The ceramics materials density defined by artificial neural networks

Abstract

Predicting the ceramic materials properties and designing the desired microstructures characteristics are very important objectives in ceramic samples consolidating process. The goal of our research is to calculate the density within consolidated BaTiO3-ceramic samples for different consolidation parameters, like sintering temperature, using obtained experimental data from the material’s surface, by applying back propagation neural network (BP). This method, as a very powerful tool, provides the possibility to calculate the exact values of desired microelectronic parameter at the level of the grains’ coating layers. The artificial neural networks, which have biomimetic similarities with biological neural networks, propagate the input signal forward, unlike the output signal, designated as error, which is propagated backwards spreading throughout the whole network, from output to input neuron layers. Between these two neuron layers, there are usually one or more hidden layers, where the grains of the sintered material are represented by network neurons. Adjustable coefficients, called weights, are forward propagated, like input signals, but they modify the calculated output error, so the neural network training procedure is necessary for reducing the error. Different consolidated samples density values, measured on the bulk, substituted the errors, which are calculated as contribution of all network elements, thus enabling the density calculation of all constituents of ceramic structure presented by neural network. In our future research we plan to increase the number of neurons and hidden layers in order to improve this method to become even more accurate and precise.