Please join the Hampton Roads section for a discussion of a new approach to maneuvering models based on machine learning. Traditional maneuvering models use a series of coefficients based on a Taylor series expansion of the equations of motion to describe the motion of a vehicle. The accuracy of these models is based on the number of coefficients used to describe the motion. Designers tend to limit motion to a single plane which reduces the number of coefficients required, but these techniques fail to capture nonlinear effects due to motion in multiple planes. In order to more accurately capture these nonlinear effects, additional coefficients are required, which significantly increases the number of experiments that must be run to establish these coefficients. Here, a methodology for directly establishing the forces and moments on a maneuvering body based on the instantaneous state of the vehicle is demonstrated using machine learning. Artificial neural networks are used to create prediction models for the forces and moments as functions of the angle of attack, side-slip angle, and non-dimensional angular velocity based on training data from computational fluid dynamic simulations using NavyFoam. These methods are shown to provide strong correlation with experimental data while providing significant reductions in computational cost.
Speaker Name & Bio
Jessica Puodziunas, Student, Virginia Tech
Jessica Puodziunas is a hydrodynamics engineering co-op who has been working at Newport News Shipbuilding (NNS) since May 2019. She is a junior at Virginia Tech studying Aerospace Engineering with a minor in Naval Engineering. Jessica has spent most of her time at NNS researching the use of machine learning to predict the maneuverability of submarines and establishing best practices using different meshing techniques for various NNS projects.
$15 - Student
$20 - Member
$30 - Non-member
Please register by Monday, January 20th!