Real-time modelling of incremental multi-pass flow forming by a hybrid, data-based model

The incremental flow forming process features a large number of process parameter combinations that can be varied from pass to pass or during a pass. In the future however, a more efficient utilization of this large number of process parameter combinations and a compensation of process disturbances could be required. This is due to a rising demand for increasing the part complexity, e.g. by graded property structures or a more complex geometry. In this context, innovative approaches like closed-loop property control and optimal control are advantageous, but require fast process models of flow forming that are not state of the art. This paper thus proposes a new modelling approach of multi-pass flow forming especially taking the transfer behavior between process parameters and wall thickness evolution from pass to pass into focus. A hybrid modelling approach is developed that combines knowledge about the incremental process character with empirical data regression to a basic analytic relation. The basic relation is further extended by a multi-layer neural network to enhance the overall model accuracy. This hybrid modelling approach is finally validated using experimental data. Thus, it is shown that a suitable model structure was found in context of a future closed-loop control or optimal control for multi-pass flow forming.