TY - JOUR

T1 - Deep learning algorithm for supervision process in production using acoustic signal

AU - Safaei, Mahmood

AU - Soleymani, Seyed Ahmad

AU - Safaei, Mitra

AU - Chizari, Hassan

AU - Nilashi, Mehrbakhsh

PY - 2023/10

Y1 - 2023/10

N2 - In an industrial environment, accurate fault diagnosis of machines is crucial to prevent shutdowns, failures, maintenance costs, and production downtime. Existing methods for system failure prevention are often unsatisfactory and expensive, prompting the need for alternative approaches. Acoustic signals have emerged as a new method for predicting machine component lifespan, but recognizing relevant features and distinguishing them from noise remains challenging. To address the aforementioned challenges, we present a comprehensive model that integrates various components to enhance the accuracy and effectiveness of machine process identification. The proposed model incorporates a deep learning algorithm, which enables the forecasting of machine operation based on acoustic signals. In addition, we employ a customized Continuous Wavelet Transformation (CWT) technique to convert the acoustic signals into CWT images, preserving vital information such as signal amplitude. This transformation allows for a more comprehensive analysis and representation of the acoustic data. Furthermore, a Convolutional Neural Network (CNN) is utilized as a powerful classifier to accurately classify and differentiate between different machine processes based on the extracted features from the CWT images. By combining these elements, our model provides a robust and efficient framework for machine process identification using acoustic signals. Testing our model on a dataset generated from the Institute for Manufacturing Technology and Machine Tools (IFW) for the Gildemeister machine (CTX420 linear), we achieve over 97% accuracy in discovering and early detecting emerging faults and machine processes based on acoustic signals.

AB - In an industrial environment, accurate fault diagnosis of machines is crucial to prevent shutdowns, failures, maintenance costs, and production downtime. Existing methods for system failure prevention are often unsatisfactory and expensive, prompting the need for alternative approaches. Acoustic signals have emerged as a new method for predicting machine component lifespan, but recognizing relevant features and distinguishing them from noise remains challenging. To address the aforementioned challenges, we present a comprehensive model that integrates various components to enhance the accuracy and effectiveness of machine process identification. The proposed model incorporates a deep learning algorithm, which enables the forecasting of machine operation based on acoustic signals. In addition, we employ a customized Continuous Wavelet Transformation (CWT) technique to convert the acoustic signals into CWT images, preserving vital information such as signal amplitude. This transformation allows for a more comprehensive analysis and representation of the acoustic data. Furthermore, a Convolutional Neural Network (CNN) is utilized as a powerful classifier to accurately classify and differentiate between different machine processes based on the extracted features from the CWT images. By combining these elements, our model provides a robust and efficient framework for machine process identification using acoustic signals. Testing our model on a dataset generated from the Institute for Manufacturing Technology and Machine Tools (IFW) for the Gildemeister machine (CTX420 linear), we achieve over 97% accuracy in discovering and early detecting emerging faults and machine processes based on acoustic signals.

KW - Acoustic

KW - Deep learning

KW - Fault diagnosis

KW - Production

UR - http://www.scopus.com/inward/record.url?scp=85166482751&partnerID=8YFLogxK

U2 - 10.1016/j.asoc.2023.110682

DO - 10.1016/j.asoc.2023.110682

M3 - Article

AN - SCOPUS:85166482751

VL - 146

JO - Applied soft computing

JF - Applied soft computing

SN - 1568-4946

M1 - 110682

ER -