Predicting Dynamic Process Limits in Progressive Die Sheet Metal Forming

Abstract

Abstract Tool makers have a limited selection of tools and are afforded limited flexibility during progressive die try-outs when attempting to identify suitable process control parameters and optimize throughput. The performance of a given tooling design hinges on selecting a suitable stroke rate for the press. Cost efficiencies are realized when operating a press at higher stroke rates, but risk subjecting the sheet metal strip to larger, uncontrolled oscillations, which can lead to collisions and strip-misalignment during strip progression. Introducing active control to the strip feeder and lifters can offer increased flexibility to tool makers by allowing the strip progression to be fine-tuned to reduce strip oscillations at higher stroke rates. To alleviate uncertainties and assist in fine-tuning the process control parameters, machine learning models, such as an artificial neural network, are constructed to predict whether a given set of process parameters will lead to a collision or strip-misalignment during the strip progression. The machine learning models are trained using a dataset of FEA simulations which model the same progressive die operation using different process control inputs for the feeder, lifter and press. The machine learning models are shown to be capable of predicting the outcome of a given process permutation with a classification accuracy of about 87 % and assist in identifying the dynamic process limits in the progressive die operation.