Metallic parts created by laser fusion additive manufacturing (AM) processes display spatially varying polycrystalline microstructures whose morphology is largely driven by the localized temperature history field in proximity to the deposited material melt, a function of the deposition processes and parameters. This microstructure morphology determines the materials local response, e.g. modulus and strength, making its prediction of utmost importance to design engineers. However in order to resolve microstructural features at the resolution required for material property prediction (i.e., at the melt pool scale), high spatial resolution of the temperature field is required. These highly refined thermal simulations typically result in prohibitively large thermal history datasets that are burdensome to store and computationally expensive to analyze. Examples of refined thermal simulations can take days to solve, and produce output files in excess of 100 gigabytes, which not only cause time delays but also memory issues on certain machines.

These challenges motivate the need for a computationally tractable methodology to link process to structure as a step towards relating process to properties. To this end, we present an exploration of the link between thermal process and microstructure by employing dimensionality reduction techniques and neural networks. A clustering algorithm is used to reduce the dimensionality of large datasets of temperature field time histories from finite element simulations. Clustering of the thermal history data will provide valuable insight into the underlying structure of the temperature history field by storing data with similar features into groups. Representative microstructure morphology for each cluster is obtained by performing a kinetic Monte Carlo grain growth simulation. Voxelized representations of the microstructure are then encoded by latent vectors to extract invariant microstructural features using deep learning techniques. Lastly, a neural network is trained to find a mapping between the reduced thermal dataset and the encoded microstructures, resulting in a computationally efficient mapping between process and structure. Additionally, performing the mapping between the reduced datasets rather than in the original feature space has the advantage of identifying mappings between the invariant features of the latent spaces which can provide meaning insight into the process-structure link.

The proposed methodology is demonstrated on a non-trivial geometry for the Laser-Direct Energy Deposition of a steel material system. Insights into the underlying data structures will be presented along with an analysis of predictive capabilities versus method efficiency. This method will begin the path forward towards the development of novel efficient analysis techniques for AM components, providing a needed tool to AM engineers and increasing the viability of AM components for industry.


Exploration of Process-Structure Link in Laser Fusion Additive Manufacturing via Clustering and Neural Networks

Division / Sponsors: 
Aerospace, Bioengineering, Applied Mechanics, Design Engineering Division, Advanced Energy Systems, Manufacturing Engineering Division, Noise Control and Acoustics, Nondestructive Evaluation, Diagnosis, and Prognosis (NDPD), Nuclear Engineering Division, Process Industries Division, Fluids Engineering, Heat Transfer, Materials, Microelectromechanical Systems, Safety Engineering and Risk Analysis, Computers & Information in Engineering


Paper Proceedings: https://asmedigitalcollection.asme.org/IMECE/IMECE2020/volume/84478

IMECE2020 - Virtual Conference

Supporting the need among researchers worldwide for additional online and multimedia content, ASME now offers the ASME Conference Video Collection. This platform will collect and provide long-term access to conference presentation videos, encompassing all technology areas within ASME's technical conference and events program. About 30,000-40,000 engineers across all mechanical engineering and related disciplines participate in ASME conferences each year. This includes authors of approximately 6,000 to 8,000 technical papers published by ASME, the majority of whom also submit video presentations for our virtual conferences. The video collection will be available as an add-on to existing product packages or as a separate subscription.