Quality detection system for filling infusion bottle production

[0054] Example 1

[0055] figure 1 A schematic structural diagram of a defective sorting and rejecting device for filling infusion bottles 1038 provided by the present invention, such as figure 1 As mentioned above, the device includes: a light source, an industrial camera, a robotic arm 101 and a host computer 102, wherein,

[0056] The industrial camera and the robotic arm 101 are respectively connected to the host computer 102 through Gigabit Ethernet; the maximum transmission rate of the Gigabit Ethernet data interface can reach 1000Mb/s, which is much higher than the transmission rate of USB2.0 and 1394. The industrial camera in the invention is a variable magnification area array CCD color camera. CCD color camera integrates photoelectric conversion, charge storage, charge transfer, and signal reading. Using charge as a signal, it has the advantages of low power consumption, low voltage operation, and no lag. Whether it is static acquisition or dynamic acquisition, it can be obtained without Distorted high-quality images; moreover, the variable magnification area CCD color camera allows the operator to photograph moving objects with one exposure at any shutter speed, which can acquire two-dimensional image information and measure images intuitively. Compared with a linear CCD camera to acquire a two-dimensional image, it must be equipped with devices such as scanning motion and grating, so that the image acquisition time is long and the measurement efficiency is low, and the image accuracy is affected by the scanning motion accuracy. Finally, the RGB components in the image output by the variable magnification area array CCD color camera are different, which can ensure the clarity and sensitivity of the image, and present the image information truly and completely. From traditional monochrome imaging to color imaging, the requirements for color feature extraction in the detection of heterochromatic objects in machine vision technology are improved, as well as the color resolution of image acquisition systems. It can be understood that in the present invention, the existing variable magnification area array CCD color camera and the matching host computer 102 can be used, and the image recognition algorithm in the host computer 102 uses the existing neural network image recognition algorithm. .

[0057] figure 2 A schematic structural diagram of the manipulator 205 in the defective sorting and rejecting device for filling infusion bottles 1038 provided by the present invention, such as figure 2 As shown, the robotic arm 101 is used for gripping the infusion bottle 1038 and grabbing the infusion bottle 1038 to the detection station. The hydraulic telescopic mechanism is composed of a base, a column, an extension arm, and a piston hydraulic cylinder. The column is equipped with a piston hydraulic cylinder, which can complete the reciprocating linear motion. It can be translated up and down, and the flexible hand is installed at the end of the extended arm to grab the infusion bottle.

[0058] The shape of the manipulator 205 is similar to that of a human hand, and a pair of fingers is used as an example for introduction. One finger of a pair of fingers includes a fixed segment 2053 and a free segment 2051. The fixed segment 2053 of the finger is distributed in a C-shaped grip shape. The interior of the fixed segment 2053 is a hollow cavity structure for accommodating the contracted free segment 2051. The free section 2051 is a cylindrical hollow cavity structure made of soft silicone material, and the cavity of the free section 2051 on each finger is connected to the inflating and deflating device 2054 through the air passage 2052 .

[0059] Specifically, the C-shaped grip formed by a pair of finger fixing segments 2053 may be that the fixing segments 2053 are bent toward each other to form a semicircular structure, or a 2/5 ring structure. The sum of the lengths of the free segment 2051 and the fixed segment 2053 of the two fingers is equal to the outer circumference of the infusion bottle 1038 . The flexible manipulator is used to grab the infusion bottle, which avoids the damage of the infusion bottle caused by the direct contact between the infusion bottle and the steel manipulator, and improves the utilization rate of the secondary recovery of the infusion bottle. For example, a traditional electromagnetic ejector or a suction-cup sorting manipulator can also achieve this effect. However, although the suction-cup sorting manipulator increases the fault tolerance, it is easy to cause multiple sorting. Although the electromagnetic ejector is more accurate, it is also easy to cause damage to the infusion bottle. The embodiment of the present invention utilizes a flexible mechanical claw, and the flexible material can better fit the infusion bottle, thereby reducing damage to glass products and improving recycling efficiency. After using the flexible mechanical claw that can be opened and closed, there can be a considerable gap distance for fault tolerance, which greatly improves the fault tolerance of sorting and rejection.

[0060] The detection station includes a white foreign body detection station 103 and a black foreign body detection station 104, wherein, image 3 A schematic structural diagram of the white foreign matter detection station 103 in the defective sorting and rejecting device for filling infusion bottles 1038 provided by the present invention, as shown in image 3 As shown, the white foreign object detection station 103 includes a black light-absorbing background plate 1031, a first detection area 1032, a first light blocking plate 1033, a first industrial camera 1034 and a bottom light source 1035. There is a first detection area 1032 between them, and a first opening 1036 is arranged on the first light blocking plate 1033 at the position facing the first industrial camera 1034, and the size of the first opening 1036 is the same as that of the infusion bottle 1038; the first industrial camera 1034, the first opening 1036, the infusion bottle 1038 and the center of the black light-absorbing background plate 1031 are arranged collinearly; the bottom light source 1035 is arranged facing the bottom of the infusion bottle 1038 to illuminate the bottom of the infusion bottle 1038 with light;

[0061] Figure 4 A schematic structural diagram of the black foreign body detection station 104 in a defective product sorting and rejecting device for filling infusion bottles 1038 provided by the present invention, such as Figure 4 As shown, the black foreign object detection station 104 includes: a backlight 1041 , a second detection area 1042 , a second light blocking plate 1043 and a second industrial camera 1044 , and the second light blocking plate 1043 is provided with a second opening. 1045, the size of the second opening 1045 is the same as that of the infusion bottle 1038.

[0062] Before the robotic arm 101 grabs the infusion bottle 1038, the gas filling and deflating device draws out the gas in the cavity of the free segment. When there is no gas in the free segment, the free segment shrinks into the cavity inside the fixed segment. At this time, the robotic arm 101 controls the robotic arm 205 to move toward the infusion bottle 1038. Since the fixed segments of each finger are distributed in a C-shaped grasping shape, the robotic arm 101 controls the fixed segment of the robotic arm 205 to be attached to the outside of the infusion bottle 1038 in a grasping shape. The surface, and then the inflation and deflation equipment inflates the free section, which is soft silicone, such as the silicone material used to make pacifiers. When the free section is inflated, the free section continuously expands and protrudes from the cavity in the fixed section, and then clamps Hold the infusion bottle 1038. It should be emphasized that after the free segment is fully expanded, its shape is curved, and the free segment and the fixed segment on the same finger are connected end to end to form a semicircle. The two opposite fingers form a ring structure after the card holds the infusion bottle 1038 . In practical applications, the manipulator 205 may have six fingers, and when gripping the infusion bottle 1038 , the two sides of the infusion bottle 1038 are respectively held by three fingers.

[0063] The control methods of the manipulator 101 and the manipulator 205 can be implemented using existing control algorithms, and the present invention only protects their specific mechanical structures and the layout methods of the detection devices.

[0064] The robotic arm 101 places the infusion bottle 1038 in the white foreign body detection station 103 and the black foreign body detection station 104 respectively for foreign body detection. Taking the white foreign body detection station 103 as an example, the bottom light source 1035 emits a red light beam to the bottom of the infusion bottle 1038. The light beam enters the glass bottle and spreads in the bottle body. At the same time, some light enters the liquid medicine in the bottle. When there are cracks in the body or there are foreign objects in the liquid, the light transmission will change. Against the background of the black background, the change is more obvious. This phenomenon is captured by the industrial camera, and the industrial camera will capture the image signal. It is transmitted to the host computer 102, and the host computer 102 invokes the existing image recognition algorithm to identify whether there are points with pixel value variation in the black background, and then judge whether there are cracks or foreign objects according to these points. In the white foreign object detection station 103, the red LED ring light source model OPT-RI9215 and the bottom light supply method are used. The reflectivity of the white foreign object is less than the refractive index, and it appears as a bright spot on the image. The gray value is large, which is easy to achieve and background separation. The LED light source is durable and has stable performance, which is suitable for long-term use. The panel light source emits evenly, and the acquired image background is consistent and has high quality.

[0065] The working principle of the black foreign body detection station 104 is similar to that of the white foreign body detection station 103 , and details are not described herein again in this embodiment of the present invention. In the black foreign object detection station 104, the red LED panel light source model OPT-FLC360240 and the back light supply method are used. Realize the segmentation with the background.

[0066] Further, in the present invention, a variable magnification area array CCD color camera is used, which can automatically adjust the shooting magnification according to the size of the infusion bottle 1038, and group and retain images under different magnification lenses in the detection link to achieve high magnification and high resolution. Under the foreign body detection, improve product quality. Through eyepiece imaging of different magnifications, the small targets in the liquid medicine can be further identified.

[0067] Example 2

[0068] On the basis of Example 1, after the detection of black foreign objects and white foreign objects is completed, Example 2 adds the following zero-degree incubator 201, spray box 202 and third industrial camera 203, wherein,

[0069] The zero-degree incubator 201 accommodates the infusion bottle 1038 to be detected. The zero-degree incubator 201 is a box-type structure, and its side is provided with a side cover that can be automatically opened and closed. When the mechanical arm 101 triggers the limit switch, the side cover drives the structure. Turn on the power to open the side cover, and the robotic arm 101 extends into the zero-degree incubator 201 to take out the infusion bottle 1038, and moves away from the zero-degree incubator 201. When the set position is reached, the limit switch is released and the side cover is closed. Specifically, the limit switch can be arranged on the slide rail where the carrying platform of the robotic arm 101 is located. When the carrying platform moves to the set position, the carrying platform presses the limit switch and maintains the pressed state. The position of the carrying platform at this time enables the robotic arm 101 to take out the infusion bottle 1038 .

[0070] In order to reduce the damage to glass products and improve the recycling efficiency, the bearing platform is designed with a double-track ball screw equipped with a stepping motor, and a unipolar stepping motor drive circuit is used to drive the motor. To precisely control the stepper motor, the rotation of the motor drives the carrying platform to move, and drives the manipulator on the carrying platform to grab. Further, one side of the carrying platform is equipped with a defective sorting table, and the other side is placed with a robotic arm, which can be sorted and carried according to different defective categories, and moved to different defective processing mechanisms through sliding rails, which effectively improves the The fault tolerance of the traditional infusion bottle defective sorting and rejection work is suitable for a more extensive and detailed defective product processing and recycling process.

[0071] Figure 5It is a schematic cross-sectional structure diagram of the interface of the spray box 202 in a defective sorting and rejecting device for filling infusion bottles 1038 provided by the present invention. As shown in 5, the spray box 202 is a box-type structure with an open top, and the spray A number of ultrasonic atomizing nozzles are evenly distributed on the inner wall of the box 202 to spray the atomized liquid on the infusion bottle 1038 . Specifically, the zero-degree incubator 201 is a box-type ultrasonic atomizing nozzle with a shutter on the top, which atomizes the ethanol and sprays it on the outer wall of the infusion bottle 1038 . Further, in order to spray more evenly, the ultrasonic atomizing nozzles are distributed on the four vertical surfaces of the spray box 202 .

[0072] In order to facilitate the collection of dripping ethanol, or the ethanol that is not contaminated by the infusion bottle 1038, the bottom 2021 of the spray box 202 is funnel-shaped, and a collection hole is opened at the lowest point, and the ethanol flows down along the bottom 2021 of the funnel-shaped structure. , and flows into the collection box through the collection hole 2024. Furthermore, in order to facilitate spraying the atomized liquid to the bottom of the infusion bottle 1038 , ultrasonic atomizing nozzles are evenly spaced in the funnel-shaped bottom of the spray box 202 .

[0073] Furthermore, the fan 2023 can be stirred on the funnel-shaped part of the spray box 202 to further increase the flow rate of the ethanol after atomization, which is convenient for the liquefaction and adsorption of ethanol on the infusion bottle 1038 at zero temperature, and can also affect the outer surface of the infusion bottle 1038 at the same time. Play the role of disinfection and sterilization. When the atomized droplets of ethanol meet the outer surface of the low temperature infusion bottle 1038 , they can be quickly adsorbed on the outer surface of the infusion bottle 1038 , forming an uneven ground glass-like surface on the outer surface of the infusion bottle 1038 . When part of the light forms a light spot on the infusion bottle 1038 due to full emission, the frosted glass-like surface can diffuse the emitted light, thereby weakening the intensity of the light. The light spot in the picture captured by the third industrial camera 203 will be significantly reduced, which is beneficial for the upper computer 102 to identify the crack of the glass bottle body.

[0074] It should be emphasized that in the present invention, the transfer of the infusion bottle 1038 between the various processes is realized by the robotic arm 101, and the control algorithm of the robotic arm 101 can use the existing technology, and the embodiment of the present invention will not be repeated here. .

[0075] The embodiment of the present invention can also be used for automatic production lines of glass canned beer bottles and yogurt bottles.

[0076] Example 3

[0077] Based on Embodiment 1 or 2 of the present invention, Embodiment 3 of the present invention is a method for detecting defective products in the production process of canned infusion bottles reasonably based on big data. More objective, precise and standardized. Effectively reduce the missed detection rate and false detection rate in the sorting and rejection of defective infusion bottles, thereby improving the accuracy of detection. Moreover, automated and intelligent detection can greatly improve the detection efficiency of the production line, greatly shorten the detection time, and save costs. The system is a high-speed online detection system integrating machine vision, precision machinery and real-time control. It mainly realizes the detection of bottle mouth breakage, the detection of contamination on the bottle mouth, the bottom of the bottle and the bottle wall, as well as the detection and rejection of residual liquid in the bottle. and other functions.

[0078] Image 6 A schematic flowchart of a quality detection method for filling infusion bottle production provided by the embodiment of the present invention, such as Image 6 shown, methods include:

[0079] S601: Use a robotic arm to transfer the sample infusion bottle to the quality inspection system described in any one of Embodiments 1-2, obtain sample image data output from the white foreign body detection station and the black foreign body detection station, respectively, and obtain sample image data from the sample The structural features of the sample infusion bottle and the quality features of the infusion bottle are extracted from the image data as the first sample set.

[0080] Specifically, when shooting each sample image, the following methods can be used: using real-time shooting for the sample infusion bottle A to obtain two or more sample images;

[0081] Obtain the spot size and spot position in each sampled image, deduct the spot area from the sample image according to the spot size and spot position, and then superimpose the remaining parts of the sample image to obtain a superimposed image. When superimposing, an average value can be taken for the RGB three-channel data of the same pixel respectively, and the average value can be taken as the RGB three-channel value of the pixel point at the same position after the superposition. Then judge whether the superimposed image is defective. If so, send an instruction to the robotic arm to make the robotic arm transport the sample infusion bottle to the zero-degree incubator and spray box in turn, and use the third industrial camera to take a more complete sample image of the sample infusion bottle. . In practical applications, the third industrial camera can capture several sampled images during the volatilization process of atomized alcohol, for example, capture a sampled image every 1 second, and then filter out the sampled image with the most quality features and the superimposed image Use as a sample image after overlaying. Then, continue to perform the step of acquiring several sample images by using the real-time photographing method for each sample infusion bottle, so as to obtain a large number of sample images.

[0082] Further, when obtaining the superimposed image, it is possible to cut each sample image to obtain several blocks; compare several blocks at the same position, and screen out the most structural features and/or quality features of the infusion bottle. The target tiles of each position are stitched together according to their respective positions to obtain a superimposed image.

[0083] S602: Add the first sample set to the current sample set, compare the structural features of the sample infusion bottle in the current sample set with the pre-marked prior structural features of the sample infusion bottle, and compare the quality features of the infusion bottle with the pre-marked prior structural features of the sample infusion bottle, respectively. The prior quality features of the sample infusion bottles are compared respectively, the first similarity between the structural features of the sample infusion bottles and the prior structural features, and the second similarity between the quality features of the sample infusion bottles and the prior quality features When the similarity is less than the preset threshold, adjust the parameters of the upper computer recognition model.

[0084] In the first iteration, the current sample set is empty, and the first sample set can be used as the current sample set. When the quality inspection system obtains the structural characteristics of each sample infusion bottle, such as length, height, neck diameter and other data, as well as quality characteristics, such as scratch length, crack length, crack position, number of black foreign objects, number of white foreign objects, etc.; The system has pre-entered the prior structure features and prior quality features for the sample infusion bottle, and used the prior structure features and prior quality features as the labels of the corresponding samples in the current sample set.

[0085] S603: Transfer the sample infusion bottle to the quality inspection system again, obtain the sample image data output by the white foreign body detection station and the black foreign body detection station respectively, and extract the structural features of the sample infusion bottle and the infusion solution from the sample image data. The quality characteristics of the bottle are used as the second sample set, and the second sample set is added to the current sample set, and the execution of returning the structural characteristics of the sample infusion bottle in the current sample set and the pre-marked prior structure characteristics of the sample infusion bottle, and adding the infusion bottle. The quality features of the sample infusion bottle are compared with the pre-marked prior quality features of the sample infusion bottle, until the first similarity and the second similarity are both greater than the preset threshold, and a trained target model is obtained.

[0086] Further, the upper computer can also add random disturbance to the cutting angle of the robotic arm for each gripping of the sample infusion bottle. For example, when gripping the sample infusion bottle for the first time, it is gripped along the horizontal plane from the 6 o'clock direction, and the next time it can be gripped. The second clamping is performed in the direction of 6+ random numbers, and so on, in order to increase the discrimination between sample images, improve the quality of the samples, and thus avoid the model underfitting caused by the high approximation of the sample images in the current sample set. It should be noted that the random number can be a positive integer between 0-12, or a decimal.

[0087] S604: Then use the target model to perform quality inspection of the infusion bottle.

[0088] The application embodiment can intelligently extract the structural feature root of the infusion bottle, classify and identify the infusion bottle according to the target structure feature, and then carry out targeted identification. While realizing the automatic calculation of the feature weight, automatic screening and feature classification of the detection information can also be performed, and then Perform model training. During model training, the sample infusion bottle is continuously and repeatedly poured into the quality detection system. Due to the grasping angle and error of the robotic arm, the same sample infusion bottle can generate countless samples. Therefore, in the embodiment of the present invention The capacity of the current sample set used for training the model can be expanded easily and automatically, and more accurate training of the target model can be achieved.