AM Process Simulations Using Finite Element Analysis

Finite element analysis (FEA) techniques are increasingly popular tools for predicting how the output of different manufacturing processes changes with changing process parameters, geometry, and/or material. Commercial software packages such as SYSWELD, COMSOL, Moldflow, ANSYS, and DEFORM are used to predict the results of welding, forming, molding, casting, and other processes.

AM technologies are particularly difficult to simulate using finite element predictive tools. For example, the multi-scale nature of metal powder bed fusion approaches such as metal laser sintering and electron beam melting are incredibly time consuming to accurately simulate the use of FEA-based physics. AM processes are inherently multi-scale in nature, and fine-scale finite element meshes that are 10 pm or smaller in size are needed to accurately capture solidification physics around the melt stream, whereas the total component size could be 10,000 times larger than the element size.

In three dimensions, this means that if we apply a uniform 10 pm mesh size, we would need 10^s elements in the first layer and more than 10¹² elements in total to capture the physics for a single part that fills much of a powder bed. Since rapid movement of a point heat source is used to create parts, capturing the physics requires a time step of 10 ms or less during laser/electron beam melting, which for a complete build would require more than 10¹⁰ total time steps. To solve this problem, it would take billions of years for a number of elements to take this manytime phases on a fairly high-speed supercomputer. Thus, to date all AM simulation tools are limited to predictions of only a small fraction of a part, or very small, simplified geometries. Using existing FEA tools, several researchers are looking for ways to make assumptions whereby they can cut and paste solutions from simplified geometries to form a solution for large, complex geometries.

This approach has the advantage of faster solution time; however, for large, complex geometries, these types of predictions do not accurately capture the effects of changing scan patterns, complex accumulation of residual stress, and localized thermal characteristics. In addition, minor changes to the input conditions may invalidate simplified solutions. Thus, a simulation infrastructure that can quickly build up a “new” response to any arbitrary geometry, input condition, and scan pattern is the ultimate goal of a predictive AM simulation tool. Researchers have recently begun using dynamic, multi-scale moving mesh to accelerate FEA analysis for AM. These types of multi-scale simulations are many orders of magnitude faster than standard FEA simulations. However, multiscale simulations alone are still too slow to allow complete part simulation, even on the world’s fastest supercomputers. A new computational approach for AM is therefore needed, which can extend the FEA beyond its historical capabilities.

3DSIM, a new software start-up, is trying to do that for the AM industry. The software tools of 3DSIM include:
1. A new approach to formulating and resolving multi-scale moving mesh
2. A novel, finite element-based autosolver that predicts thermal evolution and residual stress very quickly in low thermal gradient regions
3. An insignificant number truncation Cholesky module, which eliminates the “null multiplication” calculations that occur when the finite element matrices are fixed, and

4. A stand-alone approach to the identification of periodicity in AM computations to enable feed-forward “insertion” of solutions in regions where periodicity is present and the solution is already known from a previous stage of time. These approaches are reported to reduce the time to solve large- scale AM problems by orders of magnitude. Taken together, these tools should make the full part of the problem solvable on a desktop GPU-based supercomputer in less than a day when algorithms are fully implemented in a combined software infrastructure. If realized, the ability to predict how process parameters and material changes affect part accuracy, distortion, residual stress, microstructure, and properties would be a significant advance for the AM industry.