Autonomous Phase Identification in X-ray Diffraction: A Hybrid Approach with Bayesian FusionNet and Feature-Optimized Ensemble Learning

X-ray diffraction (XRD) plays a pivotal role in material characterization, offering valuable insights into crystalline structures. This study introduces a comprehensive framework for autonomous phase identification through a machine learning-guided approach. The proposed methodology comprises four key stages. In the pre-processing phase, raw data undergoes meticulous cleaning to eliminate noise, followed by normalization and smoothing procedures to ensure data integrity. Feature extraction involves a multi-faceted approach. Peak identification meticulously captures critical features such as peak position, intensity, and width within XRD patterns. Statistical features, encompassing mean, standard deviation, skewness, and kurtosis, provide a robust characterization of the dataset. The incorporation of Discrete Wavelet Transform further enriches the feature space by capturing both high and low-frequency information. For feature selection, a Hybrid Optimization Approach, combining the Kookaburra Optimization Algorithm (KOA) and White Shark Optimizer, is employed. This ensures an optimal subset of features for subsequent analysis. Phase identification is facilitated by a Bayesian FusionNet, integrating the strengths of Improved GhostNetV2, Bayesian Neural Network (BNN), and Feedforward Neural Network (FNN). The outcomes from these models are aggregated by taking the mean, enhancing the reliability and accuracy of phase identification. This innovative framework not only automates phase identification in X-ray diffraction but also showcases the efficacy of a hybridized machine learning approach, amalgamating optimization algorithms and neural networks for enhanced performance and interpretability. The proposed methodology holds significant promise for advancing material science research and facilitating efficient analysis in diverse applications.

Article file ID not found.

The authors declare no conflict of interest.

Not applicable

The datasets used in this study are openly available at [repository link] and the source code is available on GitHub at [GitHub link].

This work did not receive any external funding.