This study addresses the limitations of existing deep learning-based ship cross-section topology optimization methods, which are restricted to single-section ship structures. A fast topology optimization method for general ship cross-sections is proposed.

The proposed method employs automated parametric modeling and computation techniques to construct a large-scale, structurally diverse dataset of ship cross-section static analysis and topology optimization results. This dataset enables deep supervised learning to train the neural network to rapidly generate reasonable topology optimization configurations for various ship cross-sections. Furthermore, to address the challenge of directly applying neural network predictions in engineering analysis, an algorithm was developed to automatically reconstruct finite element models from the binarized density tensor output by the neural network, thus overcoming the limitations of element removal methods and ensuring mechanical consistency between network predictions and traditional iterative calculation results.

Experimental results demonstrate that applying the proposed method to predict the topological configuration of various ship cross-sections reduces computation time by two orders of magnitude, with an average prediction accuracy exceeding 90%. Sampling inspection results indicate that the finite element models reconstructed based on the network predictions avoid stress concentration, with less than 3% deviation in mechanical performance compared to traditional iterative calculations, further verifying the reliability of the proposed method.

The proposed method provides a general solution for the rapid topology optimization of ship cross-sections, reduces ship design costs, and possesses significant engineering value.

This paper proposes a hull transverse web topology optimization method based on UNet for application in the optimization design of complex ship structures.

Taking the transverse web of a very large crude carrier (VLCC) as the research object, a UNet topology optimization surrogate model is first created according to optimization mathematical principles. The finite element grid physical quantity is then mapped to the tensor to obtain the dataset for model training. Finally, the intersection over union (IoU) method is used to evaluate the training results, and the method is compared with the solid isotropic material with penalization (SIMP) method in terms of topology configuration.

The results show that this method can quickly output the material layout of the design domain, and compared with SIMP topology optimization, it can obtain the topology configuration more efficiently.

The proposed topology optimization method can provide a new design method for ship transverse web structures.