A numerical simulation and experimental calibration for the heat transfer analysis of additive manufacturing by blown powder technologies for metal components has been recently carried out with successful results [1]. However, dealing with the computational cost of this simulation is still an open question. The purpose of this work is to address this unsolved problem with a strategy to transform the serial implementation devised in [1] into a parallel one, that can efficiently exploit the computational resources of a supercomputer.

Virtual design process and validation of products built with metal additive manufacturing technologies with powder methods is not possible without a coupled thermomechanical numerical analysis tool that is capable of predicting the final distortion and the residual stresses of a piece and gets round a slow and expensive experimental campaign. However, the numerical analyses of metal AM processes are a remarkable challenge because they involve growing geometries, complex constitutive nonlinear thermomechanical laws, different scales in space and time and, most importantly, dealing efficiently with the massive computational cost of these simulations. This Master Thesis establishes the foundation stone of an innovative high performance scientific framework that will bring at last a reliable and computationally efficient answer to the current industrial needs. More precisely, it presents a parallel finite-element framework for the heat transfer analysis of metal additive manufacturing with powder methods. The main ingredient of this parallel finite-element model consists of a finite-element activation procedure that allows one to follow the energy input from the laser in space and time in a parallel environment. This moving heat source also governs the evolution of the geometry during the printing process. That is why the procedure is also responsible for the update of the computational domain. This model has been implemented in an advanced highly-performing object-oriented research code (FEMPAR), after thoroughly redesigning an existing serial implementation from another standard and procedural research code (COMET). The numerical experiments show that this novel framework reproduces in a parallel environment the behaviour of the original serial implementation, marking not only the achievement of the objective of this Master Thesis, but also an important milestone in the long-term goal of creating a high performance scientific tool for these kind of simulations.