论文部分内容阅读
我们曾用化学键参数和模式识别方法研究合金相的晶型规律,取得较好结果。但所用的参数(金属半径和元素的电负性)系半经验参数,其物理意义和数值准确性两方面均有不够严格之处。在本工作中,我们试用描述原子核外电子各壳层结构的几个数值(组分原子的价电子数Z_1,Z_2;价电子主量子数n_1,n_2次内层的d电子数nd_1,nd_2等)作为人工神经网络输入值,用已知合金相晶型训练求得多种合金相形成条件的人工神经网络判据。我们用AB价型的合金相样本(包括CsCl型、NaTl型、CoSn型、AuCu型、CrB型),AB_2型合金相样本(包括Laves相、AlB_2型、CuAl_2型)以及AB_3型合金相样本(包括A15型、AuCu_3型)训练3层人工神经网络。求得晶型判据后试行对新发现的AB型、AB_2型和AB_3型合金相的晶型作计算机预报。其中DyMg_2,LuAl_2,NdMg_2预报为
We have used chemical bond parameters and pattern recognition method to study the phase morphology of the alloy phase, and achieved good results. However, the parameters used (the radius of the metal and the electronegativity of the element) are semi-empirical parameters, both of which have less stringent physical and numerical accuracies. In this work, we try to describe several atomic values (the valence electrons Z_1, Z_2 of valence electron main quantum number n_1, the d electrons n2 inner number of inner layer nd_1, nd_2, etc. ) As an input value of artificial neural network, artificial neural network criterion of forming conditions for a variety of alloy phases is obtained by training the known alloy phase crystal. We use AB alloy samples (including CsCl, NaTl, CoSn, AuCu, CrB), AB 2 type alloy samples (including Laves phase, AlB 2 type and CuAl 2 type) and AB 3 type alloy samples Including A15 type, AuCu_3 type) training three-layer artificial neural network. After obtaining the crystal form criterion, the crystal form of the newly discovered AB, AB_2 and AB_3 alloy phases was predicted by computer. Among them, DyMg_2, LuAl_2, NdMg_2 are predicted as