Development and performance of data-driven models for the prediction of the high-temperature fatigue life of alloy 617

Abstract

Regularised linear regression (RLR) and recurrent neural network (RNN) data-driven models for the prediction of the high-temperature (850 °C, 950 °C) fatigue life of Alloy 617 are developed, trained, and tested using available experimental datasets. The predictions of these data-driven models are compared with the semi-empirical Coffin-Manson and Goswami models. It is shown that the data-driven models can match, or in some cases even outperform, the semi-empirical models while providing the advantage that they are temperature independent. However, it is also shown that these data-driven models cannot extrapolate accurately beyond the experimental data used for their development and training.