Hybrid physics-data driven method for dynamic tool wear prediction

Existing tool wear prediction models often exhibit limited adaptability to general machining processes, frequently resulting in inaccuracies or even failures. Moreover, comprehensive collection and analysis of cutting data under complex working conditions remain challenging. To overcome these limita...

Full description

Saved in:
Bibliographic Details
Published in:Journal of the Brazilian Society of Mechanical Sciences and Engineering Vol. 47; no. 9; p. 430
Main Authors: Ran, Yawei, Yu, Weiwei, Wei, Xubing, Geng, Junhao, Li, Yongkang
Format: Journal Article
Language:English
Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.09.2025
Springer Nature B.V
Subjects:
Accuracy
Adhesive wear
Crack propagation
Cutting tools
Data acquisition
Decision making
Deep learning
Digital twins
Engineering
Fault diagnosis
Machinery
Machining
Manufacturing
Mechanical Engineering
Neural networks
Optimization
Prediction models
Real time
Root-mean-square errors
Simulation
Technical Paper
Tool wear
Wear mechanisms
Working conditions
Deep learning
Digital twin
Dynamic monitoring
Temporal convolutional neural network
Tool wear
ISSN:1678-5878, 1806-3691
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Existing tool wear prediction models often exhibit limited adaptability to general machining processes, frequently resulting in inaccuracies or even failures. Moreover, comprehensive collection and analysis of cutting data under complex working conditions remain challenging. To overcome these limitations, this study proposes a parallel dual-prediction framework integrating a self-attention temporal convolutional network (SATCN) with a digital twin (DT) mechanism model, enabling dynamic tool wear monitoring in complex environments. Firstly, A DT-enabled data acquisition system and virtual modeling platform were established to capture real-time acceleration sensor data and tool wear measurements throughout the machining process. Secondly, A DT mechanism model was developed across four dimensions (geometric, physical, behavioral, and rule-based) to simulate tool wear dynamics. A twin data model leveraging SATCN was designed to extract critical features from sensor signals. Thirdly, Sensor-derived analytical results continuously update and refine the DT mechanism model, ensuring precise wear quantification. Finally, Experimental validation utilized a digital-physical integration platform, with comparative assessments against single-model approaches employing Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), and Coefficient of Determination (R 2 ). The parallel hybrid-driven model demonstrated superior performance: 50.71% lower MAPE, 59.51% reduced average RMSE, and 3.64% higher average R 2 compared to standalone data-driven models. These metrics conclusively validate the enhanced accuracy and robustness of the proposed methodology in tool wear detection.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:1678-5878
1806-3691
DOI:10.1007/s40430-025-05727-2