Scripts to reproduce results in Cardoza 2026 paper "Efficient derivative computation for unsteady fatigue-constrained nonlinear aero-structural wind turbine blade optimization".
In order to run any of the scripts in this repository you must first install Julia 1.11.x and the required packages. The Manifest.toml pins all dependencies to the exact versions used in the paper. To reproduce the environment:
# In the Julia REPL from the repo root:
using Pkg
Pkg.activate(".")
Pkg.instantiate()This will download all packages at the exact versions we used, including the unregistered packages (DynamicStallModels, GXBeam, WATT, GXBeamCS, OpenFASTTools) which are pinned by git commit hash in the manifest.
Note on Snopt: The optimization scripts (optimize.jl, random_start.jl) use Snopt, which requires a commercial license for the underlying solver. If you do not have a Snopt license, you can use IPOPT via SNOW instead (results will differ). The Snopt.jl package itself will still be installed by Pkg.instantiate() — only running it requires a license.
The scripts provided here are:
- verify.jl - A comparison of OpenFAST and WATT for a turbulent inflow case
- optimize.jl - An optimization of a simplified version of the NREL 5MW using sparse parallelized forward mode derivatives.
- Ndesignvars.jl - A comparison of the different forward differentiation techniques while varying the number of dense columns (chord and twist).
- Nwind.jl - A comparison of the different forward differentiation techniques while varying the number of sparse columns (number of wind speeds).
- *_reverse.jl - timing the reverse mode approach for Ndesignvars and Nwind.
- random_start.jl - Run an optimization from a randomly perturbed design from the initial (designed to be run in a slurm job array).
- parse_random_start_optimizations.jl - From a set of files labeled "perturbation_x.log" (The outputs of a random start optimization piped to a log file) analyze converged and failed designs.
In order to run the all of the scripts, a module is provided in this repository that contains functions for differentiation and objective/constraint function analysis. If instantiating the environment above doesn't make the code accessible, simply add ./UnsteadyOpt in the Julia package manager when the local environment is active.