Enabling Large-scale MDO Through Adjoint Sensitivity Analysis

This presentation by Professor Joaquim Martins of the University of Michigan was made to the NAFEMS Optimisation Working Group Special Interest Group.

Presentation Synopsis: The combination of gradient-based optimization and efficient methods for computing the required gradients (sensitivities), enables the solution of large-scale design optimization problems. Adjoint methods compute sensitivities for large numbers of design variables efficiently and accurately. MDO problems require coupled sensitivities, which provided the motivation to developed the coupled adjoint method. This method was first applied to high-fidelity aerostructural optimization and has recently been generalized in the OpenMDAO framework.