Scholarlink

<p>The importance of direct current motors is crucial in many industrial applications, and their malfunctioning can lead to operational interruptions and to financial loss, accordingly. The present research introduces a novel approach that transforms audio signals to Mel spectrograms and subsequently applies convolutional neural networks (CNNs) to interpret them. The model is optimized through an enhanced version of the elk herd optimizer. A thorough comparative analysis across different metaheuristic algorithms to optimize the CNN parameters indicates that the proposed approach leads to superior performance. Various experimentation scenarios are evaluated, such as targeting malfunction detection (binary classification), as well as malfunction identification (multi-class classification) and the grayscale and the color Mel spectrograms are used, in turn. The results of the CNN variants optimized by the various metaheuristics are compared to the ones of the proposed approach in all scenarios, and their superiority is statistically validated. The study also highlights the role of optimization algorithms in increasing the quality of the results obtained by machine learning models in real-world applications.</p>

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

Evaluation Performance of Metaheuristics-Tuned Convolutional Neural Networks for Direct Current Motor Using Mel Spectrograms

  • 期刊论文

  • 发表日期：

    2025年12月20日

查看全文
  • Aleksandar Petrovic,
  • Catalin Stoean,
  • Ruxandra Stoean,
  • Luka Jovanovic,
  • Nebojsa Bacanin,
  • Vladimir Simic,
  • Milos Antonijevic,
  • Miodrag Zivkovic,
  • Stefana Janicijevic

摘要

The importance of direct current motors is crucial in many industrial applications, and their malfunctioning can lead to operational interruptions and to financial loss, accordingly. The present research introduces a novel approach that transforms audio signals to Mel spectrograms and subsequently applies convolutional neural networks (CNNs) to interpret them. The model is optimized through an enhanced version of the elk herd optimizer. A thorough comparative analysis across different metaheuristic algorithms to optimize the CNN parameters indicates that the proposed approach leads to superior performance. Various experimentation scenarios are evaluated, such as targeting malfunction detection (binary classification), as well as malfunction identification (multi-class classification) and the grayscale and the color Mel spectrograms are used, in turn. The results of the CNN variants optimized by the various metaheuristics are compared to the ones of the proposed approach in all scenarios, and their superiority is statistically validated. The study also highlights the role of optimization algorithms in increasing the quality of the results obtained by machine learning models in real-world applications.