On the Development of a Digital Twin for Underwater UXO Detection Using Magnetometer-Based Data in Application for...

Area of Science:

• Geophysics
• Machine Learning
• Computational Modeling

Background:

• Underwater unexploded ordnance (UXO) detection using magnetometers is challenging.
• Machine learning (ML) offers potential but requires extensive, costly datasets.
• Limited availability of relevant data hinders ML model training for UXO detection.

Purpose of the Study:

• To propose numerical modeling as a method for generating datasets for UXO detection.
• To validate the accuracy of numerical models against physical tests.
• To develop a methodology for creating a comprehensive UXO/non-UXO dataset.

Main Methods:

• Utilizing numerical modeling, specifically the finite element method, to simulate magnetic signatures.
• Conducting experiments to compare numerical model outputs with physical test results.
• Developing and applying a methodology for generating a dataset for discriminating UXO from non-UXO objects.

Main Results:

• High compliance between numerical model predictions and physical test outcomes was achieved.
• A simplified computational model reduced calculation time by nearly three times without compromising quality.
• A detailed methodology for dataset generation was presented, including assumptions for UXO and non-UXO objects.

Conclusions:

• Numerical modeling is a viable and cost-effective approach for generating datasets for ML-based UXO detection.
• The finite element method provides accurate simulations for underwater UXO identification.
• The developed methodology facilitates the creation of robust datasets essential for advancing UXO detection technologies.