(中国科学院上海冶金研究所， 上海 200050)
摘 要: 固相生, 成金属间化合物的液相分层二元系在Miedema的Δφ*-Δnws1/3图中，分布在Δφ*＞1.3的区域，然而在该区域混杂有固相没有金属间化合物的液相分层二元系。为了消除这种混杂现象，应用扩展的Miedema合金元胞模型研究了金属液相分层二元系固相能否生成金属间化合物的规律。在由原子参数Δφ*与Δnws1/3及ΔZ张成的多维空间中，上述两类二元系各自分布在特定的区域。据此结果，以Δφ*, Δnws1/3, ΔZ， RA/RB作为人工神经网络的输入特征量，采用误差反向传递算法，利用经已知样本集训练的人工神经网络对上述二元系的会溶温度和偏晶温度进行预报。预报结果与实测结果符合较好。
关键字: 二元合金系 液相分层 人工神经网络 会溶温度 偏晶温度
（Shanghai Institute of Metallurgy,Chinese Academy of Sciences, Shanghai 200050, P. R. China）
Abstract:In the Miedema's Δφ*-Δnws1/3 figure, the representative points of monotectic binary alloy system with compound-forming in solid state distribute in the region of Δφ*＞1.3, however, in the same region that of without compound-forming in solid state co-distribute. In order to eliminate the mingled phenomenon, an extended Miedema's cellular model of alloys has been applied to study the regularities of whether two metals can combine into compound or not in solid state of monotectic binary alloy systems. In the multi-dimensional space spanned by atomic parameters, Δφ*, Δnws1/3 and ΔZ, the representative points of monotectic binary alloy systems with and without compound-forming distribute in different regions separately. In addition, by using the error back-propagation and four parameters, Δφ*, Δnws1/3, ΔZ and RA/RB as the input features of an artificial neural network, the consolute temperature and monotectic equilibrium temperature of these monotectic binary alloy systems are predicted by the artificial neural network trained by known data of phase diagrams. The predicted results are in good agreement with the experimental ones.
Key words: binary alloy system monotectic system artificial neural network
consolute temperaturemonotectic equilibrium temperature
