Metamodelling with Artificial Neural Networks by using High Performance Computing infrastructures

Three dimensional modelling of sophisticated metal forming processes are usually very time consuming. One of the approaches allowing to shorten computing time is creation of metamodel, which is much simpler version of the model and usually prepared by using one of the black box methods. Although the metamodel is fast, its creation is very demanding, while it requires many executions of the model to obtain input and output data relations. This crucial step of metamodel creation can be accelerated through distribution of calculations on many computing nodes of High Performance Computing (HPC) infrastructures. This paper presents approach to metamodelling with Artificial Neural Network (ANN), which besides obtaining data, requires another demanding step of training. The way of ANN training with different configurations by using HPC infrastructures is also described in details. Afterwards, the results, obtained for case study analysing on crankshaft curvature after forging, are presented.