文章信息/Info

Title:

Inverse Design of Ternary Gold Alloy Materials with Low Resistivity

文章編號:

1674-3962(2021)04-0251-06

作者:

王向東1; 徐鵬程1; 盧天1; 劉秀娟2; 陸文聰1; 2

（1．上海大學 材料基因組工程研究院，上海 200444）（2．上海大學理學院，上海 200444）

Author(s):

WANG Xiangdong1; XU Pengcheng1; LU Tian1; LIU Xiujuan2;  LU Wencong1; 2

(1. Materials Genome Institute，Shanghai University，Shanghai 200444，China) (2. College of Sciences，Shanghai University，Shanghai 200444，China)

關鍵詞:

模式識別; 最佳投影識別; 三元金合金材料; 電阻率; 機器學習

Keywords:

pattern recognition;  optimal projection recognition;  ternary gold alloy materials;  resistivity;  machinelearning

分類號:

TG146.3+1;TG132.2+1;TP181

DOI:

10.7502/j.issn.1674-3962.202010008

文獻標志碼:

A

摘要:

金及其合金特別是三元金合金在電接觸材料領域得到了廣泛的應用。由于三元金合金組分和配比的復雜性，如何高效地設計具有低電阻率的三元金合金電接觸材料仍然是一個挑戰。提出了一種低電阻率三元金合金材料的逆向設計方法，該方法將機器學習的定性分類方法（模式識別最佳投影）與定量預測方法（XGBoost）相結合，設計出比已有三元金合金材料電阻率更低的新材料。采用最大相關最小冗余（mRMR）結合XGBoost算法篩選出建模的特征變量；利用模式識別逆投影方法設計了3個低電阻率三元金合金候選樣本，即AuZr1.95Cu0.52、AuZr1.12Cu4和AuSc1.86Cu2.75，并通過XGBoost模型估算了候選樣本的電阻率。結果表明，根據模式識別逆投影方法設計的樣本具有較低的電阻率，其電阻率負對數（-lg ρ）預報值分別為6.718，6.707和6.701，均超過了原始數據集-lg ρ的最大值6.68。該研究方法作為材料逆向設計的參考方法，有助于實驗數據的統計規律挖掘，可以加快新材料設計。

Abstract:

Gold and its alloys, especially ternary gold alloys, have been widely used in the field of electrical contact materials. Due to the complicated components and ratios of ternary gold alloy, how to efficiently design electrical contact materials of ternary gold alloy with low resistivity is still a challenge. In this work, new ternary gold alloy materials with lower resistivity were designed based on the inverse design method combining qualitive method (optimal projection of pattern recognition) with quantitative method (XGBoost). The critical features were screened out by using the maximum relevant minimum redundancy (mRMR) integrated with the XGBoost algorithm. Three candidate samples with lower resistivity, i.e., AuZr1.95Cu0.52，AuZr1.12Cu4 and AuSc1.86Cu2.75 were designed by using the inverse projection of pattern recognition method developed in our laboratory, and the resistivity of the candidate samples was estimated by the XGBoost model. The results indicate that the predicted negative logarithms (-lg ρ) of designed samples are 6.718, 6.707 and 6.701, respectively, exceeding the maximum value of 6.68 in the original data set. As a reference method for material inverse design, this research method is helpful for mining the statistical regularities in experimental data, and can accelerate the design of new materials.