A Study on Optimal Sensor Placement of Multi-Linked Floating Offshore Structure for Prediction Accuracy Improvement of Structural Response Using Distortion Base Mode Method

Abstract

Recently, a study on digital twin has been actively conducted to evaluate the structural intensity of ships and offshore structures. Previously, conservative structural safety was secured through design with a high safety factor and periodic inspections, but recently, using digital twin technology, it is possible to evaluate the structural intensity in real time. By synchronizing measured sensor data in real time with a digital twin model and performing simulation, evaluation of structural intensity such as structural response distribution, structural defect detection, actual fatigue life according to the sea state is performed, and further, accident prevention, maintenance plan can be established. Structural intensity evaluation through precise computational numerical analysis requires a lot of computational cost to use as a digital twin model that require faster simulation speeds. Therefore, a digital twin model is built by applying various techniques such as order reduction method and deep learning that can secure low computational costs. In this study, optimization of measured sensor placement was performed to improve prediction accuracy of reduction order model about ships and offshore structure. The target structure was a multi-linked floating offshore structure, and the bending stress was predicted by order reduction model based on distortion base mode. A structural response database was established through fluid-structure coupled analysis, and distortion base modes were selected using mode orthogonality and autocorrelation coefficients. Although it costs a lot of computational time to evaluate performance of all possible sensor placement combinations, the optimization technique saved about 8 times the time cost, and the root mean square error related in prediction accuracy with resulted in a sensor placement was reduced by about 84.0% compared to numerical analysis results. In addition, it was confirmed that the measured sensor data in the model test had a 28.6% improved prediction performance compared to the previously set sensor placement.