Résumé:
Anomalous sound detection (ASD) is the task to identify whether the sound emitted from a
target machine is normal or anomalous. Automatically detecting mechanical failure is an essential
technology in the fifth industrial revolution, including artificial intelligence -based factory
automation. Prompt detection of machine anomaly by observing its sounds may be useful for
machine condition monitoring. The goal of this work is to explore two different classes of ASD
systems : Unsupervised and Supervised ASD using the long-mel energies features. The unsupervised
ASD approach consists of a deep AutoEncoder, whereas, the Supervised ASD system
implements the outlier exposed strategy based on a deep Residual Neural Network (ResNet) .
We have performed multiple experiments using a huge dataset which consists of 54, 254 sound
files, and supported our analysis and discussion with numerous statistical tests to analyze and
compare the two ASD systems. We have dedicated experiments for investigating the impact
of varying some hyperparameters of the autoencoder architecture, like the code (or bottleneck)
size Our experimental findings indicate that the deep residual network (ResNet) outperforms
the autoencoder model.
Keywords: Anomaly Sound Detection, Deep Learning, Statistical Test, Feature Extraction,
Machine Condition Monitoring.
