Related Experiment Video
Updated: Oct 20, 2025
08:27
Author Spotlight: Efficient Image Recognition Using Directional Gradient Histogram Technique and Support Vector Machines
Published on: January 5, 2024
1.3K
A Novel Machine Learning-Based Methodology for Tool Wear Prediction Using Acoustic Emission Signals
Juan Luis Ferrando Chacón1, Telmo Fernández de Barrena1, Ander García1
1Vicomtech Foundation, Basque Research and Technology Alliance (BRTA), Mikeletegi 57, 20009 Donostia-San Sebastian, Spain.
Sensors (Basel, Switzerland)
|September 10, 2021
Summary
This study introduces a new machine learning approach for real-time tool wear monitoring using acoustic emission signals. The method significantly improves wear prediction accuracy by optimizing feature extraction and selection.
Area of Science:
- Engineering
- Materials Science
- Data Science
Background:
- Real-time monitoring of industrial asset condition, particularly tool wear, is crucial for cost reduction and minimizing scrap in machining.
- Traditional mathematical models for tool wear prediction are complex and require in-depth system knowledge.
- Machine learning (ML) models are increasingly used, but acoustic emission (AE) signal processing for tool wear prediction presents challenges in feature extraction and threshold selection.
Purpose of the Study:
- To develop an advanced methodology for accurate real-time tool wear prediction using acoustic emission signals.
- To address the complexities of AE signal interpretation, threshold definition, and optimal frequency band selection.
- To enhance the performance of ML models in tool wear estimation through optimized feature engineering.
Main Methods:
- A novel methodology employing multi-threshold count feature extraction at a multiresolution level via wavelet packet transform was proposed.
- Redundant and non-optimal feature maps from AE signals were generated and subsequently optimized using recursive feature elimination.
- Random forests regression was utilized for estimating tool wear, with the methodology validated on 19NiMoCr6 steel turning data.
Main Results:
- The proposed method effectively extracts and optimizes features from complex AE signals for tool wear prediction.
- Comparison with other ML algorithms demonstrated the superiority of the developed approach.
- A significant reduction in predicted root mean squared error by 36.53% was achieved, indicating enhanced prediction accuracy.
Conclusions:
- The proposed methodology offers a robust and accurate solution for real-time tool wear monitoring in machining processes.
- The integration of wavelet packet transform, multi-threshold counts, recursive feature elimination, and random forests regression provides a powerful tool for industrial applications.
- This approach overcomes key challenges in AE signal processing, leading to substantial improvements in prediction performance and cost-efficiency.
