Intelligent design of multi-layered variable stiffness composite structure based on transfer learning

Variable stiffness composite structures offer more flexible design space than thin-walled metal structures and have greater potential for vibration-resistant design. When faced with multiple new types of design problems, the complex modelling and analysis procedures frequently prove to be both time-consuming and costly in terms of optimization. In this study, an innovative multi-layered variable stiffness (MVS) composite structure with high design flexibility is proposed, with images representation for curvilinearly stiffened paths, non-uniform layouts, and fiber and layup angles. Moreover, an intelligent optimization method based on transfer learning is proposed for addressing a variety of factors affecting dynamic design, including boundary types, structural features, and dynamic responses. The objective of the transfer learning model is to facilitate the inheritance and sharing of variable stiffness features, thereby enabling the efficient design of new problems with limited datasets. The validation of different examples shows that the transfer learning can effectively acquire the structural features from the existing source domain datasets, thereby significantly reducing the data for some new target domains by approximately 50 %. In comparison to the initial constant stiffness (CS) structures, the different optimized configurations indicate that the MVS composite structures are capable of effectively enhancing the dynamic responses by 10 %∼146 % for natural frequency and dynamic compliance. Furthermore, the MVS optimized configuration displays superior dynamic responses in some problems, when compared to the CS optimized configuration.