Research On Early Warning Of Marine Structure Damage Based On Monitoring Database

Offshore platforms are the infrastructure for mining marine resources.After years of service,the platforms are prone to structural damage such as aging,corrosion and even fracture of components.By monitoring the health of the offshore platform and promptly detecting and warning before the structural damage is further expanded,the safety of the offshore platform structure and staff can be effectively guaranteed.At present,the mainstream research method for offshore platform damage is to analyze the changes of structural modal parameters based on the collection of structural response data to determine the occurrence of damage.Its advantage is that only a small amount of data is needed for analysis,but it is vulnerable to environmental noise and leads to inaccurate judgment results.If intelligent methods can be applied to this field on the basis of a large amount of monitoring data,the accuracy and robustness of damage identification can be improved,and the practicability of the method can be enhanced.Based on the long-term monitoring data of real ocean platforms,this paper combines the modal parameter extraction method in the traditional method with the intelligent learning method.First,wavelet packet decomposition and Hilbert-Huang Transform are used to extract damage-sensitive features from structural response data.The advantages and disadvantages of the two methods are analyzed,and the wavelet packet decomposition method is further analysis to filter out the wavelet function and decomposition level which are suitable for offshore platform monitoring data.Secondly,in view of the situation that there is no explicit sample label in the monitoring data of the offshore platform,the unsupervised fuzzy mean clustering algorithm and the single classification support vector machine algorithm are used to successfully identify the changes of the model parameters in the two-degree-of-freedom dynamic simulation system.Combined with the method,a two-stage damage identification method is proposed,which solves the problem that many current research methods cannot be directly applied to real platform damage identification due to the need for data labels.This method can successfully identify the damage of the FPSO.Then,after identifying the damage to the structure,the problem of damage location is further studied.In a complex structure,different damage locations may cause similar changes in the inherent characteristics of the structure,making it impossible to accurately analyze the damage location from the monitoring data of a single sensor.Based on the sensor array,this paper trains the damage discrimination model for each position of the sensor.By comparing the prediction results of each model,the damage location of the grid steel beam structure can be roughly located.In order to further improve the accuracy of positioning,the convolutional auto-encoding neural network is applied to the damage location,which effectively improves the accuracy of the model's recognition of the actual damage location and reduces the proportion of false alarms for other non-damaged locations.Finally,the structural damage identification method is integrated into the offshore platform monitoring data management system,which realizes the integration of monitoring data management,visual query,feature analysis and damage warning.