Intelligent empty can detection device for ring-pull can production line and detection method of intelligent empty can detection device

Abstract

The invention discloses an intelligent empty can detection device for a ring-pull can production line and a detection method of the intelligent empty can detection device. An input station, a tank opening detection station, a tank body inner wall detection station, a tank body inner bottom detection station, a tank body outer wall multi-detection station, a tank body bottom outer side multi-detection station and an output station are sequentially arranged on the circumference of the tank body, a positioning assembly capable of being matched with tank bodies with different diameters is arranged, and the input station is provided with a height and diameter detection sensor; a digital twinborn system is innovatively introduced, a'physical detection device-virtual twinborn model 'real-time mapping system is constructed, and detection full-process visualization, data full-link tracing and virtual simulation optimization are achieved. According to the invention, all-directional defect detection of the zip-top can is realized, the dynamic optimization capability of detection parameters, the equipment fault pre-judgment efficiency and the production collaboration level are improved through the digital twin technology, the method is adaptive to cans of different specifications, the detection precision and generalization are remarkably improved, and the method is suitable for a high-standard zip-top can production line.

Description

Technical Field

[0001] This invention belongs to the field of quality inspection technology, specifically relating to an intelligent detection device and method for empty cans used in beverage can production lines. Background Technology

[0002] In the aluminum can production line, empty cans are transported onto the line by logistics forklifts. After the outer packaging film and tape are removed by an automatic unpacking and untagging machine, they enter the depalletizer. The production date and batch number are printed on the bottom of the can by an inkjet printer, and unqualified products are rejected by the ECI (Empty Can Inspection System). Qualified empty cans are rinsed with brewing water and then enter the filling machine to wait for processing. At the same time, sake is transported to the filling machine through a sterile, fully sealed, insulated pipeline. After being purged with carbon dioxide to replace the air, the filling process is completed. The can lids are precisely conveyed by a fully automatic capping machine to the sealing machine, where they are sealed with the filled cans. Subsequently, the cans pass through the FCI (Filling Quality Inspection System) for liquid level height, airtightness, and sealing quality checks to ensure that the sealing parameters meet the standards. Products that pass the inspection are put into the pasteurization machine for heat treatment. After sterilization, they pass the FCI re-inspection a second time. The water stains on the can mouth and body are dried by the air knife. Finally, the packaging machine completes the packing. After the full box is weighed and visually inspected, the box is printed with code, stacked and wrapped with film and traceability barcode is affixed.

[0003] The paper "Design of Intelligent Detection System for Empty Cans" (Li Guangmei, Gou Yanyan, Liu Guoling, Packaging & Food Machinery, 2017, 35(06)) addresses the potential problems of product leakage, spoilage, and contamination caused by defects in empty cans before filling. It designs an intelligent detection system for empty can defects, focusing on the design of a negative pressure conveying system, a visual inspection optical path, and a defect detection system. The system has three detection stations for the can bottom, mouth, and body. When the can is conveyed by the conveyor belt, it transitions from a regular chain to a negative pressure chain. Small holes are opened on the negative pressure base plate, which, under the action of a negative pressure fan, adheres to the empty can, forming a stable conveyor. When the empty can triggers the photoelectric sensor, the system activates the lighting source and simultaneously starts the camera to acquire an image. After acquisition, the image is preprocessed to filter out noise and enhance image information. The above methods have the following problems: ① The lack of tank positioning makes it easy for the angle between the tank and the light source to deviate, affecting the accuracy of image analysis results; ② The defect detection of the tank station uses a top blue ring light and a 4mm wide-angle lens to take images of the inside of the tank from above to determine whether there are wrinkles or foreign objects attached to the tank. This method cannot detect whether there are scratches on the outer wall of the tank. At the same time, the inner wall area has a curved structure, and the image formed is a compressed image. Moreover, it is difficult to ensure uniform lighting effect and surface reflection exists. Therefore, it is more effective in judging the accuracy of some areas of dents and wrinkles; ③ The bottom inspection uses a top white ring light and a 12mm lens to take images. The structure of the reinforcing ribs will seriously interfere with the imaging of the bottom area, affecting the accuracy of bottom deformation detection. In addition, it is impossible to detect whether there are scratches on the outside of the bottom of the tank.

[0004] The market offers standard cans with a diameter of 65–67 mm, as well as smaller cans (250 ml and below) with a significantly reduced diameter of 53–54 mm. There are also variations such as short and stout cans and slim cans, with diameters generally ranging from 50 mm to 70 mm. The different can diameters pose certain challenges to the design of can positioning. The height of cans with different capacities usually varies significantly, but is generally between 110 mm and 130 mm. The difference in can height has a certain impact on image analysis methods.

[0005] Existing testing equipment has the following limitations: Low visualization: The physical testing process relies on manual inspection and observation, making it impossible to intuitively grasp the real-time status of each component (such as whether the positioning block is accurately fitted, whether the light source is attenuating). Troubleshooting requires disassembling the equipment one by one, which is inefficient. Data fragmentation: The testing data and equipment operation data of each station are stored in a scattered manner and do not form a correlation link, making it impossible to quickly trace the specific testing process of a non-conforming tank (such as which station missed the test, whether the light source parameters were abnormal at that time). High parameter debugging cost: When new specification tanks are put into operation, the light source, camera, and positioning parameters need to be repeatedly debugged on the physical equipment. A large number of non-conforming products are generated in the process, and the debugging cycle can last for several hours. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings and deficiencies of the existing technology and to provide an intelligent detection device and method for empty cans used in beverage can production lines.

[0007] The technical solution adopted by this invention is as follows: An intelligent empty can detection device for a beverage can production line includes a detection conveying device, wherein the detection conveying device comprises, in sequence along the conveying direction: Enter the workstation; The can opening inspection station is equipped with a light shield and a light source and camera set inside the light shield to capture images of the can opening area for detecting can opening defects. The tank inner wall inspection station is equipped with a light shield and a light source and camera set inside the light shield to capture images of the tank's circumferential inner wall area for detecting defects in the tank's inner wall. The bottom inspection station inside the tank is equipped with a light shield and a light source and camera set inside the light shield to take images of the bottom area inside the tank for detecting defects inside the tank. The first inspection station on the outer wall of the tank is equipped with a light shield and a light source and camera set inside the light shield. The light source is planar diffused light, which takes images of the area of ​​the inner wall of the tank around the circumference of the tank from two opposite sides of the tank, and is used to detect defects on the outer wall of the tank. The second inspection station on the outer wall of the tank is equipped with a light shield and a light source and camera set inside the light shield. The light source is planar diffused light, which takes images of the inner wall area around the tank from the other two opposite sides of the tank, and is used to detect defects on the outer wall of the tank. The third inspection station on the outer wall of the tank is equipped with a light shield and a light source and camera set inside the light shield. The light source consists of four sets of high-angle strip oblique light sources distributed in a square direction, which take images of the tank's circumferential inner wall area from all four sides of the tank, and are used to detect defects on the outer wall of the tank. The first inspection station on the outer side of the tank bottom is equipped with a light shield. A light source and a camera are located below the tank. The light source is planar diffused light, which takes an image of the outer area of ​​the tank bottom to detect defects on the outer side of the tank bottom. The second inspection station on the outer side of the tank bottom is equipped with a light shield. A light source and a camera are located below the tank. The light source consists of four sets of obliquely distributed LED light sources that capture images of the outer area of ​​the tank bottom to detect defects on the outer side of the tank bottom. Output station.

[0008] The detection conveying device is a disc-shaped detection conveying wheel, which includes a chassis. The input station, tank opening detection station, tank inner wall detection station, tank inner bottom detection station, tank outer wall first detection station, tank outer wall second detection station, tank outer wall third detection station, tank bottom outer first detection station, tank bottom outer second detection station, and output station are arranged sequentially at 30° intervals around the chassis, and two wheel gaps are formed between the output station and the input station.

[0009] The chassis is fixed with twelve sets of positioning components arranged in a ring. The positioning components have a released state and a positioned state. During the detection process, the twelve sets are simultaneously in the released state, and the positioning components are a certain distance away from the bottom of the can. During the movement, the twelve sets are simultaneously in the positioned state, and the positioning components are in contact with the bottom of the can.

[0010] The positioning assembly includes a positioning mounting ring fixedly connected to the chassis, an adjusting ring disposed on the positioning mounting ring, and a positioning block disposed between the positioning mounting ring and the adjusting ring. The adjusting ring is fitted with the positioning mounting ring in an inner and outer sleeve for limiting and relative rotation. A pressure cap is fixedly connected to the chassis, which is positioned above the adjusting ring to limit its movement. Several positioning blocks are provided and evenly distributed, including a positioning part, a guide part, a T-shaped limiting groove, and an adjusting rod. The positioning part is located at the inner end in a relatively radial direction, the guide part is inclined and located at the upper end of the positioning part, and the T-shaped limiting groove cooperates with the T-shaped guide strip on the positioning mounting ring. The lower end of the adjusting ring is provided with a corresponding number of equally distributed variable diameter arc-shaped adjusting grooves, and the adjusting rod extends into the variable diameter arc-shaped adjusting grooves.

[0011] The outer periphery of the adjusting ring is a first external gear section, and the inner side of the twelve positioning components is provided with an adjusting gear. The adjusting gear is provided with a second external gear section. The second external gear section of the adjusting gear meshes with the first external gear section of the adjusting ring. The adjusting gear is connected to an adjusting drive mechanism for driving the adjusting gear to rotate.

[0012] The pressure cap is disc-shaped, and its outer periphery extends above the adjustment ring of the twelve positioning components.

[0013] The input station is equipped with a diameter detection sensor; the position parameters of the positioning component in the loose state and the positioning state are set according to the diameter signal detected by the diameter detection sensor.

[0014] It includes an input wheel, an output wheel, and a limiting wall; the input wheel, the detection conveyor wheel, and the output wheel are arranged in an isosceles triangle; the input wheel has several input arc grooves evenly distributed around its circumference, the input station is adjacent to the input wheel, and the input wheel is higher than the detection conveyor wheel; the output wheel has several output arc grooves evenly distributed around its circumference; the output station is adjacent to the output wheel, the output wheel is higher than the detection conveyor wheel, and a push-out mechanism is provided at the bottom of the output station.

[0015] The input station is equipped with a height detection sensor. The images captured afterward are classified based on the height specification signal detected by the height detection sensor, and then analyzed according to the image analysis parameters of the specific category.

[0016] Also includes; The digital twin system includes a digital ID card allocation module, a data acquisition and transmission module, a twin modeling module, and a real-time mapping module; The digital ID card allocation module is connected to the input station and assigns a unique virtual digital ID card code to each can entering the input station. The data acquisition and transmission module is communicatively connected to the sensors, cameras, and drive mechanisms at each workstation in the detection and conveying device, and receives signals in real time. The twin modeling module is used to construct a three-dimensional twin model of each can with a unique virtual digital ID code based on the data from the data acquisition and transmission module, and to change the three-dimensional twin model in real time according to the signals transmitted by the sensors, cameras and drive mechanisms at each station in the detection and conveying device. The real-time mapping module is used to map and display the three-dimensional twin models of all the cans in the detection and conveying device, which are constructed by the twin modeling module, onto the three-dimensional structure of the detection and conveying device.

[0017] The empty can detection method described above for an intelligent empty can detection device used in a beverage can production line is adopted. The beneficial effects of this invention are as follows: This invention provides an intelligent empty can detection device and method for use in aluminum can production lines, enabling comprehensive detection of aluminum cans and making it suitable for aluminum can production lines with high product standard requirements. Simultaneously, this invention innovatively introduces a digital twin system, constructing a real-time interactive system of "physical detection device - virtual twin model." Through digital ID allocation, real-time data acquisition, virtual precise modeling, and full-process mapping and traceability, it achieves full-process visualization of detection, dynamic parameter optimization, and early fault prediction, significantly improving detection accuracy and production stability. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, obtaining other drawings based on these drawings without creative effort still falls within the scope of the present invention.

[0019] Figure 1 This is a schematic diagram of the intelligent empty can detection device for a beverage can production line provided by the present invention, viewed from below.

[0020] Figure 2 A partial cross-sectional view for inspecting the conveyor wheel.

[0021] Figure 3 This is a schematic diagram of the positioning block.

[0022] Figure 4 This is a schematic diagram of the regulating ring structure; Figure 5 This is a schematic diagram illustrating the interaction between modules in a digital twin system. In the diagram, 100 is the input wheel; 110 is the input arc groove; 200 is the detection conveyor wheel; 210 is the chassis; 220 is the positioning mounting ring; 230 is the positioning block; 231 is the positioning part; 232 is the guide part; 233 is the T-shaped limiting slide groove; 234 is the adjusting rod; 240 is the adjusting ring; 241 is the variable diameter arc adjusting groove; 242 is the first external gear part; 250 is the adjusting gear; 260 is the pressure cap; 300 is the output wheel; 310 is the output arc groove; 400 is the limiting wall; A is the input station; B is the tank opening detection station; C is the tank inner wall detection station; D is the tank inner bottom detection station; E is the tank outer wall first detection station; F is the tank outer wall second detection station; G is the tank outer wall third detection station; H is the tank bottom outer side first detection station; I is the tank bottom outer side second detection station; J is the output station. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings.

[0024] like Figure 1 As shown, the input wheel 100, the detection conveyor wheel 200, and the output wheel 300 are arranged in an isosceles triangle. The input wheel 100 has 12 input arc grooves 110 evenly distributed around its circumference. The detection conveyor wheel 200 includes a chassis 210 and 12 sets of positioning components fixed on the chassis 210 and evenly distributed around its circumference. The output wheel 300 has 12 output arc grooves 310 evenly distributed around its circumference. A limiting wall 400 is provided between the input wheel 100 and the output wheel 300, and has two arc limiting walls that respectively cooperate with the outer walls of the input wheel 100 and the output wheel 300, so that the can is limited to the input arc groove 110 and the output arc groove 310.

[0025] The inspection conveyor wheel 200 has, in a clockwise direction, an input station A, a can opening inspection station B, a can inner wall inspection station C, a can inner bottom inspection station D, a can outer wall first inspection station E, a can outer wall second inspection station F, a can outer wall third inspection station G, a can bottom outer first inspection station H, a can bottom outer second inspection station I, an output station J, and a wheel space between the input station A and the output station J. The input station A is adjacent to the input wheel 100, which is higher than the inspection conveyor wheel 200, so that the cans automatically fall into the positioning component of the input station A under the action of gravity. The output station J is adjacent to the output wheel 300, which is higher than the inspection conveyor wheel 200, and has a push-out mechanism (not shown in the figure, but a pneumatic rod can be used) at the bottom to move the cans at the output station J upward, away from the positioning component of the output station J, and into the output arc groove 310 of the output wheel 300.

[0026] Input station A is equipped with a height detection sensor and a diameter detection sensor. Each time it restarts, it detects the first aluminum can that enters input station A and obtains the height and diameter specifications of the corresponding batch. The images captured afterward are classified according to the height specifications, and then analyzed according to the image analysis parameters of the specific categories. The motion parameters of the positioning component are adjusted according to the diameter specifications to make the positioning component match the diameter of the aluminum can. Specifically, height detection sensors can employ camera capture combined with image recognition technology; Specifically, the diameter detection sensor can be an infrared through-beam sensor array. Edge recognition is performed based on the occlusion signal collected by the infrared through-beam sensor array, and the tank diameter D (diameter = distance between two critical points) is calculated through the on / off critical points of adjacent sensors. Station B for can opening inspection is equipped with a lifting light shield. Inside the light shield are a light source and a camera located at the top. The light source is a coaxial white ring light source (to reduce edge diffuse reflection); the camera has 2 megapixels and a 12mm lens (for vertical overhead shots of the can opening) and is used to inspect the edge and inner rim of the can opening. Station C, the inner wall inspection station of the tank, is equipped with a liftable light shield. Inside the light shield are a light source and a camera. The light source is a set of ring-shaped LED light sources (diameter matched to the tank opening, power 30W). The axis of the light source coincides with the axis of the tank. During inspection, it is moved above the tank opening, and the light enters the inner wall obliquely (using diffuse reflection to illuminate the curved surface of the inner wall). The camera consists of 4-6 miniature cameras (with wide-angle lenses) evenly arranged around the circumference. The lens axis forms an angle of 30°~45° with the axis of the tank, and the shooting range covers different quadrants of the inner wall (for example, 4 cameras, each camera is responsible for a 90° area), used to detect scratches and wrinkles on the inner wall of the tank. The bottom inspection station D inside the tank is equipped with a lifting light shield. Inside the light shield are a light source and a camera located at the top. The light source consists of 4 groups of equally spaced, obliquely incident LED light sources (30° incident angle, obliquely pointing towards the edge of the tank to create a fixed shadow on the reinforcing ribs). The camera is a 5-megapixel camera (16mm lens, vertically shooting the bottom of the tank). By using the "standard reinforcing rib template" and the real-time image for differential processing, the interference from the reinforcing ribs is filtered out, and the deformed / foreign object area is preserved. The first inspection station E on the outer wall of the can is equipped with a lifting light shield. Inside the light shield are two sets of light sources and cameras located on the side walls on both the inner and outer sides of the can. Light source: backlight (white planar diffused light LED), camera: 2-megapixel camera (8mm lens, one camera for each set of light sources). The second inspection station F on the outer wall of the can is equipped with a lifting light shield. Inside the light shield are two sets of light sources and cameras located on the side walls on the left and right sides of the can. Light source: backlight (white flat diffused light LED), camera: 2-megapixel camera (8mm lens, one camera for each set of light sources). The third inspection station G on the outer wall of the tank is equipped with a lifting light shield. Inside the light shield are four sets of light sources and cameras distributed in a square. The light source is a high-angle strip light source (60° incident angle, obliquely shining on the side of the tank to highlight the shadow of the scratch). The camera is a 2-megapixel camera (8mm lens, one camera for each set of light sources). The first inspection station H on the outer side of the bottom of the tank is equipped with a lifting light shield. The corresponding area on the chassis 210 is transparent. A light source and a camera are fixed below the chassis 210. The light source is a bottom backlight (white LED, which shines through the bottom of the tank from below the chassis); the camera is a 2-megapixel camera (12mm lens, which takes pictures of the outside of the bottom of the tank from below the bottom plate, and uses scratches to block the backlight to form dark patterns). The second inspection station I on the outer side of the bottom of the tank is equipped with a lifting light shield. The corresponding area on the chassis 210 is transparent. A light source and a camera are fixed below the chassis 210. Four sets of obliquely distributed LED light sources (30° incident angle, obliquely shining into the inside from the bottom edge of the tank, so that the reinforcing ribs produce a fixed shadow) are used. The camera is a 5-megapixel camera (16mm lens, vertically shooting the bottom of the tank). By differentiating the "standard reinforcing rib template" with the real-time image, the interference of the reinforcing ribs is filtered out, and the deformed / foreign object area is preserved.

[0027] The can is fed into the input arc groove 110 at the bottom of the input wheel 100. The input wheel 100 rotates counterclockwise to transport the can to a position close to the detection conveyor wheel 200. Under the action of gravity, the can falls into the input station A. The detection conveyor wheel 200 rotates clockwise and passes through the can opening detection station B, the inner wall detection station C, the inner bottom detection station D, the outer wall first detection station E, the outer wall second detection station F, the outer wall third detection station G, the outer bottom first detection station H, and the outer bottom second detection station I in sequence. After reaching the output station J, the ejection mechanism (not shown in the figure, but a pneumatic rod can be used) moves the can at the output station J upward, away from the positioning component at the output station J, and into the output arc groove 310 of the output wheel 300.

[0028] The light-shielding covers for each workstation can be integrated into one unit. The arc-shaped cover can simultaneously cover the tank opening inspection workstation B, the tank inner wall inspection workstation C, the tank inner bottom inspection workstation D, the tank outer wall first inspection workstation E, the tank outer wall second inspection workstation F, the tank outer wall third inspection workstation G, the tank bottom outer first inspection workstation H, and the tank bottom outer second inspection workstation I. Seven partitions block the light source of adjacent inspection workstations, forming eight inspection chambers corresponding to the above-mentioned inspection workstations, reducing costs while ensuring the synchronization of the inspection actions of each workstation.

[0029] like Figure 2 As shown, the positioning assembly includes a positioning mounting ring 220 fixedly connected to the chassis 210, an adjusting ring 240 disposed on the positioning mounting ring 220, and a positioning block 230 disposed between the positioning mounting ring 220 and the adjusting ring 240. The adjusting ring 240 and the positioning mounting ring 220 are fitted together to limit the movement of the adjusting ring 240 relative to the positioning mounting ring 220. A pressure cover 260 is fixedly connected to the chassis 210. The outer periphery of the pressure cover 260 extends above the adjusting ring 240 of the 12 positioning assemblies, forming an upper limit on the adjusting ring 240. Three positioning blocks 230 are provided and are evenly spaced in a triangular distribution. Figure 3 As shown, the device includes a positioning part 231, a guide part 232, a T-shaped limiting slide groove 233, and an adjusting rod 234. The positioning part 231 is located at the inner end of the relatively radial direction. The guide part 232 is inclined and located at the upper end of the positioning part 231. The T-shaped limiting slide groove 233 cooperates with the T-shaped guide strip (not shown in the figure) arranged radially on the positioning mounting ring 220 to ensure that the positioning block 230 can only slide in a directional direction. The adjusting rod 234 is a longitudinally extending round rod. Figure 4As shown, the lower end of the adjusting ring 240 has three equally spaced, triangularly distributed variable-diameter arc-shaped adjusting grooves 241. Each variable-diameter arc-shaped adjusting groove 241 is a planar spiral. The adjusting rod 234 extends into the variable-diameter arc-shaped adjusting groove 241, causing the three positioning blocks 230 to slide radially inward or outward simultaneously when the adjusting ring 240 rotates. The outer periphery of the adjusting ring 240 is a first external gear section 242. The inner side of the 12 sets of positioning components is provided with adjusting gears 250. The adjusting gears 250 are annular and have a second external gear section and an internal gear section, as shown... Figure 2 As shown, the second external gear part of the adjusting gear 250 meshes with the first external gear part 242 of the adjusting ring 240. The adjusting gear 250 is provided with an adjusting drive mechanism that cooperates with the internal gear part. Specifically, the adjusting drive mechanism is a rotary drive motor, which drives the adjusting gear 250 to rotate clockwise or counterclockwise through gear transmission. By controlling the number of rotations of the rotary drive motor, the position of the positioning block 230 can be precisely controlled.

[0030] The positioning part 231 of the positioning block 230 is positioned at a height corresponding to the position of the reinforcing rib at the bottom of the can. The action parameters of the positioning component are adjusted according to the diameter detected by the diameter detection sensor. The positioning component has a released state and a positioned state. In the released state, the positioning part 231 is a certain distance away from the bottom of the can for easy detection. In the positioned state, all three positioning parts 231 are in contact with the reinforcing rib at the bottom of the can, aligning the can with the center and preventing the can from vibrating and tipping over during movement. The positions of the released and positioned states of the positioning component are adjusted accordingly for each diameter. Specifically, in the released state, the positioning part 231 is a certain distance away from the bottom of the can with the largest diameter.

[0031] After each station's inspection is completed, 12 positioning components move synchronously to the positioning state, aligning the can with the center. The inspection conveyor wheel 200 rotates 30° clockwise. After moving to the next station, the 12 positioning components move synchronously to the release state for inspection.

[0032] Through the structural design of the positioning components described above, the actions of the entire detection process are synchronized and consistent.

[0033] To achieve visualization of the entire detection process, the system provided in this application embodiment also includes a digital twin system. Specifically, the digital twin system includes a digital ID card allocation module, a data acquisition and transmission module, a twin modeling module, and a real-time mapping module; The digital ID card allocation module is connected to the input station and assigns a unique virtual digital ID card code to each can entering the input station. The data acquisition and transmission module is communicatively connected to the sensors, cameras, and drive mechanisms at each workstation in the detection and conveying device, and receives signals in real time. The twin modeling module is used to construct a three-dimensional twin model of each can with a unique virtual digital ID code based on the data from the data acquisition and transmission module, and to change the three-dimensional twin model in real time according to the detection signals transmitted by the sensors, cameras and drive mechanisms at each station in the detection and conveying device. The real-time mapping module is used to map and display the three-dimensional twin models of all the cans in the detection and conveying device, which are constructed by the twin modeling module, onto the three-dimensional structure of the detection and conveying device.

[0034] Specifically, the digital ID card allocation module assigns a unique virtual digital ID code to each can that falls into the input station. This unique virtual digital ID code is tracked from the time the can falls into the input station until the end of the inspection, ensuring that each can has a unique virtual digital ID code in the digital twin system. Specifically, in one embodiment of this application, the digital ID card allocation module has a built-in encoding algorithm that generates a format of "device number timestamp sequence number," where: the device number is used to distinguish different inspection devices (e.g., "ECI-01" for inspection device #1 on production line); the production batch is used to synchronize the production batch information of the cans (e.g., "20240501" represents the production batch on May 1, 2024); the timestamp is accurate to the second (e.g., "102359" represents 10:23:59 on May 1, 2024); and the sequence number is an incrementing sequence number starting from 0001 within each batch, ensuring unique coding within a single batch. When the can falls into input station A, the photoelectric sensor next to input wheel 100 triggers a signal, and the digital ID card allocation module generates a code, which then has a unique virtual digital ID card code in the digital twin system.

[0035] The data acquisition and transmission module communicates with the sensors, cameras, and drive mechanisms at each station in the detection and conveying device through the edge gateway and transmission module, and collects the data of equipment operation and the detection data of each station in real time.

[0036] The twin modeling module is used to construct a 3D twin model, which can be based on Unity3D. The 3D twin model is generated based on a unique virtual digital ID code generated by the digital ID allocation module, i.e., a 3D standard aluminum can model containing a unique virtual digital ID code. Simultaneously, the position data of this 3D standard aluminum can model is generated, and the 3D twin model is updated in real time based on data from the data acquisition and transmission module. Specifically, this includes, but is not limited to: automatically adjusting the model size by calling a preset aluminum can model template after receiving the height H and diameter D from the data acquisition and transmission module; updating the position data of the 3D standard aluminum can model in real time based on the drive signal of the drive mechanism of the detection conveyor wheel 200; and generating a red highlighted mark at the corresponding position of the virtual can based on the defect detection results, displaying defect information and dimensions next to the mark.

[0037] The real-time mapping module pre-stores a 3D schematic diagram of the detection and conveying device, and maps the 3D twin model constructed and modified in real time by the twin modeling module onto the 3D schematic diagram of the detection and conveying device.

[0038] The specific method for detecting empty bottles using the intelligent empty can detection device for the beverage can production line described above is as follows: Step 1: System Initialization and Template Configuration 1. Start the digital twin system, load the pre-stored 3D model of the detection and conveying device, and complete the initial alignment between the virtual model and the physical device by using the "three-point alignment method" (selecting the center of the chassis, the center of the input wheel, and the center of the output wheel as reference points); 2. The data acquisition and transmission module establishes communication connections with all sensors and drive mechanisms, and the edge gateway completes parameter configuration (such as sampling frequency and transmission protocol). 3. The digital ID card allocation module initializes the encoding rules and synchronizes the production batch information of the cans to be tested (e.g., "20240501"). Step 2: Input the coding binding and specification modeling of workstation A Batch specifications are obtained by using height and diameter detection sensors, and dynamic standard templates are generated by combining self-learning function. Generative Adversarial Network (GAN) is introduced to enhance defect samples, providing a benchmark for image processing in subsequent workstations. 1.1. 3D Twin Model Construction and Digital Encoding Binding: When the aluminum can is transported to input station A by input wheel 100, the photoelectric sensor triggers a signal, and the digital ID card allocation module generates an code (e.g., "ECI-20240501-102359-0001"). The 3D twin model then generates a 3D standard aluminum can model with a unique virtual digital ID card code, and simultaneously generates its position data. 2. Batch Specification Inspection: The height H and diameter D of the cans entering the input station A are inspected using height and diameter sensors. The height is detected using image recognition technology, while the diameter is detected by calculating the can diameter through an infrared through-beam sensor array and edge recognition. 3. Dynamic standard template self-learning: Triggering conditions: When the production line starts, batches are switched, or manual triggering occurs, the system enters "self-learning mode" and continuously checks the first 10 confirmed qualified tanks (which can be filtered by previous manual re-inspection or high-confidence automatic judgment results). ① Template generation process: Simultaneously acquire images of 10 qualified tanks at each inspection station (tank opening, inner wall, inner bottom, outer wall, outer bottom), and record the original image data; preprocess each group of images: denoise by Gaussian filtering, normalize grayscale (compress to the range of 0-255), and complete image alignment based on the center of the tank opening and the bottom reinforcing rib (to eliminate individual positional deviations); use a pixel-level mean fusion algorithm to generate a "dynamic standard template" for this batch (e.g., the tank opening template is the average grayscale distribution of 10 qualified tank opening images, and the inner bottom template contains the average reinforcing rib shadow features), and the template is automatically bound to the height (H) and diameter (D) parameters of the current batch; 3D twin model correction: The 3D twin model is automatically corrected based on the tank height H and diameter D; ① 4. GAN Defect Sample Enhancement: A generative adversarial network (GAN) is trained based on historical defect samples (covering at least 8 types of defects such as can mouth damage and inner wall wrinkles, with ≥50 real samples for each type); the generative network takes qualified can images as input and synthesizes images with simulated defects (such as scratches and deformations of different shapes); the discriminative network optimizes the generation effect so that the texture and lighting characteristics of the synthesized defects are highly consistent with the real defects; the GAN synthesized samples (≥500 per type) are mixed with real samples to train the defect segmentation model for subsequent workstations; 5. Parameter adaptation: Based on the height H and diameter D, match the image ROI parameters of subsequent inspection stations (such as the ROI height of the tank opening inspection is adjusted proportionally with H) and the positioning component action parameters from the specification database; the positioning component adjusts the position of the positioning block according to the diameter D, maintaining the positioning state (fitting against the bottom reinforcing rib of the tank) during the movement, and switching to the released state during inspection (maintaining a certain distance from the bottom of the tank). Step 3: Tank opening inspection station B (defect detection of tank opening edge and inner edge) 1. Image acquisition: Drive the light shield to descend, activate the top coaxial white ring light source, and use a 2-megapixel camera (12mm lens) to vertically capture images of the tank opening area; 2. Defect detection: ① Image alignment: Align the acquired image with the dynamic standard template by ORB feature point matching (extracting no less than 500 feature points), and after correction, the rotation deviation is ≤0.5° and the translation deviation is ≤1 pixel; ② Defect segmentation: Input the aligned image into the pre-trained UNet defect segmentation model (the training set contains real and GAN synthetic samples), and output a binarized segmentation mask (the defect region is 1 and the background is 0). ③ Feature measurement: Perform connected component analysis on the segmentation mask to extract the area of ​​the defect region (accuracy ≤ 0.01 mm²). ④ Judgment: If a defect area ≥ 0.5 mm² exists, it is judged as "damaged edge of can opening" or "foreign object / deformation on the inner edge" (unacceptable, NG); otherwise, it is judged as acceptable (OK). The result will be sent to the data acquisition and transmission module, which will then forward it synchronously to the twin modeling module and the real-time mapping module: the twin modeling module will add a green border (marking "can opening acceptable") or a red highlighted defect mark (indicating can opening defect information) to the virtual can opening area; Step 4: Tank Inner Wall Inspection Station C (Inner Wall Wrinkles and Adhered Foreign Matter Detection) 1. Image Acquisition: Drive the light shield to descend until it completely covers the can, isolating it from external light source interference; move the ring LED light source to a preset position above the can opening, and the light enters the inner wall of the can at an angle, illuminating the curved surface of the inner wall evenly with the help of diffuse reflection; control 4-6 miniature cameras (with wide-angle lenses) evenly arranged around the circumference to shoot simultaneously: the lens axis is at an angle of 30°~45° with the axis of the can, to obtain multi-view images of the inner wall; 2. Defect detection algorithm: Image preprocessing: ① Multi-view image stitching: Images taken by 4-6 cameras are stitched together using the SIFT feature matching algorithm (stitching error ≤ 1 pixel) to merge the images of each region into a complete 360° unfolded inner wall image; ② Distortion correction: Image distortion caused by lens angle deviation is eliminated through perspective transformation. The image ratio is corrected based on the dynamic inner wall template generated by self-learning in this batch (ensuring that the size of the unfolded inner wall surface is consistent with the actual size); ③ Illumination equalization: Adaptive histogram equalization is used to eliminate the difference in brightness caused by uneven diffuse reflection and improve image contrast; Defect region segmentation: Input the corrected unfolded image into the pre-trained UNet defect segmentation model (the training set contains real inner wall defect samples and simulated scratches and wrinkles synthesized by GAN), and output a binary segmentation mask (in the mask, "1" represents the defect candidate region and "0" represents the normal region). Defect feature measurement: Connectivity analysis is performed on the segmentation mask, and feature parameters of the defect candidate region are extracted through the region attribute measurement algorithm: Scratches (long strip dark areas) are measured for their area (converted to actual size, accuracy ≤0.01mm²) and aspect ratio (ratio of major axis to minor axis); Wrinkles (continuous alternating light and dark stripes) are measured for their area, continuous length and stripe spacing. Acceptance judgment: The judgment logic based on dynamic standard template is invoked to compare the feature parameters with preset thresholds: If there is a long strip area (scratches) with an area ≥ 0.3 mm² and an aspect ratio ≥ 10:1, or a stripe area (wrinkles) with an area ≥ 1 mm² and a continuous length ≥ 5 mm, it is judged as unacceptable (NG); if the feature parameters of all defect candidate areas do not exceed the threshold, it is judged as acceptable (OK); the twin modeling module will add a green border (marking "tank inner wall acceptable") or a red highlighted defect mark (annotating the tank inner wall defect information) to the virtual tank inner wall area (perspective). Step 5: Inspection station D inside the tank (inner bottom deformation and foreign object detection) 1. Image acquisition: Drive the light shield to move downwards, activate 4 sets of 30° angled LED light sources distributed in a ring (to create fixed shadows on the reinforcing ribs), and use a 5-megapixel camera (16mm lens) to take an image of the inner sole from above; 2. Defect detection: ① Image alignment: Aligned with the dynamic insole template using the phase correlation method, with a rotation error ≤0.5°; ② Defect segmentation: The UNet model segments deformed regions (grayscale anomalies) and foreign objects (dark areas); ③ Feature measurement: Extract the area of ​​the defect region and the aspect ratio of the circumscribed rectangle; ④ Judgment: If there is an area ≥ 2mm² and aspect ratio ≤ 3:1 (deformation), or a dark area ≥ 0.5mm² (foreign object), it is judged as NG; otherwise it is OK; the twin modeling module will add a green border (marked as "qualified") or a red highlighted defect mark (marking the defect information at the bottom of the tank) to the bottom area of ​​the virtual tank (perspective). Step Six: Tank Outer Wall Inspection Station E / F (Outer Wall Contour Defect Inspection) 1. Image acquisition: Drive the light shield to move downwards, station E starts the front and rear planar diffuse light source, station F starts the left and right planar diffuse light source, and use a 2-megapixel camera (8mm lens) to capture images of the four-way outer wall. 2. Defect detection: ① Image stitching: Stitch the four-way images into a 360° unfolded view of the outer wall, and align it with the dynamic outer wall template; ② Defect segmentation: UNet model segments concave areas (contour abrupt changes) and large-area deformations (regional grayscale anomalies); ③ Feature measurement: Calculate the distance from the contour to the standard template and the area difference between the actual contour and the standard contour; ④ Judgment: If there are 3 consecutive pixels with a distance ≥ 0.3mm (depression), or an area difference ≥ 5mm² and a distribution range ≥ 10mm (large area deformation), it is judged as NG; otherwise, it is OK; the twin modeling module will add a green border (marking "tank outer wall qualified") or a red highlighted defect mark (marking tank outer wall defect information) to the outer wall of the virtual tank. Step 7: Tank exterior wall inspection station G (exterior wall minor scratch inspection) 1. Image acquisition: Drive the light shield to move downwards, activate four sets of 60° high-angle strip oblique light sources, and use a 2-megapixel camera to capture images of the four-way outer wall; 2. Defect detection: ① Image stitching: Four-way image stitching based on SIFT feature matching, with an error ≤ 1 pixel; ② Shadow enhancement: High-frequency components are enhanced using the Laplacian operator, and contrast is improved with Gamma correction (γ=0.8); ③ Defect segmentation: The UNet model segments long strip-shaped scratch shadows; ④ Feature measurement: Extract the aspect ratio and area of ​​the scratch; ⑤ Judgment: If there are axial / near axial scratches with an aspect ratio ≥10:1 and an area ≥0.5mm², they are judged as NG; otherwise, they are OK. The twin modeling module will add a green border (marking "Tank outer wall qualified") or a red highlighted defect mark (marking the tank outer wall defect information; if there is a "Tank outer wall qualified" mark, it will be deleted). Step 8: Inspection station H on the outer side of the tank bottom (outer bottom scratch inspection) 1. Image Acquisition: Drive the light shield to move downwards, activate the bottom plane diffuse light source (transmitted from below the chassis), and use a 2-megapixel camera (12mm lens) to take an upward-facing image of the outer sole; 2. Defect detection: ① Geometric correction: The outsole image is corrected to a perfect circle by perspective transformation (based on the center of the positioning component); ② Defect segmentation: UNet model segments scratches and dark lines (Top-Hat transform enhances contrast); ③ Feature measurement: Extract the length and width of the scratch; ④ Judgment: If the total length of the cumulative trace is ≥10mm or the length of a single trace is ≥8mm (width ≥0.1mm), it is judged as NG; otherwise, it is OK; the twin modeling module will add a green border (marking "bottom of tank is qualified") or a red highlighted defect mark (marking the defect information on the bottom of tank) to the outside of the virtual tank bottom; if there is a "bottom of tank is qualified" mark, it will be deleted. Step Nine: Inspection Station I on the outer side of the tank bottom (outer bottom deformation inspection) 1. Image acquisition: Drive the light shield to move downwards, activate four sets of 30° oblique LED light sources (obliquely shining from the bottom edge of the can), and use a 5-megapixel camera (16mm lens) to take an image of the outer bottom from an upward angle; 2. Defect detection: Image alignment: Aligned with the dynamic outsole template via ORB feature point matching; Defect segmentation: UNet model segments deformed regions (differential grayscale anomalies); Feature measurement: Calculate the deviation and area of ​​the circumcircle radius of the defect region from the standard radius; Judgment: If there is an area (protrusion / depression) with a radius deviation ≥0.3mm and an area ≥3mm², it is judged as NG; otherwise, it is OK; the twin modeling module will add a green border (marked as "bottom of tank is qualified") or a red highlighted defect mark (marking the defect information on the bottom of tank) to the outside of the virtual tank bottom; if there is a "bottom of tank is qualified" mark, it will be deleted. Step 10: Output the results for station J: The system aggregates the results from all workstations, and the empty can only proceed to the next production stage when all workstations have determined the result to be "OK". The dynamic standard template and GAN model parameters are updated regularly and regenerated after each power-on restart to ensure long-term detection accuracy and avoid the impact of changes in light source brightness or camera exposure at each workstation on the detection.

[0039] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as ROM / RAM, disk, optical disk, etc.

[0040] The above description discloses only preferred embodiments of the present invention and should not be construed as limiting the scope of the present invention. Therefore, equivalent variations made in accordance with the claims of the present invention are still within the scope of the present invention.

Claims

1. An intelligent empty can detection device for a beverage can production line, characterized in that, The device includes a detection conveying device, which has the following components sequentially along the conveying direction: Enter the workstation; The can opening inspection station is equipped with a light shield and a light source and camera set inside the light shield to capture images of the can opening area for detecting can opening defects. The tank inner wall inspection station is equipped with a light shield and a light source and camera set inside the light shield to capture images of the tank's circumferential inner wall area for detecting defects in the tank's inner wall. The bottom inspection station inside the tank is equipped with a light shield and a light source and camera set inside the light shield to take images of the bottom area inside the tank for detecting defects inside the tank. The first inspection station on the outer wall of the tank is equipped with a light shield and a light source and camera set inside the light shield. The light source is planar diffused light, which takes images of the area of ​​the inner wall of the tank around the circumference of the tank from two opposite sides of the tank, and is used to detect defects on the outer wall of the tank. The second inspection station on the outer wall of the tank is equipped with a light shield and a light source and camera set inside the light shield. The light source is planar diffused light, which takes images of the inner wall area around the tank from the other two opposite sides of the tank, and is used to detect defects on the outer wall of the tank. The third inspection station on the outer wall of the tank is equipped with a light shield and a light source and camera set inside the light shield. The light source consists of four sets of high-angle strip oblique light sources distributed in a square direction, which take images of the tank's circumferential inner wall area from all four sides of the tank, and are used to detect defects on the outer wall of the tank. The first inspection station on the outer side of the tank bottom is equipped with a light shield. A light source and a camera are located below the tank. The light source is planar diffused light, which takes an image of the outer area of ​​the tank bottom to detect defects on the outer side of the tank bottom. The second inspection station on the outer side of the tank bottom is equipped with a light shield. A light source and a camera are located below the tank. The light source consists of four sets of obliquely distributed LED light sources that capture images of the outer area of ​​the tank bottom to detect defects on the outer side of the tank bottom. Output station.

2. The intelligent empty can detection device for a beverage can production line according to claim 1, characterized in that: The detection conveying device is a disc-shaped detection conveying wheel, which includes a chassis. The input station, tank opening detection station, tank inner wall detection station, tank inner bottom detection station, tank outer wall first detection station, tank outer wall second detection station, tank outer wall third detection station, tank bottom outer first detection station, tank bottom outer second detection station, and output station are arranged sequentially at 30° intervals around the chassis, and two wheel gaps are formed between the output station and the input station.

3. The intelligent empty can detection device for a beverage can production line according to claim 2, characterized in that: The chassis is fixed with twelve sets of positioning components arranged in a ring. The positioning components have a released state and a positioned state. During the detection process, the twelve sets are simultaneously in the released state, and the positioning components are a certain distance away from the bottom of the can. During the movement, the twelve sets are simultaneously in the positioned state, and the positioning components are in contact with the bottom of the can.

4. The intelligent empty can detection device for a beverage can production line according to claim 3, characterized in that: The positioning assembly includes a positioning mounting ring fixedly connected to the chassis, an adjusting ring disposed on the positioning mounting ring, and a positioning block disposed between the positioning mounting ring and the adjusting ring. The adjusting ring is fitted with the positioning mounting ring in an inner and outer sleeve for limiting and relative rotation. A pressure cap is fixedly connected to the chassis, which is positioned above the adjusting ring to limit its movement. Several positioning blocks are provided and evenly distributed, including a positioning part, a guide part, a T-shaped limiting groove, and an adjusting rod. The positioning part is located at the inner end in a relatively radial direction, the guide part is inclined and located at the upper end of the positioning part, and the T-shaped limiting groove cooperates with the T-shaped guide strip on the positioning mounting ring. The lower end of the adjusting ring is provided with a corresponding number of equally distributed variable diameter arc-shaped adjusting grooves, and the adjusting rod extends into the variable diameter arc-shaped adjusting grooves.

5. The intelligent empty can detection device for a beverage can production line according to claim 4, characterized in that: The outer periphery of the adjusting ring is a first external gear section, and the inner side of the twelve positioning components is provided with an adjusting gear. The adjusting gear is provided with a second external gear section. The second external gear section of the adjusting gear meshes with the first external gear section of the adjusting ring. The adjusting gear is connected to an adjusting drive mechanism for driving the adjusting gear to rotate.

6. The intelligent empty can detection device for a beverage can production line according to claim 4, characterized in that: The pressure cap is disc-shaped, and its outer periphery extends above the adjustment ring of the twelve positioning components.

7. The intelligent empty can detection device for a beverage can production line according to claim 4, characterized in that: The input station is equipped with a diameter detection sensor; the position parameters of the positioning component in the loose state and the positioning state are set according to the diameter signal detected by the diameter detection sensor.

8. The intelligent empty can detection device for a beverage can production line according to claim 3, characterized in that: It includes an input wheel, an output wheel, and a limiting wall; the input wheel, the detection conveyor wheel, and the output wheel are arranged in an isosceles triangle; the input wheel has several input arc grooves evenly distributed around its circumference, the input station is adjacent to the input wheel, and the input wheel is higher than the detection conveyor wheel; the output wheel has several output arc grooves evenly distributed around its circumference; the output station is adjacent to the output wheel, the output wheel is higher than the detection conveyor wheel, and a push-out mechanism is provided at the bottom of the output station.

9. The intelligent empty can detection device for a beverage can production line according to claim 1, characterized in that: This also includes digital twin systems. The digital twin system includes a digital ID card allocation module, a data acquisition and transmission module, a twin modeling module, and a real-time mapping module; The digital ID card allocation module is connected to the input station and assigns a unique virtual digital ID card code to each can entering the input station. The data acquisition and transmission module is communicatively connected to the sensors, cameras, and drive mechanisms at each workstation in the detection and conveying device, and receives signals in real time. The twin modeling module is used to construct a three-dimensional twin model of each can with a unique virtual digital ID code based on the data from the data acquisition and transmission module, and to change the three-dimensional twin model in real time according to the signals transmitted by the sensors, cameras and drive mechanisms at each station in the detection and conveying device. The real-time mapping module is used to map and display the three-dimensional twin models of all the cans in the detection and conveying device, which are constructed by the twin modeling module, onto the three-dimensional structure of the detection and conveying device.

10. An empty can detection method using an intelligent empty can detection device for a beverage can production line as described in any one of claims 1-9.

Images