Erliang Liu, Cong Liu, Yuhang Du, Baiwei Zhu, Limin Shi

Research on Tool Wear State Recognition Method Based on Multi-Scale Feature Extraction and Deep Residual Network Fusion

Introduction

Research on tool wear state recognition method based on multi-scale feature extraction and deep residual network fusion. Enhance machining with a Multi-scale ResNet deep learning model for accurate tool wear state recognition. It uses CWT on cutting signals, achieving 93.8% accuracy and outperforming traditional methods.

Abstract

To enhance the intelligence of machining processes, accurate recognition of tool wear states has become a key issue in the manufacturing field. However, due to the non-stationary and high-dimensional nature of cutting signals, traditional methods face significant challenges in feature extraction and state classification. In the context of cutting processes, challenges such as difficulty in identifying tool wear states and the complex composition of monitoring information features persist. To address these issues, this paper proposes a deep learning model that integrates multi-scale feature extraction with a residual connection network (Multi-scale ResNet). Specifically, cutting vibration signals are processes using continuous wavelet transform (CWT), which enables the conversion of time-frequency information into images. The proposed deep learning model is then used for feature extraction and state identification. The proposed model is validated through cutting experiments conducted on γ-TiAl alloys. Experimental results show that the Multi-scale ResNet model achieves higher recognition accuracy than traditional models such as convolutional neural networks – support vector machines (CNN–SVM), Transformer, and ResNet in the initial and normal wear stages. It effectively mitigates misjudgments associated with initial and normal wear, achieving a prediction accuracy of 93.8 %, a recall rate of 94.2 %, and an F1 score of 94 %. This model offers a novel and effective approach for tool wear state monitoring, contributing to improved cutting processing efficiency and increased intelligence in production.

Review

This paper presents a timely and relevant contribution to the critical domain of intelligent manufacturing by addressing the persistent challenge of accurately recognizing tool wear states. The authors propose a novel deep learning approach, Multi-scale ResNet, specifically designed to overcome the difficulties associated with the non-stationary and high-dimensional nature of cutting signals. The integration of multi-scale feature extraction with residual connections holds significant promise for robust feature learning, particularly in complex industrial environments. The initial premise and the chosen methodology are well-aligned with the demands for enhanced intelligence and efficiency in modern machining processes. The methodology is well-articulated, detailing the crucial step of converting cutting vibration signals into time-frequency images via Continuous Wavelet Transform (CWT) before feeding them into the Multi-scale ResNet model. This CWT-based preprocessing is a sound approach for capturing the intricate temporal and spectral characteristics of wear-related signals. The validation conducted on γ-TiAl alloys, a material known for its challenging machinability, lends credibility to the experimental results. The reported performance metrics—93.8% prediction accuracy, 94.2% recall, and an F1 score of 94%—are impressive, especially given the explicit success in mitigating misjudgments during the critical initial and normal wear stages. The comparative analysis against established models like CNN-SVM, Transformer, and conventional ResNet further solidifies the superiority and practical applicability of the proposed Multi-scale ResNet. Overall, this research offers a compelling and effective solution for an enduring problem in manufacturing. The Multi-scale ResNet model represents a significant step forward in tool wear monitoring, demonstrating enhanced accuracy and reliability, particularly in the nuanced early stages of wear that are often difficult to distinguish. The implications for improving cutting processing efficiency, reducing downtime, and fostering greater intelligence in production lines are substantial. This work not only provides a novel algorithmic framework but also underscores the practical benefits of advanced deep learning techniques in industrial applications, positioning it as a valuable contribution to the field of intelligent manufacturing and predictive maintenance.

Full Text

Comments

You need to be logged in to post a comment.