6. DAKOTA-OpenFOAM optimization loop | Derivative-free optimization

Tavsif:

In this demo case, we use Salome for geometry and mesh generation, OpenFOAM 11 as the black box solver, bash scripting and Python for automatic post-processing, Python for plotting, and DAKOTA 6.19.0 for optimization and orchestration. The video covers the blunt body shape optimization case, part 4, and demonstrates various derivative‑free optimization methods such as Efficient Global Optimization (EGO), Mesh Adaptive Search (MADS), Single Objective Genetic Algorithm (SOGA), and pattern search. It shows how to launch optimization studies, analyze outcomes, and interpret results. 00:00 Introduction - Preliminaries 02:32 Choosing an optimization method. General guidelines and decision matrix. 04:40 An excellent book on optimization by Vanderplaats 06:15 DAKOTA's input file - Some derivative-free methods 07:30 Efficient global optimization EGO - One of my favorite methods in DAKOTA 12:05 Let's launch an optimization study using the EGO method 15:25 Let's take a look at the outcome of the EGO method 18:15 What a beautiful test case 20:31 Mesh adaptive search method MADS 23:15 Let's launch an optimization study using the MADS method - I selected the wrong method 25:33 Let's launch an optimization study using the MADS method - The right method this time 26:53 Let's take a look at the outcome of the MADS method 32:20 Single objective genetic algorithm SOGA 34:11 Fine tuning the methods' parameter 37:00 Let's launch an optimization study using the SOGA method 39:32 Let's take a look at the outcome of the SOGA method 41:39 SOGA method - Where is the best solution? 44:30 Warning: do not use biased data to construct meta-models. 46:31 SOGA method - What iteration corresponds to the best solution? 48:58 Pattern search method or asynch_pattern_search in DAKOTA 49:58 Let's launch an optimization study using the asynch_pattern_search method 51:00 Let's take a look at the outcome of the asynch_pattern_search method 52:30 Final remarks - Main takeaways