Graph-native modeling
Generating model evaluations and derivatives from a common graph, with efficient execution for coupled and time-dependent physics.
Methods
Research themes in which we develop new computational methods, theories, algorithms, and frameworks for large-scale design optimization.
Unsteady potential-flow adjoints
Differentiable potential-flow analysis with reduced memory costs for steady aircraft models and evolving aerodynamic wakes.
MDO under uncertainty
Reducing uncertainty-propagation cost through computational graph structure, response sensitivities, and dimension reduction for multidisciplinary design.
Distributed optimization
Coordinating disciplinary subproblems to manage memory, improve convergence, and satisfy shared design constraints.
Nonlinear programming
Using analysis convergence, directional curvature, and modular algorithms to improve the solution of coupled engineering design problems.
Projection-based model reduction
Adapting reduced simulation bases to changing designs while retaining the accuracy needed for structural and aerodynamic optimization.
Scientific machine learning
Combining physical models and sensitivity information with training data to predict spatial fields for engineering design.
Geometry parameterization
Connecting design variables to component geometry while preserving the dimensions and relationships that define a configuration.
Geometric non-interference constraints
Enforcing the component clearances needed for physically feasible configurations without impairing optimizer conditioning.
Mesh deformation
Updating aircraft analysis meshes through large shape and component-placement changes while retaining derivatives for gradient-based design.
Multiphysics coupling
Connecting disciplinary solvers through shared field representations and transfer operators that preserve the required physical relationships.
In development