This paper was produced for the 2019 NAFEMS World Congress in Quebec Canada

Resource Abstract

Usage of handbook solutions [1] or crack propagation lifing models available in tools such as NASGRO [2], AFGROW [3] or FASTRAN [4] is quite common among lifing engineers in various industries (piping, aerospace, power generation, etc.) due to ease of use, quick runtimes, ability to address complex loading missions and perform probabilistic life assessments. Depending on the lifing application, each industry or individual company has developed standards or internal practices that are recommended or enforced to regulate this process and further improve it. Utilizing three-dimensional finite element modeling to accurately capture geometry and automatically perform a crack propagation simulation is an opportunity for improvement, however it comes with a higher runtime cost than usage of generic models available in life assessment tools [2-4]. A procedure that combines the advantages of both types of models (three-dimensional finite element modeling and generic crack models) is a more attractive route for the overall lifing process development. A Gaussian Process (GP) machine learning model is trained based on 3D Finite Element Simulations to relate crack size, shape and crack loading conditions to the corresponding mode I stress intensity factors required in a crack propagation life assessment. Three examples are provided to show capabilities of the proposed GP-based procedure along with verification against full three-dimensional crack propagation simulation and validation against test data.