These key driving metrics – when cars need a servicing, air filters need changing, wheel balancing is required, or a new set of tires are needed – can be controlled by using a smartphone thanks to new software developed by computer scientists at the Massachusetts Institute of Technology.
It is hoped that the software will be commercialized and used by car manufacturers within the next two years. The idea is to provide drivers with useful diagnostic information within just a few minutes. This would not only be with the driver’s own car, but in any car that the driver drives or is a passenger in. This is possible because the software assesses everything about a car’s health through analyzing the sounds and vibrations that the car makes. A smartphone detects this through a standard microphone and accelerometer.
Such an application is not simply a cool gadget for drivers to have, it also brings economic benefits for the private driver and for businesses. The developers estimate that their software, in the form of a smartphone app, would save a typical car driver $125 and a truck driver around $600 per year.
In trials, according to lead researcher Dr. Joshua Siegel, the software performed optimally when the hosting smartphone is placed on to the dashboard of the vehicle. Here the accuracy rating, when compared with car service records, was assessed as above 90 percent. Moreover, there were no ‘false positives’ (that is faults signaled by the software which were later found to be incorrect).
As to how this works, with an air filter, an engine’s sounds provide telltale signs of how clogged an air filter is becoming. Big data analytics can then indicate when the filter needs to be changed. Changing a filter at the optimal time improves an engine’s performance. The software gained experience of this through machine learning, where thousands of sounds were captured and interpreted. For this the researchers used Mel-Cepstrum, Fourier and Wavelet features as input to construct a classification model. The software was then programmed to apply feature ranking to select the best-differentiating features for air filter sounds.
The development of the software has been reported to the journal Engineering Applications of Artificial Intelligence. The peer reviewed paper is titled “Air filter particulate loading detection using smartphone audio and optimized ensemble classification.”
