| Due to the complex marine environment and human factors,offshore platforms will inevitably be damaged during service.Therefore,the use of appropriate structural health monitoring technology(especially vibration-based damage detection technology)plays a vital role in ensuring the safety of offshore operations on offshore platforms.At present,many detection methods have been proposed to provide detection means for early warning of structural damage or abnormality.At the same time,due to the development of computing power and sensing technology in recent years,machine learning,especially deep learning algorithms,have become feasible and have been widely used in vibration-based structural damage detection.However,the application of deep learning to research on large and complex structures is still a bit less.This article explores the possibility of using a one-dimensional convolutional neural network to identify marine platform damage.Through the numerical simulation of the jackettype offshore platform under the excitation of the regular wave and the random wave in different directions,the damage location and the damage degree identification are carried out considering the different damage positions and the influence of noise.In order to improve the feature extraction ability and anti-noise ability of the convolutional neural network,a data preprocessing method based on convolution and deconvolution is proposed.On this basis,the physical model of the jacket-type offshore platform was studied with sinusoidal excitation and random excitation.At the same time,random decrement and convolutional neural network are combined to identify the damage of the offshore platform in the case of irregular waves.The research results show that the convolutional neural network method has good performance in identifying damage to offshore platforms,regardless of whether it is regular or irregular waves,single damage or multiple damage,and different wave directions.The data preprocessing of fusion convolution and deconvolution has better noise reduction and feature extraction functions in the case of noise,and its recognition accuracy in the case of noise is almost 100%.In addition,this paper also compares the three methods of convolutional neural network,long short-term memory neural network,convolutional neural network and long shortterm memory neural network in the identification of marine platform damage.The results show that the convolutional neural network is better than the other two methods in recognition accuracy and recognition efficiency.At the same time,under noisy environment and irregular waves,random decrement and convolutional neural networks are combined to carry out research on offshore platform damage identification.The results show that the fusion of random decrement and convolutional neural network methods have very good recognition performance for the damage location and damage degree of the offshore platform.The experimental results of the offshore platform show that the proposed method can well identify the damage of the offshore platform model regardless of whether it is random excitation or sinusoidal excitation. |
