Fig. 2: Machine-learning-enhanced APT strategy to find CSRO and its application in large-scale APT simulation data. | Nature Communications

Fig. 2: Machine-learning-enhanced APT strategy to find CSRO and its application in large-scale APT simulation data.

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

Fig. 2

a Flowchart of the proposed ML framework to find CSRO within APT data. Procedure to build three classes of crystal structures and generate relevant synthetic z-SDMs: b Generating supercell. c Atoms shift from theoretical sites in x, y, z directions. d Randomly discarding some of the atoms. e Examples of simulated z-SDMs of Fe-Fe and Al–Al pairs. Relevant crystal cells are enclosed. Test of the obtained ML-APT recognition model in large-scale Fe-Al APT simulation: f Morphology maps of the simulated CSRO domains and detected ones via the proposed model using 1 × 1 × 1 nm3 scanning cubes. The size and colour of one circle denote the number of atoms within one domain. g Number densities versus APT-counted atoms corresponding to simulated and recognized CSRO domains. The result from the chemically-randomized dataset (Methods) is compared, and the Pearson’s correlation coefficients (PCC) are listed in the inserted table.

Back to article page