Neural Networks Approach to Optimization of Steel Alloys Composition

Abstract

Abstract. The paper presents modeling of steels strength characteristics in dependence from their alloying components quantities using neural networks as nonlinear approximation functions. Further, for optimization purpose the neural network models are used. The gradient descent algorithm based on utility function backpropagation through the models is applied. The approach is aimed at synthesis of steel alloys compositions with improved strength characteristics by solving multi-criteria optimization task. The obtained optimal alloying compositions fall into martenzite region of steels. They will be subject of further experimental testing in order to synthesize new steels with desired characteristics.

Multi-criteria optimization of steel alloys for crankshafts production

“…In [10] we've trained five neural network (NN) models for each one of enumerated five streghth characteristics. All models structure is the same: multi-layered without feedback connections with 8:40:1 structure (8 inputs for 8 alloying elements and one output corresponding to one of the characteristics).…”

“…However, it appeared that having that much input and output variables (11 alloying elements and 6 strength characteristics) do not allow obtaining single neural network model able to fit with the same accuracy all output variables and solving of multi-criteria optimization task showed some controversial results. In [10] we continue our work with increased data set and accounting for lower number of variables. Moreover, we have created separate neural network model for each one of the considered strength characteristics.…”

Analysis and Optimization of the Properties of High-Strength Austenitic Steels by Approximation of a Primary Database

“…That is why most often the research is limited only to the influence of the chemical composition in the same mode of heat treatment on the complex of properties. The approaches developed by us [4][5][6], characterized by the use of large quantitative information by approximation with neural networks, can be used for design when a large database with relationships between composition and properties at a fixed level is available for the studied representative the same heat treatment.…”

Study of the Adhesion Strength of Plasma-Sprayed Coatings Using Statistical Methods

“…The aim of this study was to research the influence of technological modes of plasma spraying of powder coatings of alumina modified with titanium oxide Nano powder on adhesion strength. Various methods are used to optimize the technological modes of coating application [10,11,12]. Mathematical modeling was carried out in the following sequence: constructing an experimental plan, performing an experiment and a preliminary statistical analysis of its results, constructing a mathematical model based on the experimental results and analyzing the quality of the model.…”