Automatic green heart broad bean peeling and color sorting machine system

Abstract

This invention relates to the field of agricultural product peeling technology, and discloses an automatic peeling and color sorting machine system for green-heart broad beans. The system includes a frame and, sequentially arranged on the frame, a feeding mechanism, a primary coarse peeling mechanism, a visual inspection mechanism, and a secondary precision peeling mechanism. The primary coarse peeling mechanism performs an initial peeling process to remove most of the outer skin. The visual inspection mechanism collects image information of the broad beans after the initial peeling. The secondary precision peeling mechanism controls the execution components based on the image information collected by the visual inspection mechanism to target and remove residual inner skin at specific locations. The primary coarse peeling removes only most of the outer skin while retaining some inner skin to protect the green heart and cotyledons. The secondary precision peeling, relying on the image data from the visual inspection, uses a multi-degree-of-freedom micro-manipulator to target and remove residual inner skin, ensuring that the residual inner skin rate is minimized while precisely avoiding the green heart area, effectively improving the integrity of the green-heart broad beans and the quality of the finished product.

Description

Technical Field

[0001] This invention relates to the field of agricultural product peeling technology, specifically to an automatic peeling and color sorting system for green-hearted broad beans. Background Technology

[0002] Currently, in the field of green broad bean peeling and processing, traditional equipment mostly adopts a single peeling mode, mainly divided into two core structures: friction type and rolling type. These are widely used in large-scale broad bean processing production lines, and their core design concept is to remove the outer skin by directly applying mechanical force to the surface of the broad beans. The advantages of this type of equipment are its simple structure and low manufacturing cost, which can meet basic peeling needs. However, as the market's requirements for the quality of green broad beans continue to increase, its inherent design flaws have gradually become apparent, making it unable to adapt to the processing demands for high quality and high integrity.

[0003] Traditional single-peeling methods suffer from difficulty in controlling peeling precision and cannot achieve graded processing of the outer and inner skins. Because the outer skin of green-hearted broad beans is relatively hard and the inner skin is tightly adhered to the green cotyledons, if the mechanical force is set too high, it will directly damage the inner green cotyledons, causing the beans to break, darken in color, and significantly reduce the quality of the finished product. If the mechanical force is set too low, the outer skin cannot be completely removed, and some inner skin will remain on the surface of the beans, requiring secondary manual sorting, increasing labor costs, and affecting the product's appearance and taste. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an automatic peeling and color sorting system for green broad beans, which solves the problem that traditional single peeling modes have difficulty in controlling peeling precision, resulting in broken beans or failure to remove the outer skin.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an automatic peeling and color sorting machine system for green broad beans, comprising a frame and, sequentially arranged on the frame, a feeding mechanism, a primary coarse peeling mechanism, a visual inspection mechanism, a secondary precision peeling mechanism, an air separation mechanism, a color sorting and grading mechanism, and a discharge collection mechanism. The primary coarse peeling mechanism is used to perform initial peeling of the broad beans conveyed by the feeding mechanism to remove most of the outer skin and expose the inner skin. The visual inspection mechanism is installed downstream of the primary coarse peeling mechanism to collect image information of the broad beans after the initial peeling. The secondary precision peeling mechanism... The mechanism is located downstream of the visual inspection mechanism and is communicatively connected to it. The secondary precision peeling mechanism controls the execution component to target and remove residual inner skin at specific locations based on the image information collected by the visual inspection mechanism. The air separation mechanism is located below the secondary precision peeling mechanism and uses density difference to separate the detached skin from the bean paste. The color sorting and grading mechanism is used to perform final quality sorting on the bean paste after air separation. The feeding mechanism, primary coarse peeling mechanism, visual inspection mechanism, secondary precision peeling mechanism, air separation mechanism, and color sorting and grading mechanism are all electrically connected to the central control system.

[0006] Preferably, the visual inspection mechanism includes an inspection box, an image acquisition module and a light source illumination module installed inside the inspection box. The image acquisition module includes a multispectral linear array camera, which contains a visible light imaging unit and a near-infrared imaging unit. The visible light imaging unit is used to acquire the skin color and texture information of the broad bean surface, and the near-infrared imaging unit is used to penetrate the surface of the broad bean to acquire internal component information to identify the distribution of green cotyledon areas and residual endothelial cells. The light source illumination module includes a diffused light cover arranged around the transmission path, and the light cover contains a specific wavelength light source corresponding to the near-infrared imaging unit band.

[0007] Preferably, the secondary precision peeling mechanism includes a support, a multi-degree-of-freedom targeting execution component mounted on the support, and a position calibration conveyor belt. The multi-degree-of-freedom targeting execution component includes several sets of micro-manipulators arranged along the conveying direction. Each set of micro-manipulators has a high-pressure jet nozzle or a flexible grinding head at its end. The micro-manipulators are connected to the central control system and receive coordinate data generated by the vision inspection mechanism. The central control system drives the micro-manipulators to adjust the spatial angle and force of the end tool according to the coordinate data, so as to physically remove or air-jet peel off the inner skin of a specific location on a single broad bean.

[0008] Preferably, the air separation mechanism includes a vibrating feeder, a vertical air duct, a variable frequency fan, and a settling separation chamber. The vibrating feeder evenly feeds the material processed by the two-stage precision peeling mechanism into the bottom of the vertical air duct. The variable frequency fan is installed at the top or side of the vertical air duct and generates an upward vertical airflow. The airflow speed is adjustable from five meters per second to ten meters per second. The settling separation chamber is provided with a light impurity outlet and a broad bean outlet. By utilizing the difference in density between the broad bean shell and the broad bean, the shell is carried into the light impurity outlet by the airflow, while the broad bean falls into the broad bean outlet.

[0009] Preferably, the color sorting and grading mechanism includes a color sorting box, a full-spectrum color camera installed inside the color sorting box, a background plate, and a multi-stage rejection actuator. The full-spectrum color camera acquires images of both the front and back sides of individual broad beans after air separation. The multi-stage rejection actuator includes at least two sets of pneumatic spray valves arranged sequentially along the material conveying direction. The opening and closing frequency and duration of the pneumatic spray valves are controlled by the central control system according to the sorting grade, sorting the broad beans into whole green-heart retained products, green-heart superior products, ordinary products, and rejected products.

[0010] Preferably, the central control system integrates a dynamic parameter optimization module, which stores a quality prediction model based on machine learning. The central control system acquires the overall quality data of the current batch of raw materials provided by the visual inspection agency and the green heart ratio parameter required by the customer order in real time. The quality prediction model calculates the optimal sorting threshold based on the input overall quality data and sends the instruction to the color sorting and grading agency and the secondary precision peeling agency to dynamically adjust the peeling intensity of the secondary precision peeling agency and the rejection criteria of the color sorting and grading agency.

[0011] Preferably, the primary coarse peeling mechanism includes a friction peeling roller and an elastic support roller arranged in relative rotation. The surface of the friction peeling roller is covered with a layer of diamond abrasive, and the surface of the elastic support roller is covered with a rubber layer. The gap between the friction peeling roller and the elastic support roller is controlled by an electric adjustment mechanism. The electric adjustment mechanism automatically fine-tunes the gap size based on the peeling cleanliness data of the previous batch of materials fed back by the visual inspection mechanism, so that the primary coarse peeling mechanism removes most of the outer skin of the broad beans and retains some of the inner skin to protect the internal green cotyledons.

[0012] Preferably, the light source illumination module is equipped with a strobe control circuit, which is synchronously triggered with the line scanning signal of the multispectral line array camera. The light emitted by the light source illumination module is diffused through a diffuser to form a uniform surface light source that illuminates the detection area. The light source illumination module also includes a polarizing filter group for eliminating reflections on the surface of the broad beans. The polarizing filter group is installed in front of the light source emitting end and the receiving lens of the multispectral line array camera.

[0013] Preferably, a material tracking encoder is provided between the visual inspection mechanism and the secondary precision peeling mechanism. The material tracking encoder is connected to the conveyor belt drive shaft and records the displacement of the conveyor belt in real time. The central control system uses the displacement data of the material tracking encoder to synchronize the position of the single broad bean feature identified by the visual inspection mechanism with the corresponding actuator in the secondary precision peeling mechanism in time and space, so as to ensure that the actuator accurately acts on the visually identified residual inner skin area.

[0014] Preferably, the dynamic parameter optimization module is equipped with a historical database and a self-learning unit. The historical database is used to store the peeling characteristic parameters and corresponding optimal process parameters of broad bean raw materials from different origins and seasons. The self-learning unit periodically analyzes the final product qualification rate and rework rate data fed back by the color sorting and grading institution. When the rework rate exceeds the preset value, the self-learning unit uses the backpropagation algorithm to correct the weight parameters of the quality prediction model, so that the system can adapt to the slow changes in raw material characteristics.

[0015] This invention provides an automatic peeling and color sorting system for green-fleshed broad beans. It has the following beneficial effects:

[0016] 1. This invention uses a first-stage coarse peeling process to remove most of the outer skin while retaining some of the inner skin to protect the green heart cotyledons. The second-stage precise peeling process relies on image data from visual detection and uses a multi-degree-of-freedom micro-robotic arm to target and remove the remaining inner skin. This ensures that the residual inner skin rate is minimized and that the green heart area is precisely avoided, effectively improving the integrity of the green heart soybeans and the quality of the finished product.

[0017] 2. The visual inspection mechanism of this invention integrates visible light and near-infrared imaging units, enabling it not only to acquire the color and texture of the broad bean surface but also to penetrate the surface layer for non-destructive testing of the internal green core distribution and residual inner skin. Combined with polarized filtering and synchronous stroboscopic illumination technology, it effectively eliminates surface reflection interference, providing a reliable and abundant image data foundation for subsequent precise peeling.

[0018] 3. The primary coarse peeling mechanism of this invention does not operate with fixed parameters. The gap between its peeling rollers can be automatically fine-tuned through an electric adjustment mechanism based on the peeling cleanliness data of the previous batch fed back by the vision system. This closed-loop control based on effect feedback ensures that the coarse peeling stage can stably achieve the removal of most of the outer skin while retaining some of the inner skin to protect the cotyledons.

[0019] 4. This invention can automatically and in real time adjust key parameters such as the primary peeling gap, secondary peeling intensity, and color sorting threshold according to the current raw material characteristics and order requirements, enabling the entire machine to have self-adaptive capabilities. By setting a material tracking encoder between the vision system and the actuator and establishing a synchronous matching algorithm, the system can accurately correlate the feature position of each broad bean identified by the vision system with the moving secondary peeling actuator in time and space. Attached Figure Description

[0020] Figure 1 This is a system architecture diagram of the present invention. Detailed Implementation

[0021] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Please see the appendix Figure 1 This invention provides an automatic peeling and color sorting system for green broad beans, including a frame and, sequentially arranged on the frame, a feeding mechanism, a primary coarse peeling mechanism, a vision inspection mechanism, a secondary precision peeling mechanism, an air separation mechanism, a color sorting and grading mechanism, and a discharge collection mechanism. The primary coarse peeling mechanism performs initial peeling on the broad beans conveyed by the feeding mechanism to remove most of the outer skin and expose the inner skin. The vision inspection mechanism is installed downstream of the primary coarse peeling mechanism to collect image information of the broad beans after the initial peeling. The secondary precision peeling mechanism is located downstream of the vision inspection mechanism. Downstream of the inspection unit and connected to the visual inspection unit, the secondary precision peeling unit controls the execution components to target and remove residual inner skin at specific locations based on the image information collected by the visual inspection unit. The air separation unit is located below the secondary precision peeling unit and uses density difference to separate the detached skin from the bean paste. The color sorting and grading unit is used to perform final quality sorting on the bean paste after air separation. The feeding unit, primary coarse peeling unit, visual inspection unit, secondary precision peeling unit, air separation unit, and color sorting and grading unit are all electrically connected to the central control system.

[0023] Specifically, the feeding mechanism uses a spiral feeding method, with a variable frequency motor driving the spiral blades to rotate, uniformly and stably conveying the green-hearted broad beans to be processed to the primary coarse peeling mechanism. The feeding speed can be dynamically adjusted according to the processing capacity of subsequent processes to avoid congestion caused by excessive feeding or idle running of the equipment due to insufficient feeding. The core function of the primary coarse peeling mechanism is to perform the initial peeling of the broad beans conveyed by the feeding mechanism. Its design goal is to remove most of the outer skin of the broad beans while exposing only the inner skin, avoiding direct damage to the green cotyledons inside, thus laying the foundation for subsequent precise peeling. The visual inspection mechanism is installed downstream of the primary coarse peeling mechanism, precisely docking with the discharge end of the primary coarse peeling mechanism. This ensures that every broad bean after the initial peeling enters the inspection area. This mechanism collects image information of the broad beans in real time, providing accurate data support for the secondary precise peeling. The secondary precision peeling mechanism is located downstream of the vision inspection mechanism. It establishes a communication connection with the vision inspection mechanism via industrial Ethernet, enabling it to receive image analysis data transmitted by the vision inspection mechanism in real time. Based on this data, it precisely controls the actuators to target and remove residual inner skin from specific locations on individual broad beans. The air separation mechanism is located below the secondary precision peeling mechanism, with its inlet corresponding to the outlet. Utilizing the density difference between the broad bean skin and the bean kernel, it achieves efficient separation of the skin and kernel through airflow, removing impurities generated during the peeling process. The color sorting and grading mechanism performs final quality sorting on the bean kernels after air separation, classifying them into different grades based on indicators such as the proportion of green heart, color, and integrity to meet the needs of different customers.

[0024] The visual inspection mechanism includes an inspection box, an image acquisition module installed inside the inspection box, and a light source illumination module. The image acquisition module includes a multispectral linear array camera, which contains a visible light imaging unit and a near-infrared imaging unit. The visible light imaging unit is used to acquire information on the skin color and texture of the broad bean surface, while the near-infrared imaging unit is used to penetrate the surface of the broad bean to acquire information on the internal components in order to identify the distribution of the green cotyledon area and the residual endothelial layer. The light source illumination module includes a diffused light cover arranged around the conveying path, and the light cover contains a specific wavelength light source corresponding to the band of the near-infrared imaging unit.

[0025] Specifically, the visual inspection mechanism includes an inspection chamber, an image acquisition module, and a light source module. The inspection chamber adopts a closed structure with an internal light-shielding layer to prevent external light from interfering with the inspection accuracy. Transparent protective plates are installed at the inlet and outlet ends of the inspection chamber to both ensure material transport and prevent dust from entering the chamber and damaging the equipment. The image acquisition module is fixedly installed at the top inside the inspection chamber and uses a multispectral linear array camera. This camera integrates a visible light imaging unit and a near-infrared imaging unit, which work synchronously and collaboratively to ensure data integrity and synchronization. The visible light imaging unit uses a CMOS sensor to accurately acquire the color and texture information of the broad bean's surface, identifying residual areas of the skin through texture differences and distinguishing the boundary between the inner skin and cotyledons through color differences. The near-infrared imaging unit uses an InGaAs sensor, which can penetrate the surface tissue of the broad bean to acquire internal component information. Utilizing the difference in absorption coefficients between the green cotyledons and the inner skin in the near-infrared band, it accurately identifies the location and size of the green cotyledon area, as well as the distribution and thickness of residual inner skin. The image processing algorithm of the image acquisition module adopts a machine vision-based multi-feature fusion recognition algorithm. First, the acquired visible light and near-infrared images are preprocessed. Gaussian filtering removes image noise, and histogram equalization enhances image contrast. Then, image registration accurately aligns the two images. Subsequently, color, texture, and grayscale features are extracted, and a support vector machine (SVM) classifier is used to identify and classify these features. Finally, the coordinates, area, and relevant parameters of the residual inner skin and the green heart region are output, providing accurate positioning data for secondary precise peeling. The light source illumination module is fixedly installed inside the detection chamber, arranged in a ring around the material conveying path. It includes a diffused light cover and a specific wavelength light source. The diffused light cover, made of milky white acrylic, has an arc-shaped structure that evenly scatters the light emitted by the light source, forming a uniform surface light source that illuminates the detection area, avoiding direct light that could cause reflections on the broad bean surface and affect image acquisition accuracy. The light source inside the lampshade is a near-infrared LED light source, whose wavelength matches the working wavelength of the near-infrared imaging unit, ensuring that the near-infrared imaging unit can efficiently collect information inside the broad bean. At the same time, it is equipped with a visible light LED light source to supplement visible light illumination, ensuring that the image collected by the visible light imaging unit is clear and detailed.

[0026] The secondary precision peeling mechanism includes a support frame, a multi-degree-of-freedom targeting execution component mounted on the support frame, and a position calibration conveyor belt. The multi-degree-of-freedom targeting execution component includes several sets of micro-manipulators arranged along the conveying direction. Each set of micro-manipulators has a high-pressure jet nozzle or a flexible grinding head at its end. The micro-manipulators are connected to the central control system and receive coordinate data generated by the vision inspection mechanism. The central control system drives the micro-manipulators to adjust the spatial angle and force of the end tools according to the coordinate data, so as to physically remove or air-jet peel off the inner skin of a specific location on a single broad bean.

[0027] Specifically, the secondary precision peeling mechanism includes a support frame, a multi-degree-of-freedom targeting actuator, and a position calibration conveyor belt. The support frame is fixedly mounted on the machine frame, providing stable support for the entire mechanism. The position calibration conveyor belt is fixedly located in the middle of the support frame, extending along the material conveying direction. Its surface is equipped with anti-slip protrusions, and it adopts variable frequency speed control, which can dynamically adjust the running speed of the conveyor belt according to the detection speed of the vision inspection mechanism and the action speed of the targeting actuator, ensuring stable material conveying on the conveyor belt without deviation. The multi-degree-of-freedom targeting actuator is fixedly mounted on the top of the support frame, directly above the position calibration conveyor belt. It includes several sets of micro-manipulators evenly arranged along the conveying direction. Each set of micro-manipulators works independently without interference. Their number matches the material arrangement density on the position calibration conveyor belt, ensuring that each broad bean corresponds to a set of micro-manipulators. Each set of micro-manipulators adopts a 3-axis linkage structure, driven by a servo motor, and has high-precision position adjustment capabilities, enabling flexible adjustment of three-dimensional angles in space. The end effector of the miniature robotic arm can be equipped with either a high-pressure jet nozzle or a flexible grinding head, depending on the actual peeling requirements. The high-pressure jet nozzle is suitable for removing thinner, less adhesive residual inner skin, while the flexible grinding head is suitable for removing thicker, more adhesive residual inner skin. The two end effectors can be automatically switched based on the residual inner skin thickness parameters output by the vision inspection mechanism. The miniature robotic arm establishes a communication connection with the central control system via a bus, receiving in real time the coordinate data of the residual endothelial layer and the parameters of the green heart region generated by the vision inspection mechanism. The central control system has a built-in motion control algorithm, which uses a PID control algorithm to adjust the movement speed and force of the miniature robotic arm. Based on the coordinate data of the residual endothelial layer, the micro-robotic arm is driven to quickly adjust the spatial angle and position of the end effector, so that the end effector is precisely aligned with the residual endothelial layer area. Then, a high-pressure airflow is ejected through a high-pressure nozzle, and the impact force of the airflow is used to peel off the residual endothelial layer, or the residual endothelial layer is gently ground away by the low-speed rotation of the flexible grinding head. Throughout the peeling process, the central control system monitors the force of the miniature robotic arm in real time, and avoids the green heart cotyledon area based on the position parameters of the green heart region, ensuring that the green heart cotyledon is not damaged, while ensuring that the residual endothelial layer is completely removed.

[0028] The air separation mechanism includes a vibrating feeder, a vertical air duct, a variable frequency fan, and a settling separation chamber. The vibrating feeder evenly feeds the material processed by the two-stage precision peeling mechanism into the bottom of the vertical air duct. The variable frequency fan is installed at the top or side of the vertical air duct and generates an upward vertical airflow. The airflow speed can be adjusted from five meters per second to ten meters per second. The settling separation chamber is equipped with a light impurity outlet and a broad bean outlet. By utilizing the difference in density between the broad bean skin and the broad bean, the skin is carried into the light impurity outlet by the airflow, while the broad bean falls into the broad bean outlet.

[0029] Specifically, the air separation mechanism is used to achieve efficient separation of the peeled skin and soybean kernels. This mechanism includes a vibrating feeder, a vertical air duct, a variable frequency fan, and a settling chamber. The vibrating feeder is fixedly installed below the secondary precision peeling mechanism, with its inlet precisely aligned with the outlet of the secondary precision peeling mechanism. It receives the material processed by the secondary precision peeling mechanism. The vibrating feeder uses the principle of electromagnetic vibration. By adjusting the vibration frequency, the material is fed evenly and continuously into the bottom of the vertical air duct, preventing material accumulation at the inlet and ensuring uniform distribution of material within the vertical air duct. The vertical air duct adopts a cylindrical structure, with its inner diameter designed according to the material processing capacity to ensure uniform airflow within the duct without eddy current generation. A feed hopper is installed at the bottom of the vertical air duct, and the top connects to the settling chamber. A variable frequency fan is installed on the side or top. The variable frequency fan uses a centrifugal fan, which can dynamically adjust the fan speed according to the density changes of the material and the separation requirements, thereby adjusting the airflow speed in the vertical duct. The airflow speed adjustment range is five to ten meters per second. Its adjustment algorithm adopts a fuzzy control algorithm. The central control system collects the purity data of the separated material in real time and automatically adjusts the fan speed based on the data feedback to ensure stable separation effect. The settling separation chamber is fixedly installed at the top of the vertical duct. It is equipped with a guide plate inside to guide the flow direction of airflow and material. The upper part of the settling separation chamber is equipped with a light impurity outlet, and the lower part is equipped with a bean paste outlet. The light impurity outlet is connected to a dust collection device to collect the separated shell impurities. The bean paste outlet is connected to the feed inlet of the color sorting and grading mechanism to transport the separated bean paste to the next process.

[0030] The color sorting and grading mechanism includes a color sorting box, a full-spectrum color camera installed inside the color sorting box, a background plate, and a multi-stage rejection actuator. The full-spectrum color camera acquires images of both the front and back sides of individual broad beans after air separation. The multi-stage rejection actuator includes at least two sets of pneumatic spray valves arranged sequentially along the material conveying direction. The opening and closing frequency and duration of the pneumatic spray valves are controlled by the central control system according to the sorting grade, sorting the broad beans into whole green-heart retained products, green-heart superior products, ordinary products, and rejected products.

[0031] Specifically, the color sorting and grading mechanism is used for the final quality sorting of soybeans after air separation. It includes a color sorting box, a full-spectrum color camera, a background plate, and a multi-stage rejection actuator. The color sorting box has a closed structure with internal light-shielding and dust-proof devices to prevent external interference and ensure sorting accuracy. The feed end of the color sorting box connects to the soybean outlet of the air separation mechanism, and the discharge end connects to the discharge collection mechanism. The material is conveyed uniformly along the conveyor within the color sorting box. The full-spectrum color camera is fixedly installed at the top and bottom of the color sorting box to capture images of both sides of the soybeans, ensuring no blind spots. The frame rate can be dynamically adjusted according to the material conveying speed to ensure clear acquisition of color and shape information from both sides of each soybean. The background plate, made of matte white material, is fixedly installed below the conveyor to effectively highlight the color differences of the soybeans and prevent background reflections from affecting sorting accuracy. Image information captured by a full-spectrum color camera is transmitted to a central control system. The central control system uses a deep learning-based image recognition algorithm to preprocess the images, extracting color features, green heart ratio, and integrity features of the soybeans. These features are then classified and identified using a convolutional neural network (CNN) model to determine the quality grade of each soybean. Multi-stage rejection actuators are fixedly installed inside the color sorting box, located on the side of the conveying device. At least two sets of pneumatic spray valves are sequentially arranged along the material conveying direction. The number of pneumatic spray valves matches the grading level, with each set corresponding to the rejection and sorting of a specific quality grade. The opening and closing frequency and duration of the pneumatic spray valves are precisely controlled by the central control system according to the sorting level, ensuring accurate rejection of soybeans of the corresponding grade. After being rejected by the corresponding pneumatic spray valves, soybeans of different grades enter their respective discharge channels and are ultimately transported to different collection bins in the discharge collection mechanism.

[0032] The central control system integrates a dynamic parameter optimization module, which stores a quality prediction model built based on machine learning. The central control system acquires the overall quality data of the current batch of raw materials provided by the visual inspection agency and the green heart ratio parameter required by the customer order in real time. The quality prediction model calculates the optimal sorting threshold based on the input overall quality data and sends the instructions to the color sorting and grading agency and the secondary precision peeling agency to dynamically adjust the peeling intensity of the secondary precision peeling agency and the rejection criteria of the color sorting and grading agency.

[0033] Specifically, the central control system is the core of the entire green broad bean automatic peeling and color sorting machine system. It is responsible for the coordinated operation of various mechanisms, parameter adjustment and data processing. The central control system uses a PLC as the core control unit and an industrial touch screen as the human-machine interface. Operators can set production parameters, view equipment operating status, retrieve production data and fault alarm information through the touch screen. It also supports connection to a host computer, enabling remote monitoring and data management. The dynamic parameter optimization module integrated within the central control system is key to achieving adaptive system operation. This module stores a quality prediction model built on machine learning. The model uses data on the characteristics of green broad beans from different origins and seasons, including moisture content, particle size, outer skin thickness, and inner skin adhesion, as well as equipment process parameters, including primary coarse peeling gap, secondary peeling intensity, air separation speed, and color sorting threshold, as training samples. The model is trained using a BP neural network algorithm, and the model's weight parameters are continuously adjusted during training to ensure the model's prediction accuracy. The central control system acquires the overall quality data of the current batch of raw materials, including average particle size, average moisture content, initial outer skin thickness, and peeling cleanliness after the first peeling, from the visual inspection agency in real time through sensors. At the same time, it receives the green heart ratio parameters required by the customer order input by the operator and feeds these data into the quality prediction model. The quality prediction model, based on the input overall quality data and green core ratio parameters, calculates and analyzes to output the optimal sorting threshold and process parameters. It then sends instructions to the color sorting and grading mechanism and the secondary precision peeling mechanism, dynamically adjusting the peeling force of the secondary precision peeling mechanism, the motion parameters of the micro-robotic arm, and the rejection criteria and control parameters of the pneumatic spray valve of the color sorting and grading mechanism. This ensures that the equipment process parameters are precisely adapted to the characteristics of the current batch of raw materials, effectively improving the finished product qualification rate and reducing the rework rate.

[0034] The primary coarse peeling mechanism includes a friction peeling roller and an elastic support roller that are arranged in relative rotation. The surface of the friction peeling roller is covered with a layer of diamond abrasive, and the surface of the elastic support roller is covered with a layer of rubber. The gap between the friction peeling roller and the elastic support roller is controlled by an electric adjustment mechanism. The electric adjustment mechanism automatically fine-tunes the gap size based on the peeling cleanliness data of the previous batch of materials fed back by the visual inspection mechanism, so that the primary coarse peeling mechanism removes most of the outer skin of the broad beans and retains some of the inner skin to protect the green cotyledons inside.

[0035] Specifically, the primary coarse peeling mechanism is used for the initial peeling of broad beans. Its core design goal is to remove most of the outer skin while retaining some of the inner skin, protecting the green cotyledons and laying the foundation for subsequent precise peeling. This mechanism includes a friction peeling roller, an elastic support roller, and an electric adjustment mechanism. The friction peeling roller and the elastic support roller are mounted on the frame in a parallel arrangement, forming a peeling gap. The material passes through the peeling gap, and the outer skin is removed under the relative rotation and friction of the two rollers. The friction peeling roller is made of stainless steel with a layer of diamond abrasive on its surface, which can effectively remove the outer skin of the broad beans while avoiding excessive grinding that could damage the inner skin and cotyledons. The elastic support roller is made of rubber with a flexible rubber layer on its surface. When the broad beans pass through the peeling gap, the elastic support roller can adaptively contract according to the size of the broad beans, ensuring that each broad bean receives uniform friction while avoiding damage to the cotyledons caused by rigid contact. The friction peeling roller is driven by a variable frequency motor, and the elastic support roller is linked to the friction peeling roller through gears to ensure the friction effect. The rotation speed can be dynamically adjusted according to the characteristics of the raw materials. An electric adjustment mechanism, connected to the friction peeling roller and the elastic support roller, controls the peeling gap between the two rollers. It uses a ball screw drive structure and is driven by a servo motor. The electric adjustment mechanism is electrically connected to the central control system, receiving real-time peeling cleanliness data from the previous batch of materials from the visual inspection mechanism. The central control system uses a PID control algorithm to automatically fine-tune the peeling gap based on the peeling cleanliness data. When the peeling cleanliness is too low, the gap is reduced to increase the friction force; when the peeling cleanliness is too high, resulting in excessive removal of the inner skin, the gap is increased to reduce the friction force. This ensures that the primary coarse peeling mechanism can stably and efficiently complete the initial peeling task, providing a good material foundation for the subsequent secondary precision peeling.

[0036] The light source illumination module is equipped with a strobe control circuit, which is synchronously triggered with the line scanning signal of the multispectral line array camera. The light emitted by the light source illumination module is diffused through the lamp cover to form a uniform surface light source that illuminates the detection area. The light source illumination module also includes a polarizing filter group for eliminating reflections on the surface of broad beans. The polarizing filter group is installed in front of the light source emitting end and the receiving lens of the multispectral line array camera.

[0037] Specifically, the light source illumination module is a crucial component of the visual inspection mechanism. Besides a diffused light cover and a specific wavelength light source, this module also includes a flicker control circuit and a polarizing filter group. The flicker control circuit synchronizes with the line scan signal of the multispectral linear scan camera via a synchronization trigger module. Its working principle is as follows: the flicker control circuit receives the line scan synchronization signal sent by the multispectral linear scan camera and, based on the frequency of the synchronization signal, controls the lighting and extinguishing of the specific wavelength light source, ensuring that the flicker frequency of the light source is consistent with the line scan frequency of the camera. This ensures that the light source provides stable and uniform illumination during each line scan, avoiding image blurring and ghosting caused by material movement and continuous light source illumination, thus improving the clarity and detail of image acquisition. The flicker control circuit can dynamically adapt to the line scan frequency of the camera to ensure synchronization accuracy. The light emitted by the light source illumination module, after being scattered by the diffused light cover, forms a uniform surface light source that illuminates the inspection area, providing a stable lighting environment for image acquisition. The polarizing filter assembly consists of two sets of polarizing filters, installed at the light source emitter and in front of the receiving lens of the multispectral linear array camera, respectively. The polarization directions of the two sets of polarizing filters are perpendicular to each other, and their core function is to eliminate specular reflections on the surface of broad beans. Because the surface of broad beans has a certain degree of smoothness, specular reflections occur when illuminated by a light source. These reflective areas cause bright spots in the image, obscuring the texture and color information of the broad bean surface and affecting image recognition accuracy. By installing a polarizing filter at the light source's emitting end, the emitted light becomes linearly polarized. When this light shines on the broad bean surface, the specular reflection remains linearly polarized, with its polarization direction aligned with that of the polarizing filter at the light source's emitting end. Meanwhile, the diffuse reflection from the broad bean surface is unpolarized. Furthermore, by installing a polarizing filter perpendicular to the camera's receiving lens, the linearly polarized light from the specular reflection is effectively blocked, allowing only the diffusely reflected unpolarized light to enter the camera. This completely eliminates reflection from the broad bean surface, enabling the camera to capture clear and complete information about the bean's surface and interior, providing reliable data support for subsequent image recognition and targeted peeling.

[0038] A material tracking encoder is installed between the visual inspection mechanism and the secondary precision peeling mechanism. The material tracking encoder is connected to the conveyor belt drive shaft and records the displacement of the conveyor belt in real time. The central control system uses the displacement data of the material tracking encoder to synchronize the position of the single broad bean feature identified by the visual inspection mechanism with the corresponding actuator in the secondary precision peeling mechanism in time and space, so as to ensure that the actuator accurately acts on the residual inner skin area identified by the vision.

[0039] Specifically, to ensure the secondary precision peeling mechanism accurately targets the visually identified residual inner skin area, achieving precise targeted peeling, a material tracking encoder is installed between the visual inspection mechanism and the secondary precision peeling mechanism. This ensures temporal and spatial synchronization between visual inspection data and the secondary peeling actuator, resolving positional offset issues during material transport and improving the accuracy of targeted peeling. The material tracking encoder is an incremental encoder, fixedly connected to the drive shaft of the position calibration conveyor belt via a coupling. It can collect the conveyor belt's rotation angle in real time, thereby calculating the conveyor belt's displacement and accurately reflecting its operating status. The material tracking encoder is electrically connected to the central control system, transmitting the collected displacement data to the central control system in real time. The central control system has a built-in material tracking algorithm that converts the displacement data into the material's position coordinates on the conveyor belt. Simultaneously, it combines this data with the conveyor belt displacement data acquired by the visual inspection mechanism to establish a correspondence between the visual inspection coordinates and the actual position coordinates of the conveyor belt. When the vision inspection mechanism identifies a residual inner skin area on a broad bean and outputs its coordinates in the image, the central control system uses displacement data from the material tracking encoder to calculate in real time the actual position of the broad bean on the conveyor belt, as well as the time it takes for the bean to reach the corresponding micro-manipulator in the secondary precision peeling mechanism. Simultaneously, based on the conveyor belt's operating speed, the central control system dynamically adjusts the timing and position of the micro-manipulator, achieving time and space synchronization between the residual inner skin area identified by the vision inspection mechanism and the actuator in the secondary precision peeling mechanism. Furthermore, the central control system also calibrates the conveyor belt's operating speed in real time based on the displacement data from the material tracking encoder. When the conveyor belt slips or experiences speed fluctuations, it promptly adjusts the speed of the variable frequency motor to compensate for displacement deviations, ensuring that the actuator accurately targets the visually identified residual inner skin area, avoiding incomplete peeling or damage to the cotyledons due to positional errors.

[0040] The dynamic parameter optimization module is equipped with a historical database and a self-learning unit. The historical database is used to store the peeling characteristic parameters of broad bean raw materials from different origins and seasons, as well as the corresponding optimal process parameters. The self-learning unit periodically analyzes the final product qualification rate and rework rate data fed back by the color sorting and grading institution. When the rework rate exceeds the preset value, the self-learning unit uses the backpropagation algorithm to correct the weight parameters of the quality prediction model, so that the system can adapt to the slow changes in raw material characteristics.

[0041] Specifically, the historical database uses industrial solid-state drives for storage, storing the peeling characteristic parameters and corresponding optimal process parameters of broad bean raw materials from different origins and seasons. Peeling characteristic parameters include the moisture content of the raw materials, particle size distribution, outer skin thickness, inner skin adhesion, and green heart percentage. Optimal process parameters include the primary coarse peeling gap, friction peeling roller speed, secondary precision peeling force, micro-robotic arm motion parameters, air-separation wind speed, and color sorting threshold. It also stores data such as the finished product qualification rate and rework rate for each batch, providing data support for optimizing the quality prediction model. The historical database supports real-time data updates, queries, and statistics. Operators can retrieve production data and raw material characteristic data from different batches as needed for production analysis and parameter adjustments. The self-learning unit establishes communication connections with the historical database, quality prediction model, and color sorting and grading mechanism, periodically analyzing and processing production data. The self-learning cycle can be set according to the production batch, typically performing a self-learning analysis after every 10 batches of production. The core algorithm of the self-learning unit employs backpropagation (BP). It first retrieves raw material characteristic parameters, process parameters, and corresponding finished product pass rates and rework rates from the historical database. Using the finished product pass rate and rework rate as evaluation indicators, it analyzes the deviation between the process parameters output by the current quality prediction model and the actual optimal process parameters. When the rework rate exceeds a preset value, the self-learning unit uses the backpropagation algorithm to correct the weight parameters and thresholds of the quality prediction model based on the deviation data, adjusting the model's calculation logic to more accurately output the optimal process parameters according to the characteristics of the current batch of raw materials. Simultaneously, the self-learning unit stores new raw material characteristic parameters and optimized process parameters in the historical database, enriching the database's data volume and further improving the model's prediction accuracy and adaptability. Through continuous optimization by the self-learning unit, the system can adapt to slow changes in raw material characteristics. Even if parameters such as the moisture content and inner skin adhesion of the raw materials change, the system can automatically adjust the process parameters, reducing manual parameter adjustments by operators, lowering labor intensity, and improving production efficiency.

[0042] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An automatic peeling and color sorting system for green-fleshed broad beans, characterized in that, The system includes a frame and, sequentially mounted on the frame, a feeding mechanism, a primary coarse peeling mechanism, a visual inspection mechanism, a secondary precision peeling mechanism, an air separation mechanism, a color sorting and grading mechanism, and a discharge collection mechanism. The primary coarse peeling mechanism performs initial peeling on the broad beans fed by the feeding mechanism to remove most of the outer skin and expose the inner skin. The visual inspection mechanism is installed downstream of the primary coarse peeling mechanism to collect image information of the broad beans after the initial peeling. The secondary precision peeling mechanism is located downstream of the visual inspection mechanism and is communicatively connected to it. The secondary precision peeling mechanism controls the execution components to target and remove residual inner skin at specific locations based on the image information collected by the visual inspection mechanism. The air separation mechanism is located below the secondary precision peeling mechanism and uses density difference to separate the detached skin from the bean kernels. The color sorting and grading mechanism performs final quality sorting on the bean kernels after air separation. The feeding mechanism, primary coarse peeling mechanism, visual inspection mechanism, secondary precision peeling mechanism, air separation mechanism, and color sorting and grading mechanism are all electrically connected to a central control system.

2. The automatic peeling and color sorting system for green broad beans according to claim 1, characterized in that, The visual inspection mechanism includes an inspection box, an image acquisition module installed inside the inspection box, and a light source illumination module. The image acquisition module includes a multispectral linear array camera, which contains a visible light imaging unit and a near-infrared imaging unit. The visible light imaging unit is used to acquire the skin color and texture information of the broad bean surface, and the near-infrared imaging unit is used to penetrate the surface of the broad bean to acquire internal component information to identify the distribution of green cotyledon areas and residual endothelial cells. The light source illumination module includes a diffused light cover arranged around the transmission path, and the light cover contains a specific wavelength light source corresponding to the wavelength band of the near-infrared imaging unit.

3. The automatic peeling and color sorting system for green broad beans according to claim 1, characterized in that, The secondary precision peeling mechanism includes a support frame, a multi-degree-of-freedom targeting execution component mounted on the support frame, and a position calibration conveyor belt. The multi-degree-of-freedom targeting execution component includes several sets of micro-manipulators arranged along the conveying direction. Each set of micro-manipulators has a high-pressure jet nozzle or a flexible grinding head at its end. The micro-manipulators are connected to the central control system and receive coordinate data generated by the vision inspection mechanism. The central control system drives the micro-manipulators to adjust the spatial angle and force of the end tools according to the coordinate data, so as to physically remove or air-jet peel off the inner skin of a specific location on a single broad bean.

4. The automatic peeling and color sorting system for green broad beans according to claim 1, characterized in that, The air separation mechanism includes a vibrating feeder, a vertical air duct, a variable frequency fan, and a settling separation chamber. The vibrating feeder evenly feeds the material processed by the two-stage precision peeling mechanism into the bottom of the vertical air duct. The variable frequency fan is installed at the top or side of the vertical air duct and generates an upward vertical airflow. The airflow speed can be adjusted from five meters per second to ten meters per second. The settling separation chamber is equipped with a light impurity outlet and a broad bean outlet. By utilizing the difference in density between the broad bean shell and the broad bean, the shell is carried into the light impurity outlet by the airflow, while the broad bean falls into the broad bean outlet.

5. The automatic peeling and color sorting system for green broad beans according to claim 1, characterized in that, The color sorting and grading mechanism includes a color sorting box, a full-spectrum color camera installed inside the color sorting box, a background plate, and a multi-stage rejection actuator. The full-spectrum color camera acquires images of both the front and back sides of individual broad beans after air separation. The multi-stage rejection actuator includes at least two sets of pneumatic spray valves arranged sequentially along the material conveying direction. The opening and closing frequency and duration of the pneumatic spray valves are controlled by the central control system according to the sorting grade, sorting the broad beans into whole green-heart retained products, green-heart superior products, ordinary products, and rejected products.

6. The automatic peeling and color sorting system for green broad beans according to claim 1, characterized in that, The central control system integrates a dynamic parameter optimization module, which stores a quality prediction model based on machine learning. The central control system acquires the overall quality data of the current batch of raw materials provided by the visual inspection agency and the green heart ratio parameter required by the customer order in real time. The quality prediction model calculates the optimal sorting threshold based on the input overall quality data and sends the instruction to the color sorting and grading agency and the secondary precision peeling agency to dynamically adjust the peeling intensity of the secondary precision peeling agency and the rejection standard of the color sorting and grading agency.

7. The automatic peeling and color sorting system for green broad beans according to claim 1, characterized in that, The primary coarse peeling mechanism includes a friction peeling roller and an elastic support roller arranged in relative rotation. The surface of the friction peeling roller is covered with a layer of diamond abrasive, and the surface of the elastic support roller is covered with a rubber layer. The gap between the friction peeling roller and the elastic support roller is controlled by an electric adjustment mechanism. The electric adjustment mechanism automatically fine-tunes the gap size based on the peeling cleanliness data of the previous batch of materials fed back by the visual inspection mechanism, so that the primary coarse peeling mechanism removes most of the outer skin of the broad beans and retains some of the inner skin to protect the green cotyledons inside.

8. The automatic peeling and color sorting system for green broad beans according to claim 1, characterized in that, The light source illumination module is equipped with a strobe control circuit, which is synchronously triggered with the line scanning signal of the multispectral line array camera. The light emitted by the light source illumination module is diffused through a diffuser to form a uniform surface light source that illuminates the detection area. The light source illumination module also includes a polarizing filter group for eliminating reflections on the surface of broad beans. The polarizing filter group is installed in front of the light source emitting end and the receiving lens of the multispectral line array camera.

9. The automatic peeling and color sorting system for green broad beans according to claim 1, characterized in that, A material tracking encoder is installed between the visual inspection mechanism and the secondary precision peeling mechanism. The material tracking encoder is connected to the conveyor belt drive shaft and records the displacement of the conveyor belt in real time. The central control system uses the displacement data of the material tracking encoder to synchronize the position of the single broad bean feature identified by the visual inspection mechanism with the corresponding actuator in the secondary precision peeling mechanism in time and space, so as to ensure that the actuator accurately acts on the residual inner skin area identified by vision.

10. The automatic peeling and color sorting system for green broad beans according to claim 1, characterized in that, The dynamic parameter optimization module is equipped with a historical database and a self-learning unit. The historical database is used to store the peeling characteristic parameters and corresponding optimal process parameters of broad bean raw materials from different origins and seasons. The self-learning unit periodically analyzes the final product qualification rate and rework rate data fed back by the color sorting and grading institution. When the rework rate exceeds the preset value, the self-learning unit uses the backpropagation algorithm to correct the weight parameters of the quality prediction model, so that the system can adapt to the slow changes in raw material characteristics.

Images