What if a machine could tell you it was going to break before it stopped working?
Researchers at Michigan State University are developing an artificial intelligence system called AI Mechanic that can listen to machines, learn what normal operation sounds like and identify changes that could signal a problem. The goal is to identify problems early enough that equipment can be repaired before a failure causes greater damage, downtime or disruption. For consumers, the potential benefit is simple: equipment that lasts longer withfewer unexpected breakdowns and requires less money and time for repairs.
Josh Siegel, an associate professor in the College of Engineering, was inspired to create AI Mechanic as a results of his love ofrestoring cars and trying to diagnose problems by listening to how an engine sounds. Today, the research has expanded far beyond vehicles to include road maintenance equipment, construction equipment, cement mixers, excavators, hydraulic pumps, generators and manufacturing equipment.
“I think improving the useful service life of things that you already own is a key value proposition,” said Siegel. “If we do our jobs well, your washing machine is not going to break, your microwave is going to run forever, you’re not going be stuck in traffic as long in construction sites because the excavator that’s on-site isn’t going break down nearly as frequently.”
The research combines artificial intelligence with an understanding of the underlying physics of machines. Siegel’s research group includes students and researchers from multiple engineering and computing disciplines, and the team examines recordings for characteristic frequencies and physical signals associated with components as they begin to fail.
“What are the characteristic frequencies that we’re looking for?” Siegel said. “What does it mean when a bearing really fails?
The system also can account for differences between recording devices. Researchers have developed methods to normalize samples collected from devices such as Android phones, iPhones, laptops and field recorders so the recordings can be used together.
Rather than requiring massive amounts of data from every machine, the researchers are developing methods that identify and rank the most useful information while discarding data that does not contribute to understanding the machine’s condition.
AI doesn’t necessarily need massive amounts of data to understand a machine. As AI drives demand for enormous amounts of data, MSU researchers are developing a system designed to diagnose machines using far less data.
“That ability is increasingly important as the amount and cost of storing data continue to grow,” said Siegel. “Instead of simply collecting enormous amounts of information and figuring out what to do with it later, the researchers are working to extract meaningful information from fewer samples. For example, taking 10 gigabytes of information and identifying the 10 kilobytes that matter.”
AI Mechanic isn’t just about fixing your car. It’s about keeping the infrastructure and equipment behind everyday life running.
The researchers are now working with partners across multiple industries, including automotive manufacturing, consumer packaged goods manufacturing and commercial food-service infrastructure. The team is developing models and analytics based on data collected from these partners.
The next step is commercialization. Siegel said the intent is to develop AI Mechanic into a company offering “diagnostics as a service,” working with companies to learn what normal operation looks like for their equipment and then helping identify and avoid abnormal conditions.
The technology has received support from the Michigan Translational Research and Commercialization, or MTRAC, Advanced Transportation andAgBio hubs, as well as co-investment from Spartan Innovations. The project has also received pass-through funding from the U.S. Economic Development Administration through a subaward from the University of Michigan, with co-investment from the Michigan Economic Development Corporation.
Ultimately, Siegel said, the technology is intended to make equipment maintenance something people barely notice.
“We reduce cost, we reduce complexity and we reduce friction in invisible ways that mean you get to spend more time doing what you love and less money fixing things that you don’t want to have to think about,” said Siegel.
