Study on process parameter optimization for electron beam melting additive manufacturing

• Main Author: Wang, Chengcheng
• Other Authors: Tor Shu Beng
• Format: Thesis-Doctor of Philosophy
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Engineering::Mechanical engineering
• Online Access: https://hdl.handle.net/10356/147705
• Institution: Nanyang Technological University

Electron beam melting additive manufacturing (EBM-AM), also known as selective electron beam melting (SEBM), is a mainstream metal additive manufacturing (AM) technique that employs an electron beam to process metallic powder feedstocks in vacuum environment (~ 10-4 - 10-5 mbar) at elevated temperatures (~ 500 - 1100 °C). Moreover, due to the high absorptivity and large penetration depth (~ 3 - 4 times of layer thickness) of electrons, SEBM is capable of fabricating parts with high build rates (up to 80 cm3/hr), minimized oxidation and residual stresses, and mechanical properties comparable to their wrought forms while superior to cast counterparts. However, the list of available materials for processing by SEBM is limited and the conventional method to obtain the optimized SEBM process parameters is by design of experiments (DoE), which is systematic but deficient in time and cost. This study aims at optimizing the process parameters for different metallic materials fabricated by SEBM with superior mechanical properties. Specifically, both the conventional DoE and the emerging machine learning (ML) methods are applied to discover the optimal process window and establish the paramount process-microstructure-property (PMP) relationships. In the first part, DoE method is employed to optimize the process parameters for SEBM. Stainless steel 316L (SS316L), an austenitic stainless steel popular for laser-based metal AM but less investigated for SEBM, is fabricated by SEBM from the feedstocks with two different powder size distributions. The SS316L parts with flat top build surface and high relative density (≥ 99%) are successfully built from the two kinds of powders. The samples fabricated from fine powders (nominal size ~ 20 - 63 μm) exhibit typical columnar-grain microstructure and obvious anisotropy in tensile properties, which are commonly seen in AMed metals and alloys. In contrast, the samples built from coarse powders (nominal size ~ 45 - 106 μm) show near-equiaxed-grain microstructure and isotropy in tensile testing, leading to higher tensile strength (ultimate tensile strength of 634.1 ± 8.6 MPa) and ductility (elongation of 62.8 ± 4.6%) than most of their AM-built counterparts. Secondly, fabrication of equimolar NiTi, an important shape memory alloy (SMA), is attempted by SEBM from pre-mixed elemental powders but to no avail due to the self-sustaining high-temperature synthesis (SHS) that occurs during processing. In the second part, a ML method is leveraged for fast and precise process optimization with small datasets for SEBM. Firstly, Ti-6Al-4V is chosen to build the baseline ML models, because it is one of the most important lightweight engineering materials and also the most extensively studied alloy in AM. Ti-6Al-4V samples are fabricated by SEBM with 56 sets of parametric combinations. Various ML models are integrated to predict the optimal process window, where samples are built with ~ 5% higher in ultimate tensile strength and ~ 87% faster in build rate than those processed by Arcam’s default parameter setting. Besides, a novel ML-centered tetrahedral framework is proposed to map the paramount PMP relationships over the entire parameter space with deep learning (DL). Secondly, to verify the generalization ability of this method, SS316L is fabricated with only 24 sets of parametric combinations. Transfer learning (TL) is employed to adopt the trained weights from the Ti-6Al-4V baseline models for SS316L datasets. TL could significantly improve the performance of the models and determine the process window for SS316L with high accuracy and short time. This work has demonstrated that the ML method could speed up discovering the optimal process window up to 5 times faster and 10 times cheaper for SS316L. Moreover, the significance and potential industrial applications of the ML-centered tetrahedral framework are depicted in the general discussion chapter of the thesis. Last but not least, the established ML-empowered methodology for process parameter optimization based on SEBM processing in this study can also be applied to other metal AM processes, such as laser-based powder bed fusion, directed energy deposition and binder jetting.