Abstract
To address the data quality degradation in precision machine tool cutting processes caused by low sensor signal fidelity and severe zero-drift, this paper proposes a standardized signal processing pipeline aimed at providing high-reliability data support for tool wear monitoring. First, precise signal segmentation is achieved using sliding-window variance analysis. Second, zero-drift is compensated via 7th-order polynomial fitting. Subsequently, a three-stage composite denoising chain—comprising wavelet soft-thresholding, Hampel filtering, and Butterworth low-pass filtering—is constructed to enhance signal quality. Finally, a 76-dimensional multi-domain feature vector covering time, frequency, and time-frequency domains is extracted. Experimental results demonstrate that the processed features, such as the peak-to-peak values of cutting force and spindle torque, exhibit high consistency with the tool VB wear curve, with significantly enhanced monotonicity and improved signal-to-noise ratio (SNR). This algorithm-agnostic front-end module provides a high-quality, noise-suppressed data foundation for subsequent high-precision intelligent tool wear state prediction, and is directly compatible with deep learning or hybrid-driven prognostic frameworks.