Research On Online Damage Identification Of Offshore Platform Based On Machine Learning

The ocean is rich in resources and it is an important source of human energy supply.Offshore platform is an indispensable infrastructure for marine energy development.In complex and changeable marine environment,the structure of offshore platform may be damaged by fatigue or corrosion.If these damage can not be found in time,it will lead to serious consequences.At present,the damage identification research on offshore platforms is mainly based on environmental loads,and modal analysis is used to identify and analyze the response data.In recent years,traditional machine learning and deep learning methods have developed rapidly.If these intelligent methods can be applied to this field,the efficiency and accuracy of damage identification of offshore platforms will be greatly improved,and the security of platform structures will be improved.In view of the development trend of intelligent health monitoring of offshore platform,this thesis combines traditional machine learning and deep learning with damage identification of offshore platform,and the research is mainly reflected in three aspects.Firstly,the unsupervised dimension reduction method is used to reduce the dimension of the extracted structural features,and the data is mapped from the high-dimensional space to the low-dimensional space,and the visual analysis of the data distribution is carried out.Secondly,the anomaly detection methods based on density,clustering and single classification are used to judge the damage of the structure through the detection of outliers in the monitoring data.Thirdly,based on the Generative Adversarial Networks(GAN),an nonlinear system characteristic change identification network called NLS-GAN is proposed,which integrates deep learning method into damage identification of offshore platform.These methods learn from the monitoring data under normal working conditions,and the damage of the new data can be evaluated after the diagnosis model is obtained.At the same time,aiming at online identification in the process of structural long-term monitoring,a solution of online learning is given,and an online boundary incremental learning method is proposed.Online boundary incremental learning method enables the model to continuously learn and update from real-time monitoring data.On the premise of ensuring the accuracy of the model,the pressure of data storage can be reduced and the learning efficiency can be improved.For the above damage identification methods,the second-order nonlinear dynamic system simulation experiments are used to verify the effectiveness of the methods.After the parameters of the simulation system are changed,these methods can find the changes of the inherent characteristics of the system from the output response of the system.For the proposed online boundary incremental learning method,this thesis compares it with other online methods,and finally proves that this method can achieve the best online learning results.In this thesis,the proposed method is applied to the damage identification of FPSO single point mooring structure monitoring data,and the identification results are in agreement with the real situation.