Pavel Karban | Searcharxiv

arXiv2020

Semi-Analytical Solution for a Multi-Objective TEAM Benchmark Problem

Benchmarking is essential for testing new numerical analysis codes. Their solution is crucial both for testing the partial differential equation solvers and both for the optimization methods. Especially, nature-inspired optimization algorithm-based solvers, where is an important study is to use benchmark functions to test how the new algorithm may perform, in comparison with other algorithms or fine-tune the optimizer parameters. This paper proposes a novel semi-analytical solution of the multi-objective T.E.A.M benchmark problem. The goal of the benchmark problem is to optimize the layout of a coil and provide a uniform magnetic field in the given region. The proposed methodology was realized in the open-source robust design optimization framework Ārtap, and the precision of the solution is compared with the result of a fully hp-adaptive numerical solver: Agros-suite. The coil layout optimization was performed by derivative-free non-linear methods and the NSGA-II algorithm.

math.NA↗

arXiv2019

Artap: Robust Design Optimization Framework for Engineering Applications

The main goal of the Artap project is to provide an extensive infrastructure for robust design optimization, where usually many different numerical solvers have to be used together and the impact of the manufacturing uncertainties have to be minimized. Artap is an open-source software platform, developed jointly with the coupled numerical field solver, Agros Suite. Artap ensures interfaces for a broad collection of optimization algorithms (genetic and evolutionary algorithms, various interfaces to libraries such as Nlopt, Bayesopt, etc .), tools for machine learning (neural networks, Gaussian processes, etc. ), finite element solvers (Agros Suite, Comsol, Multiphysics, deal.II). The implemented tools offers an easy and straightforward solution not only for robust design optimization but parameter identification, model order reduction, and shape optimization, as well. Moreover, Artap provides automatic parallelization of the optimization process. The paper presents the structure of the framework and technologies powering the project. The main features of Artap are demonstrated on an induction brazing process design tasks.

math.OC↗