Domain knowledge-driven decomposition-based large-scale optimization for ship cabin structures

Abstract: When conducting fine-tuned optimization design for ship cabin structures, s longitudinal and transverse girders with variable stiffness are inevitably considered. This leads to a large number of design variables in the optimization problem for cabin structures. Objectives This paper proposes a domain-knowledge-driven large scale optimization algorithm for ship cabin structures based on a decomposition optimization framework that combines domain mechanical knowledge with general black-box optimization algorithm. Methods The proposed algorithm groups the design variables into location variables and size variables and decomposes the original problem into a series of low-dimensional subproblems. Due to the monotonicity and locality of each bending stress, shear stress and deformation constraint, the subproblems with larger constraint margins are prioritized for optimization. All of the location variables are grouped into one subproblem, and the corresponding subproblem's objective function is to maximize the minimum constraint margin. Each girder size variables are separately grouped, and the corresponding subproblem's objective function is the weight of the cabin structures. Additionally, a surrogate model is introduced to predict the constraints of each subproblem quickly, and the sample infill criterion is adopted only to the constraint surrogate model. Results The experiment results show that the algorithm can reduce the overall weight of the case in this work by 43.5% compared to the upper bound. Conclusions the proposed algorithm has a higher optimization efficiency and can obtain better optimization solution compared to both the differential evolution algorithm with directly using finite element method and general black-box optimization algorithm.