A self-adaptive dynamic-adjustment springback compensation algorithm for single-stamping forming of double-curved hull plates based on neural network

Abstract

Springback compensation of double-curved hull plates faces challenges of low accuracy and high difficulty, often requiring multiple stamping operations to approximate the target surface. This work proposes a new self-adaptive dynamic-adjustment springback compensation algorithm for double-curved hull plates. Unlike the traditional step-by-step approximation approach, this algorithm enables single-stamping compensation. Additionally, a springback ratio prediction method based on Bayesian optimization (BO) and Backpropagation (BP) neural network is proposed to enhance the performance of the compensation algorithm. The compensation algorithm dynamically adjusts the compensation surface according to the springback ratios predicted by the BP neural network, automatically calculates the error between the target surface and the surface after springback based on the compensation surface (compensated springback surface), and finally provides a compensation surface that meets the error requirements. The finite element (FE) simulation and experimental validation for the single-stamping forming of double-curved hull plates have been conducted. The results demonstrate that the springback ratio prediction method proposed exhibits relatively high accuracy, offering more reliable springback predictions for the compensation algorithm. When compared with an existing method, the new springback compensation algorithm demonstrates superior accuracy in single-stamping forming.