Global sensitivity analyses of a selective laser melting finite element model: influential parameters identification

Abstract

Selective laser melting, an additive manufacturing technique that allows 3D printing of metal components, has recently gained great interest. Despite its growing popularity, this technology continues to suffer from deficiencies in standards and qualifications, factors that limit wide industrial use. Numerical modeling is currently a standard tool in production engineering for both process optimization and a more comprehensive understanding of the process physics. However, inherent to any model implementation and computation are various sources of uncertainties and errors. It is of major importance to identify them and assess whether their influences on outputs is significant. This can be accomplished using sensitivity analysis. To determine the parameters that most influence variability in the computational results, global sensitivity analyses were performed using an in-house developed nonlinear finite element model of the selective laser melting process. The computational load was limited by utilizing a 2D model for a single-layer simulation. The studies were performed on 26 process parameters including material properties, their dependencies on temperature, laser-related parameters, and boundary conditions, among others. Computed maximal temperatures and melt pool widths and lengths were obtained. Two sensitivity analyses were performed using the elementary effect method: one included process parameters and the other excluded them. Among the 26 input parameters tested, 16 did not show significant effects in either study. By including process parameters, they were found to be the most influential. By excluding them, the significant influence of the emissivity coefficient on output variability was revealed. These results evidence the parameters that should be given higher priority in modeling, the sources of error to be considered during validation, and insights into which parameters should be prioritized for further studies, both experimental and computational.