Machine Hearing for Industrial Acoustic Monitoring using Cochleagram and Spiking Neural Network

This paper presents a bio-inspired machine learning framework, which aims to mimic the human hearing functionalities, for industrial acoustic monitoring. It involves firstly modelling the functionality of the cochlea, which is an essential part of the inner ear. This is accomplished by extracting important time-frequency information of the acoustic signals through cochleagrams. Then, to emulate more closely the neural activities in the brain when processing information, a bio-plausible Spiking Neural Network (SNN) is applied for pattern recognition. Finally, the proposed method is verified with acoustic data collected from machine bearings with healthy and faulty conditions. The initial feasibility study has demonstrated the viability and the efficacy of the proposed “machine hearing” approach for industrial acoustic monitoring applications.