Bendability assessment of advanced high strength steels thanks to artificial neural networks

Abstract Thanks to their very important strength in addition to their considerable formability and ductility, the newly developed Advanced High Strength Steel (AHSS) allow carmakers to save mass on the vehicles they produce. However, before industrial implementation by OEMs, any grade has to be approved. The bending test defined in VDA238-100 [1] test specification is a key indicator of the material ductility and is increasingly requested in the scope of their homologation. The test consists in bending a square sample of sheet material with a knife. The sample is supported by two rolls. The aim of this test is to determine the maximum angle reachable by the sample before it breaks. We can also derive the maximum strain reachable by the material before fracture initiation. Mainly 2 different angles are defined: the angle after springback, that can be measured manually and the angle at maximum load, which is to be considered in the VDA238-100. The test specification gives instructions on the determination of the bending angle. Nevertheless, the approach is purely geometric and only relies on the parameters of the test. No influence of the tested material is taken into account. This approach can be improved because for AHSS that are very ductile and have a bending angle over 100°, the proposed formula can provide unphysical results. For this reason, the innovative proposed approach is to calculate the bending angle at maximal force and to determine the fracture strain thanks to Artificial Neural Networks, one of the most promising tools for resolution of engineering problems. This new approach considers a large panel of materials tested experimentally or numerically and significantly improves the prediction.