Rational Design Of Composition And Process For High-Performance Copper Alloys Via Machine Learning

High-performance copper alloys are key materials for many fields in the electronic information industry,such as leadframes for ultra-large-scale integrated circuits and connectors for high-end electronic components,with good mechanical and electrical properties.With the increase in circuit integration and miniaturization of electronic components,higher comprehensive performance requirements have been put forward for the alloy.However,the complex composition,the long and parametric processing,and the complicated composition-process-property relationship lead to difficulties in the composition and process design of highperformance copper alloys.The rapid development of data-driven machine learning methods for materials design in recent years has not only improved research and design efficiency and reduced costs,but also significantly improved material performance,opening up new avenues for the design of high-performance copper alloys.However,at present,there are still challenges in the field of alloy design with machine learning methods represented by copper alloys,such as the modeling method is mainly for direct modeling of data,which cannot explain the mechanism of action of alloy elements;it is difficult to add elements that do not exist in the sample to the machine learning model to design new alloys;new alloys cannot be directly modeled due to lack of process data,and it is difficult to rapidly design the process parameters;the integrated design of components and processes has high variable dimensions.So there is an urgent need to develop machine learning methods for materials that meet design needs.In this study,high-performance copper alloys were taken as the object,various new machine learning methods for rational composition design and efficient process optimization were developed,and various high-performance copper alloys and their preparation processes were developed.Below are key research findings:An alloy factor screening method that organically combines correlation screening,recursive elimination and exhaustive method was proposed to find out the key physical and chemical factors that affect the properties of the alloy,which realized the data-driven interpretation of the mechanism of element action and realized the mechanism modeling,breaking through the optimal elemental screening and new alloy discovery puzzles.Taking the composition design of solid solution strengthened copper alloys as an example,the key alloy factors affecting the performance were clarified as the mean and variance of absolute electronegativity,the variance of atomic radius and the variance of nuclear electron distance.Taking the key alloy factors as input,strength and electrical conductivity prediction models with errors less than 2%and 7%were constructed,respectively.And based on this,a new solid solution strengthing element Indium for copper alloys was discovered,and four new solid solution strengthened conductive copper alloys with Indium content below 0.7 wt%were designed and prepared.The comprehensive performance was significantly improved over the level of existing solid solution strengthened conductive copper alloys.Combining the mechanism modeling with the Bayesian optimization method,a multi-objective performance active learning design strategy was proposed,which breaks through the simultaneous improvement of the two contradictory properties of the mechanical and electrical properties of the multivariate complex alloy.Five key alloy factors affecting alloy hardness and six key alloy factors affecting electrical conductivity were obtained by screening alloy factors,and hardness and electrical conductivity prediction models with errors less than 7%and 9%were established,respectively.Then a multi-objective utility function was proposed,and the Bayesian optimization algorithm was used to design the copper alloy composition,and the Cu-1.3Ni-1.4Co-0.56Si-0.03Mg alloy with excellent comprehensive performance was designed through experimental iteration.The ultimate tensile strength and electrical conductivity reach 858MPa and 47.6%IACS,which are superior to Cu-Ni-Co-Si series high-strength and medium-conductivity copper alloys reported in the literature.Combining mechanism modeling with interpretive methods,a physically explainable machine learning method was developed to realize element substitution design,breaking through the bottleneck of sustainable and green development restricted by the extensive use of rare and harmful elements.The success of this design strategy is demonstrated in reducing the content of the scarce element Co in the C70350 series alloy(e.g.Cu-1.3Ni-1.4Co-0.56Si-0.03Mg).Accordingly,an ultra-low Co content Cu-1.95Ni-0.5Co-0.6Si-0.2Mg-0.1Cr alloy was designed by substituting a trace amount of Cr for Co followed by optimizing the content of all alloying elements.Although the Co content was reduced by 64%from 1.4 wt%to 0.5 wt%,the alloy’s properties whose ultimate tensile strength is 850 MPa and electrical conductivity is 47.2%IACS remain comparable to those of the C70350 alloy.Aiming at the problem that new alloys cannot be directly modeled due to lack of process data,Bayesian optimization iterations were used to realize the rapid design of process parameters for the designed Cu-1.95Ni-0.5Co-0.6Si-0.2Mg-0.1Cr alloy,two types of process parameters were rapidly designed to achieve 745 MPa,244 HV and 56.9%IACS for strength,hardness and electrical conductivity,and 862 MPa,287 HV and 48.9%IACS,respectively.The tensile strength and conductivity products were further improved by 5.7%and 5.1%based on empirical trial-anderror optimization.The developed screening method was used to simultaneously screen key alloy factors and key process parameters that affect performance for dimensionality reduction,and combine empirical knowledge and machine learning collaborative strategies to break through the problem of high dimensionality in the integrated design of components and processes.Three key alloy factors and six key process parameters affecting hardness,three key alloy factors and six key process parameters affecting electrical conductivity were screened,and hardness and electrical conductivity prediction models with errors less than 9%were established.The Cu-2.5Ni-0.3Co-0.6Si-0.1Cr-0.1Mg-0.6Zn-0.02Ca-0.02P alloy with lower Co content and its process parameters were designed with the help of empirical knowledge,and the measured tensile strength and electrical conductivity reached 843 MPa and 47.4%IACS.The machine learning composition and process design methods,the high performance copper alloy and preparation processe developed in this research have been implemented and applied in the partner companies.This paper offers new avenues for composition design and process optimization of high performance alloys,enriching machine learning methods in the field of material design,and promoting the development of high performance copper alloy industry.