Design Optimization of Submarine Torispherical Domes using Data Information Integrated Neural Networks

: Torispherical end closures are critical structural components in submarine pressure hulls exposed to severe hydrostatic loads. Predicting their collapse strength remains computationally challenging due to the nonlinear interaction between crown elastic buckling and knuckle local buckling, coupled with high sensitivity to geometric imperfections. This study proposes an advanced structural synthesis framework tailored for the scantling optimization of submarine pressure hulls, leveraging Data Information Integrated Neural Networks (DINN) to capture the complex nonlinear buckling characteristics of torispherical end closures. A parametric finite element study generated 2,304 simulation cases by systematically varying crown radius, knuckle radius, thickness, and initial imperfections for three high-tensile steel grades (HY-80, HY-100, HY-130). The DINN architecture explicitly constrains the solution space by embedding the first-order physical derivatives (proxied by statistical correlation) into the loss function. This ensures that the model adheres to the monotonicity requirements inherent in shell buckling (e.g., must decrease as increases), even in sparse data regions. This approach prevents the model from learning spurious patterns and ensures robust generalization even in sparse data regions. The trained DINN model demonstrates superior predictive accuracy ( ) and physical plausibility compared to standard data-driven models. The results confirm that the proposed framework provides a highly reliable and computationally efficient digital surrogate for the structural design of submarine pressure hulls.