東華大學圖書館 |

語系: 繁體中文

說明(常見問題)

回圖書館首頁

手機版館藏查詢

登入

回首頁

切換: 標籤 | MARC模式 | ISBD

FindBook

Google Book

Amazon

博客來

Physics-Informed Data-Driven Models for Ship Response Prediction Using Global Wave Data.

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	Physics-Informed Data-Driven Models for Ship Response Prediction Using Global Wave Data./
作者:	Schirmann, Matthew L.
出版者:	Ann Arbor : ProQuest Dissertations & Theses, : 2021,
面頁冊數:	277 p.
附註:	Source: Dissertations Abstracts International, Volume: 83-05, Section: B.
Contained By:	Dissertations Abstracts International83-05B.
標題:	Naval engineering. -
電子資源:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28845437
ISBN:	9798471102460

Physics-Informed Data-Driven Models for Ship Response Prediction Using Global Wave Data.
Schirmann, Matthew L.

Physics-Informed Data-Driven Models for Ship Response Prediction Using Global Wave Data. - Ann Arbor : ProQuest Dissertations & Theses, 2021 - 277 p.

Source: Dissertations Abstracts International, Volume: 83-05, Section: B.

Thesis (Ph.D.)--University of Michigan, 2021.

This item must not be sold to any third party vendors.

Time-and-place specific wave model data, traditional ship response prediction tools, and onboard measurements present an opportunity to train data-driven models for improved motion and structural response predictions. In a real-world implementation, these predictive models would furnish vessel owners and operators with information to support underway, maintenance, and deployment decisions. As demonstrated in this work using over 16,000 30-minute windows from two operational sister ships, although data-driven models are powerful enough to outperform traditional seakeeping and structural response predictions in many instances, retaining physics-based information in these models is paramount to consistent performance improvement. To further incorporate physics-based information, a novel, neural network structure was developed that leveraged shared layers to enforce consistent physics in multidirectional wave data for response prediction. Unlike traditional NN structures trained in this work, this shared-layer approach allowed variation in the number of wave directions considered between samples of multidirectional wave data, as demonstrated using three unique wave data sources with 3 to 24 discrete spectral directions defined for a given time and location. The developed structure also allowed robust wave data interpolation in space and time. In addition to constructing physics-informed data-driven models, a study was conducted regarding the significance of wave data source selection for response prediction and fatigue damage tracking. This study considered a full year of data from several wave data sources at three separate locations to quantify the resultant differences in response predictions and accumulated fatigue damage. Additionally, a novel approach was developed for RAO corrections using full- or model-scale seakeeping data and Gaussian process regression. Like other models in this work, the developed approach leveraged physics to enforce that the RAO correction reverted to physics-based RAOs in regions of the input space far from train samples. A final demonstration combining this RAO correction approach with the shared-layer neural network was performed using the full dataset from the two sister vessels. The results of this work demonstrate significant promise for real-world implementation of data-driven response prediction models, especially those that retain physics-based information.

ISBN: 9798471102460Subjects--Topical Terms:

3173824
Naval engineering.
Subjects--Index Terms:

Data-driven models

Physics-Informed Data-Driven Models for Ship Response Prediction Using Global Wave Data.
LDR:03823nmm a2200421 4500 001 2347543
005 20220801062232.5
008 241004s2021 ||||||||||||||||| ||eng d
020 $a 9798471102460
035 $a (MiAaPQ)AAI28845437
035 $a (MiAaPQ)umichrackham003849
035 $a AAI28845437
040 $a MiAaPQ $c MiAaPQ
100 1 $a Schirmann, Matthew L. $3 3686811
245 1 0 $a Physics-Informed Data-Driven Models for Ship Response Prediction Using Global Wave Data.
260 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2021
300 $a 277 p.
500 $a Source: Dissertations Abstracts International, Volume: 83-05, Section: B.
500 $a Advisor: Collette, Matthew D.;Gose, James W.
502 $a Thesis (Ph.D.)--University of Michigan, 2021.
506 $a This item must not be sold to any third party vendors.
506 $a This item must not be added to any third party search indexes.
520 $a Time-and-place specific wave model data, traditional ship response prediction tools, and onboard measurements present an opportunity to train data-driven models for improved motion and structural response predictions. In a real-world implementation, these predictive models would furnish vessel owners and operators with information to support underway, maintenance, and deployment decisions. As demonstrated in this work using over 16,000 30-minute windows from two operational sister ships, although data-driven models are powerful enough to outperform traditional seakeeping and structural response predictions in many instances, retaining physics-based information in these models is paramount to consistent performance improvement. To further incorporate physics-based information, a novel, neural network structure was developed that leveraged shared layers to enforce consistent physics in multidirectional wave data for response prediction. Unlike traditional NN structures trained in this work, this shared-layer approach allowed variation in the number of wave directions considered between samples of multidirectional wave data, as demonstrated using three unique wave data sources with 3 to 24 discrete spectral directions defined for a given time and location. The developed structure also allowed robust wave data interpolation in space and time. In addition to constructing physics-informed data-driven models, a study was conducted regarding the significance of wave data source selection for response prediction and fatigue damage tracking. This study considered a full year of data from several wave data sources at three separate locations to quantify the resultant differences in response predictions and accumulated fatigue damage. Additionally, a novel approach was developed for RAO corrections using full- or model-scale seakeeping data and Gaussian process regression. Like other models in this work, the developed approach leveraged physics to enforce that the RAO correction reverted to physics-based RAOs in regions of the input space far from train samples. A final demonstration combining this RAO correction approach with the shared-layer neural network was performed using the full dataset from the two sister vessels. The results of this work demonstrate significant promise for real-world implementation of data-driven response prediction models, especially those that retain physics-based information.
590 $a School code: 0127.
650 4 $a Naval engineering. $3 3173824
650 4 $a Artificial intelligence. $3 516317
650 4 $a Computer engineering. $3 621879
650 4 $a Geographic information science. $3 3432445
653 $a Data-driven models
653 $a Digital twin
653 $a Structural health monitoring
653 $a Motion forecasting
653 $a Neural networks
653 $a Physics-based models
690 $a 0468
690 $a 0800
690 $a 0464
690 $a 0370
710 2 $a University of Michigan. $b Naval Architecture & Marine Engineering. $3 3346711
773 0 $t Dissertations Abstracts International $g 83-05B.
790 $a 0127
791 $a Ph.D.
792 $a 2021
793 $a English
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28845437