The Massachusetts Institute of Technology (MIT) is developing an app that uses a phone's microphone and accelerometer to diagnose a vehicle's potential problems. It will enable wannabe mechanics and the safety-conscious, but could also be useful to others.
MIT researchers imagine a scenario where a ride-share passenger who is using the app tells the driver that the car might need a filter change or some new spark plugs. The reality would carry the technology much further, though, as everyday car owners could use it to become more intimately familiar with the nuances of their vehicle and its needs. It would also empower used car buyers.
The app is outlined in a series of papers published in the journal Engineering Applications of Artificial Intelligence with co-authors Joshua Siegel, PhD, and Sanjay Sarma, Professor of Mechanical Engineering at MIT, along with others. It outlines the development of the app's diagnosis algorithms. The high sensitivity of today's smartphones, Siegel explains, allows detecting relevant signals "without needing any special connection."
A smartphone in the hand, for example, can detect the vehicle's exhaust note, some pings, and so forth. A phone mounted to the vehicle's dashboard can do even more. The accuracy of diagnostic results so far, Siegel says, is better than 90 percent. The app is 100 percent accurate for misfire detection, he adds, and false positives are not happening at all.
The latest publication from the researchers outlines the app's results in diagnosing air filter status. An engine's sounds, they've found, can reveal how clogged that intake filter might be and when it needs to be changed. When a filter needs changing, Siegel says, the app can tell the car is starting to "snore." While humans can't differentiate the difference in sound over all other engine noises, a phone's microphone can.
The team is testing and adding new diagnostic systems one at a time. They then try to fool the app's AI by adding or changing the sounds it's receiving or by muddying the sound waters with background noise. The team rents perfectly good cars and then induces problems, attempts to have the AI diagnose them, fixes them, and moves on. "I've rented cars and given them new air filters, balanced their tires, and done an oil change," Siegel says.
Data combinations are the key to the app's AI and its ability to diagnose a problem accurately. For example, to determine if tires are overinflated or going bald, the app monitors the phone's GPS to determine speed. Vibration data is then added to determine how fast the wheels are turning. That can be used to calculate the wheel's diameter, which can then be compared to the diameter expected if the tire were new and properly inflated. Multiple recordings of sound from the vehicle can be used over time to diagnose further problems as well.
A prototype of the app is being developed for field testing and is expected to begin trials in the next six months, the MIT team says. A commercial version should be on the market a year or so after that. It will be offered through a startup company called Data Driven, founded by Dr. Siegel.