### Skapa referens, olika format (klipp och klistra)

**Harvard**

Larsson, A. (2002) *Methods for Inversion of Arc Welding Process Control*. Göteborg : Chalmers University of Technology

** BibTeX **

@mastersthesis{

Larsson2002,

author={Larsson, Anna},

title={Methods for Inversion of Arc Welding Process Control},

abstract={ABB Corporate Research is developing a new tool to be able to predict a weld profile for an arc-welding robot, Virtual Arc. For a certain set of inputs the program will predict a corresponding set of outputs describing the weld profile. To be able to program a robot off-line is of great importance for the cost reductions in a producing company since the programming otherwise takes a lot of valuable producing time.
As a part of the Virtual Arc project this thesis aims to find the inverse to the existing system, that is to produce an algorithm that takes a desired weld profile and finds the corresponding input. Such a feature would make the tool even easier to use, whereas the user would only have to decide on the desired result and the system would calculate how to obtain this result.
The model in Virtual Arc is built on physics and a Bayesian Neural Network and it is here within that the first problem (or challenge) can be viewed, since an inversion of such a system is a one to many mapping and one output can correspond to numerous different inputs.
In this thesis two main methods are considered to solve this problem: Non Linear Programming (NLP) and Iterative Learning Control (ILC). Both methods are iterative methods but with a somewhat different philosophy.
The NLP method has shown very good results and can find the inverse with a quite high accuracy. The method was verified by applying a large amount of test cases to it, and all indicate on a stable algorithm. Using the ILC method though, a solution to the problem was not reached. The main reason for this is that the method requires a linear model whereas some linearisation must be applied and a new such approximation must be constructed in each iteration.
As an extension to the initial problem some investigation has also been done on the possibilities of using fewer of the weld profile parameters. For the user it would be of large interest if the parameters were optional so that the efforts could be put on exactly those parameters interesting in the specific case.
},

publisher={Institutionen för maskin- och fordonssystem, Mekaniska system, Chalmers tekniska högskola},

place={Göteborg},

year={2002},

keywords={Inverse, Neural Networks, Nonlinear Programming, Iterative Learning Control, Arc Welding, Off-Line Programming},

note={60},

}

** RefWorks **

RT Generic

SR Print

ID 23216

A1 Larsson, Anna

T1 Methods for Inversion of Arc Welding Process Control

YR 2002

AB ABB Corporate Research is developing a new tool to be able to predict a weld profile for an arc-welding robot, Virtual Arc. For a certain set of inputs the program will predict a corresponding set of outputs describing the weld profile. To be able to program a robot off-line is of great importance for the cost reductions in a producing company since the programming otherwise takes a lot of valuable producing time.
As a part of the Virtual Arc project this thesis aims to find the inverse to the existing system, that is to produce an algorithm that takes a desired weld profile and finds the corresponding input. Such a feature would make the tool even easier to use, whereas the user would only have to decide on the desired result and the system would calculate how to obtain this result.
The model in Virtual Arc is built on physics and a Bayesian Neural Network and it is here within that the first problem (or challenge) can be viewed, since an inversion of such a system is a one to many mapping and one output can correspond to numerous different inputs.
In this thesis two main methods are considered to solve this problem: Non Linear Programming (NLP) and Iterative Learning Control (ILC). Both methods are iterative methods but with a somewhat different philosophy.
The NLP method has shown very good results and can find the inverse with a quite high accuracy. The method was verified by applying a large amount of test cases to it, and all indicate on a stable algorithm. Using the ILC method though, a solution to the problem was not reached. The main reason for this is that the method requires a linear model whereas some linearisation must be applied and a new such approximation must be constructed in each iteration.
As an extension to the initial problem some investigation has also been done on the possibilities of using fewer of the weld profile parameters. For the user it would be of large interest if the parameters were optional so that the efforts could be put on exactly those parameters interesting in the specific case.

PB Institutionen för maskin- och fordonssystem, Mekaniska system, Chalmers tekniska högskola,

LA eng

OL 30