Implementation and Validation of a Fully Mechanistic Fatigue Modeling Approach in a High Performance Computing Framework

Abstract

Abstract Current approaches of fatigue evaluation of nuclear reactor components or other safety critical structural systems use S∼N curve based empirical relations which may have large uncertainty. This uncertainty may be reduced by using a more mechanistic approach. In the proposed mechanistic approach, material models are developed based on the evolution of material behavior under uniaxial fatigue experiments and implement those models into 3D finite element (FE) calculations for fatigue evaluation under multiaxial loading. However, this approach requires simulating structures under thousands of fatigue cycles which necessitates the use of high performance computing (HPC) to determine fatigue life of a large component/system within reasonable time frame. Speeding up the FE simulation of large systems requires the use of a higher number of cores, which is extremely costly, particularly when a commercial FE code is used. Also, commercial software is not necessarily optimized for use in an HPC environment. In this work, an open source parallel computing solver along with a multi-core cluster is used to scale up the number of cores. The HPC-based mechanistic fatigue modeling framework is validated through evaluating fatigue life of a pressurized water reactor surge line pipe under idealistic loading cycles and comparing the simulation results with observations from uniaxial fatigue experiment of 316 stainless steel specimen.