Abstract
Reliable tool wear monitoring in CNC machining remains challenging due to nonlinear degradation dynamics, regime variability, and limited labelled data. Existing approaches—physics-based models, single-sensor signal analysis, and purely data-driven machine learning—each exhibit structural limitations in terms of adaptability, robustness, or interpretability. This paper introduces a conceptual hybrid thermo–electrical AI architecture that integrates multimodal signal fusion with physics-guided structural priors. New approach called FUSAID is fusion-based signal AI diagnostics. Tool degradation is modelled as a latent thermo–mechanical process manifested through coupled temperature and spindle current responses. The proposed framework embeds physics-constrained feature construction and regularized learning objectives within a multimodal representation learning pipeline to improve cross-regime generalization and prediction stability. Rather than proposing a proprietary algorithm, this study establishes a structured architectural foundation for physics-informed multimodal diagnostics in intelligent machining environments. Comparative analysis demonstrates that the hybrid paradigm bridges the gap between theoretical consistency and data-driven adaptability, offering improved robustness, interpretability, and generalization under variable cutting conditions.