Physics Informed Data Fusion Model for Uncertainty Quantification in Atmospheric Entry Vehicle Dynamic Stability

This paper presents an advanced methodology for enhancing the accuracy of trajectory prediction of atmospheric entry vehicles by combining experimental and numerically simulated data. The study utilizes sparse oscillation observation from the ballistic range during the deceleration phase of entry vehicles. It aims to enhance the accuracy of predictions by integrating these ballistic range observations with available free-flight computational fluid dynamics (FF-CFD) data. The proposed method consists of two main components. First, Bayesian optimization-based Gaussian processes seamlessly combine the ballistic range and FF-CFD datasets. Second, the proposed method utilizes the equations of motion to refine prediction accuracy and uncertainty further. This is achieved through a Physics-Informed Gaussian Neural Network (PI-GNN), which interfaces with the system’s governing differential equation to optimize dynamic stability, leading to a more robust understanding of the vehicle’s landing. This integrated approach represents a novel strategy for uncertainty quantification in atmospheric entry vehicle dynamics, enhancing the reliability and precision of predictive models.