You’re driving down the road and your car starts to make a “plink, plink” sound. Or maybe it’s more of a “pring, pring?” Is it the gearbox? The clutch? The alternator? The AC? Is your engine about to fall out and go careening down the road?
Many of us are in the dark when it comes to our cars. We rely on manuals and mechanics to tell us when something needs replacing or what’s wrong when something’s broken.
But what if your smartphone could diagnose your car instead? MIT researchers have developed an app they say can analyze a car’s noises and vibrations and tell if the air filter’s clogged or the wheels are imbalanced.
The app can “empower everyday drivers to be their own Click and Clack,” says developer and mechanical engineer Joshua Siegel, referring to the hosts of the long-running NPR program Car Talk, who could famously diagnose car problems by listening to callers imitate whatever strange noise their car was making.
“Growing up in Detroit, I was surrounded by car culture,” Siegel says. “I stood in awe of the friends and family surrounding me who had a knack for being able to identify subtle problems within vehicles, from slight changes in pitch to minute vibrations in the suspension…I reasoned that if trained individuals could detect these problems accurately, mobile phones possessing the same ‘sensors’ as people could be adapted to give anyone that ‘Motor City Knack.’”
The app works by utilizing smartphones’ microphones and accelerometers, as well as their GPS systems. The microphone can be used to “hear” the whistling sound of a clogged air filter. A GPS can monitor a car’s speed which, when combined with vibration data, can tell if tires are properly inflated. The app uses machine learning to learn what sounds and vibrations signify what problems. In testing, its accuracy was above 90 percent, the team says.
To develop the app, Siegel and his colleagues rented numerous kinds of cars and temporarily "broke" them, inducing the kinds of problems they wanted to study. Then, before returning them, they’d put them back in tip-top shape with tire rotations, oil changes and so on.
A paper about the work was recently published in the journal Engineering Applications of Artificial Intelligence.
The app’s powers are currently limited to certain common problems that can be easily detected by smartphone sensors, including wheel imbalance, engine misfires, improper tire pressure and clogged air filters.
“We can’t yet replace the neighborhood mechanic,” Siegel says. “That’s because some problems require more nuanced fault tracing, or occur intermittently, or might not have a repeatable, characteristic vibration pattern.”
Perhaps unsurprisingly, some mechanics are skeptical of just how how much an app can really do.
Charles Sanville, a master certified Volkswagen technician from outside Raleigh, North Carolina, says that a given car problem might present as a "plink" in the majority of cars, but a significant minority of cars will make a totally different sound, despite having the same problem. This is where an experienced mechanic is needed.
When Sanville is diagnosing a noise in the air-conditioning, for example, he’ll first sit in the driver’s seat to listen, then move to the passenger seat, then stick his head under the dashboard, then change all the settings on the climate control system, all to see if the noise changes.
“While a microphone on an app can detect those changes, you still have to have someone who knows how to do that,” Sanville says. “I think that’s the big disconnect between how vehicles are diagnosed in a shop, in the real world, and how they’re diagnosed in the laboratory.”
Sanville says most mechanics are eager to embrace new technologies, and he believes apps could play a large role in diagnostics one day.
“But I don’t think we’re there,” he says.
Still, Siegel and his team estimate the app could save car drivers some $125 a year, and save truck drivers in the neighborhood of $600 a year. It could also save gas by making sure cars are running efficiently, and help drivers avoid blowouts and breakdowns.
Siegel founded a startup called Data Driven to bring his idea to market. A prototype app will be ready for field testing in about six months, and he's aiming to have a commercial version a year later.