, to maximize quality of service, Lach will develop algorithms that manage energy consumption based not just on currently available energy, but also on its predicted future availability. “The idea is to provide a more dependable self-powered system that provides sustained service in the face of the uncertainties of energy harvesting,” Lach said.
Reducing Energy Consumption in Buildings
As Whitehouse points out, a full 40 percent of the nation’s energy budget flows through the energy meters of its buildings — residences, offices and even shopping malls. “From a conservation point of view, reducing building energy consumption is really low hanging fruit,” he said. “There is a great deal of waste that we can easily eliminate. The issue is doing it practically and efficiently.”
The obstacle to using cyber-physical systems for this task is what Whitehouse refers to as the billion building challenge. Thanks to advances in low-power sensing and wireless communications, it is now possible to instrument buildings to monitor their energy consumption. The cost for a program on a meaningful scale, however, would be prohibitive. Even at $100 per building, a very low estimate, tracking the billion buildings on the planet would cost $100 billion dollars.
Working with Assistant Computer Science Professor Hongning Wang, Whitehouse is taking advantage of the fact that there are standardized construction practices and usage patterns associated with specific building types. By taking a sampling of buildings in each type, metering each of the subcircuits that supply current to individual rooms or appliances and adding static data like number of occupants, building size and location, Whitehouse and Wang hope to collect a dataset that they can use to derive the energy use of buildings not in their dataset.
“We are using machine learning to develop new mathematical techniques that will enable us to find essential similarities among buildings and identify the predictive components that can be applied to other buildings,” Whitehouse said. Building occupants are already able to upload their historical energy bills to a website Whitehouse and Wang have created and receive an appliance-level energy breakdown that can help them reduce their energy consumption. The two investigators estimate that, compared to sensing in all buildings, this approach could lower the cost of energy monitoring by at least a factor of 100.
A Modular Approach to Wearable Monitoring
Because they are attached to the skin, wearables devices like smartwatches and fitness trackers represent the future of medical monitoring. One of the impediments to their more rapid adoption, according to Stankovic, is that there is no common set of tools that designers could use to develop apps for them. “Right now, everyone is starting from scratch,” he said. “This didn’t seem very scientific or efficient to me. People are trying to solve the same problems over and over with varying degrees of success.”
Stankovic proposes a modular approach to developing these human-in-the-loop cyber-physical systems that would consist of a set of basic underlying algorithms that can be applied in a variety of contexts. As a starting point, he has created five core modules. The modules include one that can be used to analyze human motion that doesn’t depend on direction and one that can be used when direction is important. “In the first case, you can eliminate some of the information generated by the wearable’s built-in accelerometer, gyroscope and magnetometer,” he said. “Other modules can be applied to situations where the speed of the human action is an issue and where it is not.”
Using a modular approach, he hopes to develop an app for detecting hand-washing by hospital clinicians that works better than an existing one-off solution. He has deployed an initial version of the app in the pediatric intensive care unit at the UVA Medical Center. One of the early challenges he discovered is distinguishing hand-washing from diaper-changing. “There is nothing magical about the five modules we started with,” he said. “As we are applying these modules to new applications, we expect to refine them and add more modules to our core group.”
Building on Established Strengths
These three NSF awards to highlight the wisdom of the Engineering School’s decision to make making cyber-physical systems a focus for leadership and excellence.
“Securing these proposals takes us to the next step, putting us in a better position to leverage our core CPS strength to bring in more people and more ideas from around the school,” Lach said. “Each successive generation of proposals will be more collaborative and more discipline agnostic."
Lach expects to see some big team proposals in cyber-physical systems in the coming year, especially as UVA Engineering continues to add new faculty in this area.