5.1. Fish cake grasping and sorting test
Cracking and sorting tests for fish cake processing were performed using the automated algorithm described above. In this test, the speed of the belt conveyor conveying fish cakes was set to 500 rpm, while the grabbing speed of the parallel robot was set to 800 mm/s. Three types of fish cakes were used in the test, characterized by: smaller or similar distances from the center point to the corners, longer shapes, and larger fish cake areas. Figure 9 illustrates the fishcake model used in this study. The first type of fish cakes include cuboid, heart-shaped, sphere, round, oval, triangle, etc. The second type is long and thick cylindrical fish cakes, and the third type is flat square fish cakes.
For the first type of fish cake, a single suction cup is used for grasping, as shown in Figure 10a. The second type of fish cake is grabbed using two suction cups, as shown in Figure 10b. The third type of fish cake uses all three suction cups for grasping, as shown in Figure 10c. Afterwards, the fish cakes of each class were put into their respective buckets. Use a suction cup to hold the type 1 fish cake and place it into the first bucket, as shown in Figure 11a. The fish cakes from category 2 were clamped by two suction cups and sorted into the second bucket, as shown in Figure 11b. The fish cakes from category 3 were clamped by three suction cups and sorted into the third bucket, as shown in Figure 11c.
5.2. Test result analysis
As mentioned before, the image recognition and grasping test of the fish cake in Figure 8 was performed 10 times for each shape. Tables 1 and 2 summarize the results of these tests. Table 1 summarizes the image recognition success rate of fish cakes. Items marked 1, 2, and 3 indicate that they are classified into categories 1, 2, and 3, respectively. “-” indicates that the image recognition failed and was not classified into any category. Items that were not recognized are shown in dark color and items that were recognized as another category are shown in gray. Cuboids, spheres, circles, and triangles have always been considered the first category. However, heart and oval shapes are recognized as the first category with 80% probability. The Heart encountered recognition failure twice, but when it was not recognized, it was classified as Level 1 and had no gripping issues. Oval was twice identified as category 2 due to its elongated shape. Items classified as long, thick, and cylindrical were all correctly identified as category 2, while flat squares were uniformly identified as category 3. In this case, all the problems seem to stem from an insufficient data set. Input fish cakes may be incorrectly classified into different groups based on factors such as light, shape, illumination, etc., or recognition results may not be generated. It is evaluated that augmenting the dataset by incorporating more diverse examples can effectively solve this problem.
Table 2 summarizes the catching and sorting success rates of fish cakes. Instances that failed to crawl and classify are indicated by a dark color and an “X”. When the capture is successful but somewhat unclear due to issues such as fish cake width and material, it is marked with gray and “△”. The capture and classification success rate was 100% for all fish cakes except the oval ones.
As mentioned before, recognition failure occurred for the heart; however, since it was scheduled to be classified as Category 1, there was no problem with grabbing and classifying it. In the case of ellipses, grasping and classification failed twice each. Despite being classified as Category 2, grasping was successful using two suction cups, aligned with the shape of the ellipse.
Catching Level 2 fishcakes can prove challenging because of their elongated nature, and even the slightest deviation in the center point can alter their shape. For long and thick cylindrical fish cakes, it is easy to have small positional errors when grasping. However, since the fish cakes are thick and long, it is relatively easy to grasp the thicker fish cakes. In contrast, the grasping success rate of long fish cakes with thin shapes and cylindrical fish cakes with irregular surfaces decreased significantly.
The thinnest of the long fish cakes was about 15mm thick, while the diameter of the clamp in this test was 20mm. To reliably catch thin or uneven fish cakes in the future, it may be necessary to consider changing to a smaller diameter suction cup. During this transition, it was crucial to choose a suction cup that could securely hold the heaviest fish cake tested (which weighed 86 grams). If the diameter of the suction cup is too small, the suction power may be weak. Additionally, reducing the diameter increases the pressure applied, but too much pressure can cause the suction cup to collapse. Therefore, the appropriate pressure level must be determined to ensure effective gripping without affecting the functionality of the suction cup.
Category 3, flat and square, has a large surface area and stable grasping with three suction cups, and the success rate of grasping and sorting is 100%.
Source link