REPORT BRIEF
Machine Vision Inspection of VF Display Boards
April 9, 1998
CENTER FOR ENGINEERING EDUCATION AND PRACTICE
SCHOOL OF ENGINEERING, UNIVERSITY OF MICHIGAN-DEARBORN
 
AUTHOR(S): Yi Lu Murphey
Department of Electrical and Computer Engineering
PARTNER(S): Anthony Tisler
Jabil Circuit, Inc.
ASSISTANT(S): Tie-Qi Chen, Jie Chen
Research assistant

 
 
BACKGROUND Vacuum Florescent (VF) Displaying boards widely used in the automobile industry to display information about vehicle's status. The displays are illuminated by circuit boards specially designed to disclose specific information including speed, mileage, fuel level, compass heading, and etc. These displays are the only part of the circuit board seen by the automobile operator and therefore high quality of the display is required to insure excellent report from the customers. Once the circuit board is complete, the entire board must be tested for its proper functions. VF inspection is to various displaying patterns of each VS board on the production. VF displays are mounted directly onto the circuit board using pin through hole and surface mounting. In the current manufacturing, the screen test of VF display is visually inspected by a functional test operator. The functional test forces the display to show pre-specified patterns which are verified by the test operator. The functional test lasts about 4 to 8 seconds per circuit board and in that time an operator is expected to inspect at least two display patterns of VF board. Two major problems related to manual inspection:
  1. Currently because of time limit, an operator can inspect only two display modes which is not enough to insure the display board is functionally good. Ideally every possible display pattern should be inspected. However, the number of display combinations that can be inspected is bounded by how fast a human can inspect them.
  2. The nature of the visual inspection is repetitive and tedious. It is impossible to expect a person to maintain peak awareness during an 8-hour period.
The VF displaying functions similarly to most calculator displaying. Different boards have different display contents. Figure 1 shows the image of an electronic compass/temperature/ trip board used in one type of automobile. The entire display of a VF board consists of a number of fields. Each field is comprised of segments. A group of components whose state is independent of other components is called a segment. Each segment in the board can be in two states: on or off. The display in Figure 1 has 18 fields, for illustration purpose, each is bounded by a rectangular box. A field can have more than one segment. For examples, the fields "0" and "8" in this image have multiple segments. A field is called a single field if it has only one segment, i.e. all the components within the field always are in the same state, either all on or all off.


Figure 1 A VF display of compass/temperature/trip board.

 
OBJECTIVES The objective of this project is to develop computer vision algorithms combined with machine learning techniques to inspect various VF displaying boards to:
  • permit inspection of all possible display patterns without impeding product line rates;
  • eliminate human error in the inspection process.

    •  
    APPROACH Automatic visual inspection is one of the primary applications of computer vision. Visual inspection has broad applications in industry automation and covers the full range of technical difficulty in computer vision.

    Because of its diverse application environment, it has been recognized that there is no pervasive generic solution in machine vision, each application requires a careful study of alternatives and perhaps even the invention of a new technique. Our approach to solving the VF displaying board is illustrated in Figure 3, which presents the overall view of the machine vision system designed for VF board on-line inspection.

    The system consists of two major procedures, learning and inspection. The learning procedure is applied to every new type of VF boards and produces a symbolic list of specification for the inspection procedure. The learning procedure can be executed off the production line. For every VF board on the production line, the inspection procedure will generate pass or fail signal based on image analysis of various test patterns. At the same time, the inspection procedure also saves the image that contains the defect evidence for further verification. The following subsections describe in more detail the learning and inspection procedures.

    For a new type of VF board, the "learning procedure" is employed to learn the board features from a good compass board of the type, and the algorithm generates a symbolic representation of the compass board, which is referred to as Inspection Reference List (IRL). The "learning procedure" employs a number of image processing techniques including binarization, tilt detection and correction, and computation of connected components.

    The inspection procedure runs on the production line. The VF boards on the production lines come in batches of the same type. Before a different batch of VF boards comes on the production line, the inspection procedure will read in the corresponding test reference list file which contains the master template and the multiple test pattern lists for this type of VF boards generated by the learning procedure. The inspection procedure is guided by the test reference list.
     
    RESULTS We have implemented the system described above on a PC computer under the Window NT/95/3.1x operating system. The learning subsystem has been deployed at the Jabil Circuit, Inc. and the test subsystem has been integrated into the Jabil test system.
    CONCLUSIONS We have presented a machine vision system for reliable inspection of various types of defects of VF boards. The system has two procedures, the learning and inspection procedures. The machine vision system is robust to a plant environment where lighting condition can vary and shadows can occur in images, and the testing boards can be tilted and be placed in various positions. During the learning process that is performed off-line, the system attempts to learn the characteristics of each test pattern for every new type of VF board. The output of the learning procedure is a symbolic description of the test features of the particular type of VF boards. The inspection procedure is performed on the production line. Therefore the inspection procedure is designed to be extremely efficient in computational time.
    IMPACT This project has a significant impact on industry since it automates the inspection of the VF boards in manufacturing lines, speeds up the test procedure which ultimately leads to higher product yield. The project involves a number of students. During the project period, students visited the assembly plant at the Jabil Circuit, Inc., and learned to solve practical problems. The result of this project can be used in a number of existing courses including machine vision, image processing, and intelligent systems.

    To start a demo.

    For more information and software, please contact Professor Yi Lu.

    Back

    This page is prepared by Jie Chen, May 11, 1998.