Dual-branch convolutional neural network with attention modules for LIBS-NIRS data fusion in cement composition...

Area of Science:

• Materials Science
• Analytical Chemistry
• Spectroscopy
• Artificial Intelligence

Background:

• Cement composition (CaO, SiO2, Al2O3, Fe2O3) is critical for strength and durability.
• Real-time monitoring of cement components ensures optimal raw material ratios.
• Laser Induced Breakdown Spectroscopy (LIBS) and Near Infrared Spectroscopy (NIRS) offer rapid, non-destructive analysis but have limitations (matrix effects, spectral overlap).

Purpose of the Study:

• To develop a novel fusion method integrating LIBS and NIRS data for enhanced cement component quantification.
• To overcome the individual limitations of LIBS and NIRS through synergistic data analysis.
• To achieve accurate and stable real-time cement composition analysis.

Main Methods:

• A dual-branch convolutional neural network with an attention module (DBAM-CNN) was developed.
• The dual-branch CNN extracts atomic (LIBS) and molecular (NIRS) information.
• Attention modules refine feature weights for capturing spectral fingerprint information; SHAP analysis was used for feature interpretation.

Main Results:

• The DBAM-CNN method demonstrated superior performance compared to existing fusion strategies and single spectroscopic techniques.
• The model achieved high precision in real-time cement composition analysis.
• SHAP analysis confirmed the method's ability to highlight key LIBS and NIRS features for improved quantitative outcomes.

Conclusions:

• The DBAM-CNN effectively integrates complementary LIBS and NIRS data, enhancing cement composition analysis.
• This approach addresses information redundancy and feature loss, offering a reliable solution for real-time monitoring.
• The study advances spectroscopic data fusion techniques, paving the way for improved cement quality control.