Bayesian Calibration for Large-Scale Fluid Structure Interaction Problems Under Embedded/Immersed Boundary Framework

Seamless integration of observation data with computational models starts to play a significant role in improving the prediction for Fluid‐Structure Interaction (FSI) problems (e.g., patient-specific hemodynamic modeling). The integration can be formulated as a Bayesian calibration problem, which has been widely applied in inverse analysis and uncertainty analysis. In this talk, we focus on Bayesian calibration for large-scale fluid-structure interaction systems that feature large structural deformations. We aim to address three major challenges: 1) observation data are noisy; 2) the FSI solvers are given as a black box, or the associated numerical methods are not differentiable (e.g., immersed/embedded boundary methods and fracture mechanics models); 3) each forward FSI evaluation is computationally expensive for real-world applications. In this regard, a new Bayesian framework built on unscented Kalman filter/inversion is developed. The approach is derivative-free and non-intrusive, and it can efficiently calibrate and provide uncertainty estimations of FSI models with noisy observation data. We demonstrate and validate the framework by successfully calibrating the model parameters of a piston problem and identifying the damage field of an airfoil under transonic buffeting.