Scholarlink

Background <p>Accurate and timely detection of drill bit failure is essential for enhancing the reliability, safety, and productivity of CNC machining operations. Conventional fault-diagnosis approaches either rely heavily on handcrafted signal features or deep learning models that require sequence inputs or image transformations, limiting interpretability and adaptability to real industrial conditions.</p> Purpose <p>This study aims to develop and validate a hybrid deep learning model capable of leveraging both handcrafted features and deep feature learning to achieve high-accuracy, interpretable fault diagnosis of drill bit conditions in CNC machining environments.</p> Methods <p>A novel dual-stream architecture, GRN-AttnNet, is proposed. It independently processes vibration-signal features extracted from the time domain and frequency domain using Gated Residual Networks (GRN). A feature-wise self-attention mechanism is integrated to dynamically learn feature importance across domains, capturing high-order nonlinear interactions without relying on raw sequence data or signal-to-image transformations. Experiments were conducted on vibration data collected from a vertical milling center (XTRON-544) equipped with a piezoelectric accelerometer. Multiple drill bit failure modes—including outer corner wear, flank wear, crater wear, chisel edge wear, margin wear, and edge chipping—were evaluated. Performance was compared with standard baseline classifiers.</p> Results <p>GRN-AttnNet achieved a classification accuracy of 98%, outperforming all baseline models tested. The dual-stream GRN architecture and attention mechanism improved feature representation and interpretability, enabling robust differentiation across multiple drill bit failure modes.</p> Conclusion <p>The proposed GRN-AttnNet model effectively bridges handcrafted feature engineering with advanced deep learning, delivering interpretable and highly accurate fault diagnosis for CNC drill bit monitoring. Its strong performance and domain-aware feature learning demonstrate its potential for deployment in real industrial machining environments to enhance reliability and operational efficiency.</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

A Dual-Stream Gated Attention Network for Fault Diagnosis of Drill Bit Using time and Frequency Domain-based Vibration Features

  • 期刊论文

  • 发表日期：

    2025年11月25日

查看全文
  • Chandan M. N.,
  • Himadri Majumder,
  • Avinash Badadhe

摘要

Background

Accurate and timely detection of drill bit failure is essential for enhancing the reliability, safety, and productivity of CNC machining operations. Conventional fault-diagnosis approaches either rely heavily on handcrafted signal features or deep learning models that require sequence inputs or image transformations, limiting interpretability and adaptability to real industrial conditions.

Purpose

This study aims to develop and validate a hybrid deep learning model capable of leveraging both handcrafted features and deep feature learning to achieve high-accuracy, interpretable fault diagnosis of drill bit conditions in CNC machining environments.

Methods

A novel dual-stream architecture, GRN-AttnNet, is proposed. It independently processes vibration-signal features extracted from the time domain and frequency domain using Gated Residual Networks (GRN). A feature-wise self-attention mechanism is integrated to dynamically learn feature importance across domains, capturing high-order nonlinear interactions without relying on raw sequence data or signal-to-image transformations. Experiments were conducted on vibration data collected from a vertical milling center (XTRON-544) equipped with a piezoelectric accelerometer. Multiple drill bit failure modes—including outer corner wear, flank wear, crater wear, chisel edge wear, margin wear, and edge chipping—were evaluated. Performance was compared with standard baseline classifiers.

Results

GRN-AttnNet achieved a classification accuracy of 98%, outperforming all baseline models tested. The dual-stream GRN architecture and attention mechanism improved feature representation and interpretability, enabling robust differentiation across multiple drill bit failure modes.

Conclusion

The proposed GRN-AttnNet model effectively bridges handcrafted feature engineering with advanced deep learning, delivering interpretable and highly accurate fault diagnosis for CNC drill bit monitoring. Its strong performance and domain-aware feature learning demonstrate its potential for deployment in real industrial machining environments to enhance reliability and operational efficiency.