This article describes the application of machine vision techniques that can be applied to automatically detect incorrectly placed labels in a manually loaded injection molding process. In this application, simple techniques involving counting features are used to check orientation along a particular direction. If the part is incorrectly placed, the vision system sends a signal through its input and output (I/O) system, and an external device such as a buzzer or warning light will be activated. The use of more advanced techniques such as character recognition and template matching are also discussed with specific applications in inspecting the quality of print.
In order to segment a region into foreground and background, it is necessary to threshold an image. The threshold.T. basically sets a boundary between pixels that are considered dark. i.e. l(m.n)
The global characteristics of the image derived from the histogram are used to modify individual pixel values. Contrasting in point operations is usually achieved by simple scaling. In global operations, the approach is slightly different.A technique known as histogram equalixation is used to redistribute pixel values in order to produce a uniform histogram. In an image with m rows and n columns, and with a bit resolution of r. an ideal histogram would be uniform with (m x n /2^sup r^) pixels at each gray level.
The spatial distributions of the pixels are usually changed to deliberately achieve a desired effect. Examples of geometric operations include magnification of images, rotations and transformations. In general, these operations involve mapping functions which would transform a set of pixels at a location (x,y) to another location (x',y').
Frame-based operations basically utilize more than one image to perform or achieve a desired effect.17'18 An example of such an operation commonly used in inspections is a point-by-point comparison or subtraction of one pixel from another. A newly constructed image can then be passed as good or bad, depending on how it compares with the original or some reference set of pixels. Another algorithm that may be used is one that compares distances between similar features.This algorithm, known as template matching, uses a known image's data as a training set.The data will typically consist of distances between features as vectors, and/or their similarities. 17An unknown image vector data is compared to that of the training set. A threshold is set to define level of success for which the unknown image compares with the trained data. In this paper, template matching was used to identify print quality on plastics. For example casting mould,mold making,plastic mold etc.
In order to obtain prints on a plastic molding, most processes employ a pre-printed plastic template to be placed in the mold cavity, like the one shown in Figure 1. Figure 2 shows an example of complete molded parts with prints showing.The template is inserted in the mold cavity, usually upside down so as to obtain the correct orientation of the print. Once the template is secured in the cavity, the mold is closed and the molding is done over the print. The part is then retrieved and manually inspected for quality. Usually, this inspection process is not done on all the parts and at times, a fatigued operator may miss discovering a poorly printed part before it is shipped to the customer.
In order to avoid heavy penalties that are usually imposed for supplying defective parts, a smart vision system can be used to ensure that the template is placed in the proper orientation before the mold halves are closed. A schematic of how this can be done automatically is illustrated in Figure 3.To test it, a prototype was developed with a DVT(TM) series 600 camera.18 The camera was set up to capture the image as soon as the template was placed in the mold half. Next, a number of algorithms (or tools) were used to check whether the print orientation was correct.Two linear feature count tools were employed. One was set at the top of the left edge of the print and the other at the bottom left. The lower tool was set to identify existence of a line (dark feature with a specified minimum thickness in pixels in this case) and the upper one set not to identify any dark or bright features. Using Framework(TM) software,18 it is fairly easy to apply and set these tools to perform the required inspection without having to carry out elaborate image processing and analysis.The result of each feature count was then digitally sent to the camera I/O board as "Pass" or "Fail." Using the software, one can toggle through the digital I/O settings to configure this.The camera board's signal was then sent to a programmable logic controller's (PLC) input. If the template was inserted wrongly, the PLC was programmed to set off an audible alarm (buzzer) and turn on a warning light. Figure 4 is an illustration of the application of the linear feature count tools used to perform this inspection.
During the setting up of this test, the feature count tools were tested severally (at least ten times) under fixed lighting and optical conditions. This was used to determine the tolerance levels for the feature sizes. In a real-life application, the number of tests would have to he increased because of possibility of noise interference such as varying lighting conditions with time, or other sources such as electrical signals. With well-set tolerance levels on minimum and/or maximum feature sizes, it is possible to achieve 100% efficiencies in the inspection processes.

Automated Inspection for A Case Study

Partager cet article