Our research is mainly focused on computational modeling of additive manufacturing (AM), particularly on high-fidelity multi-physics modeling, which is an interdisciplinary field of computational mechanics and advanced manufacturing. Currently, we are interested in the mechanical applications of AM in the areas like aerospace, so most of the materials in our research are metals and some are composites. As illustrated in Figure 1, our research is aimed at 1) thoroughly understanding the fundamental relationships from the AM processes, to microstructure evolution, and finally to mechanical properties, by leveraging high-fidelity multi-physics modeling and experimental validation; and 2) guiding AM manufacturing parameter selection and machine development in an intelligent way by using computer simulation to reduce/replace most of the costly trial-and-error experiments. The past three-year-long pandemic has highlighted the importance of intelligent, resilient, and flexible manufacturing in the long-term sustainable development of human society, where the core strength stems from the digitalization and intelligence of manufacturing and the computational modeling of our research focus is no doubt a key enabler.
Figure 1 Schematic of our research focus: computational modeling of metal AM
Specifically, we have developed one of the most comprehensive (in the world) and high-fidelity/accurate series of computational models to understand the process-structure-property relationships and to guide the design and selection of AM machines and manufacturing parameters, as illustrated in Figure 1, including the multiphase flow models to simulate the laser-powder-molten pool interactions covering both (1) powder and (2) molten pool dynamics, (3) phase field and cellular automaton models to reproduce the microstructure evolutions at both the grain and dendrite scales, and (4) crystal plasticity models to predict the mechanical properties and thermal/residual stresses. While our group also works on integrating physics-based simulation and data-driven modeling/analysis as well as applying AM and modeling in other areas, we only summarize our core research achievements in below.
(1) Modeling powder dynamics, in Figure 2. There are two major procedures in powder-bed-based AM processes: powder spreading and melting/binding. The dynamic behaviors of powder particles in both procedures are highly complex but critical to the product quality. We achieved the most comprehensive understanding (to date) of the powder spreading mechanisms by leveraging high-fidelity modeling and experiment, revealed the major effects related to powder size, and corrected the long-standing inaccurate intuition that mixing coarse and fine powders can always increase the powder bed density [Acta Materialia 179 (2019): 158-171]. We further extended the model to different powder spreading machines and powder shapes, e.g., roller-type spreading of mixed polymer powder and short fiber for AM of composites [International Journal of Machine Tools and Manufacture 153 (2020): 103553; ESI Top 1% highly cited]. These unprecedented insights provide clear guidelines on powder size selection and AM machine design.
The phenomena of powder spattering and denudation during laser powder bed fusion AM process have been clearly observed in experiments, but the physical mechanisms were not well understood. After two years’ thinking and programming, we reported the first multi-phase flow model to reproduce powder spattering and denudation, and provided thorough understanding of the underlying physics, by explicitly simulating the thermal and momentum interactions between powder, vapor and ambient gas [Acta Materialia 196 (2020): 154-167]. This is the foundation to reduce defects caused by spattering. Furthermore, we reported the first high-fidelity model of binder jetting reproducing both the binder flow (impingement, spreading and penetration) and powder movement (collision, spattering and agglomeration) simultaneously [Acta Materialia 260 (2023): 119298].
Figure 2 Modeling of powder dynamics: (a) spreading, (b) spattering, and (c) binder jetting
(2) Modeling molten pool dynamics, including laser/electron beam powder bed fusion and directed energy deposition. As Figure 3 shows, we have made original contributions to remarkably advance the model in almost every aspect as summarized below, so that our model is one of the most accurate and powerful in the world, partially evidenced by winning 7 awards in molten pool tests in 2022 NIST AM Bench Simulation Challenges (the biggest winner) and being generously sponsored by a world-leading CFD software company.
(i) Physically-informed Heat Source Models. For electron beam, we developed the first physically-informed heat source model by simulating the electron-atom interactions [Acta Materialia 115 (2016): 403-412], considering the material compositions and electron beam properties, where the energy absorption and distribution are accurately predicted based on quantum mechanics equations instead of being intuitively assumed and then calibrated. For laser, we implemented the ray-tracing model to track the multiple reflection/absorption of each laser ray, considering the material compositions, local incidence angle, and laser wavelength and polarization, where the energy absorption and distribution keep fluctuating dynamically and vary with process parameters.
(ii) Rigorous Evaporation Model Incorporating Chemical Compositions. Serious evaporation and strong recoil pressure due to the high-intensity energy input play a dominant role in the molten pool dynamics. After realizing the drawbacks of the commonly used evaporation model derived in 1970s, we derived a new evaporation model from fundamental thermodynamics, particularly incorporating the chemical compositions [Physical Review Applied 14 (2020): 064039], which proved superior in predicting the keyhole dynamics and keyhole pore formation when validating against the high-speed X-ray imaging results [Nature Communications 10 (2019): 3088, ESI Top 1% highly cited; npj Computational Materials 8 (2022), ESI Top 1% highly cited].
Figure 3 Modeling of molten pool dynamics
(iii) Multi-Component/Material with Chemical Reactions. The first scenario is oxidation, which is inevitable in AM. To understand its impacts and guide powder reuse, we developed the first high-fidelity model to reproduce oxidation (also other gas-liquid metal reactions), oxygen mass transport, and the effect on molten pool flow during the AM process [Acta Materialia 249 (2023): 118824; Most downloaded in last 90 days]. The second scenario is AM of mixed powders for particle-reinforced composites. To simulate the evolution of reinforcing particles, we developed the first high-fidelity models for both micro-scale and nano-scale particles considering the size effect [Acta Materialia 235 (2022): 118086; International Journal of Plasticity 164 (2023): 103591]. The third scenario is AM of mixed powders for in-situ alloying, which is our on-going work. These models will be powerful tools for synergistic design of alloy compositions and AM process parameters, which is currently lacking but imperative.
(iv) External energy fields. Adding ultrasound in directed energy deposition AM can effectively refine grain structure and improve mechanical properties. We were the first to report a high-fidelity model reproducing the molten pool dynamics under ultrasound [Applied Physics Reviews 9 (2021): 021416, IF: 15, Featured article, AIP Scilight]. We also developed a thermoelectric magnetohydrodynamic model incorporating the electrodynamics with the Seebeck effect into the molten pool dynamics model, to simulate how external magnetic field influences the molten pool flow and reduce defects [Physical Review Applied 15 (2021), 064051]. These provide practical guidance on setting external fields in AM.
(3) Modeling microstructure (grain and dendrite) evolution, in Figure 4. We developed a phase field model to simulate the grain evolutions in AM, from nucleation, growth, to coarsening [npj Computational Materials 7(2021)], and we also incorporated the nano-particle induced heterogeneous nucleation and pinning effect on grain boundaries to understand how nano-particles refine grains in AM [Additive Manufacturing 47(2022): 102286]. To understand the solidification defects due to dendrite growth, we developed a multi-grid Cellular Automaton model to simulate dendrite growth and shrinkage pore formation [Additive Manufacturing 47 (2021): 102284; Computational Mechanics 69 (2022), 133-149]. Particularly, we pioneered in two-way coupling the model with a computational fluid dynamics model to simulate the dendrite growth in the molten pool, thereby revealing the impact of melt flow and locally varied solute concentration on the dendrite growth and formation of new grains in front of dendrites with various orientations [Additive Manufacturing 55 (2022): 102832]. These models have provided valuable guidance for the design of new AM superalloys with intrinsic hot cracking resistance [Acta Materialia 258 (2023): 119193].
Figure 4 Modeling of microstructure evolution at grain and dendrite scales
(4) Multi-scale modeling of thermal/residual stresses, in Figure 5. With the intention to understand micro-cracking, we developed the first micro-scale crystal plasticity model to simulate the thermal stress evolutions within each grain, by incorporating both the grain evolution from phase field simulation and temperature profiles from the molten pool dynamics model [Journal of the Mechanics and Physics of Solids 161 (2022): 104822]. We further used the model to reveal the dislocation structure formation under thermal stress at high temperature [Journal of the Mechanics and Physics of Solids 173 (2023): 105235], which is of great interest in materials field. At meso-scale, besides the temperature profiles, the realistic geometry including rough surfaces and internal voids is also implemented to reproduce the thermal stress concentrations [Materials & Design 196 (2020): 109185], which can rationally explain cracking. Leveraging the simulation and experimental results by our collaborators, we were the first to demonstrate the compression-tension thermal stress cycles being the origin of high-density dislocations in AM metals [Materials Research Letters 8 (2020): 283-290; 2022 MRL Impact Award]. By implementing the distributed inherent strain extracted from meso-scale simulations, the macro-scale thermal stress model can accurately and rapidly predict the thermal distortion and strain distribution of the product, which is an urgent demand in industry. Moreover, we developed a data-driven prognostic model to provide rapid but equally accurate prediction of temperature field to accelerate the entire simulation chain [Computer Methods in Applied Mechanics and Engineering 392(2022): 114652]. These research have received wide recognition and awards including two awards in property-related tests in 2022 NIST AM-Bench Simulation Challenges, and got sponsored jointly by CADFEM and ANSYS.
Figure 5 Multi-scale modelling of thermal stresses with temperature and structure input
The video below (which was an invited online seminar for Imperial College London) provides a brief introduction of some representative works of the group. More videos about our research (including recorded conference presentations & invited talks) can be found at our YouTube Channel “NUS Yan Group”: https://www.youtube.com/channel/UC_0KjxWp1t-o2fMWnJxliag