Scholarlink

Purpose <p>This study employs LeNet5 - a specialized CNN architecture to detect faults in cutting tool condition by utilizing vibration signals and adapting a pre-trained model from its source domain.</p> Method <p>In the initial stage, a large dataset of vibration signals was procured, wherein stable and clearly defined vibrations were indicative of normal tool conditions, while fluctuations were observed in faulty conditions. The data underwent pre-processing procedures, following which a pre-trained model was employed for the extraction of feature representations. This deep neural network was subsequently subjected to fine-tuning using the provided dataset. The study then utilizes the pre-trained model on a larger dataset to extract feature representations. The acquired knowledge is subsequently applied to train and evaluate a target model on a smaller dataset. This process expedited the convergence of the model and augmented its ability to detect faults.</p> Results <p>Once the model was refined, it was deployed to actively monitor vibrations in real-time. Faults or deviations from typical tool conditions were promptly identified and flagged, enhancing the ability to prevent unforeseen tool failures and effectively enhance the overall maintenance of machinery.</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

Adapting LeNet-5 with Transfer Learning for Fault Diagnosis in Cutting Tools Using Vibration Signals

  • 期刊论文

  • 发表日期：

    2025年6月10日

查看全文
  • Keshav Jatakar,
  • Varsha Shah,
  • Sachin Kandharkar,
  • A. D. Patange

摘要

Purpose

This study employs LeNet5 - a specialized CNN architecture to detect faults in cutting tool condition by utilizing vibration signals and adapting a pre-trained model from its source domain.

Method

In the initial stage, a large dataset of vibration signals was procured, wherein stable and clearly defined vibrations were indicative of normal tool conditions, while fluctuations were observed in faulty conditions. The data underwent pre-processing procedures, following which a pre-trained model was employed for the extraction of feature representations. This deep neural network was subsequently subjected to fine-tuning using the provided dataset. The study then utilizes the pre-trained model on a larger dataset to extract feature representations. The acquired knowledge is subsequently applied to train and evaluate a target model on a smaller dataset. This process expedited the convergence of the model and augmented its ability to detect faults.

Results

Once the model was refined, it was deployed to actively monitor vibrations in real-time. Faults or deviations from typical tool conditions were promptly identified and flagged, enhancing the ability to prevent unforeseen tool failures and effectively enhance the overall maintenance of machinery.