Our paper on ``Federated learning of predictive models from federated Electronic Health Records''...
ENG faculty developing sensors that stay in touch
By Tim Stoddard
|John Baillieul, Christos Cassandras, and Yannis Paschalidis (seated, from left) of ENG’s Center for Information Systems Engineering are developing sensor network control systems with far-reaching applications in manufacturing, homeland security, and space exploration. Their interdisciplinary team includes University of Massachusetts, Amherst, faculty Abhi Deshmukh, Weibo Gong (standing, from left), and Robert Gao (seated, far right). Photo by Fred Sway|
When an unbalanced washing machine starts dancing across the floor, a sensor inside detects the wild gyrations and halts the spin cycle. In the future, says Christos Cassandras, washing machines and other household appliances will go a step further: they’ll call a repairman before a part breaks.
Cassandras, an ENG professor of manufacturing engineering, says that appliances will be only one of many applications for so-called sensor networks, which feature a constellation of miniature devices that monitor a wide range of things, from vibrations to temperature to toxic chemicals and biological weapons. Cheap, lightweight, and built with tiny radios, the sensors will talk to one another, coordinating their reconnaissance and relaying their findings to a distant computer. An aging washing machine, for example, might notify the manufacturer of an imminent problem over a wireless Internet connection. The owner might then receive an e-mail from the dealer to schedule a service visit for a part that may not break for several weeks.
Sensor networks promise to improve everything from manufacturing to homeland security to endangered species management. But while the technology for building tiny, cheap, and intelligent sensors is rapidly emerging, Cassandras says, the challenge now is to develop theories for managing the flow of information among hundreds or even thousands of networked sensors. With an interdisciplinary team of engineers at ENG’s Center for Information Systems Engineering (CISE), Cassandras and his colleagues have received a prestigious award from the National Science Foundation’s Division of Design, Manufacturing, and Industrial Innovation to develop methods of managing complex sensor networks.
With the award’s $2.5 million over the next five years, the BU team will work with colleagues at the University of Massachusetts, Amherst, to design sensors that communicate with one another, learn on the job, and tolerate extreme environments. The team includes Yannis Paschalidis, an ENG associate professor of manufacturing engineering, David Castañon, an ENG professor of electrical and computer engineering, and John Baillieul, an ENG professor and chair of the department of aerospace and mechanical engineering.
Initially, the CISE team will go to the chalkboard, hashing out the theories and principles underlying sensor networks. “We’ll be conceptualizing these new technologies and abstracting them to pictures, equations, and computer-based simulations,” Cassandras says. “One of our main tasks is to develop algorithms — systematic and methodical ways of doing something more efficiently — to make good decisions and get this information and technology to people who can implement it.”
Down the road, members of the team will also develop a test bed at BU to experiment with sensor network designs. To start, however, the UMass faculty will be principally involved in building and testing sensor hardware. Researchers there have been designing ball bearings with tiny sensors that monitor vibrations and send a wireless signal when the ball bearing is breaking down and needs to be changed.
The CISE team believes sensor networks will play an important role in large-scale industrial settings. In factories, for instance, sensors will be embedded in tools and machines to identify problems before they cause catastrophic failures. A variety of sensors could be used to keep track of fuel and raw materials at every step of production, says Paschalidis, providing a real-time inventory that would dramatically improve planning and supply chain management. The smart washing machine is an example of the potential for postmanufacturing product monitoring. “What we’re headed for is a day in which products essentially will repair themselves,” Baillieul says, “or communicate with the factory about what’s gone wrong and what needs to be done.”
Locating personnel and mobile equipment such as forklifts within an industrial campus can be a major bottleneck to efficiency and safety, Cassandras says. By tagging these mobile elements with radio frequency identification tags (RFIDs), similar to the EZ-pass units used on the Massachusetts Turnpike, companies would have a real-time map of the location of each item. RFIDs are also expected to have a major impact in retail. Instead of bar codes, items in the supermarket and department stores will have tiny RFIDs that identify themselves to sensors on the shelves, giving companies valuable real-time inventories.
The major challenge for the CISE team will be integrating different kinds of information from sensors that are built to monitor different things. “Computers talk to each over the Internet,” Cassandras says. “That’s fascinating, but relatively simple. What we’re dealing with in sensor networks are heterogeneous devices that are taking qualitatively different measurements at different rates. It’s not just the exchange of information. There’s also the element of motion in the different sensors, and the fact that they are essentially speaking different languages.”
As the CISE team wrestles with the management of sensor networks, it is prepared to adapt to the rapidly evolving sensor technologies that will undoubtedly develop in the coming years. “As a group, we’ve done a lot of work in other areas that are cutting edge,” says Cassandras. “But this is really one situation where we don’t know what’s going to happen in fundamental technology three years down the line.”