Active learning-driven control point optimization method for efficient modeling of complex stiffened curved shells

The mesh-mapping technique has recently garnered significant attention due to its ability to model stiffened curved shells accurately. The precision of the modeling depends on the number of control points, particularly for complex stiffened curved shells. However, using too many control points results in a time-consuming process of model training and mapping, thereby impeding the structural optimization efficiency. To overcome this challenge, this study proposes an active learning (AL) function that considers the mechanical response and uncertainty of stiffened curved shells. This function is applied to select control points of the radial basis function (RBF) model. The control points with the most substantial AL function value are added to the control point set, and the process is iterated until convergence is achieved. Additionally, the proposed method is compared to other AL strategies to confirm its high efficiency. Furthermore, a data-driven optimization approach, based on the deep neural network method, is employed to minimize the weight of the complex stiffened curved shells. Finally, two engineering examples are presented to verify the effectiveness of the proposed method, which reduces the modeling time by 83.87% and 66.29%, respectively, in comparison to using all control points in mesh-mapping. Additionally, the total time of modeling and optimization is decreased by 42.63% and 59.35%, respectively. In conclusion, the proposed control point optimization method based on active learning-driven for complex stiffened curved shells has substantial potential in efficient modeling, significantly reducing the time of meshmapping modeling, and improving engineering's optimization efficiency.