Year of Publication


Degree Name

Doctor of Philosophy (PhD)

Document Type

Doctoral Dissertation




Electrical and Computer Engineering

First Advisor

Dr. YuMing Zhang


Combining human welder (with intelligence and sensing versatility) and automated welding robots (with precision and consistency) can lead to next generation intelligent welding systems. In this dissertation intelligent welding robots are developed by process modeling / control method and learning the human welder behavior.

Weld penetration and 3D weld pool surface are first accurately controlled for an automated Gas Tungsten Arc Welding (GTAW) machine. Closed-form model predictive control (MPC) algorithm is derived for real-time welding applications. Skilled welder response to 3D weld pool surface by adjusting the welding current is then modeled using Adaptive Neuro-Fuzzy Inference System (ANFIS), and compared to the novice welder. Automated welding experiments confirm the effectiveness of the proposed human response model.

A virtualized welding system is then developed that enables transferring the human knowledge into a welding robot. The learning of human welder movement (i.e., welding speed) is first realized with Virtual Reality (VR) enhancement using iterative K-means based local ANFIS modeling. As a separate effort, the learning is performed without VR enhancement utilizing a fuzzy classifier to rank the data and only preserve the high ranking “correct” response. The trained supervised ANFIS model is transferred to the welding robot and the performance of the controller is examined. A fuzzy weighting based data fusion approach to combine multiple machine and human intelligent models is proposed. The data fusion model can outperform individual machine-based control algorithm and welder intelligence-based models (with and without VR enhancement).

Finally a data-driven approach is proposed to model human welder adjustments in 3D (including welding speed, arc length, and torch orientations). Teleoperated training experiments are conducted in which a human welder tries to adjust the torch movements in 3D based on his observation on the real-time weld pool image feedback. The data is off-line rated by the welder and a welder rating system is synthesized. ANFIS model is then proposed to correlate the 3D weld pool characteristic parameters and welder’s torch movements. A foundation is thus established to rapidly extract human intelligence and transfer such intelligence into welding robots.