A novel mooring system anomaly detection framework for SEMI based on improved residual network with attention mechanism and feature fusion

The structural safety of mooring line is of paramount importance for maintaining the stability of floating structure and personnel health. Once mooring line failure occurs, it may lead to catastrophic consequences. Realtime monitoring and damage identification of mooring line integrity provide an early warning and response to mitigate potential risks and losses. This paper presents a motion-based mooring line anomaly detection framework, combining continuous wavelet transform, multi-scale feature fusion, and squeeze-and-excitation residual network (namely CWT-FFSeResNet). The framework aims to identify different degrees of mooring line damage in a semi-submersible platform (SEMI). Extensive numerical simulations under various sea conditions provide motion response data for different mooring line damage states. Subsequently, time-series motion data is converted into a time-frequency image, and feature fusion stacks images of three motions from the same time period on channel, forming a whole sample to represent the state of a mooring line. Compared with other existing models, the model shows a perfect performance in terms of accuracy and efficiency. Based on the test results of insufficient samples, the model indicates the potential to be established at a smaller time consuming. In addition, test experiments with different Gaussian noise levels demonstrated relatively satisfactory noise robustness of proposed method.