A deep learning model optimized by Bayesian Optimization with Hyperband for fast prediction of the elastic properties of 3D tubular braided composites at different temperatures

Three dimensional (3D) tubular braided composites are widely used in various industries due to their excellent mechanical properties and lightweight characteristics. However, traditional numerical and experimental methods face challenges in predicting mechanical properties quickly and accurately due to factors such as ambient temperature, component materials, and geometric parameters. To address this issue, this paper combines deep neural networks (DNN) and two-scale finite element analysis to accelerate the solution speed. The dataset is first obtained through a two-scale finite element model with temperature based on micro-CT. Then, the mapping model of macroscopic compression elastic properties and the influencing factors of material properties is established by DNN and Bayesian Optimization with Hyperband (BOHB) hyperparameter optimization algorithm. The rapid prediction of axial compression elastic properties of 3D tubular braided composites under different ambient temperatures, component materials, porosities, braiding angles and fiber volume contents is achieved. Finally, the accuracy of the predicted results of the constructed model is verified by experiments. Highlights: A BOHB optimized deep learning model coupled with a finite element framework is proposed Fast prediction of elastic properties of 3D tubular braided composites at different temperatures The accuracy of the prediction results of the constructed model is verified by experiments.