On-going Research I: Statistical Learning for Nonlinear Dynamical Systems
[NSF CAREER: Domain-Aware Statistical Learning (CMMI 2143695)]
Objectives: to develop a physics-informed statistical learning approach capable of
learning nonlinear dynamics by utilizing data generated from computer models as well as the fundamental physics laws of nonlinear dynamics, and
predicting nonlinear dynamics for new parameters (i.e., out-of-sample prediction) in a computationally-efficient manner that needs to be significantly faster than direct numerical solutions.
Application: physics-informed statistical modeling approach for aircraft-UAV (Unmanned Aerial Vehicle) collision severity assessment, which aims to predict aircraft surface deformation due to collisions under a range of impact parameters (e.g., impact attitude, speed, position, altitude, etc.).
Challenges: (i) conventional high-fidelity Finite Element Analysis (FEA) is computationally too expensive; (ii) the data-driven model needs to incorporate fundamental structural dynamics for such a high-stake applications.
Methodology:
Impact
Illustration:
FEA simulation (generated in ~28 hours)
Statistical predictions (generated in <2 mins)