Software EDAOPT for Experimental Design, Analysis and Multiobjective Robust Optimization
OPT-i International Conference on Engineering and Applied Sciences Optimization 2014
Jānis Auziņš, Aleksandrs Januševskis, Jānis Januševskis, Eduards Skuķis

Abstract. The Machine Dynamics Laboratory (MDL) of Riga Technical University has broad experience in the area of design of physical and numerical experiments, metamodeling and optimization of machines and structures. The researchers of MDL proposed the first space filling Latin hypercube (LH) type experimental designs in 1977. Here we present the latest version of software EDAOpt with the newest methods implemented. The software EDAOpt provides all phases of experimental optimization: 1) design of experiments, 2) building a mathematical model on the basis of experimental results, 3) multiobjective and robust optimi-zation using approximated models as objective and constraint functions, 4) validation of results. Methods implemented in all phases are: 1. Design of experiments: classic factorial and Response surface designs, LH designs optimized according to different space filling criteria, including Mean Square Error (MSE) criterion, sequential non-LH type MSE-optimal designs and orthogonal D-optimal designs for high-order polynomial approximations. 2. Approxima-tion methods: multivariate polynomials, kriging, locally weighted 1-3 order polynomials and high-order orthogonal Legendre polynomials penalized according to generalized thin-plate potential energy criterion.3. Optimization methods: multistart simulated annealing, stochastic search. Optimization objective functions and constraints may depend on standard deviations (STD) of approximated responses. STDs are calculated by Monte Carlo simulation. The results of multiobjective optimization are presented as a Pareto frontier. Monte Carlo simulation and metamodeling are used for robust optimization and post-optimization analysis. The effectiveness of software is demonstrated by solving the classic two-bar test problem and practical problem of experimental optimization – robust identification of elastic parameters of composite material structural elements.


Atslēgas vārdi
Design of experiments, Metamodeling, Surrogates, Approximation, robust optimization.

Auziņš, J., Januševskis, A., Januševskis, J., Skuķis, E. Software EDAOPT for Experimental Design, Analysis and Multiobjective Robust Optimization. No: OPT-i International Conference on Engineering and Applied Sciences Optimization, Grieķija, Kos, 4.-6. jūnijs, 2014. Athens: National Technical University, 2014, 101.-123.lpp. ISBN 978-960-99994-5-8. ISSN 2241-9098.

Publikācijas valoda
English (en)
RTU Zinātniskā bibliotēka.
E-pasts: uzzinas@rtu.lv; Tālr: +371 28399196