Publication Details
Abstract
Thermal-fatigue life prediction of gas-turbine discs is challenging because plasticity, creep, and progressive strain accumulation interact under extreme thermomechanical loading. This study compares constant-bias, linear, and Gaussian-process corrections applied to deformation-based crack-initiation life predictions for five disc cases. The correction inputs were maximum temperature and total strain range, and model performance was assessed through leave-one-disc-out evaluation. The uncorrected deformation criterion yielded a mean absolute error of 8.0 cycles and a root-mean-square error of 11.9 cycles. These values were reduced by Gaussian-process correction to 4.9 and 6.9 cycles, respectively, while the two-input linear model achieved lower errors of 2.2 and 2.4 cycles. Damage decomposition also revealed that in some discs, one-sided creep accumulation dominated while in other discs, cyclic-creep dominated, indicating that similar total damage can result from different mechanisms. The findings demonstrate the value of residual correction and highlight the importance of comparing Gaussian-process models with simpler alternatives in deformation-based disc-life prediction.