Machine Learning Assisted Design And Optimization Of Microstructure And Properties Of Cobalt-based Superalloys

γ’-strengthened Co-based superalloys have high melting points and great hightemperature creep resistance,which is regarded as one of the strong competitors of a new generation of high-temperature structural materials.However,its low γ’solvus temperature,high density,inferior stability of γ/γ’ phases and insufficient oxidation resistance limit its development and application.Superalloys for practical applications often have more than seven alloying elements,and there are complex interactions among them,which leads to the use of first-principles calculation,thermodynamic calculation,and trial and error experimental methods,showing apparent inefficiency in the research of multi-component γ’-strengthened Co-based superalloys.The implementation of material genetic engineering has promoted the application of data-driven machine learning(ML)technology in the field of material,and provided a new method to accelerate the development of highperformance γ’-strengthened Co-based superalloys.Under the guidance of material genetic engineering,this thesis uses ML to study the composition design,γ’coarsening behavior and multi-performance optimization of Co-based superalloys.In the problem of designing and optimizing the multi-component composition and properties of Co-Ni-Al-Ti-Ta-W-Cr superalloys,a material design strategy integrating the screening of the multiple predicted properties from ML models,global optimization algorithm and experimental feedback is employed.The targeted properties are γ’ solvus temperature,γ’ volume fraction,microstructure without detrimental phases,density,oxidation resistance,heat-treatment window and freezing range.After 3 rounds of experiments,the Co-36Ni-12Al-2Ti-4Ta-1 W-2Cr alloy is designed and its γ’ solvus temperature is 1266.5℃.After long-term aging at 1000℃ for 1000 hours,the microstructure only contains γ/γ’ phases without detrimental phases,the γ’ volume fraction is 74.5%,and the density is 8.68 g cm-3.After isothermal oxidation at 1000℃ for 100 hours,protective alumina is formed on the surface of the alloy,and the weight gain is 1.99 mg cm-2,which indicates the alloy has good oxidation resistance.After characterization and calculation of partitioning coefficients,it is found that Ta atoms strongly enter γ’ phase,which has a positive effect on the γ’ solvus temperature.During the iteration,the content of Ta in the alloys is always high.However,Cr atoms strongly enter the γ phase,which reduces the γ’ solvus temperature.The content of Cr in the alloys also decreases with iteration.This is consistent with the influence of these elements on the γ’ solvus temperature derived from literature,which indicates that MI.models have captured this internal interaction,providing guidance for the material design and property optimization of Co-based superalloys.In the problem that the γ’ precipitates of Co-based superalloys coarsen rapidly and the classical kinetic theory cannot explain the internal mechanism,a quantitative framework of microstructure based on a deep learning image segmentation model is established to extract the γ’ microstructural information from the microstructure image of Co-based superalloys,and extracted γ’ microstructure data are used to build ML models.In order to discover the key factors affecting the rapid γ’ coarsening,a feature selection method that can reflect material knowledge is developed to determine the optimal feature combination.This method is used to find descriptors affecting the γ’ coarsening in Co-based superalloys,and Young’s modulus difference and valence electron number are the main descriptors.Combined with the symbolic regression,a mathematical expression that can predict the γ’ size in Co-based superalloys is developed.It can be found the effect of Young’s modulus difference of alloy is the most obvious,and the larger the value is,the smaller the γ’ size is.This expression is also verified by the experimental data of other alloys,and it is effective to control the γ’ size in Co-based superalloys.The action forms of these descriptors can provide guidance to control the γ’ size of Cobased superalloys to avoid the rapid y’ coarsening.On the basis of composition design and property optimization,in the problem that the data of multiple material properties of Co-based superalloys is small and these properties need to be improved,a more efficient multi-objective optimization algorithm that can simultaneously optimize multiple material properties is developed.Based on the important microstructural parameters and domain knowledge found in the research of γ’ coarsening,the γ’ volume fraction,size and shape,which can significantly affect the creep properties of superalloys,are selected as targeted properties,and new multi-component Co-based superalloys with great microstructure at 1100℃ are designed.After experimental verification,the new alloys Co-30Ni-11Al-2Ti-1 W-3.5Ta-5Cr-1 Re,Co-30Ni-10.5Al-2.5Ti-2W3.5Ta-5Cr and Co-30Ni-11Al-1.5Ti-1.5W-4Ta-5Cr have high γ’ volume fraction(>54%),small γ’ size(<480 nm)and high cuboidal γ’ fraction(>77%).In addition,the γ’ solvus temperature of these alloys is higher than 1190 ℃,and the density is lower than 8.9 g cm-3.After isothermal oxidation at 1100℃ for 100 hours,continuous and dense alumina is formed on the surface of alloys.The experimental results show that the multi-objective optimization algorithm is effective,and because its mathematical basis is universal,it can be popularized and applied to multi-objective properties optimization problems of other materials.