Audio-based detection of malfunctioning machines using deep convolutional autoencoders
In this paper, a modular deep convolutional autoencoder with a dense bottleneck structure is developed to perform the task of unsupervised anomaly detection in machine operating sounds. The proposed model consists of multiple sub-networks with identical encoder-decoder structures, trained to learn a mapping function between different mel-scaled frequency bands. Experiments are conducted on the recently introduced MIMII (Malfunctioning Industrial Machine Inspection and Investigation) open benchmark dataset. Results demonstrate that the proposed model yields improved fault detection performance in terms of the Area Under Curve (AUC) metric compared to the baseline approach.
Authors: Iordanis Thoidis (Aristotle University of Thessaloniki), Marios Giouvanakis (Aristotle University of Thessaloniki) and George Papanikolaou (Aristotle University of Thessaloniki)
