Shugo Kaneko

Shugo Kaneko

He/Him

Aircraft MDAO Researcher

MDO Approaches for Design-Trajectory Optimization

PhD research, 2023 - 2024

Overview

Simultaneous design and trajectory optimization, also called open-loop control co-design (CCD), aims to concurrently optimize the design of a physical system (e.g., aircraft sizing and wing shape) and its trajectory (e.g., climb flight path). The ultimate goal is to improve the design and performance by capturing the coupling between vehicle design and its trajectory, or more generally, dynamical behavior.

In this project, we investigated computationally-efficient approaches for simultaneous design-trajectory optimization.

MDO approaches

Simultaneous design-trajectory optimization can be solved by SAND approach (direct collocation method) or MDF approach (direct shooting method).

SAND and MDF approaches

Furthermore, when coupling mid- or high-fidelity design analysis (e.g., vortex-lattice method for wing aerodynamics) to trajectory/dynamics model, we can use either direct coupling or surrogate-based coupling.

Coupling approaches

Therefore, we have choices of:

  • SAND or MDF
  • direct coupling or surrogate-based coupling

when solving CCD problems with computationally-expensive design analysis. The goal of our project is to compare the computational efficiency of these approaches and provide general guidelines in selecting the MDO approaches.

Test problem

As a test problem, we solved simultaneous optimization of the wing aerostructural design and its climb trajectory for fixed-wing UAVs.

Wing design and climb trajectory optimization

We used OpenAeroStruct for the wing aerostructural analysis, which couples vortex-lattice method and 1D beam finite-element model. For trajectory, we used Dymos as an implementation of the Radau collocation method. Both OpenAeroStruct and Dymos are built on top of the OpenMDAO framework, therefore, the coupling is done seamlessly and we have access to the analytic coupled derivatives.

Results

Here is one of the results, where we compared the optimization wall times for different initial guesses and two optimizers (SNOPT and IPOPT). Wall time results

Key findings are

  • The selection of direct or surrogate-based coupling strongly depends on the number of coupling variables from dynamics (trajectory) to design analysis.
  • SAND (collocations) are generally more robust and efficient.
  • MDF is more compatible with surrogate-based coupling.

Further discussion can be found in our paper.

Relevant publications