Alloys-by-Design Strategies Using Stochastic Multi-Objective Optimization: Initial Formulation and Results
Abstract
The objective of this research was to develop and demonstrate a technique for multi-objective optimization of the chemical composition of steel alloys with the use of an existing experimental database, The technique consists in the organization and execution of an iteration optimization experiment, which results in a set of Pareto optimum compositions of steel. The technique is based on the use of algorithms of response surface building known as IOSO. The response surfaces are built in accordance with existing experimental information. In a set of experiments the information on alloy properties in Pareto set neighborhood is accumulated, which makes it possible to increase the accuracy of results obtained, At each iteration of this technique a set of alloy compositions is formed, which are assumed to be Pareto optimal, and for which an experiment should be carried out. For this work, algorithms of artificial neural networks (ANN) were used that utilized radial-basis functions modified in order to build the response surfaces. The modifications consisted in the selection of ANN parameters at the stage of their training that are based on two criteria: minimal curvature of response surface, and provision of the best predictive properties for given subset of test points.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 08, 2003
- Accession Number
- ADA416083
Entities
People
- G. Muralidharan
- George S. Dulikravich
- Igor N. Egorov
- Vinod K. Sikka
Organizations
- University of Texas at Arlington