In a manufacturing plant, a nuclear reactor, a traffic network and a host of other complex systems, millions of sensors are constantly collecting torrents of data.
But that data, in and of itself, is not knowledge, says Kaibo Liu, an associate professor of industrial and systems engineering at the University of Wisconsin-Madison.
“The data is like a message,” he says. “How can we better transform the data into knowledge, into smart decisions?”
Liu is working to do just that by harnessing data science methods—and developing new ones—to improve quality in manufacturing systems, refine maintenance decisions in nuclear power plants and other smart and connected systems, and enhance sensor system design, data acquisition, monitoring, diagnosis and prognostics across a range of fields.
While the specific topics and techniques vary from project to project, there’s a clear connective thread for all of Liu’s work: Each endeavor is rooted in a plausible scenario drawn directly from industry with practical values. He says he’s trying to bridge the gap between fundamental data science research and industry needs, two perspectives that haven’t historically connected.
“I’m interested in industrial data science, which means before I develop something, I need to make sure some people need it with practical impact,” says Liu, who’s landed early career awards from the Institute of Industrial and Systems Engineers, the Society of Manufacturing Engineers and the American Society for Quality. “Otherwise, I don’t want to develop it.”
Liu has teamed up with industry partner 3M on several projects, including developing new methods to identify faulty code and spot anomalies in data from manufacturing production lines, indicating problems, and leveraging a technique called transfer learning to examine production processes.
He’s also employing transfer learning, a machine-learning tactic that leverages insights from one dataset to analyze similar samples, to model maintenance concerns in nuclear reactors with funding from the U.S. Department of Energy (DOE). It’s the second DOE grant he’s earned to apply his prognostic modeling to nuclear equipment, representing the continued growth of a line of research that started with aircraft engine maintenance. Liu has also led several projects funded by Department of Defense agencies, such as the Office of Naval Research, U.S. Army Corps of Engineers and the Air Force Office of Scientific Research, to investigate advanced predictive maintenance strategy for complex smart and connected systems, a longstanding critical issue for military applications.
Since joining UW-Madison in 2013, Liu has analyzed the neurodegenerative process in Alzheimer’s disease, detected solar flares on NASA images of the sun, modeled and predicted real-time travel demand, and more. He also recently coauthored a paper in the Journal of Machine Learning Research, considered a top journal in the machine learning field, examining quality in crowdsourcing for collusion detection, a more recent interest within his research group.
But each of those disparate topics requires a nuanced solution from within the data science toolkit Liu began building as a PhD student at Georgia Tech, where he pivoted from studying optimization and took courses in computer science and statistics. He earned a master’s degree in the latter and gained exposure to machine learning before the field exploded. Across his research portfolio, Liu uses a variety of artificial intelligence and machine learning techniques, including transfer learning, deep learning and reinforcement learning.
“I’m not a person to say I only use this type of method,” he says. “It’s like you only have one axe and then you try to use that axe to try to cut all kinds of trees.”
Liu is passing on his insights through ISyE 412, Fundamentals of Industrial Data Analytics, an undergraduate course he developed that uses a blended learning approach. He’s also taught an online version for working engineers through the College of Engineering’s Interdisciplinary Professional Programs and created an online knowledge-sharing hub, called IERA, specifically for industrial engineers. In 2020, the Institute of Industrial and Systems Engineers (IISE) honored Liu with its Innovations in Education Award.
A year later, Liu received IISE’s Award for Technical Innovation in Industrial Engineering, mainly due to his work with 3M. Both sides are keen to extend their ongoing collaboration, which has prompted several of Liu’s graduate students to complete summer internships at 3M and the company to hire graduate Honghan Ye (PhDIE ’21) as a data scientist. Liu calls such industry research partnerships a win-win for both sides.
“They can share with you what they really need and the challenges they’re facing,” he says. “I think this is one of the most important parts of research, because the problem formulation and practical justification is really half of the effort. A lot of research, if your problem formulation is wrong or meaningless—even though you develop something, you prove something—I don’t think it has any applicability. Working with those industry partners, they can really inspire me.”