A post-exposure creep life and service-condition assessment model for DZ411 superalloy using microstructural state descriptors and symbolic regression

To address the nonlinear coupling between microstructural degradation and creep fracture in superalloy turbine blades, we develop a post-exposure assessment framework for DZ411 that integrates dual-mechanism damage decoupling with symbolic regression. A dual-channel supervised partial least squares (PLS) method transforms measured fracture features into a thermally induced isotropic descriptor (DT) and a mechanically induced anisotropic descriptor (Dσ), which are history-dependent statistical variables, not independent physical properties. Adding these descriptors to classical creep models under leave-one-out cross-validation (LOOCV) reveals an accuracy ceiling of fixed algebraic structures (R2 ≈ 0.966). Symbolic regression under minimal physical priors yields a compact constitutive equation with improved accuracy (R2 = 0.9785) and clear physical interpretability. A temperature-stress diagnostic map then bounds plausible service conditions after failure. Because DT and Dσ require measured current microstructural states, the model is intended for post-exposure assessment and failure attribution, not prediction from initial heat-treatment state alone.