Advanced chatter detection in internal turning for industry 4.0: Adaptive Threshold Wavelet De-noising with enhanced ICEEMDAN–Hilbert fusion using Adaptive Probabilistic Neural Network

Machine tool chatter adversely affects tool life and surface quality, making early detection essential in machining processes. However, vibration signals collected during machining are often contaminated by noise, hindering accurate chatter prediction. This study presents an advanced chatter detection framework integrating Adaptive Threshold Wavelet De-noising (ATWD), Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN), Hilbert–Huang Transforms (HHT), and an Adaptive Probabilistic Neural Network (APNN). The novelty of this work lies in the introduction of an adaptive noise term, β 2 E 2 ( η ( t ) ) , within the ICEEMDAN process, which mitigates mode mixing and ensures precise resonance frequency identification. The ATWD dynamically adjusts noise thresholds based on signal decomposition levels, achieving significant noise suppression for non-stationary signals while preserving critical chatter features. Using the Normalised Energy Rati; (NER), the most responsive Intrinsic Mode Functions (IMFs) are selected for feature analysis, leading t; improved signal decomposition. The APNN further enhances classification accuracy by dynamically adjusting network parameters, outperforming traditional PNN classifiers. Comparative analysis demonstrates that the proposed APNN achieves a classification accuracy of 99.5%, representing a substantial improvement over baseline methods. Internal turning experiments using a flexible boring bar validate the proposed methodology, showcasing its practical effectiveness and reliability. The integration of ATWD, ICEEMDAN–HHT fusion, and APNN provides a novel solution for chatter detection, offering significant advancements in noise filtering, feature extraction, and real-time classification accuracy. This methodology is particularly suited for challenging machining environments and Industry 4.0 applications, where precise and rapid chatter detection enhances tool life, reduces production costs, and improves overall machining productivity. • Robust Multi-Stage Signal Processing: The study integrates Adaptive Threshold Wavelet De-noising (ATWD), ICEEMDAN, and Hilbert-Huang Transform (HHT) to effectively suppress noise and extract chatter-specific features from non-stationary machining vibration signals. • Innovative Adaptive Noise Term: A novel adaptive noise term β 2 E 2 ( η ( t ) ) is introduced within the ICEEMDAN framework to mitigate mode mixing, enhancing the stability and interpretability of Intrinsic Mode Functions (IMFs). • Real-Time Adaptive Classification: The proposed Adaptive Probabilistic Neural Network (APNN), enhanced with a ridgelet activation function, dynamically adjusts to input features for accurate real-time chatter classification, achieving 99.5% accuracy. • End-to-End Industry 4.0 Integration: The methodology is validated with internal turning experiments using a flexible boring bar and includes a real-time feedback loop to CNC systems for dynamic adjustment of machining parameters. • Superior Performance Over Baselines: Compared to traditional methods (e.g., EMD + PNN, FFT + CNN), the proposed ATWD-ICEEMDAN-HHT-APNN pipeline significantly outperforms in signal denoising, feature extraction, and chatter detection accuracy in dynamic environments.