Error compensation and path planning for sheet metal bending process based on improved modeling algorithms

Sheet metal bending, an essential part of manufacturing, is increasingly used in transportation, aerospace, and related fields. Challenges related to forming accuracy and path generation have always been major concerns in industrial applications. A prediction model based on GA-BPNN (Genetic Algorithm-Backpropagation Neural Network) was developed, determined, and optimized to mitigate the impact of springback on the forming angle. A 9-input model, considering material classification, was adopted to further enhance prediction accuracy. An artificial potential field (APF) method based on discrete geometric calculations of main and auxiliary (DMA) points was introduced to improve the reliability of a collision-free bending path for the shape characteristics of sheet metal parts, and further optimized using the rapidly exploring random tree (RRT) method. The results indicated that the maximum errors in forming angles were reduced from 2.45 to 0.4°, and 0.15°, respectively, when applying different models. The experimental collision-free path maintained a safe distance of 2 mm, ensuring that the robotic arm operated smoothly to the target position. Additionally, this study could advance the development of a more accurate and efficient bending process in industrial settings.