ISLAB/CAISR:Fault detection using acoustic signals through anomaly detection

From ISLAB/CAISR
Title Fault detection using acoustic signals through anomaly detection
Summary Fault detection using acoustic signals through anomaly detection
Keywords Fault detection, accoustic, anomaly detection
TimeFrame Fall 2024
References [[References::[1] Coelho, Gabriel, et al. "Deep autoencoders for acoustic anomaly detection: experiments with working machine and in-vehicle audio." Neural Computing and Applications 34.22 (2022): 19485-19499.

[2] Di Fiore, Emanuele, et al.”An anomalous sound detection methodology for predictive maintenance.“ Expert Systems with Applications 209 (2022): 118324.

[3] Jombo, Gbanaibolou, and Yu Zhang "Acoustic-Based Machine Condition Monitoring—Methods and Challenges" Eng 4:1 (2023): 47-79.

[4] Pacheco-Chérrez, Josué, et al. "Bearing fault detection with vibration and acoustic signals: Comparison among different machine leaning classification methods." Engineering Failure Analysis 139 (2022): 106515.

[5] Pang, Guansong, et al. "Deep learning for anomaly detection: A review." ACM computing surveys (CSUR) 54.2 (2021): 1-38

[6] Park, Sebin, et al. "Measurement Noise Recommendation for Efficient Kalman Filtering over a Large Amount of Sensor Data" Sensors 19.5 (2019): 1168. [7] Surucu, Onur, Stephen Andrew Gadsden, and John Yawney. "Condition Monitoring using Machine Learning: A Review of Theory, Applications, and Recent Advances." Expert Systems with Applications 221 (2023): 119738. [8] MIMII Dataset: Sound Dataset for Malfunctioning Industrial Machine Investigation and Inspection]]

Prerequisites
Author
Supervisor Elena Haller, Peyman Mashhadi
Level Master
Status Open


Acoustic signal analysis has gained significant attention as a non-intrusive and effective method for fault detection in a variety of industrial applications [4,5]. Acoustic sensors (e.g., microphones) are quick and cheap to implement and do not require physical contact with the machine. In this context the fault detection is reduced to anomaly detection and there are several deep learning methods [2,3,5] that can be applied to classify normal and anomalous machine sounds. In all scenarios the main challenges are background noise and changes in machine operating conditions.

In this project, the aim is to use deep learning based anomaly detection methods (e.g. “deep isolation forest”, “variational autoencoder” [7]) in combination with noise reduction techniques [3,6] to detect faults in industrial machines [1]. The benchmark dataset to this end is Malfunctioning Industrial Machine Investigation and Inspection (MIMII) [8].