A Reduced Order Modeling in Finite Element for Rapid Qualification of Creep-Resistant Alloys

Abstract

This study outlines the application of a Reduced Order Modeling (ROM) approach for the probabilistic creep response of components subject to creep conditions. Time-dependent creep damage is unavoidably inflicted in elevated temperature. Typical operating condition fluctuations experienced during service can greatly limit creep life when compared to the ideal design conditions. To mimic the uncertainty in component, probabilistic Finite Element Analysis (FEA) can be employed; however, numerous full-field FEA simulations (103−105 trials) for probabilistic assessments are time-intensive and computationally prohibitive. To address this challenge, the computationally efficient ROM approach is introduced for probabilistic creep deformation, damage, and rupture predictions in FEA. In this approach, full-scale probabilistic simulation using a 1D model are performed, the extremum conditions retrieved, and applied in 2D/3D model simulations to capture the scatter bands of component response. The Wilshire-Cano-Stewart (WCS) model is calibrated to quintuplicate 304 Stainless steel data. Test condition, initial damage, and material property uncertainty are incorporated into the WCS model via appropriate probability distribution function (pdfs). A USERCREEP.F material model is developed for the WCS model and compiled for ANSYS FEA simulations. Deterministic simulations of the WCS model are carried out in FEA for validation. The goodness-of-fit between the prediction and experiment are observed to be satisfactory. Probabilistic predictions are executed in the 1D model to generate the creep deformation, damage, and rupture prediction. The extremum cases of ductility, rupture, and area under creep (AUC) curves are established. The extremum cases alone are simulated using a 2D model to capture the component level uncertainty. A %Error statistical analysis is performed to verify the accuracy of ROM approach and further validate the approach for proposed simulation of a complex geometry (e.g., turbine blade) at a significantly reduced computational time and memory. Future investigations will introduce stochasticity, temporal, and spatial uncertainty for component-level simulation and improved prediction.