isbel banner
About us
Research
Courses
Clinical Radiology
Industrial Liason
People
Contact Us
Join Us
Search Facility
Home Page

Introduction

Formed tube is an important component in many manufactured assemblies, particularly in the car and aerospace industries where it is used for fuel lines, brake pipes, exhaust pipes etc. Tubes often need to be bent into complex and precisely specified three-dimensional shapes and high accuracy may be required in bending tubes to fit a prescribed specification. Inspection of bent tubes is necessary for quality control, reverse engineering and precise specification of bending parameters. The last requirement arises because metal tubes often "spring back" in an unpredictable way after bending, so that the final conformation differs from that specified. The precise operating parameters of the bending machine need to be adjusted by measuring the difference between the specification and the resulting tube. Our collaborator in this project was a major supplier tube bending and measuring equipment. The company's requirement was to develop an instrument for reliable, accurate and automatic measurement of the 3D shape of formed tube. Existing measuring equipment was either expensive and slow, or labour intensive with poor repeatability. The specification for the instrument was to measure the path of the tubes in three dimensions within a tolerance of 0.2mm with minimal operator intervention. An operational requirement was to be able to specify the bending parameters for a tube in the minimum number of bend-and-measure cycles.

Technical Approach

The technical approach taken was to use stereo vision. The tube is viewed by a number of television cameras mounted in an inspection cell as shown in the picture, arranged so that any section is visible in at least three different views. The combination of views can be used to calculate the 3D position of each point on the tube in a manner analogous to human stereopsis. The technical problem in stereo vision lies in deciding which points in each of the 2D views correspond to the same points in the scene. This is known as the correspondence problem, and is usually solved by complex reasoning about the positions of points in two views. In this case, we exploited the fact that three views were available of each section, making the calculation simpler and more robust. Difficulties in interpretation arise from the fact that the surface characteristics vary from dark matte to highly reflective, and in most views some sections of the tube are hidden by others. These problems were solved in this case by using controlled illumination and making use of the fact that the tube is composed of straight sections connected by approximately circular bends. Since high positional accuracy is required, the edges of tubes needed to be detected with a precision of 0.1 pixel in the 2D views.

Calibration

An important part of the instrument was the calibration system by which the positions of points in each 2D view are related to precise 3D co-ordinates. The mathematical relationship between 2D and 3D points is determined by finding the positions in the 2D images of points on a calibration target, whose relative 3D positions are known accurately. The instrument requires to be recalibrated daily, so the target and calibration algorithm were designed to achieve the required accuracy while retaining sufficient robustness to be used in a factory environment. Careful and reliable calibration resulted in 3D positions being measured with a precision of about 0.1mm.

Prototype Instrument

The result of the project was a prototype instrument, shown in the picture. The specification for accuracy and precision has been met, as has the operational requirement. Accurate bending parameters are established after a single bend-and-measure cycle. A production engineered version is now being beta-tested. The company is continuing to develop the measurement software.

Summary

Stereo Vision has often been applied to industrial problems, but this has usually been in applications such as robot guidance or object recognition where high positional accuracy are not important. This is one of the few examples of its use in 3D metrology and is, to our knowledge, unique in being used in a factory-floor instrument.

Acknowledgements

This project was funded by EPSRC as a Teaching Company Programme. Our industrial partners were Addison Tube Forming Ltd, Bamber Bridge, Preston, Lancs

Contact

Jim Graham: Jim.Graham@man.ac.uk

Research  Courses  Clinical Radiology   Industrial liaison  Contact Us  Join Us  Search   Home

Top Page

Contact the Website Administrator with comments or queries about this website.
Copyright © 1999 - 2002 The University of Manchester.
Page last updated: 27 November, 2002