Machine learning recognition of hybrid lead halide perovskites and perovskite-related structures from X-ray...

Area of Science:

• Materials Science
• Crystallography
• Machine Learning

Background:

• Accurate crystal structure identification is vital for novel functional materials discovery.
• Traditional methods for hybrid perovskite structure analysis are time-consuming and require specialized expertise.
• Challenges include potential false positives/negatives and the complexity of perovskite-related structures.

Purpose of the Study:

• To develop a machine learning (ML) approach for rapid classification of hybrid lead halide structures.
• To identify the dimensionality of inorganic substructures, polyhedral connection types, and overall structure types.
• To simplify and expedite the interpretation of powder X-ray diffraction (XRD) data.

Main Methods:

• Utilized a machine learning decision tree classification model.
• Trained and tested the model using powder XRD data, including theoretically calculated patterns.
• Validated the ML model's performance on experimental XRD data.

Main Results:

• Achieved average accuracies of 0.76 ± 0.07 for inorganic substructure dimensionality, 0.827 ± 0.028 for lead halide connection types, and 0.71 ± 0.05 for structure types among 14 common types.
• Extended dataset to 30 structure types, yielding accuracies of 0.820 ± 0.022, 0.74 ± 0.05, and 0.633 ± 0.018, respectively.
• Experimental validation demonstrated 1.0 accuracy for dimension prediction and 0.82 for structure type prediction.

Conclusions:

• The developed ML model effectively classifies hybrid lead halide structures using powder XRD data.
• This approach significantly simplifies and accelerates the interpretation of complex crystallographic data.
• Facilitates faster exploration and discovery of novel hybrid perovskite materials.