Neural network-assisted probabilistic creep-fatigue assessment of hydrogenation reactor with physics-based surrogate model

Abstract

In order to obtain better hydrogenation reaction efficiency, the prospective hydrogenation reactor faces harsh working conditions, including the high-stress level, high-temperature environment, long creep dwell period and degradation of material resistance, leading to significant creep-fatigue interaction at the structural critical locations. The uncertainty widely existing in such design conditions further aggravates the creep-fatigue failure risk of the hydrogenation reactor structure, and makes the damage evolution and lifetime difficult to capture and predict. According to the latest appendix A15 of the R5 Procedure, a series of probabilistic analysis practices are delivered based on Volume 2/3 regarding creep-fatigue crack initiation. In this study, the probabilistic analysis for the hydrogenation reactor is implemented by the neural network-assisted creep-fatigue surrogate model under the probabilistic Linear Matching Method Framework (pLMM), with the superior balance between the computational efficiency and prediction accuracy fully achieved. The statistical distribution of structural creep-fatigue lifetime is investigated with the physics-based modelling technology, and the reliability-centred evaluation diagram and safety factors of the hydrogenation reactor are proposed for different reliability levels, which is dedicated to facilitating the risk management of crucial components in the hydrogen industry.