Accurate X-ray-based thickness determination of aluminum sheets using ACO-optimized MLP neural networks

Area of Science:

• Materials Science
• Non-destructive Testing
• Computational Physics

Background:

• Accurate aluminum sheet thickness is vital for aerospace and automotive industries.
• Traditional thickness measurement methods often require known alloy compositions, limiting their applicability.
• Variations in aluminum alloy composition can affect measurement accuracy.

Purpose of the Study:

• To develop a novel x-ray-based system for accurate aluminum sheet thickness measurement.
• To create a system that is independent of specific aluminum alloy compositions.
• To enhance measurement precision and reduce computational complexity for industrial applications.

Main Methods:

• Utilized Monte Carlo N-particle simulations for modeling x-ray interactions.
• Employed an optimized multi-layer perceptron (MLP) neural network for thickness prediction.
• Applied ant colony optimization (ACO) for efficient feature selection.

Main Results:

• Achieved precise thickness predictions for four aluminum alloys (1050, 3105, 5052, 6061) from 1 to 45 mm.
• Demonstrated high accuracy with a mean relative error (MRE) of 1.06% on test data.
• The developed system showed superior performance compared to conventional measurement techniques.

Conclusions:

• The proposed x-ray-based system provides a scalable and calibration-free solution for real-time thickness measurement.
• This novel approach overcomes the limitations of composition-dependent traditional methods.
• The system offers a robust and accurate tool for diverse industrial applications requiring precise aluminum sheet thickness determination.