Tool wear monitoring is critical for maintaining machining efficiency, product quality, and cost-effectiveness in manufacturing. Traditional data-driven approaches rely on manually engineered features, which require extensive preprocessing and domain expertise, and often lead to information loss and limited adaptability. This study proposes an alternative method that uses wavelet transform (WT)-based scalogram images and a conditional generative adversarial network (cGAN) for feature representation and data augmentation. Scalograms retain the full time-frequency structure of raw signals, eliminating the need for handcrafted features. The cGAN model synthesizes realistic scalograms, expanding dataset diversity and improving model generalization. Compared with FFT-based numerical features, the proposed method preserves more wear-related information and improves classification performance. Experimental results demonstrate that the cGAN-based approach effectively captures tool wear progression, enhances training robustness, and improves accuracy under limited data conditions. These findings highlight the potential of integrating cGAN-generated features into tool wear monitoring frameworks.

Hybrid models are increasingly employed to simulate intricate physical processes by combining domain expertise embedded in physics-based models with process measurements-based data. The combination of domain knowledge and measurement data results in hybrid models that lead to high-accuracy decisions, extrapolation capabilities, and compliance with basic physical laws. Surrogate modeling, one aspect of hybrid modeling, involves a simplified model to capture the physics in a process, emphasizing an adaptable mathematical expression. This study applies hybrid modeling to address the intricate tool wear process in precision machining by developing a recursive model using symbolic regression.

In precision machining, predicting the tool health can improve productivity, job quality, and reduce machine downtime and energy consumption. While deep learning (DL) algorithms have garnered recent interest, they lack the physics understanding associated with machining processes. To address this limitation, we present a symbolic regression framework for tool wear diagnostics. The method explores analytical symbolic expressions using health indicators and cutting settings. Tool health indicators are computed from wavelet subspaces of vibration signals by applying a distance metric to the wavelet coefficients. These indicators are strongly correlated with tool wear measurements, making them suitable for tool wear diagnostics. We applied the developed framework to the IEEE PHM 2010 data, which comprises three sets of run-to-failure machining tests conducted with three tools. The framework predicted tool wear with an$R^{2}$of 0.947 and a mean absolute error (MAE) of 0.006 mm across test sets. The results demonstrate the effectiveness of the symbolic regression approach for tool wear diagnostics, showcasing the richness of information in the indicators and the quality of the developed model$\mathbf{for}$tool wear estimation.