Improving Knowledge, Prediction, and Forecasting of Ships in Waves via Hybrid Machine Learning Methods (FORWARD)
Abstract
The objective of this project is the development and application of hybrid machine learning (ML) and modal-decomposition/reduced-ord er methods to improve knowledge, prediction, and forecasting of global/local loads, motions, and trajectories for ships operating in waves.To ensure the safety of structures, payload, and crew in adverse weather conditions, ships must have good seakeeping, maneuve rability and structural performance. In this regard, commercial and military ships must meet International Maritime Organization (IM O) Guidelines and NATO Standardization Agreements (STANAG). The prediction capability of ship performance in waves, along with the u nderstanding of the physics involved, is of utmost importance. Recent computational and experimental fluid dynamics studies have dem onstrated the maturity of computational tools for the prediction of ship performance in waves, including their assessment in extreme sea conditions. The computational cost associated with the analysis is generally very high, especially if statistical convergence o f relevant estimators is sought after and complex hydro-structural problems are investigated via high-fidelity solvers. Computations and experiments usually produce a large amount of data, whose investigation could shed light onto the problem physics. Nevertheless , data science studies for ships in waves are still very limited.We propose to investigate the combination (or hybridization) of ML approaches (such as, for instance, recurrent neural networks, RNN) with modal-decomposition/reduced-order methods (such as, for inst ance, dynamic mode decomposition, DMD) to (a) predict/forecast global/local loads, motions, and trajectories of ships operating in w aves (b) extract statistical predictive models for relevant kinematic and dynamics quantities of interest and (c) facilitate the int erpretation and explanation of ML results. The proposed approach combines methods that work well in predicting systems dynamics (su ch as RNN) with methods that offer a physical explanation via modal decomposition (such as DMD). The proposed hybrid approach is dat a driven and relies, whenever possible, on training sets from multi-information sources and heterogenous data types/scales (such as experimental data and computation results for global/local loads and/or multi-fidelity computations), providing also, whenever possi ble, the uncertainty associated with the prediction.Expected results include predicting/forecasting not only motions but also local global/loads on the hull, which can be extended to forecasting loads based on real-time data of ship motions only, fusing real-time data with pre-computed CFD results. Whenever possible, results will also include the prediction/forecast of unseen/untested geometri es/sea states/maneuvers/conditions via transfer learning. Potential applications span from seakeeping to maneuvering, from structura l safety/fatigue in heavy/extreme sea conditions to hydroacoustics. The proposed hybrid approach constitutes an original contributio n to ship hydrodynamics. ML and decomposition methods will be tested using applications of interest to ONR/ONRG and NATO Science and Technology Organization Applied Vehicle Technology task groups AVT-331 Goal-driven, multi-fidelity approaches for military vehicle system-level design, AVT-348 Assessment of Experiments and Prediction Methods for Naval Ships Maneuvering in Waves, and AVT-351 Enhanced Computational Performance and Stability & Control Prediction for NATO Military Vehicles. The proposed research is expecte d to meet the goals of the Naval S&T Research Area on Platform Design and Survivability.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 20, 2021
- Source ID
- N629092112042
Entities
People
- Matteo Diez
Organizations
- Consiglio Nazionale delle Ricerche
- Office of Naval Research
- United States Navy