Using the finite element method and artificial neural networks to predict ductile fracture in cold forming processes

Apart from the calculation of the plastic formability of metals the prediction of ductile cracks in cold forming processes is very important in order to design these processes efficiently. Therefore, many crack criteria have been developed and implemented in several FEM Programs. These criteria scale the crack prediction down to one value and they are qualified to detect the most endangered areas occurring cracks during the forming process quite well. All these criteria have two significant disadvantages: on one hand none of these criteria consider the whole forming history and on the other hand the detected critical value is not applicable to other forming processes. Therefore a new method to predict ductile fracture in cold forming processes has been developed. Various upsetting, bending and extrusion tests were designed in order to provoke a failure during the forming process. All these processes were modelled by means of the Finite Element Method to acquire the whole forming history (including the first principle stress, the equivalent stress and the equivalent strain starting with the first deformation to the first crack occurrence) for the area where the first fracture occurs. Basal in the results way a database with forming histories which all will lead to an failure during a forming process was built up. This database is used to train an artificial neural network. The artificial neural network will be able to predict a failure for new forming histories. The paper gives an overview over the use of the artificial neural network, the calculation of the forming histories and the used forming processes as well as the interaction between the Finite Element Method and the artificial neural network.