A method for optimizing process parameters in layer-based rapid prototyping

Abstract

In layer-based rapid prototyping, a volumetric object is approximated as a pile of slices with vertical walls. Process parameter selection in layer-based prototyping is a multicriteria multiparameter optimization problem. A number of criteria may be used for assessing the prototype's quality. Volumetric accuracy of shape approximation and building time are just two criteria taken in this work as an example. Criteria depend on process parameters, most commonly in a mutually contradictory manner. Model orientation and slice thickness constitute the minimum of process parameters to be considered, but others may also be added. For this reason, a neural network is used, trained by a number of input-output vectors, when analytical formulae representing the dependency of criteria on process parameters are not possible to develop and/or available numerical models take too long to execute. Neural network meta-models are used in the evaluation (cost) function of a genetic algorithm, each representing a particular criterion, and criteria are weighted according to the user's particular view. A case study is presented, referring to a wax model prototyping machine in which a particular tree for investment casting is built. A new criterion for assessing the quality of shape approximation is introduced, namely the local volumetric error per slice.