|
In
a plant that inspects 30,000 lbs of fruit an hour and removing 100% of
the foreign matter is the primary goal, trying to improve quality
control is both a necessity and a challenge. A North American blueberry producer had such a challenge
for Orus Integration, a machine-vision systems integrator in
Boisbriand, Quebec. During the spring of 2003, the customer
commissioned Orus to design a machine vision-based blueberry inspection
system to replace its current optical-based system and eventually
reduce some of the labor costs associated with the manual inspection
step. Orus Integration's team of engineers believed that a
color-based machine vision system was the way to go. "Many of our
competitors offer systems with optical inspection, but those can't use
color analysis and don't generate result data," says Louis Dicaire,
Project Manager at Orus Integration. "Initially, we could guarantee our
system would catch 94% of the foreign material. Our tests caught 97%."  The
Orus FL6500C blueberry inspection system currently uses five color
Marlin 1394 cameras connected to three Matrox Meteor-II/1394 adapter
cards. The system is powered by three Dual Xeon 3.06 GHz servers and a
single P4 1U client machine that acts as a Graphical User Interface
(GUI). The image data is analyzed by the Matrox Imaging Library (MIL);
results are sent to an Omron PCL via Ethernet cable to control the
reject mechanism. A white strobe illuminates the inspection area. There is the option of adding five more Marlin cameras to
this inspection system so that the berries are seen from above and
below. (The current five camera configuration only allows for viewing
from above.) Adding new cameras requires the addition of two more Dual
Xeon machines, bringing the CPU total to ten (see Figure 1). |
|
To
cover the 60-inch width of the conveyor, Orus' FL6500C system uses five
color Marlin 1394 cameras from Advanced Vision Technology Ltd. which
are connected to three Matrox Meteor-II /1394 adapter cards. The system
is powered by three Dual Xeon 3.06 GHz 1U servers and a single P4 1U
client machine that acts as a Graphical User Interface (GUI). The image
data is analyzed by the Matrox Imaging Library (MIL); results are sent
to an Omron PLC via Ethernet cable to control the reject mechanism. A
white strobe LED illuminates the inspection area.
First
the berries are dumped onto a vibrating conveyor whose surface is
designed with "lanes" to help the berries sort themselves into single
layer, facilitating inspection. Then the berries are transferred to a
conveyor with a textured belt that grips the berries to bring them to
rest within the first two feet of the 12-ft belt. "The textured
conveyor almost works too well!" recalls Dicaire. The vision system is
positioned over the end of the textured conveyor and captures the
images of the berries as they are "thrown" off the edge. The textured conveyor on the FL6500C moves at a rate of
600 feet/minute, so the entire system's timing is critical for optimal
operation. The system pulses every 1/8"; the 20th pulse (every 2.5
inches) triggers the cameras for a 120 µs-exposure. Inspection relies
almost solely on MIL's blob analysis module, and each blob is analyzed
according to its average hue, average brightness in the red layer, size
and roundness. Based on this criteria, ice chunks can be rejected based
on color, for example, and other foreign particles such as twigs or
insects can be rejected based on their lack of roundness and/or color.
Unripe or overripe fruit can also be caught based on their color values
or size, as can other fruit such as cranberries, which sometimes ends
up in the blueberries. An array of air jets directs the particles that fail
inspection onto a reject plate. The air jets are positioned over the
14-inch gap between the textured conveyor and a third conveyor. When
the processing locates the "bad blobs", their positions are converted
to a reject array which corresponds to the air jets which shoot the
matter out of the path of the good berries. Since the engineers at Orus are experienced MIL users,
most of the project's challenges were mechanical: assembling components
and handling the speed. Maintaining the timing for reject mechanism was
vital to the mechanics and image processing, because the engineers only
have a 20 ms window available for the analysis and creation of reject
array. If the air jets are engaged for too long, they will direct good
fruit onto the reject plate. Furthermore, the processing of each image
has to take the same amount of time, regardless of the number of
berries in the image; the processing of frames with more berries cannot
take longer. "The number of berries per image is quite random, and we
didn't want to limit the number of blobs that could be processed in a
given image," explains Dicaire. "And since Matrox optimized the
algorithm to separate the hue layer for MMX, the 20 ms time requirement
could be met." Finally, Orus felt the project should be product
independent, so the system could be easily modified for other foods. |
|
The
first advantage the FL6500C has over its competition is speed. The
system's reject mechanism is also unique; it inspects 20,000
berries/sec and relies on logic to locate and reject the bad blobs.
Flexibility is also key, since the operator has full control over the
tolerances and performance for shape and color analysis, as well as the
timing of the air jets. Finally, the user can easily find out exactly
how many blueberries are inspected and rejected in a given batch. No
product for that industry provides such a wide array of both
quantitative and qualitative results.
Currently
the FL6500C was developed for a specific customer. "But we wanted to
ensure the system was product-independent so that we could adapt it for
other food products such as coconuts or cranberries," notes Dicaire. For more information, contact Orus Integration.
Matrox Meteor-II/1394 is an IEEE 1394-to-PCI adaptor board for simplified high-performance digital video capture.
|