Scholarlink

In a multitude of industrial sectors, bearingless direct current (DC) motors hold a pivotal position, demanding meticulous maintenance protocols to avert costly breakdowns. Conventional maintenance strategies often rely on periodic inspections, which may lead to inefficiencies and unexpected downtimes. To confront these challenges head-on, this study introduces a novel methodology that seamlessly integrates audio analysis with state-of-the-art machine learning (ML) techniques. Leveraging the power of convolutional neural networks (CNNs), the research endeavors to discern intricate patterns embedded within the audio signals emitted by bearingless DC motors. With the goal of establishing better classification performance, the final layer of CNN is forwarded as an input to extreme gradient boosting (XGBoost) classifier. Furthermore, the study harnesses metaheuristic optimization algorithms to finely calibrate the parameters of the CNN and XGBoost models, introducing a bespoke optimizer specifically tailored to the objectives of this investigation. Additionally, for the purpose of conducted research, recently emerged sinh cosh optimizer (SCHO) is modified by employing hybridization technique. Rigorous testing of the proposed framework is conducted using a publicly available dataset. The introduced modified metaheuristics along with other cutting-edge algorithms used in comparative analysis exhibit commendable performance, achieving an accuracy of 100% with a cohen kappa coefficient of 1.0 with the most proficient model identified.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Hybrid CNN and XGBoost Model Tuned by Metaheuristics for DC Motor Failure Prediction Using Mel Spectograms

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Luka Jovanovic,
  • Vesna Radojcic,
  • Milos Dobrojevic,
  • Miroslav Cajic,
  • Miodrag Zivkovic,
  • Nebojsa Bacanin

摘要

In a multitude of industrial sectors, bearingless direct current (DC) motors hold a pivotal position, demanding meticulous maintenance protocols to avert costly breakdowns. Conventional maintenance strategies often rely on periodic inspections, which may lead to inefficiencies and unexpected downtimes. To confront these challenges head-on, this study introduces a novel methodology that seamlessly integrates audio analysis with state-of-the-art machine learning (ML) techniques. Leveraging the power of convolutional neural networks (CNNs), the research endeavors to discern intricate patterns embedded within the audio signals emitted by bearingless DC motors. With the goal of establishing better classification performance, the final layer of CNN is forwarded as an input to extreme gradient boosting (XGBoost) classifier. Furthermore, the study harnesses metaheuristic optimization algorithms to finely calibrate the parameters of the CNN and XGBoost models, introducing a bespoke optimizer specifically tailored to the objectives of this investigation. Additionally, for the purpose of conducted research, recently emerged sinh cosh optimizer (SCHO) is modified by employing hybridization technique. Rigorous testing of the proposed framework is conducted using a publicly available dataset. The introduced modified metaheuristics along with other cutting-edge algorithms used in comparative analysis exhibit commendable performance, achieving an accuracy of 100% with a cohen kappa coefficient of 1.0 with the most proficient model identified.