基于人工神经网络的BaTiO3陶瓷配方研究

人工神经网络具有巨量并行、结构可变、变度非线性等特点，其建立数学模型并不需要预先知道太多有关问题背景的知识，这尤其适用于陶瓷配方研究中某些机理尚未完全清楚、传统数学方法无法分析的情况，本工作将人工神经网络技术用于介电陶瓷的配方性能分析，以BaTiO3为研究对象选取了几种掺杂剂，在均匀实验设计的基础上，用BP人工神经网络对所得实验结果进行了分析，并且用图形化方式直观地表达了出来，根据实验结果，并与多重非线性回归模型相比发现，人工神经网络模型比多重非线性回归模型更加准确且能给出配方组成与性能更丰富的信息，这对于研究各组分作用规律并获得介电陶瓷多性能指标的优化配方具有重要的指导作用。

Investigation of BaTiO3 Formulation Through Artificial Neural Network Technique

Artificial neural network (ANN) technique is endowed with certain unique attributes such as the capability of universal approximation, the ability to learn from and to adapt to its environment and the ability to invoke weak assumptions about the underlying physical phenomenon responsible for generation of the input data. ANN seems to be of most interest, in particular, when a solid theoretical basis or mathematical relationship is not available in advance. In this study ANN technique is used to model the dielectric properties of BaTiO 3 based system. Based on the homogenous experimental design, the experimental results of 21 samples were analyzed by a three_layer BP network modeling. The results were also expressed by intuitive graphics. Comparison of the results from ANN model and multi_nonlinear regression (MNLR) model indicates that the BP network is a very useful tool in dealing with problems with serious non_linearity encountered in the formulation design of dielectric ceramics.