Finite element and neural network models for process optimization in selective laser sintering

An iterative method to optimize non-linear production processes is described. In contrast to classical design-of-experiment methods, it starts with a small number of experiments, based on which a preliminary data-based model is developed. From this model a vector of process parameters with (potentially) improved performance is calculated. The results of the experiment carried out with these new process parameters enlarge the database and result in an improved process model. The iteration is stopped if the requirements of the product properties are fulfilled. The method is applied to the selective laser sintering process of titanium powder. The goal of the optimization is to produce a ring with a prescribed geometry. One innovation of this approach is to feed the database with experimental results obtained from process simulation. The purpose of the simulation is to use the sintering machine only once to produce a part that is 'right first time'. The simulation is based on a three-dimensional finite element model of the selective laser sintering process. The input data are the various process parameters and the model is sufficiently detailed to predict the density and the bounding quality (sintering potential) of the manufactured part.