Detecting Clusters in Atom Probe Data with Gaussian Mixture Models

On the Use of a Cluster Identification Method and a Statistical Approach for Analyzing Atom Probe Tomography Data for GP Zones in Al-Zn-Mg(-Cu) Alloys

Early-stage clustering in two Al-Mg-Zn(-Cu) alloys has been investigated using atom probe tomography and transmission electron microscopy. Cluster identification by the isoposition method and a statistical approach based on the pair correlation function have both been applied to estimate the cluster size, composition, and volume fraction from atom probe data sets. To assess the accuracy of the quantification of clusters of different mean sizes, synthesized virtual data sets were used, accounting for a simulated degraded spatial resolution. The quality of the predictions made by the two complementary methods is discussed, considering the experimental and simulated data sets.

Quantitative three-dimensional imaging of chemical short-range order via machine learning enhanced atom probe tomography

Chemical short-range order (CSRO) refers to atoms of specific elements self-organising within a disordered crystalline matrix to form particular atomic neighbourhoods. CSRO is typically characterized indirectly, using volume-averaged or through projection microscopy techniques that fail to capture the three-dimensional atomistic architectures. Here, we present a machine-learning enhanced approach to break the inherent resolution limits of atom probe tomography enabling three-dimensional imaging of multiple CSROs. We showcase our approach by addressing a long-standing question encountered in body-centred-cubic Fe-Al alloys that see anomalous property changes upon heat treatment. We use it to evidence non-statistical B2-CSRO instead of the generally-expected D03-CSRO. We introduce quantitative correlations among annealing temperature, CSRO, and nano-hardness and electrical resistivity. Our approach is further validated on modified D03-CSRO detected in Fe-Ga. The proposed strategy can be generally employed to investigate short/medium/long-range ordering phenomena in different materials and help design future high-performance materials.

A Machine Learning Framework for Quantifying Chemical Segregation and Microstructural Features in Atom Probe Tomography Data

Atom probe tomography (APT) is ideally suited to characterize and understand the interplay of segregation and microstructure in modern multi-component materials. Yet, the quantitative analysis typically relies on human expertise to define regions of interest. We introduce a computationally efficient, multi-stage machine learning strategy to identify compositionally distinct domains in a semi-automated way, and subsequently quantify their geometric and compositional characteristics. In our algorithmic pipeline, we first coarse-grain the APT data into voxels, collect the composition statistics, and decompose it via clustering in composition space. The composition classification then enables the real-space segmentation via a density-based clustering algorithm, thus revealing the microstructure at voxel resolution. Our approach is demonstrated for a Sm-(Co,Fe)-Zr-Cu alloy. The alloy exhibits two precipitate phases with a plate-like, but intertwined morphology. The primary segmentation is further refined to disentangle these geometrically complex precipitates into individual plate-like parts by an unsupervised approach based on principle component analysis, or a U-Net-based semantic segmentation trained on the former. Following the composition and geometric analysis, detailed composition distribution and segregation effects relative to the predominant plate-like geometry can be readily mapped from the point cloud, without resorting to the voxel compositions.

Cluster characterization in atom probe tomography: Machine learning using multiple summary functions

In this work, we develop a machine learning-based method to characterize intracluster concentration (ρ_c), background concentration (ρ_b), clustering radius (r̄), and radius dispersity (δr) in simulated atom probe tomography data using multiple spatial statistics summary functions to train a Bayesian regularized neural network. We build upon previous work that utilized Ripley's K-function by incorporating additional features from nearest-neighbor spatial statistics summary functions to better characterize concentration-based metrics. The addition of nearest-neighbor based features allows for highly accurate estimates of ρ_c and ρ_b, both with 90% of the predictions within 4.0% of the real value; the root-mean-square errors are reduced by 81.5% and 92.8% from predictions using only K-function based features, respectively. Additionally, including these nearest-neighbor based features improves the ability to differentiate between r̄ and δr.

Atom probe tomography

Atom probe tomography (APT) provides three-dimensional compositional mapping with sub-nanometre resolution. The sensitivity of APT is in the range of parts per million for all elements, including light elements such as hydrogen, carbon or lithium, enabling unique insights into the composition of performance-enhancing or lifetime-limiting microstructural features and making APT ideally suited to complement electron-based or X-ray-based microscopies and spectroscopies. Here, we provide an introductory overview of APT ranging from its inception as an evolution of field ion microscopy to the most recent developments in specimen preparation, including for nanomaterials. We touch on data reconstruction, analysis and various applications, including in the geosciences and the burgeoning biological sciences. We review the underpinnings of APT performance and discuss both strengths and limitations of APT, including how the community can improve on current shortcomings. Finally, we look forwards to true atomic-scale tomography with the ability to measure the isotopic identity and spatial coordinates of every atom in an ever wider range of materials through new specimen preparation routes, novel laser pulsing and detector technologies, and full interoperability with complementary microscopy techniques.

Three dimensional cluster analysis for atom probe tomography using Ripley's K-function and machine learning

The size and structure of spatial molecular and atomic clustering can significantly impact material properties and is therefore important to accurately quantify. Ripley's K-function (K(r)), a measure of spatial correlation, can be used to perform such quantification when the material system of interest can be represented as a marked point pattern. This work demonstrates how machine learning models based on K(r)-derived metrics can accurately estimate cluster size and intra-cluster density in simulated three dimensional (3D) point patterns containing spherical clusters of varying size; over 90% of model estimates for cluster size and intra-cluster density fall within 11% and 18% error of the true values, respectively. These K(r)-based size and density estimates are then applied to an experimental APT reconstruction to characterize MgZn clusters in a 7000 series aluminum alloy. We find that the estimates are more accurate, consistent, and robust to user interaction than estimates from the popular maximum separation algorithm. Using K(r) and machine learning to measure clustering is an accurate and repeatable way to quantify this important material attribute.

Application of the Maximum Flow-Minimum Cut Algorithm to Segmentation and Clustering of Materials Datasets

Problems involving image segmentation, atomic cluster identification, segmentation of microstructure constituents in images and austenite reconstruction have seen various approaches attempt to solve them with mixed results. No single computational technique has been able to effectively tackle these problems due to the vast differences between them. We propose the application of graph cutting as a versatile technique that can provide solutions to numerous materials data analysis problems. This can be attributed to its configuration flexibility coupled with the ability to handle noisy experimental data. Implementation of a Bayesian statistical approach allows for the prior information, based on experimental results and already ingrained within nodes, to drive the expected solutions. This way, nodes within the graph can be grouped together with similar, neighboring nodes that are then assigned to a specific system with respect to calculated likelihoods. Associating probabilities with potential solutions and states of the system allows for quantitative, stochastic analysis. The promising, robust results for each problem indicate the potential usefulness of the technique so long as a network of nodes can be effectively established within the model system.

Enhancing Element Identification by Expectation-Maximization Method in Atom Probe Tomography

This paper describes an alternative way to assign elemental identity to atoms collected by atom probe tomography (APT). This method is based on Bayesian assignation of label through the expectation-maximization method (well known in data analysis). Assuming the correct shape of mass over charge peaks in mass spectra, the probability of each atom to be labeled as a given element is determined, and is used to enhance data visualization and composition mapping in APT analyses. The method is particularly efficient for small count experiments with a low signal to noise ratio, and can be used on small subsets of analyzed volumes, and is complementary to single-ion decomposition methods. Based on the selected model and experimental examples, it is shown that the method enhances our ability to observe and extract information from the raw dataset. The experimental case of the superimposition of the Si peak and N peak in a steel is presented.

Atom Probe Tomography Interlaboratory Study on Clustering Analysis in Experimental Data Using the Maximum Separation Distance Approach

We summarize the findings from an interlaboratory study conducted between ten international research groups and investigate the use of the commonly used maximum separation distance and local concentration thresholding methods for solute clustering quantification. The study objectives are: to bring clarity to the range of applicability of the methods; identify existing and/or needed modifications; and interpretation of past published data. Participants collected experimental data from a proton-irradiated 304 stainless steel and analyzed Cu-rich and Ni-Si rich clusters. The datasets were also analyzed by one researcher to clarify variability originating from different operators. The Cu distribution fulfills the ideal requirements of the maximum separation method (MSM), namely a dilute matrix Cu concentration and concentrated Cu clusters. This enabled a relatively tight distribution of the cluster number density among the participants. By contrast, the group analysis of the Ni-Si rich clusters by the MSM was complicated by a high Ni matrix concentration and by the presence of Si-decorated dislocations, leading to larger variability among researchers. While local concentration filtering could, in principle, tighten the results, the cluster identification step inevitably maintained a high scatter. Recommendations regarding reporting, selection of analysis method, and expected variability when interpreting published data are discussed.

The Application of the OPTICS Algorithm to Cluster Analysis in Atom Probe Tomography Data

Atom probe tomography (APT) is a powerful technique to characterize buried three-dimensional nanostructures in a variety of materials. Accurate characterization of those nanometer-scale clusters and precipitates is of great scientific significance to understand the structure-property relationships and the microstructural evolution. The current widely used cluster analysis method, a variant of the density-based spatial clustering of applications with noise algorithm, can only accurately extract clusters of the same atomic density, neglecting several experimental realities, such as density variations within and between clusters and the nonuniformity of the atomic density in the APT reconstruction itself (e.g., crystallographic poles and other field evaporation artifacts). This clustering method relies heavily on multiple input parameters, but ideal selection of those parameters is challenging and oftentimes ambiguous. In this study, we utilize a well-known cluster analysis algorithm, called ordering points to identify the clustering structures, and an automatic cluster extraction algorithm to analyze clusters of varying atomic density in APT data. This approach requires only one free parameter, and other inputs can be estimated or bounded based on physical parameters, such as the lattice parameter and solute concentration. The effectiveness of this method is demonstrated by application to several small-scale model datasets and a real APT dataset obtained from an oxide-dispersion strengthened ferritic alloy specimen.