Application of adaptive control theory to on-line GTA weld geometry regulation

TLDR: In this article, the use of adaptive schemes for on-line control of backbead width in the Gas Tungsten Arc (GTA) welding process has been investigated and two adaptive controllers, model-referenced adaptive control (MRAC/NL) and self-tuning control with pole placement (STC/PP), were evaluated.

Abstract: This study addresses the uses of adaptive schemes for on-line control of backbead width in the Gas Tungsten Arc (GTA) welding process. Open-loop tests using a step input current confirm the validity of a nominal first order process model. However, the time constant and gain prove highly dependent upon welding conditions including torch speed, arc length, material thickness, and other material properties. Accordingly, a need exists for adaptive controllers that can compensate for these process nonlinearities. The performance of two adaptive controllers is evaluated: Narendra and Lin’s Model-Referenced Adaptive Control (MRAC/NL), and Self-Tuning Control with Pole Placement (STC/PP). The addition of a quadratic term to the adaption mechanisms of MRAC/NL is proposed and preliminary simulations and experiments clearly demonstrate the stabilizing effect of this added term. The main experiments compare the performance of the modified MRAC/NL controller and the STC/PP controller with each other and with linear PI controller and the STC/PP controller with each other and with linear PI controller under four experimental conditions: first, where welding conditions are nominal; second, when conditions are disturbed by a step-wise increase in the torch velocity, and third, when conditions are disturbed by a step-wise increase in material thickness. In each case the experimental demonstrates the superiority of the adaptive controllers over the linear PI controller. However, the STC/PP controller exhibits high frequency control action in response to severe disturbances of material thickness and the parameter estimates it generates drift during steady-state operations. The MRAC/NL controller proves more robust under these circumstances. Analysis demonstrates that the superior performance of the MRAC/NL is due both to the inherent normalizing effect of the quadratic feedback terms and to the noise filtering properties of the adaptive mechanism.