Study Of Three-dimensional Imaging Methods For Transmission Electron Microscopy In Materials Science

Using modern transmission electron microscopes,structural features at the nanoscale in materials can be accurately characterized in two dimensions.With the development of hardware and advances in technology,how to extract quantitative three-dimensional(3D)microstructural information from two-dimensional(2D)projected images has become an important subject for materials scientists and electron microscopists.In this thesis,several improved algorithms for accurate and efficient 3D reconstruction and 3D structure parameter determination in transmission electron microscopy have been proposed through theoretical analysis,model building,simulation testing,and experimental verification.Four innovative research results are achieved as follows.(1)A 3D discrete algebraic reconstruction method IDART(improved discrete algebraic reconstruction technology)is proposed,which adds a two-dimensional constraint by image segmentation to DART,can reduce the number of parameters,and expand the scope of application even when reconstructing materials with multiple tissues.From simulation tests,IDART can obtain better reconstruction results than the existing SIRT(simultaneous iteration reconstruction technique),DART(discrete algebraic reconstruction technology),TVRDART(total variation regularized discrete algebraic reconstruction technique),or MDART(modified discrete algebraic reconstruction technology)methods.In the experimental application,the effective reconstruction of Au-Ag core-shell nanorods and the quantitative core-shell volume ratio were obtained by IDART,which verified the effectiveness and reliability of IDART in quantitative 3D discrete reconstruction.(2)A 3D electron microscope imaging technology NNART-TVM(neural network algebraic reconstruction technology-total variation minimization)based on the fully connected neural network algorithm is proposed,which can realize the precise 3D electron microscope imaging of the microstructure of nano-holes in ceramic materials and nano precipitations in aluminum alloys.NNART-TVM is improved based on NNART(neural network algebraic reconstruction technology),adding total variation minimization(TVM)constraints,reducing the number of hidden layers in the neural network,and accelerating the reconstruction speed.NNART-TVM is able to suppress both missing wedge artifacts and noise in the reconstruction results due to the incorporation of TVM,a weighted coefficient of which is used to adjust the constraint strength of regularization.When the projection image is noisy,the weight parameter of the regularization term needs to be increased.(3)A generative adversarial networks method mwGAN(missing wedge-generative adversarial networks)is proposed to suppress the typical missing wedge effect in reconstruction results.This method is different from the traditional idea of improving the reconstruction method.The idea of mwGAN to improve the reconstruction quality is to perform extensional completion on the missing projection images.By simulating a large number of complete projection images,mwGAN obtains a generator that can complete a missing projection image through unsupervised learning.Reconstructing from the incomplete projections after complementation,the missing wedge artifacts in the reconstruction results are effectively suppressed,and the three-dimensional imaging quality of nano-scale precipitates in aluminum alloys is improved.In addition,the combination of completed projection maps and optimized reconstruction algorithms,such as IDART or NNART-TVM,can further improve the reconstruction quality.(4)A volume fraction self-corrected 3D electron microscope imaging technology vfIDART was developed based on 3D electron tomography.This technology,which overcomes the difficulty of obtaining 3D microstructure parameters of hole-like in material science,can obtain quantitative parameters and its error of the 3D microstructure of materials with both accuracy and statistical significance through 3D electron tomography.Using vfIDART,the quantitative volume fractions of nano-scale precipitates in three samples of7N01 aluminum alloy(Al Zn Mg)with different treatment states were measured and corrected respectively,and the contribution of the precipitates in the material to the material strength was quantitatively calculated and evaluated.The relationship between the microscopic volume fraction of precipitates and the macroscopic yield strength of the material was investigated.