실해역 계측 기반 탄소복합재 선체 구조응답 추정에 관한 연구 = A Study on Estimation of Structural Response of Carbon Composite Hull Based on Full-Scale Sea Measurement

다국어 초록 (Multilingual Abstract)

This study aims to develop a data-driven structural response estimation methodology for a carbon-fiber-reinforced plastic (CFRP) hull by utilizing long-term full-scale sea measurement data. Conventional design or laboratory testing approaches alone are often insufficient to capture the complex and highly dynamic loads experienced by composite hulls under real operating conditions. Therefore, a full-scale measurement system was designed and installed on the test vessel, incorporating strain gauges, accelerometers, GPS-based velocity and position sensors, and ocean environmental sensors such as wave height, wind speed, and wind direction. Extensive sea trials were conducted across various operating modes—including steady cruising, acceleration, deceleration, and maneuvering—and under a wide range of seasonal and environmental conditions to obtain large-scale, high-resolution time-series datasets.

Due to the inherent characteristics of full-scale measurements, the acquired data exhibited heterogeneous sampling rates, noise contamination, signal spikes, and missing segments. To ensure the reliability of the analysis, the data were refined through synchronization, filtering, missing-data reconstruction, and segment-wise quality verification. RMS and FFT-PSD analyses were performed to investigate the dominant frequency components and dynamic characteristics of hull responses. Furthermore, correlation analyses between hull stresses and various motion and environmental variables were conducted to formulate the basis for selecting input features for prediction modeling.

Using the processed dataset, multiple time-series prediction models—RNN, LSTM, Bi-LSTM, GRU, and ARIMA—were developed to estimate structural stress components (σx, σy, τxy). Each model was configured to learn nonlinear and temporally dependent characteristics of hull behavior, and model performance was evaluated using metrics such as RMSE and R². Although the overall prediction accuracy did not reach a high level, the Bi-LSTM model consistently showed superior performance relative to the other tested models. This demonstrates the fundamental validity of using full-scale sea measurement data to estimate unmeasured structural responses through a data-driven approach.

The comprehensive measurement, preprocessing, analysis, and modeling procedures established in this study provide a foundational framework for structural condition assessment of composite hulls. Future improvements in model architecture, input variable selection, and hyperparameter optimization are expected to further enhance prediction accuracy.

목차 (Table of Contents)

• 1. 서론 1
• 2. 계측 이론 및 시계열데이터 예측기법 9
• 3. 탄소복합재 선체 및 계측 시스템 25
• 4. 계측데이터 전처리 41
• 5. 계측데이터 분석 65
• 6. 딥러닝 기반 선체 구조응답 추정 95
• 7. 결과 및 고찰 116
• Appndix 122
• Reference 126