Sunteți pe pagina 1din 4

Automatic Detection Technology of Surface Defects

on Plastic Products Based on Machine Vision


Bin Liu and Shengjin Wu

Shifang Zou

College of Mechanical and Automobile Engineering


South China University of Technology
Guangzhou, China
bliu@scut.edu.cn; wushengjin@tom.com

Dongguan Janus Group Corp., Ltd.


Dongguan, China
zoushifang@janusgroup.cn

AbstractIt is very necessary to detect surface defects on plastic


products during production process and post treatment. The
research and application of automatic detection technology of
surface defects on plastic products is supposed to greatly liberate
the human workforce, improve the automated production level,
and has wide application prospect. The development of machine
visions key technologies such as illumination system, CCD
camera, image enhancement, image segmentation, image
recognition, and so on has been explained in detail. Its
application on detection for surface defects on plastic products
such as plastic electronic components, strips, PVC building
materials, films, leather, bottles, and so on is also presented
briefly. Especially, it mainly focuses on the automatic detection
of surface defects for injection products, and the automatic
detection system is proposed. It is composed of the conveyor belt
device, the image acquisition and processing software and PLC
control device, et al.
Keywords- Surface Defects; Plastic Products; Machine Vision;
Automatic Detection

I.

INTRODUCTION

With the development of science and technology, surface


quality of plastic products is more and more required for the
follow-up industry. The detection and control of surface
defects has attracted plastic products manufacturers attention.
The automatic detection system, which is suitable for plastic
products, has significant practical significance and wide
application prospect.
Surface defect detection technology has been widely used
as an important means of quality control. In the past, this can
be done with naked-eye observation, which the job is bored,
tedious and repetitive, which may make mistake. Furthermore,

Fig. 1

978-1-4244-7739-5/10/$26.00 2010 IEEE

the long-time work would do harm to the health of workers.


Today, with the new technology of automatic detection system
based on machine vision, the same job can be done more
effectively, costless, reliable, consistent and faster.
Machine vision technology is a typical interdisciplinary
subject, which involves artificial intelligence, neurobiology,
psychophysics, computer science, image processing, pattern
recognition and many other areas. Machine vision is mainly
aimed at simulating or reproducing certain intelligent behavior
concerned with human visual using computer, processing and
understanding the information extracted from object images,
and ultimately being used in the actual defect detection and
quality control.
II.

KEY TECHNIQUES AND ACTUALITIES

The machine vision system consists of light source, optical


imaging module, image capture module, image acquisition and
digitization module, intelligent image processing and decision
module, control and implementation module as shown in Fig.1
[1]. These modules constitute an integrated industrial machine
vision system.
Although the machine vision system applies many
techniques, the key techniques are illumination system,
camera, image enhancement, image segmentation and image
recognition. The actualities of these key techniques are
described as follows.
A. Illumination System
The performance of illumination system is essential to the
machine vision system. It should be possessed of the following
characteristics [2]: (1) produce significant difference between

System structure of machine vision

required detection and non-detection areas on the object by


highlighting the target feature; (2) ensure sufficient brightness
and stability; (3) maintain the image quality when the object
changes its position.
As the detected surface is usually irregular, its roughness is
variable, some products even have electrodeposit or coating
surface, resulting of inconsistent reflection of light on the
surface of the detected products brings various difficulties in
image acquisition and processing. In order to obtain good
image quality and facilitate the analysis processing, its needed
to design special illumination systems for the products with
different surface defects.
In order to minimize the influence of surface roughness,
Tang et al. [3] analyzed the relationship between illumination
and defect expressivities, and between illumination and parts
roughness. It was found that the relation between illumination
and defect expressivities was linear. How to realize the selfadjust function of illumination was proposed to improve the
capability of defect checking system.
Qian et al.[4] designed a lamp-house employing MAX1698
chip. It settled two difficulties which formed on the strongly
reflected and complex surface. One was that the lightness of
CCD imaging menu was too strong, and the other was that the
information of defect which should be detected was always
lost. Fig.2 shows the optics principle schematic of cone LED
light source.
At present, LED lights are usually adopted as illumination
source in machine vision technology. LED lights consist of
spot lights, line lights, dome lights, ring lights, and so on,
which can be used for different types of defects for automatic
detection. Spot lights with high luminous intensity can be used
for chip detection and Mark point location. Line lights with
cylindrical lens, whose brightness is ultrahigh, is suitable for
stepless detection on a variety of production lines, applied to
linear array cameras and AOI (i.e. Automatic Online
Inspecting). Dome lights with hemispherical integral side
walls, can evenly reflect light emitted from the bottom in 360,
and are applied to surface defect detection with convexconcave and curved surface, metal and glass with strong
reflection surface. Ring lights provide different illumination
angles and color combinations, highlight the 3D information of

objects better, settle the problem of the shadow aroused by


illuminating from opposite direction, are applied to defect
detection for PCB boards, IC components, plastic containers et
al.
B. CCD Camera
Currently, CCD(Charge Coupled Device) camera is widely
applied in machine vision image acquisition process. Its
principle of operation is as follows. CCD camera captures the
images of the detected object on the imaging sensitive surface
via the optical imaging system. The image illumination signals
in the image sensitive units are changed into minority-carrier
density signals and stored there. Then, they are transferred to
the CCD shift register by transfer pulse, and moved out of the
device sequentially by driving pulse and become video signals
finally.
CCD consists of linear array CCD and area array CCD.
The linear CCD employs linear array integrated circuit as its
sensor, which is fast in processing information. The follow-up
processing circuit is simple and easy to achieve real-time
control. However, it can only acquire 1D images, and can not
process complex images. The area array CCD can capture 2D
images, access to large amount of information and process
complex images. Whereas, its processing speed is relatively
slower and the cost is higher than the linear array CCD.
The key performance parameters of CCD camera are pixel,
resolution and sensitivity. Pixel determines the clarity of the
displayed images. The performance of image details is better
while the resolution is higher. In image processing, in order to
acquire a clear image, it needs a 600-line camera. CCD
sensitivity is a sensitivity level to the ambient light, that is, the
limit light is needed for normal imaging. The illumination is
smaller which means it needs less light, and the camera is
more sensitive.
Imaging on moving objects and recording information
from the periphery of cylindrical objects will lead to blurring
of the images. In order to resolve this problem, Asundi et al.[5]
proposed TDI-CCD (Time Delay Integration Charge Coupled
Device) technology for motion detection and rotating cylinder
imaging. Experiments showed[6] that the application of TDICCD technology can capture images of cylindrical objects, and
quickly identify the edge of the detected objects and defects.

Fig. 3
Fig. 2

Optics principle schematic of cone LED light source

Experimental equipment for automatic surface defect detection of


cylindrical objects

Fig.3 shows the experimental equipment for automatic surface


defect detection of cylindrical objects.
C. Image Enhancement
Image enhancement technology is to maximize the contrast
to give optimum output for edge processing. Image
enhancement technology can be classified into two categories
[7], i.e. the methods based on image domain and the methods
based on transform domain.
Methods based on image domain are to process image data
directly, e.g. the single pixel and the smaller sub-image. The
algorithms consist of linear contrast stretch, Unsharp Masking
(UM), adaptive enhancement algorithm, histogram
equalization algorithm and a variety of optimized algorithms.
Methods based on transform domain are to process images
inside the transform domain, that is, the image data are
converted to other domain through some sort of
transformation, and then converted back to image domain after
processed. Some of the most common transformation methods
are Fourier transform, discrete cosine transform, wavelet
transform and so on. Wavelet-based adaptive enhancement,
Retinex and Curvelet transform-based enhancement method
belong to this category too.
D. Image Segmentation
Image segmentation is the technology that segments the
image into several areas with different features, and extracts
the target areas. The features can be gray-level, color, texture,
et al., and the target areas can be a single area or multi-areas
correspondingly [8].
The study of image segmentation is emphasis and difficulty
in the image engineering and machine vision technology. In
recent years, there are many algorithms being proposed and
applied to practice, such as the Markov random field [9],
expert system [10], Gribbs random field[11], Bayesian theory
[12], wavelet modulus maximum [13], hidden Markov model
[14], fractal [15], et al.
The methods of image segmentation can be classified into
three categories based on the image features[7]. The first is the
threshold method, image segmentation is achieved by
determining a certain threshold in teams of the distribution
characteristics of the images gray-level. The second is the
boundary segmentation method, it is achieved by means of
checking out the boundary of a closed area, namely
segmenting the area along the closed edge lines. The third is
the regional extraction method which is based on different
characteristics between the specific region and the background
region.
E. Image Recognition
Image recognition is to classify images after processing
and determine their categories. It is able to select target
features, measure certain parameters, extract these features and
ultimately classify them in terms of measurement results.
Therefore, image recognition is to ease images classification,
and analyze the whole image by finding out features (e.g.

shape and texture) in the areas after segmenting, i.e. feature


extraction.
There are two main factors that affect the outcome of
image recognition, that is, the selection of the target features
and the classification algorithms. The accuracy of target
features selection is crucial to recognition result and the
computation of classification algorithm.
As Gabor wavelet transform can well describe the
problems aroused from biological visual neurons, adjust its
spatial and frequency sampling characteristics accordingly in
terms of specific visual needs, obtain target features of interest,
many researchers have paid extensive attention to Gabor filter
and its application in image classification and machine vision
[16]. BLOB algorithm [17] is aimed at finding out the scope of
the gray-level mutations in an area, and determining its size,
shape, area and exact location. The algorithm which is fast and
exhibits good real-time can be employed in many occasions
that the image quality is very high and can be well segmented
with threshold.
III.

APPLICATIONS OF MACHINE VISION IN SURFACE DEFECT


DETECTION OF PLASTIC PRODUCTS

During the process of plastic products, there are various


defects on the surface, such as short shot, shrink marks,
welding line, streaks, spots, gas bubbles, voids, blistering,
spray marks, warp, scratches, cracks and so on. The products
defects often arise as a result from materials, process, mold,
product design and equipments, et al.
A common task in molding industries is the visual
inspection of plastic products attempting to detect any defects
present. Traditionally this task has been undertaken by process
workers, but this poses certain problems, since humans are
relatively costly, and the repetitive nature of the task can cause
problems with fatigue and loss of concentration. Recent years
have seen advances in machine vision systems, whereby the
human inspector is (at least partially) replaced and/or
supported by the automatic detection.
Prabuwono et al.[18] developed an intelligent visual
inspection system(IVIS) for bottling machine, focusing on the
development of image processing framework for defect

Fig. 4

Intelligent visual inspection system for the surface defects of plastic


bottles

detection. Fig.4 shows the intelligent visual inspection system


for the surface defects of plastic bottles. The experiment
results showed that the system was accurate enough to detect
moving objects at 106 rpm and the accuracy of the image
acquisition was 94.264%.
We mainly focus on the automatic detection of surface
defect for injection products. Fig.5 shows the system structure
schematic for surface defects automatic detection for injection
products.
The system is composed of three parts. They are described
respectively as follows.
(1) Conveyor belt device. It is available to convey the
injection products uniformly and facilitate CCD camera
capture the images dynamically.
(2) Image acquisition and processing section. It can
transfer the collected images to the industrial computer using
CCD camera and image acquisition card. The industrial
computer will process the images and send out judgment
results.
(3) PLC control device. The rejects products will be
separated from the quality ones according to the judgment
results.
The system works by conveying the injection products
through conveyor belt using linear mode transmission.
According to the principle of machine vision, the CCD camera
dynamically captures images synchronized with production
line, and transmits the image data to the industrial computer
for processing using image acquisition card timely. The
industrial computer analyzes and compares the processing
results to make appropriate judgment. Judgment results are
then transmitted to the PLC control unit. The latter locates and
marks on the reject products on the basis of results. Then the
slide driven by the motor rotates, and separates the quality
products and the reject ones into different containers,
achieving detection and sorting the products automatically.
IV.

production level, and has wide application prospect. At


present, there are some difficulties in the promotion and
application of automatic detection. Meanwhile, the application
of the automatic detection technology of surface defects on
plastic products in the practical production is not very popular,
and the research on injection products is expressly rare. In
addition, because of varied types, forms and features of the
defects, it causes much more difficulties in the research. It is
urgently needed of experts to work together to improve the
development and application level of automatic detection
technology of surface defects on injection plastic products.
[1]

[2]

[3]

[4]

[5]

[6]

[7]
[8]
[9]

[10]

[11]

CONCLUSIONS

The research and application of automatic detection


technology of surface defects on plastic products will greatly
liberate the human workforce, improve the automated

[12]

[13]

[14]

[15]
[16]

[17]

[18]
Fig. 5

System structure schematic for surface defect automatic detection of


injection products

F. Duan, Y. Wang, X. Lei, L. Wu., and W. Tan, An overview of


machine vision technology and its applications, Automation Panorama,
no. 3, pp.59-62, 2002.
W. Zhang, Q. Zhao, and D. Wang. Actualities and developing trend of
machine vision, Journal of zhongyuan university of technology, vol.
19, no. 1, pp..9-15, 2008.
H. Tang and Z. Liu, Research of illuminations self-adjusting function
for on-line checking the surface of spray-paint part, Opto electronic
technology, vol. 27, no. 2, pp.135-138, 2007.
X. Qian, X. Qu, Y. Chen, and G. Li, Design of stepless controlling
well-proportioned brightness light and its application on surface defects
detection, Manufacturing Automation, vol. 27, no. 12, pp.43-45,54,
2005.
A. Asundi, C. S. Chan, and M. R. Sajan, 360-deg profilometry: new
techniques for display and acquisition, Optical Engineering, vol. 33,
no. 8, pp.2760-2769, 1994.
M. R. Sajan, C. J. Tay, H. M. Shang, and A. Asundi, TDI imaging and
scanning moire for online defect detection, Optics & Laser
Technology, vol. 29, no. 6, pp.327-331, 1997.
Y. Zhang, Image engineering, Beijing: Tsinghua University Press, 2007.
Y. Zhang, Machine vision and image techniques, Automation
Panorama, no. 2, pp.20-25, 2009.
F. R. Hansen and H. Elliott, Image segmentation using simple Markov
random field models, Computer Graphics and Image Processing, vol.
20, no. 2, pp.101-132, 1982.
S. A. Stansfield, ANGY: a rule-based expert for automatic
segmentation of coronary vessels from digital subtracted angiograms,
IEEE-PAMI, vol. 8, no. 2, pp.188-199, 1986.
H. Derin and H. Elliott, Modeling and segmentation of noisy and
textured image using Gibbs random fields, IEEE-PAMI, vol. 9, no. 1,
pp.39-55, 1987.
P. B. Chou and C. M. Brown, The theory and practice of Bayesian
image labeling, International Journal of Computer Vision, vol. 4, no. 3,
pp.185-210, 1990.
D. Bhandary, N. R. Pal, and D. D. Majumder, Fuzzy divergence,
probability measure of Fuzzy events and image thresholding, Pattern
Recognition Letters, vol. 13, no. 12, pp.857-867, 1992.
J. L. Chen and A. Kunda, Unsupervised texture segmentation using
multichannel decomposition and hidden Markov models, IEEE-IP, vol.
4, no. 5, pp.603-619, 1995.
B. B.Chaudhuri and N. Sarkar, Texture segmentation using fractal
dimension, IEEE-PAMI, vol. 17, no. 1, pp.72-77, 1995.
Tsai, M. Du, and S. K. Wu. Automated surface inspection using Gabor
filters, International Journal of Advanced Manufacturing Technology,
vol. 16, no. 7, pp.474-482, 2000.
R. Geng, L. Cao, S.Geng, and P. Lin, Design and realization of robotvision system, Electronic instrumentation customer, no. 1, pp.36-38,
2007.
A. S. Prabuwono, R. Sulaiman, A. R. Hamdan, and A. Hasniaty,
Development of Intelligent Visual Inspection System(IVIS)for Bottling
Machine, IEEE Region 10 Annual International Conference,
Proceedings. Hong Kong: TENCON 2006.

S-ar putea să vă placă și