Taek Cho, Kyu K.
M.S. (Master of Science)
Department of Mechanical Engineering
Direct Energy Deposition (DED) is an additive manufacturing (AM) process capable of producing complicate-shaped or functionally graded components, and it is getting intense attention as a revolutionary technology to satisfy high demand in manufacturing process for the aerospace, automotive, and medical industries. However, the repeatability in geometries and properties of fabricated products is one of the most challenging issues for the DED process to be fully utilized, requiring comprehensive understanding of effect of processing conditions on the properties of fabricated parts, and development of relations among those conditions and properties. That is the motivation of this research. In this study, a novel research approach was utilized to synergistically combine a physics-based model and data-based machine learning models to develop relations among processing parameters and melt-pool/clad geometries. A two dimensional multiphysics melt-pool model was developed in COMSOL Multiphysics by incorporating mass, momentum, and energy conservation equations. The model was validated with experimental results, and it was used to predict melt-pool temperature distribution, geometries and contact angles of a single clad under the influence of buoyancy and Marangoni forces for the various processing conditions such as laser powers, travel speeds, and powder feed rates. Two machine-learning models were developed with data (i.e. total of 288 cases) generated in the melt-pool model for the wide range of operating conditions. The first machine learning model is an Artificial Neural Network (ANN) which was utilized to make fast and precise predictions on height, depth and width of the clad. The second machine learning model is a Support Vector Machine (SVM) model which was used to predict the optimal process window for the DED process. The best ANN model could predict the clad geometry with R2 = 0.94, and the best SVM model could predict the optimal process window with weighted average accuracy of 90%. Using the developed model, we could find the correlation among process parameters and melt-pool geometries, and the optimal parameters that could result in the desired clad geometries.
Tayebati, Sina, "A Theory-Supported Machine Learning Model For The Prediction of Melt Pool Geometry and Optimal Process Window in Metal Additive Manufacturing" (2021). Graduate Research Theses & Dissertations. 7718.
Northern Illinois University
Rights Statement 2
NIU theses are protected by copyright. They may be viewed from Huskie Commons for any purpose, but reproduction or distribution in any format is prohibited without the written permission of the authors.