Prediction of Ship Roll Motion using Machine Learning-based Surrogate Model

기계학습기반의 근사모델을 이용한 선박 횡동요 운동 예측

Abstract

Seakeeping safety module in Korean e-Navigation system is one of the ship remote monitoring services that is employed to ensure the safety of ships by monitoring the ship's real time performance and providing a warning in advance when the abnormal conditions are encountered in seakeeping performance. In general, seakeeping performance has been evaluated by simulating ship motion analysis under specific conditions for its design. However, due to restriction of computation time, it is not realistic to perform simulations to evaluate seakeeping performance under real-time operation conditions. This study aims to introduce a reasonable and faster method to predict a ship's roll motion which is one of the factors used to evaluate a ship's seakeeping performance by using a machine learning-based surrogate model. Through the application of various learning techniques and sampling conditions on training data, it was observed that the difference of roll motion between a given surrogate model and motion analysis was within 1%. Therefore, it can be concluded that this method can be useful to evaluate the seakeeping performance of a ship in real-time operation.

한국형 e-Navigation의 내항성 안전 모듈은 운항 중인 선박을 실시간으로 모니터링하고 내항성의 이상 상태를 사전에 경고함으로써 선박의 안정성을 확보하는 선내 원격 모니터링 서비스 중 하나이다. 일반적으로 선박설계를 위한 내항성능은 주어진 조건에서 선체 운동 시뮬레이션을 수행하여 평가하여 왔다. 하지만 운항 중 선박의 내항성능을 실시간으로 평가하기 위해 이러한 시뮬레이션을 실제 운항조건에 맞추어 수행하는 것은 계산시간의 한계로 인해 현실적이지 않다. 본 연구에서는 기계학습 기반의 근사모델을 활용하여 선박의 내항성능 평가 요소들 중 하나인 횡동요 운동특성을 합리적으로 보다 빠르게 예측하는 방법을 소개하고자 한다. 다양한 학습 기법과 데이터의 샘플링 조건을 적용하여, 얻어진 근사모델의 결과와 운동해석 결과의 오차가 거의 1% 내로 일치함을 보였다. 따라서 이러한 방법을 활용하면 선박의 실시간 내항성능을 평가하는데 효율적으로 사용할 수 있을 것으로 판단된다.

Fig. 1 Hull panel of the model ship

Fig. 2 R² of TGP model by heading sampling interval

Fig. 3 R² of PLA model by heading sampling interval

Fig. 4 R² of TGP model by speed sampling interval

Fig. 5 R² of PLA model by speed sampling interval

Fig. 6 Example of full factorization

Fig. 7 Error rate by ship’s speed and heading

Fig. 8 Comparison of actual roll RAO and predicted roll RAO at maximum error rate

Fig. 9 Average error rate of alternative models by the number of training data set

Fig. 10 Maximum error rate of alternative models by the number of training data set

Fig. 11 Error rate by ship’s speed and heading(Case 6)

Fig. 12 Comparison of actual roll RAO and predicted roll RAO at maximum Error rate(Case 6)

Fig. 13 Histogram of peak roll RAO

Fig. 14 ITTC wave spectrum

Fig. 15 Comparison of actual roll response and predicted roll response at maximum error rate

Fig. 16 Comparison of actual m₀ and predicted m₀ by ship’s heading at 18knot

Fig. 17 Comparison of actual m₂ and predicted m₂ by ship’s heading at 18knot

Table 1 Principal particulars of the model ship

Table 2 Simulation cases for test data set

Table 3 Classification guidances for ship‘s motion & strength analysis

Table 4 Sampling intervals for training data set

Table 5 Prediction accuracy of top four techniques

Table 6 Sampling intervals for training data set

Table 7 Prediction accuracy of TGP & PLA models

Table 8 Minimum sampling interval of classification guidances

Table 9 Prediction accuracy of model with minimum sampling interval

Table 10 Sampling intervals for training data set of alternative models

Table 11 Prediction accuracy of alternative models

Table 12 Comparison of actual values and predicted values for m₀, m₂, Q_p at maximum error rate