Process-based deep learning model: 3D prediction method for shot peen forming of an aircraft panel

Abstract

Shot peen-forming is a more precise method of forming aircraft panels than conventional methods. The traditional method of acquiring the process parameters relies mainly on prior theoretical knowledge and trial-and-error. Despite the finite element method's ability to replace some experimentation, it still cannot realize the design of shot peen forming processes parameters of an aircraft panel based on a known contour. This study uses an innovative model-based deep learning approach to predict aircraft panel deformation and active design the shot peening parameters. The prediction time is less than 1 second, resulting in a significant reduction in computational time. The shot peen forming process parameters and the geometric structure characteristics of the aircraft panel are divided into independent channels to establish a high-dimensional feature map, which are used to train the deep learning model. The forming contours of the 2024-T351 high-strength aluminum alloy panel are predicted under different shot peening processes. In addition, the process parameters are designed according to the known contour of the forming process. To verify the precision of the proposed method, the designed shot peen forming process is used to manufacture a single curvature aircraft panel with a curvature radius of 3500 mm. There is good agreement between the forming contour and the theoretical design contour. The maximum deformation error is less than 1 mm and its mean error is 7.8%. The mean curvature radius error is 5.668%. The proposed method provides a new and practical reference to the precise design of the shot peen-forming process.