Objectives In the fields of underwater vehicles and aerospace, ring-stiffened cylindrical shells are widely used. It is of great significance to conduct refined design of such structures to reduce structural weight while meeting strength and stability requirements. This study focuses on solving the optimization design problems of ring-stiffened cylindrical shells with each of ribs being design variables and aiming to overcome the difficulties in automatically distinguishing the buckling modes (whether it is global buckling or local buckling) and to improve the effectiveness and efficiency of the collaborative decomposition optimization algorithm assisted by domain knowledge.

Methods To achieve those goals, a ResNet neural network is trained for buckling mode image recognition of ring-stiffened cylindrical shells. Through ABAQUS parametric modeling, a large number of finite element simulations are carried out under different design variable combinations to generate buckling mode image datasets covering various conditions. These images are preprocessed and divided into training, validation, and test sets to train the ResNet101 model. Meanwhile, based on the domain knowledge of the coupling relationship between design variables and constraint quantities in ring-stiffened cylindrical shell design, specific grouping and resource allocation strategies are proposed.

Results The experimental results show that the trained ResNet neural network has an outstanding performance in identifying the buckling modes of ring-stiffened cylindrical shells, with an accuracy rate of 98%. For the optimization results of the case under consideration , the volume of the optimized solution obtained by the algorithm driven by domain knowledge is significantly reduced compared with the initial solution. Specifically, it is reduced by 38.4% compared to the initial solution, and the volume of the optimized solution obtained by this algorithm is further reduced by 7.06% on average compared with the optimization solution without domain knowledge.

Conclusions In conclusion, the combination of the neural network image recognition technology and the collaborative decomposition optimization algorithm with the help of domain knowledge is able to effectively solve the optimization problem of ring-stiffened cylindrical shell with different ribs. The high recognition accuracy of the neural network ensures the accurate calculation of stability constraint quantities, and the innovative strategies based on domain knowledge improve the optimization effect of the algorithm and provide a valuable reference for the design of ring-stiffened cylindrical shells in related fields.