TAILORING WELD GEOMETRY AND COMPOSITION IN FUSION WELDING THROUGH CONVECTIVE MASS TRANSFER CALCULATIONS
Open Access
- Author:
- Mishra, Saurabh
- Graduate Program:
- Materials Science and Engineering
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- March 17, 2006
- Committee Members:
- Tarasankar Debroy, Committee Chair/Co-Chair
Kwadwo Osseo Asare, Committee Member
Long Qing Chen, Committee Member
Judith Todd Copley, Committee Member - Keywords:
- Neural Network
Genetic Algorithm
Mass Transfer in Welding
Liquation Cracking
Effect of Sulfur - Abstract:
- Fusion welding is characterized by various complex physical processes where the heat source interacts with the material leading to rapid heating, formation of the weld pool, vigorous circulation of the liquid material in the pool, heat transfer in the entire weldment, and solidification of the molten material. All these physical processes have a profound impact on the mass transport in the weldment. Mass transport during fusion welding has a significant impact on various weld features like the weld pool geometry and the creation of weld defects like liquation cracking. For example, surface active elements like sulfur alter the fluid convection pattern in the weld pool, and lead to a high depth to width ratio under favorable welding conditions. Furthermore, the fluid convection pattern becomes very complex when two plates with different amounts of surface active elements such as sulfur are welded together. Similarly, during gas metal arc (GMA) welding of aluminum copper alloys, the amount of solute (copper) in the solidifying weld metal is a key factor in determining the susceptibility of the weld to liquation cracking. In the past two decades, numerical transport phenomena based models have provided useful information about the thermal cycles and weld pool geometry. However, no effort has been made to apply these concepts to design weld consumables, to study the weld bead shape on welding two plates with different sulfur contents and to tailor weld pool geometry to specified dimensions. The present research focuses on these unexplored areas. The research proposed here seeks to develop a quantitative understanding of mass transport during fusion welding, with special emphasis on the role of surface active elements and the effect of solute distribution on weld defects like liquation cracking. A comprehensive model, incorporating numerical three-dimensional calculations of temperature and velocity fields and solute distribution in the weld pool is developed for the proposed quantitative study. The currently available numerical transport phenomena based models are faced with some major difficulties including lack of reliability of output and limited utility because of the ability to go only in the forward direction. The output of these models is not generally reliable because of the presence of some uncertain input parameters such as arc efficiency. Furthermore, these models cannot go backwards, i.e., calculate multiple combinations of welding process variables such as current, voltage and welding speed to obtain a target weld attribute such as weld pool geometry. Also, these models are complex, require specialized training to develop and test, and consume a large amount of computer time to run. These shortcomings of the numerical transport phenomena based models are addressed in the present thesis research. New methodologies are developed in the present study where the reliability of output of the numerical transport phenomena based model is enhanced by developing a computational procedure, where the uncertain input parameters of the model are optimized by combining it with a genetic algorithm optimization model and limited volume of experimental data. The resulting numerical model providing more reliable output is validated by comparing its predictions of weld pool geometry with the corresponding experimental results. In order to reduce the computation time a neural network, trained by data generated from the numerical transport phenomena based model, is developed in the present study. Next, the neural network is combined with a genetic algorithm to go backwards, i.e., find multiple combinations of welding variables to obtain a target weld geometry. Finally, this model is applied to study solute distribution in the weld pool when joining two steel plates with different sulfur contents, and to determine the copper content of the solidifying weld metal in aluminum-copper alloys welded with different copper containing filler metals. The study identifies the factors that affect the weld pool geometry on joining two plates with different sulfur contents, and predicts the susceptibility of an aluminum-copper alloy GMA weld to liquation cracking. The specific contributions of the present thesis research include (i) development of a numerical solute transport model for fusion welding; (ii) improving the reliability of output of the numerical model; (iii) achieving computational efficiency and economy by developing a neural network trained by data generated by the numerical model; (iv) creating a bi-directional methodology where a target weld attribute like weld pool geometry can be attained via multiple combinations of input process parameters like arc current, voltage and welding speed; (v) calculating sulfur distribution during gas tungsten arc welding of stainless steel plates with different sulfur contents and predicting the resulting weld pool geometry; and (vi) calculating copper distribution during gas metal arc welding of aluminum-copper alloys by incorporating the heat and mass addition from filler metal and a non-equilibrium solidification model, and using the copper content of the mushy zone to predict the occurrence of liquation cracking.