Quality Factor Estimation Using Physics-Guided Feature Importance

The patent changes the parameter used for determining split points from purely data-driven clustering or manual setting to physics-driven parameter selection. By using parameters that reflect the actual physical mechanism of defect occurrence, the method ensures that increased model accuracy does not come at the cost of physical consistency. This resolves the contradiction by transforming the split determination criterion to align with physical reality while maintaining high accuracy.

The patent introduces physics-based knowledge as an intermediary between the data-driven model and the physical mechanism. This intermediary guides the splitting process to ensure that while the model achieves high accuracy through multiple splits, the resulting local regions remain consistent with the underlying physical mechanisms. The physics knowledge acts as a mediator that reconciles the conflicting requirements of accuracy and physical consistency.

The patent fundamentally changes the parameter selection criterion from convenience-based (clustering/manual) to physics-based. By selecting split points based on parameters that represent the actual physical mechanisms of defect occurrence, the method ensures that the constructed model is not only easy to build but also physically meaningful. This resolves the contradiction by making physics consistency the foundation of model construction rather than an afterthought.

The patent performs preliminary action by incorporating physics-based knowledge before the actual model construction and splitting process. By pre-establishing the physical mechanisms and selecting split points based on these mechanisms in advance, the method ensures that subsequent model construction automatically maintains physical consistency. This preliminary physics-based preparation prevents contradictions from arising during model construction.