Compact Acoustic Signal Portraits for Resource-Efficient Mechanical Fault Detection

Abstract

A resource-efficient semi-analytical method for mechanical fault detection via acoustic signal analysis is presented. A compact signal portrait is constructed by tracking dominant salient-frequency peaks over time, without a priori assumptions on peak locations. Peak feature statistics are aggregated into a 3D matrix that encodes nominal oscillatory modes. Fault detection is achieved by measuring the deviation between current and healthy portraits using a single decision threshold. Experiments on internal combustion engine recordings demonstrate improved separability and lower computational cost compared to empirical mode decomposition, confirming suitability for resource-constrained monitoring.