Weld quality monitoring and estimation for gas metal arc welding of aluminum.

Aluminum alloys have been increasingly used in the automotive industry as an alternative to steels to improve fuel economy, reduce emissions, and enhance vehicle safety and performance. One of the major processes for joining aluminum parts is gas metal arc welding (GMAW) because of its high productivity, flexibility and low cost. However, aluminum GMAW is prone to various defects and weld quality is commonly evaluated using manual inspection procedures, which are labor intensive and may introduce inconsistence. Furthermore, the knowledge of aluminum GMAW is insufficient to have a consistent and efficient way to evaluate weld quality. The objective of this thesis is to obtain a comprehensive understanding of aluminum GMAW process and develop quality monitoring and estimation methods, which can be used in the automated production environment to continuously improve the weld quality and system consistency. Fundamental research has been conducted to study the welding process signals with synchronized metal droplet transfer images using an online data and high-speed image acquisition system. Based on these experiments, welding signature analyses in both time and frequency domains are performed to reveal the correlation between the droplet transfer process and the welding voltage and current signals. Algorithms have been developed to characterize individual droplet transfer dynamics, detect the existence of secondary droplets and estimate droplet sizes. A model is then developed to estimate weld bead geometry based on the welding signature analysis and liquid surface modeling. The model uses a numerical liquid surface model to represent the steady-state profile of welding pool, which is updated drop by drop with information from incoming droplets obtained by analyzing the welding signals. The weld geometry estimation is applied to real welding signals in both globular and spray transfer modes and the results show a good match to the physical weld geometries. A method is also developed to detect and classify different weld defects in pulsed GMAW (GMAW-P). The method first uses a state-space model to numerically represent a stable welding process. The state-space model is then used to predict the voltage signals using the measured current signals as the inputs. Voltage prediction residuals are calculated and found to be correlated to process disturbances and different welding defects. The prediction residuals are further characterized and classified to identify different weld defects. The method is validated to have a 100% defect detection rate and an 84% overall correct classification rate for different weld defects.PhDApplied SciencesMechanical engineeringUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/126675/2/3276144.pd