Simple Data Analytics Approach Coupled with Larson–Miller Parameter Analysis for Improved Prediction of Creep Rupture Life

Machine learning of the creep rupture life dataset, which consists of test temperatures, stresses and rupture lives, received less attention in the community due to numerous physics-based and empirical models already available for the prediction of the creep rupture life, and a limited number (typically about 10 to 40) of available creep rupture life data points considered to be too small to be trained for the reliable prediction. A simple data analytics approach was developed for the quick and reliable assessment of the creep rupture life. The proposed approach involves linear regression as a major algorithm and the four features [two generic features (temperature (T) and stress (σ)) and two physics-informed features (ln σ and −1/T)], and exhibited superior creep rupture life predictions (validated by the 41 creep datasets of ferritic Cr steels) without any violation of creep phenomenology and data overfitting. In particular, the proposed approach was extremely useful to assess the fidelity of the Laron–Miller relation for a given creep rupture life dataset and to find an optimum Larson–Miller constant that minimizes a deviation from the ideal Larson–Miller relation. An analytical model was also developed based on curve fitting of Larson–Miller parameters calibrated by the optimum Larson–Miller constant. The proposed analytical model gave additional improvement in creep rupture life prediction, particularly for creep datasets, of which creep rupture lives were slightly less predicted by the data analytics approach. The two proposed models provided a synergistic effect in creep rupture life prediction when interactively used.