eprintid: 20435
rev_number: 3
eprint_status: archive
userid: 1
dir: disk0/00/02/04/35
datestamp: 2026-01-12 12:18:29
lastmod: 2026-01-12 12:18:29
status_changed: 2026-01-12 12:18:29
type: conference_item
metadata_visibility: show
creators_name: Lim, Jason Jing Wei
creators_name: Ooi, Boonyaik Yaik
creators_name: Liew, Soung Yue
creators_name: Cheng, Wai Khuen
title: Comparative Analysis of Machine Learning Techniques for Acoustic Machine Tracking under Different Signal Durations
ispublished: pub
keywords: Acoustic waves; Learning algorithms; Machine learning; Tracking (position); Legacy machine monitoring; Machine learning techniques; Machine monitoring; Operation status; Operation status tracking; Signal duration; Sound recognition; Status tracking; Tracking system; Training dataset; Internet of things
note: Cited by: 1
abstract: The concept of smart factory introduced in Industrial Revolution 4.0 raises demand for purchasing advanced machines among the industry. However, the idea to replace the outdated machines with limited functionality is not viable for small and medium-sized enterprises as those machines are working optimally through decades. Therefore, in order to alleviate low productivity output issue of the old machines, a machine activity tracking system via non-contact acoustic approach is proposed as an alternative option that saves purchasing and implementation costs altogether through simple installation of durable sensor onto the machines. The paper suggests that the alteration of acoustic signal duration potentially affect three critical attributes: tracking accuracy, training time and inference time that determine the effectiveness of the tracking system. Thus, the objective of the paper is to analyze the influence of the signal duration on the key attributes for various machine learning techniques in search of the best suited machine tracking technique. The paper also investigates the limitation of the acoustic tracking approach in terms of training dataset size. The experimental result had shown that with 4 seconds acoustic signal, the logistic regression technique with one-vs-rest scheme and liblinear solver yields the best f1-score of 0.909 and standard deviation of 0.013. The technique also performs significantly fast in terms of training and inference times for recognizing machines' sound. Additionally, most techniques grant f1-score near to 0.8 with 6 minutes training dataset. © 2022 IEEE.
date: 2022
publisher: Institute of Electrical and Electronics Engineers Inc.
official_url: https://www.scopus.com/inward/record.uri?eid=2-s2.0-85141780292&doi=10.1109%2FAiDAS56890.2022.9918703&partnerID=40&md5=8355dcd7b91c73604aff15d3480abad3
id_number: 10.1109/AiDAS56890.2022.9918703
full_text_status: none
pagerange: 1 - 6
refereed: TRUE
isbn: 9781665491648
citation:   Lim, Jason Jing Wei and Ooi, Boonyaik Yaik and Liew, Soung Yue and Cheng, Wai Khuen  (2022) Comparative Analysis of Machine Learning Techniques for Acoustic Machine Tracking under Different Signal Durations.  In: UNSPECIFIED.