Tailoring Defect Free Fusion Welds Based on Phenomenological Modeling

Open Access
Author:
Kumar, Amit
Graduate Program:
Materials Science and Engineering
Degree:
Doctor of Philosophy
Document Type:
Dissertation
Date of Defense:
May 15, 2006
Committee Members:
  • Tarasankar Debroy, Committee Chair
  • Kwadwo Osseo Asare, Committee Member
  • Long Qing Chen, Committee Member
  • Kenneth K Kuo, Committee Member
  • John W Elmer, Committee Member
Keywords:
  • Gas Metal Arc fillet welding
  • Heat transfer and fluid flow
  • Optimization
  • Neural network
  • Humping defects
  • Reliability of calculations
Abstract:
In the last few decades, phenomenological models of fusion welding have provided important understanding and information about the welding processes and welded materials. For example, numerical calculations of heat transfer and fluid flow in welding have enabled accurate quantitative calculations of thermal cycles and fusion zone geometry in fusion welding. In many simple systems such as gas tungsten arc (GTA) butt welding, the computed thermal cycles have been used to quantitatively understand weld metal phase compositions, grain sizes and inclusion structure. However, fabrication of defect free welds with prescribed attributes based on scientific principles still remains to be achieved. In addition, higher fabrication speeds are often limited by the occurrence of humping defects which are characterized by periodic bead-like appearance. Furthermore, phenomenological models have not been applied to tailor welds with given attributes. The goal of the present work is to apply the principles of heat transfer and fluid flow to attain defects free welds with prescribed attributes. Since there are a large number of process variables in welding, the desired weld attributes such as the weld geometry and structure are commonly produced by empirically adjusting the welding variables. However, this approach does not always produce optimum welds and inappropriate choice of variables can lead to poor welds. The existing transport phenomena based models of welding can only predict weld characteristics for a given set of input welding variables. What is needed, and not currently available, is a capability to systematically determine multiple paths to tailor weld geometry and assess robustness of each individual solution to achieve safe, defect free welds. Therefore, these heat transfer and fluid flow based models are restructured to predict the welding conditions to achieve the defect free welds with desired attributes. Systematic tailoring of weld attributes based on scientific principles still remains an important milestone in changing welding from almost an empirical art to a mainstream science-based technology. The ability to determine multiple welding variable sets to achieve desired weld attributes, based on scientific principles, would be an important step to achieve this goal. Furthermore, no comprehensive unified theoretical model exists today that can predict the formation of commonly occurring humping defects considering the effects of important welding variables such as the arc current, voltage, welding speed, nature of the shielding gas, electrode geometry, torch angle and ambient pressure. In this research work, a model is developed to achieve desired weld attributes and avoid high speed weld defects like humping. Three main requirements are desirable in a model for systematic tailoring of weld attributes. First, the procedure should embody an adequate phenomenological description of the complex physical processes in welding. Although the heat transfer and fluid flow models use time-dependent equations of conservation of mass, momentum and energy, the predictions of temperature fields and thermal cycles do not always agree with experimental results because the models require many input variables all of which cannot be prescribed with certainty. For example, the reported values of arc efficiency vary significantly for minor differences in the surface characteristics that are difficult to characterize for every welding process. Second, the models are designed to calculate the temperature and velocity fields for a given set of welding variables. However, very often what is needed is to determine the welding variables required to achieve a given weld attribute such as the weld geometry, cooling rate and the microstructure. The current generation of unidirectional heat transfer and fluid flow models are designed to calculate temperature and velocity fields from welding conditions and are incapable of determining welding conditions. Finally, the welding system is highly complex and involves non-linear interaction of several welding variables. As a result, a particular weld attribute such as the geometry can be obtained via multiple paths, i.e., through the use of various sets of welding variables. The current generation of numerical heat transfer and fluid flow models cannot determine alternative pathways to achieve a target weld attribute. In this thesis, a new structure of the phenomenological models is developed by combining numerical heat transfer and fluid flow models with a suitable optimization algorithm in the form of genetic algorithm. The combined model has new capabilities for bi-directional simulation where either the traditional input or the output variables can be specified. The new formulation also allows determination of multiple solutions to attain a specified weld attribute. Genetic algorithms (GA) can systematically search for multiple combinations of welding variable sets that comply with the phenomenological laws of welding physics and obtain a population of solutions following certain rules of evolution. This research represents the very first effort to adapt transport phenomena based models along with genetic algorithm based optimization model to attain defects free welds with desired attributes during gas metal arc fillet welding. Through uncommon synthesis of appropriate concepts from transport phenomena, optimization and data mining, this research work outlines a completely new direction of exceptional promise.