AutoMotorHelp – Motor Fault Detection Using Sound Patterns
Entities: Audio and Acoustics Laboratory
Machine fault detection is often done by ‘ear’. That is, by analyzing sound patterns, the proper or improper functioning of a motorized system can, in certain situations, be conveniently assessed.
Predictive machine maintenance anticipates failures, and actions can include corrective actions, system replacement, or even planned failure, as opposed to condition-based maintenance, which only reports if something is wrong ‘here and now’. This can lead to significant cost savings (problems can be detected a priori without the need to stop machines), greater predictability, and greater system availability. The Internet of Things (IoT) and big data concepts offer many opportunities for industries to manage machine maintenance permanently and in a decentralized way, partly at the edge (edge/fog computing) and partly in the cloud, with a significant reduction in costs.
This project combines machine learning and artificial intelligence algorithms, enabling the early detection and prediction of mechanical malfunctions in machines. It utilizes an on-edge computing approach with neural networks on low-cost platforms as a first level of analysis of machine sound patterns.