The process of reducing the cross-section of a wire by pulling the wire through a series of dies belongs to a special class of manufacturing processes that have a sequential nature. In each pass, the diameter of incoming wire is reduced by a certain factor. Determining number of passes and their configurations required to achieve the desired reduction while optimizing properties such as tensile strength, strain distribution, energy consumption, etc. is an optimization problem. An essential building block of this optimization problem is a model of a drawing pass that can predict the output properties for a given parameter configuration of the pass (forward inference), or its inverse, i.e. predict the configuration parameters required to achieve the desired properties.
In this paper, we present a case study on the application of Bayesian networks to address the problem of inference in multi-pass wire drawing. Also we explore the building of a generic model that can be used for any pass and compare its effectiveness with pass-specific models. Our findings so far are as follows: For forward inference, the generic model has prediction accuracy close enough to pass-specific models. Given that it can be used to solve a problem with arbitrary number of passes, this model is clearly more useful. Further, this being a generative model, it can be used for inverse inference as well.
