A thorough quantitative understanding of the MAM process requires insights from different types of computer simulations, namely: thermal, mechanical, metallurgical, and fluid-dynamics. The extremely high computational cost of such multi-physical fields simulations has been addressed in previous research by either decreasing the simulation domain size and/or by adopting gross simplifications. Instead, this project proposes to exploit the power of multiscale modelling and machine learning algorithms to realise reliable component-scale simulations for MAM. The underlying idea originates from the incremental build nature of MAM and the fact that the involved phenomena during deposition of the material increments are very similar at each location. The variation in the characteristics of the deposited increments does not originate from a significant change in the involved physics but only from a (modest) change in the imposed boundary conditions. Component-scale simulations of MAM process can therefore be considered as solving a large number of nominally similar small-scale models, each under different boundary conditions than the rest. The computational cost of such repetitive simulations can thus be significantly reduced by employing a specific class of machine learning toolboxes called 'metamodels', i.e. parsimonious statistics based learners. We have already demonstrated the effectiveness of the proposed idea for thermal analysis of selective laser melting (SLM) process.The proposed research aims to develop a flexible simulation platform which, for a given set of SLM process conditions, trains and validates a metamodel on the basis of few small-scale high-fidelity simulations that incorporate thermal, mechanical, and microstructural computations. For a given geometry and process parameter set, the platform will use the metamodel for predicting a range of relevant parameters/characteristics within the SLM build ? such as temperature history profiles, hot spots, residual stress distributions, part distortion, and microstructural parameters like grain size and texture. The prediction results will be subsequently exploited by a crystal plasticity finite element (CPFE) framework for assessing the mechanical response of SLM components. Importantly, the existing high-end experimental facilities at Empa, ETHZ and PSI will be employed for dedicated experiments to i) understand the involved physical phenomena, ii) accordingly derive relevant numerical models, and iii) ultimately evaluate the reliability and effectiveness of the proposed simulation strategy.
Marking its evolution as a central feature of the Fourth industrial revolution, additive manufacturing broadly refers to the layer-by-layer 'printing' of spatially precise on-demand 3D structural objects with enhanced design freedom relative to conventional manufacturing technologies. Effective contribution of metal additive manufacturing (MAM) to the vision of Industry 4.0 necessitates addressing key challenges related to mechanical reliability of the printed parts and their costs. Due to the lack of a better alternative, trial-and-error strategies are often adopted to optimise MAM process conditions. A deeper understanding of the various physical phenomena associated with MAM process might be achieved through numerical simulations, that would in turn enable a more systematic optimisation of process conditions required to achieve the goal of 'first-time-right' high-quality production.A thorough quantitative understanding of the MAM process requires insights from different types of computer simulations, namely: thermal, mechanical, metallurgical, and fluid-dynamics. The extremely high computational cost of such multi-physical fields simulations has been addressed in previous research by either decreasing the simulation domain size and/or by adopting gross simplifications. Instead, this project proposes to exploit the power of multiscale modelling and machine learning algorithms to realise reliable component-scale simulations for MAM. The underlying idea originates from the incremental build nature of MAM and the fact that the involved phenomena during deposition of the material increments are very similar at each location. The variation in the characteristics of the deposited increments does not originate from a significant change in the involved physics but only from a (modest) change in the imposed boundary conditions. Component-scale simulations of MAM process can therefore be considered as solving a large number of nominally similar small-scale models, each under different boundary conditions than the rest. The computational cost of such repetitive simulations can thus be significantly reduced by employing a specific class of machine learning toolboxes called 'metamodels', i.e. parsimonious statistics based learners. We have already demonstrated the effectiveness of the proposed idea for thermal analysis of selective laser melting (SLM) process.The proposed research aims to develop a flexible simulation platform which, for a given set of SLM process conditions, trains and validates a metamodel on the basis of few small-scale high-fidelity simulations that incorporate thermal, mechanical, and microstructural computations. For a given geometry and process parameter set, the platform will use the metamodel for predicting a range of relevant parameters/characteristics within the SLM build ? such as temperature history profiles, hot spots, residual stress distributions, part distortion, and microstructural parameters like grain size and texture. The prediction results will be subsequently exploited by a crystal plasticity finite element (CPFE) framework for assessing the mechanical response of SLM components. Importantly, the existing high-end experimental facilities at Empa, ETHZ and PSI will be employed for dedicated experiments to i) understand the involved physical phenomena, ii) accordingly derive relevant numerical models, and iii) ultimately evaluate the reliability and effectiveness of the proposed simulation strategy.The project will benefit from ongoing collaborations with researchers from Empa (Dr Leinenbach), ETH Zürich (Dr Marelli, Prof De Lorenzis & Prof Mazza), Bremen Univesity, Germany (Dr Mohebi), and Northwestern University, USA (Prof Wagner). The proposed research requires the involvement of three PhD students, where the laboratory of the PI at Empa has committed to funding 50% of the PhD positions.
Last updated:18.06.2022
