Improving Part Quality in Wire Arc DED Through Machining-Based Hybrid Manufacturing and Scientific Machine Learning

[Completed and presented on April, 2025] 

[Full article] [Brief description]

Supervisor: Dr. Shreyes Melkote, George W. Woodruff School of Mechanical Engineering, Georgia Institute of Technology

Introduction: 

• Wire Arc Directed Energy Deposition (DED) is a promising metal additive manufacturing process that enables the fabrication of large-scale metallic structures with high deposition rates

• However, inherent challenges such as geometric inaccuracies, inconsistent mechanical properties, microstructural inhomogeneities, and limitations in current geometry modeling hinder its broader industrial adoption

• Machining offers a promising approach for improving dimensional accuracy while simultaneously refining surface characteristics through severe plastic deformation

• Existing hybrid manufacturing strategies primarily focus on bulk deformation processes such as rolling and forging, which enhance microstructure but lack precision in geometric control

• Additionally, conventional data-driven and numerical approaches for bead geometry prediction remain either resource-intensive or computationally impractical for real-time applications

Project Aim: 

• The purpose of this project is to develop an understanding of the process-structure-property relationship in Robotic Hybrid Wire Arc DED by integrating interlayer machining interventions into the deposition process

• If successful, the proposed hybrid approach would enable more reliable, scalable, and efficient Wire Arc DED-based metal additive manufacturing with improved geometric accuracy, surface quality, and mechanical performance

Research Approach: 

• A radiused edge cutting tool is employed to induce severe plastic deformation by machining the deposited layers before subsequent layer depositions

• We hypothesize that combining interlayer machining with Wire Arc DED would refine the microstructure and improve the mechanical and geometric properties of the final structure beyond what standalone DED can achieve

• This research further develops a Physics-Informed Machine Learning framework as a computationally efficient alternative to traditional predictive modeling approaches for bead geometry prediction

Research Methodology: 

• A Hybrid Wire Arc DED system was developed using a six-axis KUKA KR 210 industrial robotic arm equipped with a welding gun for material deposition and a milling spindle for interlayer machining

• The local effects of machining-induced deformation on microstructure evolution were investigated using optical microscopy, EBSD, XRD, and Vickers microhardness testing

• The global effects of interlayer machining on geometric accuracy, surface roughness, and mechanical properties were evaluated by systematically comparing parts fabricated with and without machining interventions

• A Physics-Informed Neural Network solver was developed and trained using high- and low-fidelity deposition data to predict bead geometry with reduced computational cost

Research Objectives: 

• Investigate the local effects of machining-induced deformation on microstructure evolution in Hybrid Wire Arc DED

  • • Assess the effects of edge-radiused tool machining on the microstructural evolution of Wire Arc DED-manufactured parts

• Quantify the synergistic effects between machining-induced deformation and subsequent thermal cycling inherent to the Hybrid Wire Arc DED process

• Characterize grain refinement and phase transformation using EBSD, XRD, and optical microscopy

• Evaluate the mechanical implications of the refined microstructure through Vickers microhardness testing

• Evaluate the global effects of interlayer machining on geometric and mechanical properties

  • • Assess the impact of periodic machining interventions on dimensional accuracy, surface quality, and mechanical homogeneity

    • Examine the role of machining temperature on bead geometry evolution and material deformation

    • Compare parts fabricated with and without machining interventions across the full build height

• Develop a physics-informed predictive framework for bead geometry prediction

• Formulate governing equations and constraints for a Physics-Informed Neural Network solver

• Train and validate the model using high- and low-fidelity deposition data

• Assess model generalizability along the bead length and quantify the required computational resources compared to existing modeling approaches

A brief description of the research project can be found here.