UC Berkeley

Additive Manufacturing (AM) is an umbrella term to describe processes that build a final part by fusing together raw material layer-by-layer according to a digital 3D CAD model design. This emerging but promising suite of technologies now have the capability of generating builds in the factory floor due to recent advances in precision, process repeatability, and a vast material library, which includes thermoplastics, metals, ceramics, and composites. AM now allows for the production of highly customizable parts with complex geometries in production-ready material grades. The use of numerical methods of AM processes in industry is an economical approach for optimizing technologies efficiently and in an environmentally conscious way.

A coupled Discrete Element--Finite Difference (DE-FD) model is leveraged for solving the thermal field of 316L stainless steel in a Selective Laser Sintering (SLS) manufacturing process. The DE method models the powder particles as discrete, lumped capacitance bodies resting on a solid substrate, which is discretized in space in a uniform grid of voxels. An additional heat-flux condition enforces the coupling between both domains. Data from experiments in literature serve to validate the present model by measuring the material melt pool shape from single-track simulations.

Two practical data-driven techniques are then presented for bridging the numerical simulation into an industrial setting. (1) A statistical Uncertainty Quantification (UQ) study of the material melt pool shape due to random powder layer thickness variations is explored. A voxel-wise powder depth heuristic modulates the local amount of laser energy that is able to penetrate through the powder layer down to the substrate. (2) A Reduced Order Model (ROM) is built via a Proper Orthogonal Decomposition (POD) of the thermal field data from a single run. The POD procedure is able to compress the data from the full simulation from 588,060 degrees of freedom to about 6 degrees of freedom with a 99.8\% data recovery upon reconstruction with significant speed-up in the time integration of the simulation.