Inspection Techniques for Print Quality on Molded Plastic

The applications presented utilized tools that areGenerally, the training required for using such
developed from image processing algorithms. Insystems is minimal since most software packages
the inspection of correctly inserted printsupplied by vision systems manufacturers are
templates.it has been shown for instance that twouser-friendly. The inspection for quality also
feature detection algorithms can be applied; in thisrequires very simple tools like those that have
case canned in the software described as featurebeen demonstrated such as feature count and
detection tools. If the print template is inverted intemplate matching.
any way, whether upside down or left to right,After the molding process is over, the part is
either or both tools will return a "fail" based on theremoved from the mold cavity manually, and
prescribed dark or bright feature size requiredvisually inspected for quality. Despite this, process
across the tools. For print quality, two methodsvariations could cause minor blemishes or smears
have been described. The first is to use aon the print that are not immediately visible to
template match of a good print and compare itthe operator. Figure S shows an example of such
with other parts. This was mainly used to detecta defect with a close-up on a print revealing a
smudges, smears and poorly printed characters.small smear on the letter "d." Two methods can
Provided the degree of mismatch is adequatelybe used for this inspection.The first is to use a
defined, the template match can be used fairlytemporal operation such as the template match
adequately. In this particular case study, thedescribed in section 2.S.The training data is
template match could not be tested extensivelyobtained from a captured image of what is
because of lack of adequate samples.The secondperceived as a very good quality print.
method used algorithms for reading opticalSubsequent images can then be captured from
characters (OCR).The printed characters on aparts as they flow along a conveyor, and each
good plastic part, which are not standard OCRimage compared with the trained data. A problem
fonts, were read into an OCR tool.Throughsuch as a smear or a missing character may
software, the tool was trained to recognize thesecause a mismatch in the number and position of
as OCR fonts with varying degrees ofdark pixels that are in the image. Figure 6 is an
acceptance. Like in the case of the templateillustration of the application of this tool.
match, the OCR was not tested extensively dueAnother useful tool that would he used to inspect
to lack of adequate different productiona print is an optical character reader (OCR).
samples.For example casting mould,moldAlthough the prints are not true optical
making,plastic injection mold etc.characters, using the software, normal non-OCR
A variety of such software and hardware exist infont characters can be trained to correspond to a
the market today. The comparison between theparticular print image.After an image of a good
different software/hardware platforms is notprint is captured, using this software, the actual
intended to be the subject of this paper; howevercharacter string is typed into the OCR reader.
a comprehensive listing can be obtained from theThe reader is then "trained" to interpret the
Automated Imaging Association.20 Plastic moldingimage data as corresponding to character string
processes are widely used in the manufacturingfrom the keyboard entry. After several trials, an
of various engineering and consumer items. Theacceptance level for allowing the captured image
growth of the plastics sector has seen a slightcharacters as ones that match the corresponding
decline (-5% overall) in the U.S. since 2000 due tokeyboard characters is determined. If any of the
the increasing costs of fuel and gas, thecharacters from a subsequent part contains a
weakening of the dollar against major currencieslarge smear it would not match the trained data
in the world, and more so, the movement ofset. Additionally, if there is a missing character on
manufacturing to Asia (especially China). Thisa plastic part, the string will not match the trained
deficit has been absorbed mainly by China, Canadaone. An example of this is shown in Figure . There
and Japan. Despite this, there is still a substantialare two limitations with this tool however. The
proportion of manufacturing companies in the U.S.,first is that there ought to he adequate spacing
especially in the molding industry. Thus there is stillhetween the characters for it to work effectively.
a great need for improved process and qualitysecondly, minor smears on the prints may not
control. This paper presents a simple approacheasily he detected. Such limitations have heen
that utilizes commercially available hardware andaddressed hy the use of advanced processing
software for machine vision applications toalgorithms such as those that utilize neural-fuzzy
automate the inspection of molded plastics.classifiers.