Research on the characterization of acoustic signals and the detection of anomalies in machines and equipment

This research is dedicated to the in-depth exploration of the detection of unusual machine-generated sounds. By conducting a comprehensive analysis of both normal and abnormal sound data collected from six machines obtained from the ToyADMOS and MIMII datasets, we have laid the foundation for our study. Our investigation involved a multidimensional analysis of the acoustic profiles of each machine, examining key parameters such as root-mean-square energy, spectral center of mass, over-zero rate, mel features, MFCC (Mel-frequency cepstral coefficients), log-mel features, spectral contrast, chromaticity features, and hue center of mass. Our experimental results have unveiled significant variations among these acoustic features across different machines and sound states, providing invaluable insights for the detection of sound anomalies. This study serves as a pivotal stepping stone for further advancements in the fields of sound signal processing and anomaly detection, while also guiding the direction of future research endeavors.