小样本数据的Cu基非晶合金热力学性能软测量

Modeling with small samples for characteristic of Cu-based bulk metallic glass

针对Cu基非晶合金材料受时间、条件及制备成本的限制,导致获得的热力学性能数据较少,影响新材料的研发问题.提出一种在小样本数据情况下,仍然具有较好泛化能力的Cu基非晶合金热力学性能软测量方法,为新材料配方的优化提供模型参考.通过注入噪声,提出一种小样本数据的扩充方法,增加样本的多样性.考虑样本分布函数的未知,在准则函数中引入信息论的微分熵,建立最大熵准则的神经网络反向传播理论,获得具有较高泛化能力的数学模型.仿真分析表明:该方法可对三元Cu基非晶合金的小样本数据,建立其热稳定性及玻璃形成能力与材料配方之间的非线性关系,模型精度较高.

The samples needed for the modeling of Cu-based bulk metallic glass are short due to the experiment time,condition limitation and the resource cost.This study provided with a novel method to expend the sampling data by setting a justifiable neighborhood of fluctuation around each original data in sample space,inspired by the existing of external random error.In addition,a modeling criterion of Neural Network with Entropy is presented for considering the maximization of the uncertainty on the unknown distribution of small samples.According to the modeling of thermodynamic performance of Cu-based bulk metallic glass with small samples based on the mentioned method,the result shows that the method has higher accuracy.