Preventative medicine: Algorithms enable manufacturers to remotely anticipate machine health crises
Modern machines generate diagnostic information whenever they are in service. For example, current automobiles provide real-time data to drivers about systems such as the engine and transmission, and drivers are able to make educated decisions about when their vehicles need service.
Industrial and Systems Engineering Professor Shiyu Zhou envisions a future in which diagnostic information from multiple machines can be processed centrally so that manufacturers can more accurately monitor their systems and anticipate or diagnose problems.
Thanks to a National Science Foundation grant, he’s working to make that a reality. The NSF grant is a three-year, $200,000 commitment to Zhou’s research on teleservice systems. Also called remote monitoring systems, teleservice systems consist of an in-field unit or machine connected to a centralized service hub via a telecommunication network. These centralized hubs collect massive amounts of real-time data on the condition of everything from the vehicle’s oil life to failure events, such as when an airbag deploys.
Zhou focuses on using that constant stream of data to mathematically model accurate failure predictions and help both users and manufacturers make better, more informed service decisions. His research is part of the NSF Grant Opportunity for Academic Liaison with Industry initiative, and he is working in partnership with General Motors.
General Motors is an ideal partner for Zhou’s research because a massive teleservice system is already in place. GM’s OnStar teleservice system consists of GM vehicles, the OnStar telecommunication network, and all of the monitoring data OnStar collects. Zhou is developing algorithms based on that data to better predict when specific parts will fail. “In the back-office hub we can use an algorithm to analyze the data we collected to predict the occurrence of failure,” he says. “Then we can identify, if we are working in normal conditions, how likely it will be that the machine will fail, and when.”
Zhou says the ultimate goal of these teleservice systems is to improve the user experience, enhance product safety, lower maintenance costs and eventually gain competitive advantage for the manufacturer. For example, GM’s OnStar system constantly collects real-time data about the condition of the vehicle’s battery, and it also collects event data when the battery fails. Through a predictive algorithm, Zhou hopes to be able to alert drivers before their battery fails, and to provide GM service professionals the information they need to provide recommendations on how to service it.
Zhou sees applications for his research beyond the automotive industry, and cites Magnetic Resonance Imaging (MRI) machines—which also are increasingly connected to teleservice systems—as an example. “We can collect information from an automobile or an MRI machine and, based on that information, we can try to diagnose or predict failure events in the system,” Zhou says. “How we combine this data to build an efficient prognosis model is the goal of this project.”