Sound-Based Tool Wear Classification in Turning of AISI 316L Using Multidomain Acoustic Features and SHAP-Enhanced...

Area of Science:

• Manufacturing Engineering
• Signal Processing
• Machine Learning

Background:

• Reliable tool-wear monitoring is critical for machining quality and preventing downtime.
• Acoustic signals offer a non-contact method for assessing tool-workpiece interactions.

Purpose of the Study:

• To develop and evaluate a sound-based classification framework for identifying tool wear states in AISI 316L stainless steel turning.
• To assess the effectiveness of advanced gradient-boosting models for this classification task.

Main Methods:

• Acoustic signals were recorded during turning operations with constant cutting parameters.
• 540 multidomain acoustic features were extracted and refined using Boruta, LASSO, and SHAP analysis.
• LightGBM, XGBoost, and CatBoost classifiers were trained and evaluated using stratified 10-fold cross-validation.

Main Results:

• LightGBM and XGBoost achieved high performance with mean accuracies exceeding 0.96 and excellent PRC-AUC and ROC-AUC values (0.98-1.00).
• The models demonstrated clear separability for unworn and severe wear states, with slight wear being the most challenging class.
• The SHAP-enhanced feature selection resulted in a compact, informative feature subset.

Conclusions:

• Gradient-boosting models combined with SHAP-enhanced feature selection provide an effective, low-cost, non-contact solution for tool-wear classification in 316L turning.
• The proposed framework can enhance manufacturing quality control and reduce unscheduled downtime.