Listen carefully: How algorithms can help detect car gremlins

By Pritesh Ruthun Time of article published Jul 20, 2018

Share this article:

New York - Already, automotive technicians are able to detect what's wrong with a car’s engine without raising the bonnet. In the near future, it's expected that deep learning algorithms  will be able to tell you about a potential mechanical issue before it causes a breakdown.

Remember those old-school mechanics that would prop up a car's bonnet, ‘listen for the problem’, and sort out the issue before it turned into a nasty mechanical failure? They are harder and harder to come by these days. In fact, auto-tech has evolved, and workshop technicians (no longer called mechanics) plug in to vehicles with diagnostics tools, without having to get their hands greasy. They repair as needed, and customers drive away happy, if they can afford the fix. 

In the future though, deep-learning technologies are expected to tell consumers about a mechanical issue in a car, truck or home or industrial machine before it causes any serious problems.  

“We’re developing an expert mechanic’s brain that identifies exactly what is happening to a machine by the way that it sounds,” says Amnon Shenfeld, founder and CEO of 3DSignals, a start-up based in Israel. 3DSignals is already using machine learning to train computers to ‘listen’ to industrial machinery and diagnose problems at facilities like hydro-electric plants and steel mills.
Amnon Shenfeld, CEO of 3DSignals based in Israel
Sure, this technology will work well in heavy industries, but it’s the automotive sector that can truly leverage it to improve the motoring experience for customers. Imagine a luxury SUV that’s able to ‘listen to’ and self-diagnose a mechanical problem, report it to the dealer and, maybe, self-drive itself there for maintenance.  This sort of scenario isn’t unreal. 

Listening for changes in frequency

3DSignals in Israel is training computers to listen for changes in sound, vibration, and heat. These are just a few variables in car engines that can give you an idea of whether a car is running efficiently or about to rattle a piston loose. The hope is that the computer 'brain' can catch mechanical failures before they happen, saving on repair costs and reducing downtime.

Focused on using its machine learning tools to assimilate acoustic data, an area the company’s founder says is, “Surprisingly neglected outside of speech recognition,”  ultrasonic wireless sensors are placed near rotating parts and equipment to listen for unusual noises. The sensors feed data to signal processing tools that can then send an alert to a driver, customer, fleet manager or plant manager. With this information on the strange sound, the problem area can be investigated and maintained to avoid catastrophe. 

No more guessing games

“There is nothing more frustrating than taking your car into the mechanic with only the vaguest sense that something is wrong. You know that odd little creak or strange whine is new, but you don’t have a clue what it’s trying to tell you. You would like to know before it becomes a serious problem, and an expert is going to charge you a lot of money to find out,” says Ben Popper at The Verge.

Devices that listen and diagnose mechanical creaks and groans then can be a cost-effective solution where pricey experts are concerned. New-York-based start-up Augury, is working on listening devices with deep learning capability that can inform consumers of strange noises in their appliances and personal devices. 
The Halo device by Augury is a continuous diagnostics tool that listens for sound anomalies
Solutions such as Augury’s and 3DSignals’ could actually help (pro-active) car companies offer predictive maintenance products and services in the future.  Instead of preventative maintenance, which can help in certain instances if customers don’t mind spending money on parts that might not have needed changing, car makers can predict when maintenance is going to be needed by ‘listening’ to what the car needs to run right.

Share this article: