Study Of Phase Diagrams And Elastic Properties Of Fe-Cr-Al Alloys

With excellent oxidant resistance at high temperature and good mechanical properties,Fe-Cr-Al alloys are widely used as high temperature structure materials.Due to the issues such as the hydrogen embrittlement and high temperature oxidation appear in traditional Zirconium nuclear cladding,now the applications of Fe-Cr-Al alloys as nuclear cladding materials have been proposed.With the development of new applications,the optimization of properties is still needed.Though the research on Fe-Cr-Al alloys began at1930's,there is still a lack of systematical theoretical research on ternary phase diagrams and elastic properties.In this thesis,we investigate the phase diagrams and elastic properties of Fe-Cr-Al alloys using first-principles calculations,cluster variation method,cluster expansion method,and machine learning method.We have calculated the isothermal phase diagrams of Fe-Cr-Al alloys in fulltemperature and all-compositional ranges by combining first-principles calculations and cluster variation method.From the results of phase diagrams,we found that a new ternary ordered phase B32 appears at 600 K.The binary Fe-Al phases show an extremely high solubility for Cr,while the binary Cr-Al phase solid solution has a low solubility for Fe.In addition,we correct the phase diagram calculated within bcc lattice by introducing fcc cluster variation method,we found that the fcc correction only affect the Al-rich corner of the isothermal phase diagrams,which could reduce the phase separation at the Al-rich corner.We have employed the first-principles calculation and cluster expansion method to calculate the bulk modulus,shear modulus and Poisson's ratio of Fe-Cr-Al alloys.The results have revealed the dependencies of elastic moduli on compositions,temperature and symmetry of alloys.The bulk modulus and Poisson's ratio of Fe-Cr alloys show the linear dependencies of composition,and the solvent of Al can reduce the bulk modulus and Poisson's ratio.As the temperature rise,the bulk modulus and Poisson's ratio decrease slowly,while the shear modulus decreases significantly.The ordered phases show higher shear modulus than the disordered one with the same compositions.And also from the distribution of Poisson's ratio,we found that the disordered Fe-Cr alloys with little Al are ductile,while the Al-rich corner are brittle.Machine learning are used to predict the properties of inorganic materials.We have used the unsupervised learning k-means to obtain the Fe-Al and Fe-Cr phase diagrams,besides,indicators behaving as “order parameter” which can character the A2 and B2 phases are found.The supervised learning algorithms Extra Trees and DNN(Deep neural network)are used to established a hierarchical model which can predict the properties of ternary alloys only using binary information.The Extra Trees model performances much better than the DNN model with metric MAE(mean absolute error)on the binary test set.By choosing the proper feature vectors for bulk and shear modulus prediction models,we have obtained the predictions of bulk and shear modulus of ternary alloys with relative error within 10% compared with the results calculated from cluster expansions method.In addition,the different feature vectors between bulk and shear modulus shows that both compositions and temperature are important in affecting the bulk modulus,while the symmetry of alloys is also important in making a difference in shear modulus.