Machine Learning Framework for Industrial Machine Sound Classification in Predictive Maintenance
IEEE Access · 2025 · 13 · 154960-154975
DOI: 10.1109/access.2025.3601999 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Access, 2025. Verfasst von Nur Fatinah Md Hafiz und 4 weiteren Autor:innen. Die Arbeit wurde bisher 6-mal zitiert (Crossref) und nennt 63 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Universiti Putra Malaysia.
Abstract (Originalsprache)
Predictive maintenance, utilising anomalous sound classification, demonstrates a strong potential to identify mechanical faults in industrial machinery. This research proposes a machine learning-based framework for classifying anomalous sounds in industrial machines, with a particular focus on CT scan machines and fan units. The study utilises both real-world data from CT scan machine sound and the Malfunctioning Industrial Machine Investigation and Inspection (MIMII) dataset. It offers a comprehensive analysis of sound signal processing techniques, synthetic data generation methods, feature extraction processes, and classification using machine learning models to support predictive maintenance applications. In this research, sound data from a CT scan machine was collected using an Internet of Things (IoT) connected microphone located on the machine in a Klang Valley hospital. Due to the limited availability of faulty condition data, synthetic anomalous data for both operational and non-operational conditions were generated using a noise injection method. Features derived from Mel Frequency Cepstral Coefficients (MFCCs) and Mel Spectrogram representations were employed to analyse the sound data. The dataset for CT scan machine sounds is categorised into four distinct classes: anomalous operational sound (Aop), anomalous non-operational sound (Anop), normal operational sound (Nop), and normal non-operational sound (Nnop). In contrast, the MIMII dataset is classified into two categories: normal and abnormal. A Convolutional Neural Network (CNN) model was used for a sound classification system, achieving training accuracies of 98.22% with Mel spectrogram features and 98.12% with MFCC features. The results emphasise the possibility of using CNN-based sound classification to effectively anticipate and maintain CT scan machines. This finding also has the potential to be applied to predictive maintenance applications by detecting both normal and anomalous operating sounds in industrial machinery.
Förderung
- Universiti Putra Malaysia
Zitieren
APA
Hafiz, N. F. M., Mashohor, S., Shazril, M. H. S. E. M. A., Ali, A. M., & Rasid, M. F. A. (2025). Machine Learning Framework for Industrial Machine Sound Classification in Predictive Maintenance. IEEE Access, 13, 154960-154975. https://doi.org/10.1109/access.2025.3601999
Vancouver
Hafiz NFM, Mashohor S, Shazril MHSEMA, Ali AM, Rasid MFA. Machine Learning Framework for Industrial Machine Sound Classification in Predictive Maintenance. IEEE Access. 2025;13:154960-154975. doi:10.1109/access.2025.3601999
BibTeX
@article{Hafiz2025,
author = {Hafiz, Nur Fatinah Md and Mashohor, Syamsiah and Shazril, Mohammad Habib Shah Ershad Mohd Azrul and Ali, Azizi Mohd and Rasid, Mohd Fadlee A.},
title = {Machine Learning Framework for Industrial Machine Sound Classification in Predictive Maintenance},
journal = {IEEE Access},
year = {2025},
volume = {13},
pages = {154960--154975},
doi = {10.1109/access.2025.3601999}
}
EUVIMED