Scholarlink

<p>This paper presents a novel approach for fault diagnosis in elevator door machine system, addressing the challenges posed by complex and variable operating conditions. A robust and accurate diagnostic model is proposed, leveraging the Short-Time Fourier Transform (<i>STFT</i>) for time-frequency representation and the IncepNext deep learning architecture for feature extraction and classification. The <i>STFT</i> converts raw vibration signals into spectrograms, capturing localized time-frequency information crucial for identifying subtle fault signatures. The IncepNext model, inspired by Inception and <i>EdgeNeXt</i>, employs multi-scale convolutional kernels within Inception-like modules and asymmetric convolutions for efficient and comprehensive feature learning. This architecture effectively captures local and global patterns within the spectrograms, enabling accurate fault classification. Experimental evaluation using a real-world dataset comprising six fault types under three different operating conditions demonstrates the model’s effectiveness. The proposed approach achieves a remarkable average accuracy of 94.07%, significantly outperforming baseline models employing alternative time-frequency transformations (<i>CWT</i>, <i>GAF</i>) and backbone architectures (<i>EdgeNeXt</i>, <i>ResNeXt</i>). The results highlight the synergy between <i>STFT</i> and IncepNext, offering a robust and accurate solution for fault diagnosis in elevator door machine system with significant implications for enhancing elevator safety, reliability, and predictive maintenance strategies.</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

Fault diagnosis method of elevator door machine system based on STFT-IncepNext

  • 期刊论文

  • 发表日期：

    2025年9月16日

查看全文
  • Ji Xiaosheng,
  • Zhu Zengzhen,
  • Du Chenyu,
  • Khalil AL-Bukhaiti,
  • Anping Wan

摘要

This paper presents a novel approach for fault diagnosis in elevator door machine system, addressing the challenges posed by complex and variable operating conditions. A robust and accurate diagnostic model is proposed, leveraging the Short-Time Fourier Transform (STFT) for time-frequency representation and the IncepNext deep learning architecture for feature extraction and classification. The STFT converts raw vibration signals into spectrograms, capturing localized time-frequency information crucial for identifying subtle fault signatures. The IncepNext model, inspired by Inception and EdgeNeXt, employs multi-scale convolutional kernels within Inception-like modules and asymmetric convolutions for efficient and comprehensive feature learning. This architecture effectively captures local and global patterns within the spectrograms, enabling accurate fault classification. Experimental evaluation using a real-world dataset comprising six fault types under three different operating conditions demonstrates the model’s effectiveness. The proposed approach achieves a remarkable average accuracy of 94.07%, significantly outperforming baseline models employing alternative time-frequency transformations (CWT, GAF) and backbone architectures (EdgeNeXt, ResNeXt). The results highlight the synergy between STFT and IncepNext, offering a robust and accurate solution for fault diagnosis in elevator door machine system with significant implications for enhancing elevator safety, reliability, and predictive maintenance strategies.