RAPID DENDRITIC GROWTH INVESTIGATED WITH ARTIFICIAL NEURAL NETWORK METHOD

Rapid dendritic growth of γ-(Ni, Fe) phase, β-CoSb intermetallic compound and α-Fe phase was realized by undercooling Ni-10%Fe single phase alloy, Co-60.5%Sb intermetallic alloy and Fe-40%Sn hypomonotectic alloy to a substantial extent. Their experimentally measured dendrite growth velocities were 79.5m/s, 12m/s and 0.705m/s, corresponding to undercooling levels of 303K(0.18T_L), 168K(0.11 T_L) and 219K(0.15 T_L) respectively. Since the usual dendrite growth theory deviates significantly from reality at great undercoolings, an artificial neural network incorporated with stochastic fuzzy control was developed to explore rapid dendrite growth kinetics. It leads to the reasonable prediction that dendritic growth always exhibits a maximum velocity at a certain undercooling, beyond which dendrite growth slows down as undercooling increases still further. In the case of Fe-Sn monotectic alloys, α-Fe dendrite growth velocity was found to depend mainly on undercooling rather than alloy composition.     

A new local-world evolving network model

In some real complex networks, only a few nodes can obtain the global information about the entire networks, but most of the nodes own only local connections therefore own only local information of the networks. A new local-world evolving network model is proposed in this paper. In the model, not all the nodes obtain local network information, which is different from the local world network model proposed by Li and Chen (LC model). In the LC model, each node has only the local connections therefore owns only local information about the entire networks. Theoretical analysis and numerical simulation show that adjusting the ratio of the number of nodes obtaining the global information of the network to the total number of nodes can effectively control the valuing range for the power-law exponent of the new network. Therefore, if the topological structure of a complex network, especially its exponent of power-law degree distribution, needs controlling, we just add or take away a few nodes which own the global information of the network.     

一类权重网络的加速演化模型

采用动态形成权重网络的方法，研究了在演化过程中新增边具有加速连接情况下权重网络的拓扑特性和强度分布，给出了节点强度与度的解析表达式.分析表明，加速演化的权重模型具有明显的无标度特性.再者，只要权重网络的边权重服从某一概率分布，则在演化过程中强度择优连接与度择优连接对于网络的度分布没有影响，且与具体的概率分布无关.