Optimizing Thermographic Fatigue-Limit Estimation: GPR, SVR, and Stacking Ensembles within the Two Curves Method for C45 Steel

Passive Infrared Thermography provides an approach to estimate fatigue limits rapidly by extracting thermal parameters during loading. However, the application of the Two Curves Method (TCM) during the fatigue resistance computation may result in sparse measurements of the thermal parameter and approximating curves adopted for analysis. Based on thermal emissions measurements of C45 steel, the current study suggests a TCM-machine-learning model enhancing continuity of the approximating curves and minimizing scatters by training continuous surrogates of the experimental thermal behavior. Three other regressors are used to model thermal parameters: (i) Gaussian Process Regression (GPR), (ii) Support Vector Regression (SVR), and (iii) a stacking ensemble combinig the use of both kernel and tree learners through a regularized metamodel. The ML-augmented thermal curves are extrapolated into TCM by use of Linear-Linear and Parabola-Power-law estimates and the fatigue-limit estimates obtained are compared to Staircase references at 50 and 1 percent failure probability. Findings indicate that the suggested surrogates reduce the between strategies variation and narrow agreement between ΔT- and A-based intersections on a regular occasion, whereas uncertainty in GPR offers a viable confidence mark of the end estimate.