Composite cylindrical shells with openings represent an innovative primary structural configuration for submersibles, effectively enhancing their structural performance and, consequently, their endurance capabilities. Under hydrostatic pressure conditions, it is essential to optimize the structural design parameters to improve the buckling load and static strength margins of composite cylindrical shells with openings. However, the stress state around openings is highly complex, making simulation and computational analysis challenging and time-intensive. Therefore, leveraging advanced computational methods to assist in structural design is imperative. This study presents an experimental study on the hydrostatic pressure bearing performance of filament-wound composite shells. Based on the experimental results, an accurate finite element method (FEM) model has been established. This study has developed a data-rich framework, generating a synthetic dataset of strength performance under hydrostatic pressure through 16,996 FEM analysis. Subsequently, machine learning (ML) models have been established to predict strength performance. This study presents the validation of the superiority of Deep Neural Network (DNN) models in predicting structural strength under hydrostatic pressure, and evaluates the prediction error. Additionally, for the first time, the DNN model is applied to estimate the structural performance under hydrostatic pressure, with optimization achieved through the integration of a genetic algorithm. The introducing of DNN greatly accelerates the computational speed compared to full FEM-based optimization procedures. By optimizing representative long and short shell models separately, it is demonstrated that ML-based optimization can replace full FEM-based procedures, facilitating preliminary structural design optimization and offering more efficient design solutions.
Read full abstract