Condition-Based Monitoring and Anomaly Detection of Industrial Equipment using Autoencoder

• Main Authors: Lagazo, Daniel, de Vera, Jose Alfredo, Coronel, Andrei D, Jimenez, Joseph Mark, Gatmaitan, Emman
• Format: text
• Published: Archīum Ateneo 2021
• Subjects: Computer science; Wires; Manuals; Feature extraction; Real-time systems; Data models; Manufacturing; Auto-encoder; CNN; wire equipment; MFCC; spectrogram; reconstruction error; density score; condition-based monitoring; anomaly detection; sound pressure; Computer Sciences; Databases and Information Systems; Theory and Algorithms
• Online Access: https://archium.ateneo.edu/discs-faculty-pubs/316; https://ieeexplore.ieee.org/document/9497816
• Institution: Ateneo De Manila University

Real-time Condition-based Monitoring (CbM) of wire manufacturing equipment of a partner facility involves the manual process of listening to the sound pressure of the equipment by the personnel assigned to it. This is to prevent further damage and to mitigate costs by monitoring the earliest signs of defects in the form of anomalous sound. We augmented the facility's CbM system by deploying an acoustic recorder and by building an autoencoder that is trained using the normal sound pressure of the wire extruding machine. This paper discusses a process for sound pressure acquisition, data pre-processing and preparation, feature extraction, anomaly detection, model evaluation, and case studies of downtime incidents. The objective of this paper is to automate the monitoring of the condition of the equipment and to find possible symptoms of unhealthy sound pressure prior to the reported downtimes. A comparative analysis of density score and reconstruction error, our chosen anomaly detection techniques, is presented in this paper.