Data-driven simulation of microstructure in powder bed fusion processes

• Applicant: Professor Dr.-Ing. Henning Wessels
• Subject Area: Computer-Aided Design of Materials and Simulation of Materials Behaviour from Atomic to Microscopic Scale ; Mechanics
• Term: since 2022
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 512730472

Lack of understanding in the interaction of process, structure and properties (PSP) still prevents the additive metal production of functional graded materials (FGM). In recent years, numerous numerical simulation tools have been developed to study the PSP interaction. However, these cannot keep up with the flexibility that additive manufacturing offers developers in terms of design. The reason is the transient nature of additive manufacturing and the sheer number of influencing parameters. On the other hand, sensor technology providing real-time information about additive processes is continuously improving. In sum, there is a wide variety of models and sensors that generate data. Data-driven simulation on the basis of machine learning methods offers the possibility to combine data from simulation, metrology and possibly experiments in the sense of a digital twin. The aim of this project is therefore to develop a methodology that can predict relevant properties of additively manufactured metal parts solely on the basis of thermal measurements. In the case of powder bed fusion (PBF) processes, the real-time simulation of the following parameters is particularly necessary for the understanding of the PSP interaction: 1) The size of the melt pool to monitor and evaluate the geometric accuracy of the process. 2) The formation of the microstructure in order to evaluate the mechanical properties of PBF manufactured parts. So far, simulations of microstructure evolution are only practicable for a few powder layers and therefore not suitable for the design of FGMs. The above-mentioned quantities cannot be measured or can only be measured by destroying the manufactured component. In this project a Convolutional Neural Network (CNN) will reconstruct a 3D temperature field online from 2D thermographic data. For this purpose the CNN will be trained offline with temperature data generated by numerical simulation. In addition, the reconstructed 3D temperature field will be used to simulate the formation of the microstructure during PBF by exploiting the equivalence of CNN to finite difference stencils and cellular automata. Thus, the approach to be developed in this project offers the potential to close the gap in sensor technology and virtual process design as well as in the understanding of PSP interaction. Instead of thermographic data, simulated surface temperatures will be used in this project. It is expected that upon successful completion of this project, the use of thermographic data will involve a considerable effort in calibration. Therefore, requirements for the accuracy of measured temperatures shall be defined. The actual calibration could be done in subsequent industrial projects.

• DFG Programme: Research Grants