Neural Networks for Quantifying Laboratory Confocal Micro X-ray Fluorescence Measurements

Area of Science:

• Materials Science
• Analytical Chemistry
• Spectroscopy

Background:

• Confocal micro X-ray fluorescence spectroscopy (CMXRF) with polychromatic excitation presents significant quantification challenges.
• Complex dependencies, extensive calibration, and intricate data evaluation make CMXRF analysis time-consuming and difficult.

Purpose of the Study:

• To develop a simplified and automated method for CMXRF data quantification.
• To introduce the first application of a neural network for analyzing CMXRF data from homogeneous bulk samples.

Main Methods:

• Development of a simulation routine for generating CMXRF data of homogeneous bulk samples.
• Training a neural network on simulated data to perform quantification tasks.
• Utilizing depth profiling measurements for simultaneous analysis.

Main Results:

• The neural network successfully quantifies elemental concentrations for 53 elements.
• The AI model accurately determines sample density and surface position.
• Demonstrated substantial simplification of CMXRF data evaluation and reduced need for human input.

Conclusions:

• The developed neural network offers a significant advancement in CMXRF data analysis.
• The study highlights the potential of neural networks for feature extraction and prediction in complex spectroscopic data.
• This approach streamlines the quantification process, making CMXRF more accessible and efficient.