Three-dimensional (3D) finite element (FE) models offer highly accurate and comprehensive capabilities for predicting acoustic scattering by underwater unexploded ordnance (UXO), being able to capture such effects as target tilt and burial, seafloor interface variation, volume scattering and reverberation, and clutter. This makes them highly appealing for munition detection and mitigation efforts, both for data-model comparisons and for UXO classifier training. This project aimed to develop and validate 3D FE model templates with customizability for different target/environment scenarios.

Technical Approach

The initial template development subtask consisted of constructing various FE model test environments, assessing computational accuracy and speed, with the goal of determining a setting that maximizes accuracy and solves within a predetermined timescale. This investigation resulted in the development of two 3D FE model templates, which differ in domain truncation and regime of application, but together offer full target/environment/source customizability. Additional features common to both templates include automatic domain scaling with frequency, a Fully Scattered Field Formulation for improved model accuracy, and far-field scattering determination capability through numerically determined Green’s functions. During the template validation subtask, the project team compared FE model scattering predictions against analytic solutions and against real target scattering measurements from the 2013-2017 Target and Reverberation Experiment, BAYEX, and CLUTTEREX experiments.

Interim Results

Model truncation options considered within template development yielded strongly different results in terms of speed and accuracy in 3D test environments. Two target scattering model templates were constructed, which combined various truncation methods to produce accurate scattering results with high efficiency. Template validation studies were largely successful and demonstrated target scattering prediction capability within a variety of target and environmental contexts. Environments considered included sandy and muddy environments, rough and rippled seafloor, and seafloor slopes and shelves. Different target shapes, compositions, and burial states were also considered.


A 3D FE model template introduces a cost-effective method for rapidly generating accurate scattering models in variable environmental contexts, and therefore offers great utility for training UXO classifiers within machine learning algorithms. Synthetic data can be assembled much more quickly, easily, and inexpensively than real scattering measurements, and FE models are capable of capturing the level of environmental complexity present in real environments. Model results can be produced in large quantities as required for accurate classifier training. Additionally, machine learning algorithms can offer quantitative assessment of model-data agreement, useful for assessing model performance and improvements.