Abstract
Structural defects such as porosity have detrimental effects on additively manufactured parts which can be reduced by choosing optimal process conditions. In this work, the relationship between process parameters and lack-of-fusion (LOF) porosity has been studied for the laser powder bed fusion (L-PBF) process of the Ti‐6Al‐4V alloy (Ti64). A physics-based thermo-fluid model is used to predict LOF porosity in the multilayer multitrack PBF process. To effectively map the high-dimensional processing parameters with porosity, an active learning framework has been adopted for the optimal design of experiments. Furthermore, a customized neural network-based symbolic regression tool has been utilized to identify a mechanistic relationship between processing conditions and LOF porosity. Results indicate that combining the physics-based thermo-fluid model for PBF porosity prediction with active learning and symbolic regression can find an appropriate mechanistic relationship of LOF porosity that is predictive for a wide range of processing conditions. This mechanistic relationship was further tested for other metal AM materials systems (IN718, SS316L) through non-dimensional numbers. The presented workflow effectively explores the high-dimensional process design space for different additive manufacturing materials systems.
Original language | English (US) |
---|---|
Article number | 103500 |
Journal | Additive Manufacturing |
Volume | 68 |
DOIs | |
State | Published - Apr 25 2023 |
Keywords
- Active learning
- Additive manufacturing
- Lack-of-Fusion porosity
- Laser powder bed fusion
- Symbolic regression
ASJC Scopus subject areas
- Biomedical Engineering
- Materials Science(all)
- Engineering (miscellaneous)
- Industrial and Manufacturing Engineering