Year of Publication

2008

Document Type

Dissertation

College

Engineering

Department

Electrical Engineering

First Advisor

Bruce Walcott

Second Advisor

YuMing Zhang

Abstract

It is well known that typical welding processes such as laser welding are nonlinear although mostly they are treated as linear system. For the purpose of automatic control, Identification of nonlinear system, especially welding processes is a necessary and fundamental problem. The purpose of this research is to develop a simple and practical identification and control for welding processes. Many investigations have shown the possibility to represent physical processes by nonlinear models, such as Hammerstein structure, consisting of a nonlinearity and linear dynamics in series with each other. Motivated by the fact that typical welding processes do not have non-zeroes, a novel two-step nonlinear Hammerstein identification method is proposed for laser welding processes. The method can be realized both in continuous and discrete case. To study the relation among parameters influencing laser processing, a standard diode laser processing system is built as system prototype. Based on experimental study, a SISO and 2ISO nonlinear Hammerstein model structure are developed to approximate the diode laser welding process. Specific persistent excitation signals such as PRTS (Pseudo-random-ternary-series) to Step signal are used for identification. The model takes welding speed as input and the top surface molten weld pool width as output. A vision based sensor implemented with a Pulse-controlled-CCD camera is proposed and applied to acquire the images and the geometric data of the weld pool. The estimated model is then verified by comparing the simulation and experimental measurement. The verification shows that the model is reasonably correct and can be use to model the nonlinear process for further study. The two-step nonlinear identification method is proved valid and applicable to traditional welding processes and similar manufacturing processes. Based on the identified model, nonlinear control algorithms are also studied. Algorithms include simple linearization and backstepping based robust adaptive control algorithm are proposed and simulated.

Share

COinS