Data-enabled intelligent design framework for underwater composite cylindrical shells

By leveraging all available data, a data-enabled intelligent design framework for underwater composite cylindrical shells is proposed, which significantly reduces the required expert knowledge and computational expense. Eight mixed design variables are considered simultaneously, including seven continuous variables (geometrical ratios and layup angles) and a discrete variable (composite material type). Combined with finite element analysis, probabilistic machine learning, which incorporates Gaussian process prior and acquisition function (probability of improvement, expected improvement, lower confidence bound, and their Markov Chain Monte Carlo versions), is utilized to enhance buckling load-bearing capacity. Specific geometrical ratios, layup angles, and composite material type are determined. Data-enabled intelligent design for underwater composite cylindrical shells has been accomplished with remarkably few finite element model evaluations. Without the help of expert knowledge, layup design [89.4°/−36.2°/22.0°/−45.9°/−47.0°]_s is discovered. The angles of both the innermost and outermost layers are approximately 90°, which can effectively increase buckling load-bearing capacity. The proposed data-enabled intelligent design framework is advantageous for underwater composite cylindrical shells and can be applied to structural design problems with complex geometric configurations, constraints, and loads.