Prediction of the properties of multicomponent ceramics based on the Kohonen self-organizing network

  • Lyudmila Akhmetshina
  • Stanislav Mazurik
  • Ihor Skuratovsky

Abstract

The information possibilities of the method for predicting the values of multidimensional experimental data determined on an uneven grid are considered. The essence of the method is to use the Kohonen self-organizing network to determine the qualitative effect of independent parameters in tin oxide-based multicomponent ceramics on its non-linearity coefficient in the task of improving the parameters of electrical circuit protection elements.
There are many applications that require the application of data forecasting methods, for example, when evaluating the properties of multicomponent materials. In particular, ceramic materials are formed on the basis of several initial components and the determination of their required compositions is a laborious process, since the target properties can significantly change with a slight variation of any of the independent components. Thus, the study of multicomponent ceramic samples requires a large amount of experimental work.
Currently, an alternative is an approach based on artificial intelligence methods, in particular, the use of Kohonen’s self-organizing neural networks, which make it possible to make forecasts based on small amounts of information. The purpose of the analysis of experimental data is to determine the influence of the composition of impurities on the nonlinearity coefficient alpha of varistor ceramics: it is necessary to obtain the largest possible value of alpha at a low electric field strength E (parameters were measured at the same current density). Verification of the predictive properties was carried out in two directions - restoration of the null-values of individual parameters, including the target and forecasting of new patterns based on the values of the centroids of the clusters corresponding to the closest to the target prototype.
On the basis of intelligent forecasting using SOM, the areas of promising changes are determined for the future values of the composition of ceramics for further physical experiments. Using SOM provides the implementation of a fairly universal technology for obtaining a qualitative assessment of the dependence of the target value on independent components for a limited set of experimental data.

References

Kokhonen T. Samoorhanizuyutsya karty / Kokhonen T. [per. 3-ho anhl. yzd. V.N. Ahyeyevapid red. YU.V. Tyumentseva]//.M .: Binom. Laboratoriya znan.- 2008. - 665 s.

Yehorov A. Optymizatsiya yaskravosti zobrazhen naosnovi neyrofazzi tekhnolohiy / A. Yehorov, L. Akhmetshyna// Monohrafiya.- Vyd. Lambert. -2015. -139 c.

Yang H. Self-organizing nonlinear output (SONO): a neural network suitable for cloud patch-based rainfall estimation from satellite imagery at small scales/ H. Yang, K. L. Hsu, S. Sorooshian, and X. Gao // Water Resources Research.- vol. 41.- Article ID W03008.- 2005.

L.G.Akhmetshina. Metod interpolyatsii neravnomernykh prostranstvennykh dannykh na osnove seti Kokhonena / L. G.Akhmetshina, A. M. Akhmetshin, I. M. Udovik // Sistemní tekhnologíí̈, –2013. -Dnípropetrovs'k.– № 1. - S. 106-112.

Glot A.B. Keramika ZnO-PbO-Co3O4-B2O3 s neomicheskoj elektroprovodnostyu/ A.B. Glot, S.V. Mazurik// Neorganicheskie materialy, 1998.- T. 34.- № 2.- S. 247-250.

Glot A. B. Non-Ohmic conduction in tin dioxide based varistor ceramics/ Glot A. B., Skuratovsky I. A.// Materials Chemistry and Physics.- 2006.- Vol. 99, № 6.- P. 487-493.

Published
2020-03-27