Using Artificial Neural Networks for Short-Term Ship Motion Prediction during Deck Landings

Rotorcraft operations in marine environments are heavily limited due to the complex nature of the open seas. Launch and recovery in rough seas is especially difficult due to the irregular motion of the flight deck; pilots must wait for an opportunity where the flight deck is safe to land, the so-called quiescent period. Various quiescent period prediction methods have been explored with limited success due to the technology available at the time. With the rise in machine learning capabilities, successful short-term forecasting may be achievable. This paper explores various artificial neural networks to forecast upcoming quiescent periods using previously recorded ship motion as training data. Artificial neural networks have an advantage over other quiescent period prediction models due to their superior ability to predict highly nonlinear data. Three neural networks are assessed: the Multi-Layer Perceptron (MLP), Radial Basis Function network (RBF), and the Long-Short Term Memory network (LSTM). These models were tested on simulated ship motion data collected from a simple frigate in sea state 5. The results from this study show that the accuracy of these short-term forecasting models is similar but vary greatly with regards to computational performance. Recommendations are made for integrating the Quiescent Period Prediction (QPP) model with existing deck landing procedures. Future work includes testing the QPP model with both an Uncrewed Aerial System (UAS) flight controller and pilot in the loop simulations.