The DeepTech Lab at Michigan State University conducts research into the use of machine learning algorithms and vibroacoustic (sound and acceleration) signals to improve the maintenance of diverse physical systems.

An estimated 290 million vehicles operate below optimal efficiency due to delayed service or unaddressed faults. The DeepTech Lab has developed algorithms that can identify the make, model, and certain fault types for vehicles using audio captured by mobile devices.

Developed under the supervision of Dr. Josh Siegel, the Data-Driven Mechanic application helps everyday people benefit from the DeepTech Lab’s algorithms to improve the maintenance and care of their vehicles.

Users record audio and vibration data from their vehicle in our software with the native microphone and accelerometer on their mobile device.

When the user selects the classify option, the data are processed by the DeepTech Lab’s algorithms, and the results are displayed to the user, outlining details and detected faults of their vehicle.

When the user chooses the annotate option, they are shown a series of form pages populated with dynamic fields to fill in with information about the vehicle. After annotation, the labeled data are stored in a server for later use by the DeepTech Lab. The annotate function enables users to participate and contribute to the improvement of the DeepTech Lab’s algorithms for diagnosing vehicles and other physical systems.

Our software runs on both Android and iOS mobile devices that have a built-in microphone and accelerometer. The front end is written in JavaScript using the React Native Expo framework. The back-end server is written in Python Flask and the underlying database is MySQL.