Tool wear trend prediction using acoustic emission signals: A composite multiscale fractional diversity entropy approach

• CMFrDivEn captures tool wear dynamics and outperforms other entropy methods in clustering. • CMFrDivEn-DT achieves 98.3% early wear and 100% steady/severe wear classification accuracy. • VMD isolates ≥15 kHz tool wear bands and suppresses low-frequency noise. Tool wear is a critical factor influencing the machining quality, efficiency, and cost in numerical control machining. Accurate and timely tool condition monitoring remains challenging because of the complexity of cutting processes and the limitations of conventional sensing techniques. This study proposes a novel approach for predicting tool wear trends based on acoustic emission signals and a newly developed entropy measure, Composite Multiscale Fractional Diversity Entropy (CMFrDivEn). The method integrates Variational Mode Decomposition for signal preprocessing and CMFrDivEn for feature extraction across multiple timescales, enabling the effective characterization of tool wear states. The proposed approach was validated using the PHM Society 2010 milling dataset, with acoustic emission signals acquired under varying flank wear widths. Five wear stages were analyzed, and feature distributions were compared with those obtained from Multiscale Fractional Diversity Entropy (MFrDivEn), Fractional Diversity Entropy (FrDivEn), and standard Diversity Entropy (DivEn). The results demonstrate that CMFrDivEn provides superior clustering performance and class separability. When combined with a Decision Tree (DT) classifier, CMFrDivEn achieved classification accuracies of up to 100% for most wear stages, significantly outperforming the other entropy-based methods. These findings confirm that the proposed CMFrDivEn-DT framework offers a robust and reliable solution for tool wear monitoring in intelligent manufacturing systems.