Augmented Adaptive Filter for Real-Time Sea State Estimation Using Vessel Motions Through Deep Learning
Hamed Majidiyan, Hossein Enshaei, Damon Howe
Abstract
Abstract Given the recent rise of autonomous shipping, the knowledge of environmental seaway awareness has become imperative for automated control systems onboard and autonomous units in general. The real-time wave data enables the decision-maker to evaluate dynamic wave loads on the marine structure and efficiently monitor/operate the vessel’s conditions on and off-site. To this end, the wave buoy analogy (WBA) aims to benefit from 6 DOFs ship responses to estimate the induced wave specifications. Although this concept has been extensively scrutinized, the integration of AI and its subsets, machine and deep learning in this field is new. So, the objectives of the current research are twofold; the first is to take advantage of temporal and spectral information for seaway estimation in a deep learning model. The second is to propose a solution addressing two significant issues of WBA: the time lag between the exciting wave and corresponding estimation and stabilizing outputs. The study uses a setup to integrate an ARIMA-EMD predictive filter to adjust sampling window size adaptively on the upcoming data based on the anticipated fluctuations. A major advantage is that filter helps to reduce time lag while improving estimation stability. The response data is synthesized according to a tested semi-submersible platform at Shanghai Jiao Tong University (SJTU) research institute, where the time series is transformed into the pictorial presentation through several processing stages to feed a deep learning network. For training, the concept of transfer learning has been utilized by a pre-trained network, AlexNet. The result is promising and provides a foundation for attaining real-time wave updates in operational scenarios.