Efficient image-driven algorithms for sheet forming optimization based on deep learning.

With the increase of complexity of Computer-Aided Engineering (CAE) models and practical problems, the evaluation cost of sheet forming simulation is commonly expensive. Surrogate models have been employed for efficient evaluations but trouble the problem of inverse scattering. Moreover, the accuracy of the forming evaluation based on the widely used Forming Limit Diagram (FLD) is influenced by the non-working regions. In this study, image-processing techniques are employed. Simultaneously, two novel image-driven Generative Inverse Networks (GINs) are proposed to improve the sheet-forming design's efficiency and accuracy. Through validations, GIN Version d (GIN-V_d) is more efficient and can obtain higher accuracy. However, because the desired optimum should be given in advance, such applications might be limited. In comparison, the GIN Version g (GIN-V_g) is more flexible. [ABSTRACT FROM AUTHOR]