Study On Damage Detection Of Offshore Platform Structure Based On SVM-Machine Learning

The offshore platform has always been under harsh environmental conditions during its long service life,so damage will inevitably occur.Therefore,in order to ensure the safe operation of the platform and avoid safety threats and economic losses,it is particularly important to develop damage detection methods.The traditional damage detection methods have obvious limitations,and structural health monitoring based on vibration analysis has become a hot issue in domestic and foreign scholars.The structural health monitoring extract the features which are sensitive to structure damage through the dynamic response of the structure,and evaluating the structure safety state based on these features.However,the complex and changeable marine environment often interfere with the change of dynamic response caused by structural damage,causing the existing deterministic analysis methods to be difficult to identify structural damage.In this paper,the support vector machine algorithm is applied to damage identification of offshore platforms from the perspective of machine learning,and a specific analysis of the fracture of a brace of an in-service jacket platform is carried out through numerical simulation.This paper presents an implemented method of applying support vector machine algorithm in the field of offshore platform damage identification.Construction of feature vectors and model selection of support vector machines are discussed in detail.Besides,a preliminary study on the damage location is conducted.The selection of feature nodes and combination of data from different points are discussed during the construction of the sample set.A sample set of different feature vectors are established and the optimal feature dimensions are determined for three support vector machines with different kernels functions respectively.In the model selection of support vector machines,parameter optimization is performed on support vector machines with different kernel functions,and the influence of parameters on the generalization results are discussed.Finally,the performance of the optimized support vector machine with Gaussian kernel function on missing data and noisy data is discussed,and identification of damage location is discussed based on the optimized model and other methods.