William E. Faller

Improved Simulation of Ship Maneuvers Using Recursive Neural Networks

An improved Recursive Neural Network (RNN) maneuvering simulation tool for surface ships is described. Inputs to the simulation, cast in the form of forces and moments, are redefined and extended in a manner that more accurately captures the physics of ship motion; the new model is used to extend initial efforts toward RNN surface ship simulations. These extensions include improved formulations of propeller thrust, lift from deflected rudders, and the explicit inclusion of roll and pitch righting moments. Two maneuvers are simulated: tactical circles and horizontal overshoots. Simulation errors for the circles averaged over all maneuvers for such variables as speed, trajectory components and heading were 5% or less. The horizontal overshoot simulation errors were also 5% or less for the same variables with the exception of the transverse trajectory component. The explanation for the latter deficiency is believed to be the result of the exclusion of wind forces acting on the vehicle, which will be the subject of later work.

Uncertainty Analysis Applied to Feedforward Neural Networks

Abstract Three problems associated with uncertainty in feedforward neural network predictions are discussed. First, the uncertainty present in the input vector propagates through a trained network into the output vector, and this uncer- tainty is determined using the matrix of partial derivatives defining the change in each output with respect to the inputs. Second, because the partial derivative information conveys the relative sensitivity of a given output to each of the inputs, it can be used as a tool to determine the relevance of each of the inputs to the out prediction. Finally, the influence of random choices for training and testing data sets is investigated. The variability in these solutions provides a measure of the fossilized bias error in the network with respect to development decisions. The approaches are illustrated using examples of four-quadrant propeller predictions.

A Tool to Predict the Four-Quadrant Performance of the Wageningen B-Screw Series for Ship Performance Simulations

Abstract Feedforward neural network predictions of the four-quadrant thrust and torque behavior for the Wageningen B-Screw Series of propellers have been developed. The resulting prediction tool can be used to provide reasonable four-quadrant thrust and torque predictions for propellers, interpolating automatically between measured data.

Feedforward Neural Networks Applied to Problems in Ocean Engineering

The Maneuvering and Control Division of the Naval Surface Warfare Center, Carderock Div. (NSWCCD) along with Applied Simulation Technologies have been developing and applying feedforward neural networks (FFNN) to problems of naval interest in Ocean Engineering. A selection of these will be discussed. Together, they show the power of the nonlinear method as well as its utility in diverse applications. Experimental data describing a subset of the B-Screw series of propellers operating in all four quadrants have been reported by MARIN in the Netherlands. The data contain varying pitch to diameter ratios, expanded area ratios, number of blades and advance angle. These four variables were used to train a FFNN to predict the four-quadrant thrust and torque characteristics for the entire B-screw series over a range of beta from 0 to 360 deg. The results show excellent agreement with the existing data and provide a means for estimating 4-quadrant performance for the entire series. For submarine simulation and design, knowledge of the total forces and moments acting on the hull as a function of angle-of-attack, sideslip angle and dimensionless turning rate across a large parameter space is required. This data is acquired experimentally and/or numerically and can be used to train a FFNN to act as a Virtual Tow Tank or Virtual CFD Code. The network not only recovers the training data but also serves as a very fast, nonlinear six degree-of-freedom look-up table of the forces and moments acting on the hull throughout the parameter space described by the vehicle dynamics. Example solutions demonstrating this approach will be presented. Wave impact loads pose continuing problems for vessels in high sea states, with damage to hatches and appendages, suggesting that these loads may be greater than current design guidelines. Such forcing is complex and often difficult to estimate numerically. Experimental data were acquired at NSWC to measure the hydrodynamic loads of regular, nonbreaking waves on a plate and a cylinder while varying incident wave height, wavelength, wave steepness, plate angle and immersion level of the plate/cylinder. Predictions of wave impact forces from a FFNN trained on the experimental data will be presented.Copyright

Development of an advanced ship simulation & control system using neural networks

Initial efforts in a three-year program to develop an advanced simulation & control system for surface ships are described. The system employs a recursive neural network to simulate the motion of the vehicle in the presence of wind and waves. The faster-than-real-time response of the trained network will permit the use of advanced control techniques such as model-reference control or predictive control and the implementation of path planning for improved performance in the presence of adverse environmental conditions. Early results showing accurate simulation of a U.S. Navy ship conducting overshoot (zig-zag) maneuvers in the presence of wind are shown

Neural Network Models of Forces and Moments on a Model of LCAC

Data acquired from experiments with a 1/12 th scale model of an air-cushioned, amphibious vehicle (LCAC) towed in calm water were used to train two sets of six feedforward neural network (FFNN) models. The networks are used to model the six forces and moments acting on the hull of the vehicle as a function of available input data measured during the experiment. Each network is used to predict one force or moment component as its sole output. The first set of models uses inputs and outputs in dimensional form, whereas the second set uses a dimensionless representation. Results are presented comparing the predictions of the models with the measured data in various forms. Two error measures are used to quantify the results, an average angle measure and a correlation coefficient, and they indicate good solutions in every case. The intent is to use these models as input quantities in a larger simulation effort to develop a six degree-of-freedom, nonlinear, time domain simulation of LCAC to learn about the dynamics of the vehicle in calm water and in irregular waves.

RECURSIVE NEURAL NETWORKS: SIX DEGREE-OF-FREEDOM MANEUVERING SIMULATION CAPABILITIES

The technology exists to develop accurate, faster than real-time six degree-of-freedom (6-DOF) simulations directly for model scale and full scale submarines using recursive neural networks (RNN) and model scale and full scale trials experimental maneuvering data. Further, these techniques show no loss of fidelity for severe or emergency recovery maneuvers which include propeller backing and are dominated by forced unsteady fluid dynamics. Overall, the results show that RNN 6-DOF maneuvering simulations can provide accurate predictions of all maneuver types (crashbacks, rise jams, dive jams, rudder jams, turns, vertical and horizontal overshoots). Further, across hundreds of maneuvers, the results indicate that these techniques provide accurate predictions for both maneuvers used to develop the RNN 6-DOF simulation as well as for validation maneuvers comprised of novel control sequences. For the 250 maneuvers, shown herein, the average simulation error in depth was less than 10 ft (full scale), less than 0.25 kts (full scale), and less than one degree in pitch and roll. In addition, to the results shown, the critical components of the RNN technology are also briefly revisited herein. These include, a form of dynamic similarity, based on the time-varying speed U(t), which is required to scale the wide range of speeds and flow field conditions present across all maneuvers, and the force representations used for the control surfaces and hull. These force representations are believed to be sufficiently simple, to permit rapid development of the corresponding force terms for the simulation of missiles and fighter aircraft.