During the service life of offshore jacket platforms, the harsh marine environment leads to severe structural corrosion and damage, necessitating structural health monitoring. Ensuring the accuracy of numerical finite element models (FEM) requires critical model updating. This study introduces an improved DNN-OOA model updating method by incorporating actual structural responses into the optimization objective function and considering non-uniform corrosion of the structure. We utilized Pyansys to automatically generate large-scale datasets, simplifying the simulation process. An accurate and responsive surrogate model is generated using the improved deep neural network (DNN), and the optimal solution for the parameters to be corrected is sought through the Osprey optimization algorithm (OOA), completing the FEM updating. The main innovation of this study lies in incorporating non-uniform corrosion caused by the real marine physical environment into the model updating process. This phenomenon is employed to determine the updating range for different structural members. Furthermore, the parameters subject to updating include structural damage to the members and changes in the upper mass. Incorporating the structural response under static loading into the optimization objective function allows for a more comprehensive reflection of the structure’s dynamic and static behavior, addressing the regression confusion problem in the optimization process of purely modal frequency updating. Experimental results demonstrate that the proposed improved DNN-OOA model updating method effectively eliminates inaccuracies in simulated structural responses and mitigates the local optimum problem inherent in pure modal frequency updating. In the updated scaled jacket platform FEM, the maximum relative error of the modal frequencies is reduced to 2.624%, and the maximum error in structural response is reduced to 3.510%. This approach provides a more accurate and reliable FEM for the maintenance and safety assessment of offshore jacket platforms.
Read full abstract