Application of Adaptive Control to In-Process Weld Geometry Regulation

In this paper, the control of manufacturing processes in real-time is addressed in the context of a weld bead geometry problem. Since weld geometry dynamics are highly non-linear and non-stationary in nature, adaptive control techniques have been chosen. Self Tuning Control and Model Reference Adaptive Control are first examined experimentally for an "ideal" position servo. The servo system represents a linear, low noise, constant (but unknown) parameter process and it is shown that an adaptive control system improves performance dramatically over a linear control system except when plant deadband is intentionally introduced. On the other hand, the welding process represents a process noise contaminated, time varying, nonlinear dynamic process. It is shown that Self Tuning Control is capable of modifying "incorrect" initial plant parameters, thereby attaining a model response in the real system. However, at the same time the sensitivity to the process noise inherent in welding is evident, indicating the need for great care in designing such systems.