Cooperative control system for flexible clamping seedling feeding and multi-station parallel grafting of seedlings

Abstract

The invention discloses a cooperative control system for flexible clamping seedling feeding and multi-station parallel grafting of seedlings, and belongs to the technical field of agricultural machinery. The system comprises a core scheduling layer and a function execution layer, the core scheduling layer comprises a distributed multi-station cooperative control system, is composed of a main controller, a station slave controller, an execution unit and a TSN + EtherCAT hybrid communication network, and realizes global decision and data transmission support; the function execution layer is integrated with a bionic self-adaptive flexible clamping seedling feeding module, a multi-station parallel grafting execution module and a visual guidance and intelligent detection module, a scheduling instruction is issued through a main controller, and the modules feed back a collaborative mechanism of sensing / visual data, so that the whole process of automatic grafting is completed. The seedling body clamping flexibility and the multi-station cooperation efficiency are improved, the grafting precision is guaranteed, the modular design is suitable for multiple seedling bodies, and the device can be widely applied to facility agriculture grafting operation.

Description

Technical Field

[0001] This invention relates to the field of agricultural machinery technology, and in particular to a flexible seedling clamping and multi-station parallel grafting collaborative control system. Background Technology

[0002] With the increasing demand for large-scale and automated development in modern agriculture, the research and application of automated equipment for seedling grafting, a key technology for improving crop yield and enhancing stress resistance, is becoming increasingly important. Currently, automated seedling grafting equipment has gradually replaced manual grafting, effectively improving operational efficiency. However, in practical applications, existing technologies still have many shortcomings that urgently need to be addressed, making it difficult to meet the needs of grafting operations of different varieties with high precision and high efficiency. Specifically:

[0003] In the seedling clamping and feeding stages, existing automated grafting equipment generally adopts a rigid clamping structure, with clamping force mostly fixed or only adjustable to a limited range. This type of structure is designed based on the stem characteristics of a single variety of seedling, without considering the differences in stem hardness and diameter between different varieties and at different growth stages. In actual operation, if the clamping force is set according to thin-stemmed seedlings, thick-stemmed seedlings are easily not clamped securely, leading to problems such as detachment and misalignment during feeding. Conversely, if the force is set according to thick-stemmed seedlings, it will cause compression damage to thin-stemmed seedlings, damaging the stem's vascular tissue and affecting the grafting survival rate. Furthermore, existing clamping structures are mostly integrated designs; when adapting to seedlings of different diameters, the entire clamping mechanism must be disassembled and replaced, resulting in a complex and time-consuming debugging process, severely impacting operational continuity and the equipment's broad compatibility. In addition, existing clamping mechanisms lack effective pressure sensing and dynamic adjustment mechanisms, failing to provide real-time feedback on the clamping status and hindering closed-loop control, further exacerbating problems of clamping damage or insecure clamping.

[0004] In multi-station operation and collaborative control, existing equipment mainly falls into two categories. One is a single-station serial operation mode, where each grafting process is completed sequentially at the same station, resulting in low efficiency and failing to meet the high-efficiency requirements of large-scale production. The other type, while employing a multi-station parallel layout, lacks a precise collaborative control mechanism. The control architecture of existing multi-station equipment is mostly a simple master-slave control, with the master controller directly driving each station's execution unit without independent station slave controllers. This leads to mutual interference between global scheduling and local execution, making it difficult to guarantee the execution accuracy of each process. Furthermore, the communication networks of existing equipment mostly use a single Ethernet or ordinary fieldbus, which cannot simultaneously address the real-time nature of command issuance, the accuracy of multi-station synchronous control, and the reliability of sensor data transmission. During multi-station parallel operation, the lack of timing planning, conflict detection, and dynamic compensation mechanisms makes it prone to interference between station execution mechanisms, leading to equipment downtime, grafting failures, and other problems, severely impacting equipment operational stability. In addition, existing communication networks lack redundancy switching capabilities; if a communication link fails, the entire system will be paralyzed, further reducing the equipment's continuous operation capability.

[0005] In the visual guidance and quality inspection stages, the positioning and inspection methods of existing automated grafting equipment are relatively crude. The feeding positioning and grafting bonding rely heavily on mechanical limiting structures, which limit the position of the seedling and the bonding distance through preset mechanical gears. This cannot be adaptively adjusted according to the actual posture of the seedling and the shape of the cutting surface, resulting in low positioning accuracy and large bonding deviations, which affect the grafting quality. In terms of quality inspection, existing equipment mostly only has simple appearance defect detection functions, or directly relies on manual quality judgment. Manual judgment is affected by factors such as subjective experience and fatigue, and the judgment standards are not uniform, resulting in large fluctuations in the grafting pass rate. Simple appearance inspection cannot accurately identify core quality indicators such as grafting bonding degree and internal bonding gap, making it difficult to comprehensively guarantee grafting quality. At the same time, the vision module and control system of existing equipment mostly work independently, without forming an effective data interaction and closed-loop feedback. The visual inspection results cannot guide the control module to adjust the operating parameters in real time, further limiting the improvement of grafting accuracy.

[0006] Therefore, this application proposes a flexible seedling clamping and multi-station parallel grafting collaborative control system a. Summary of the Invention

[0007] One objective of this invention is to propose a flexible seedling clamping and multi-station parallel grafting collaborative control system. This invention can achieve non-destructive and precise clamping of seedlings of different varieties and broad-spectrum adaptability, ensuring the orderly parallel advancement of multi-station grafting processes to improve operational efficiency and stability. Simultaneously, it achieves precise control of grafting quality through full-process visual guidance and intelligent detection. In summary, existing seedling grafting equipment has significant technical deficiencies in core aspects such as clamping adaptability, multi-station collaborative efficiency, visual guidance, and quality control, making it difficult to meet the demands of modern agriculture for large-scale, high-precision, and high-efficiency grafting operations. Therefore, developing a grafting collaborative control system capable of achieving non-destructive seedling clamping, precise multi-station collaboration, and full-process visual intelligent control has become an urgent technical problem to be solved in this field.

[0008] According to an embodiment of the present invention, a flexible seedling clamping and multi-station parallel grafting collaborative control system includes a biomimetic adaptive flexible seedling clamping module, a multi-station parallel grafting execution module, a distributed multi-station collaborative control system, and a visual guidance and intelligent detection module. Each module realizes data interaction and collaborative work through a high-speed communication link.

[0009] The biomimetic adaptive flexible clamping and seedling loading module adopts a three-finger biomimetic structure, integrating a shape memory alloy driving unit, an elastic buffer unit, and a fiber optic pressure sensor array. It achieves damage-free clamping and precise seedling loading through a pressure threshold adaptive and fuzzy PID composite control algorithm. The formula for the fuzzy PID composite control algorithm is as follows:

[0010]

[0011]

[0012]

[0013] in, , , These are the initial proportional coefficient, integral coefficient, and derivative coefficient of the PID controller. , , These are the proportional coefficient correction, integral coefficient correction, and derivative coefficient correction values ​​output by the fuzzy controller, respectively. , , The input quantity is pressure deviation. Change rate of pressure deviation , , , To preset the clamping pressure threshold, Real-time clamping pressure collected by fiber optic pressure sensor array;

[0014] The distributed multi-station collaborative control system adopts a three-level control architecture consisting of a master controller, station slave controllers, and execution units. It incorporates a hybrid communication network built using a time-sensitive network and an EtherCAT bus. Through timing planning, conflict detection, and dynamic compensation collaborative scheduling algorithms, it achieves multi-station action synchronization and interference avoidance. The dynamic compensation algorithm formula is as follows:

[0015]

[0016] in, This is for the timing compensation of subsequent workstation actions. Delay time for the preceding workstation's actions. For compensation coefficient, This refers to the positioning deviation of the preceding workstation.

[0017] The visual guidance and intelligent detection module adopts a multi-camera distributed layout, uses deep learning algorithms to identify seedling varieties and detect postures, and combines 3D visual detection algorithms to detect grafting fit accuracy.

[0018] Furthermore, the gripper unit of the biomimetic adaptive flexible clamping seedling module has adaptive clamping and rapid adaptation functions. It can achieve non-damaging clamping of seedlings through a flexible contact layer and adapt to seedlings of different diameters through a quick-change clamping adaptation structure.

[0019] The gripper unit works in conjunction with the shape memory alloy drive module and the elastic buffer module. Through the pressure sensing feedback of the flexible contact layer, and in conjunction with the precise drive adjustment of the shape memory alloy drive module, the gripping force is dynamically adapted to meet the gripping needs of different varieties of seedlings.

[0020] Furthermore, the memory alloy drive unit provides precise driving power for the biomimetic adaptive flexible clamping seedling module. It can receive PWM drive commands from the clamping control module and achieve controllable extension and retraction of the drive component through precise current adjustment, thereby driving the gripper unit to complete the opening and closing action. The drive unit integrates a friction suppression structure and a stable transmission mechanism to ensure the smoothness and repeatability of the drive action and ensure the precise synchronization of the clamping action.

[0021] Furthermore, the elastic buffer unit is used to realize active buffering and pressure feedback during the clamping process. It can offset the rigid impact during the clamping process through an adjustable buffering mechanism to avoid damage to the seedling. At the same time, it integrates a distributed pressure sensing module to collect the pressure data of the contact between the gripper and the seedling in real time. After converting the pressure signal into a standard control signal, it is transmitted to the clamping control module to provide real-time pressure feedback data for the fuzzy PID composite control algorithm. Together with the buffering guidance mechanism, it ensures the stability of the clamping posture.

[0022] Furthermore, the multi-station parallel grafting execution module has the functions of orderly multi-station flow and rapid process adaptation. It realizes the intermittent and precise flow of each station carrier through the circular flow module, and ensures the positioning accuracy of each station in conjunction with the station positioning module. The execution module integrates modular functional units, and each functional unit can be quickly switched according to the grafting process requirements. Through signal interaction with the distributed collaborative control system, the coordinated linkage of each functional unit and station flow is realized, ensuring the orderly progress of the entire grafting process.

[0023] Furthermore, each functional workstation achieves its corresponding grafting process function through the collaboration of functional units and sensing units, and the workstations achieve time synchronization through a distributed collaborative control system. The specific steps are as follows:

[0024] S1, Rootstock Loading Station: Through the collaboration of the flexible clamping unit and the posture pre-correction unit, the rootstock seedlings are accurately grasped and positioned for loading, and the loading signal is fed back to the collaborative control system.

[0025] S2, Rootstock Cutting Station: Through the collaboration of the cutting control unit and the rootstock positioning unit, precise cutting of the rootstock is achieved, and the cutting parameters can be adaptively adjusted according to the seedling type;

[0026] S3, Scion feeding station: The biomimetic adaptive flexible clamping and feeding module enables precise feeding and posture correction of scion seedlings, ensuring alignment with the axis of the rootstock.

[0027] S4, Scion Cutting Station: Through the collaboration of the cutting drive unit and the attitude adjustment unit, the scion cutting surface and the rootstock cutting surface are precisely matched.

[0028] S5, Grafting and Fitting Station: Through the collaboration of dual flexible clamping and coordination units and precision displacement control unit, combined with pressure feedback, the precise fitting of rootstock and scion is achieved;

[0029] S6. Fixed curing station: Through the collaboration of the fixed execution unit and the constant temperature curing unit, the grafting parts are firmly fixed and cured;

[0030] S7, Quality Inspection Station: Through the collaboration of the vision inspection unit and the pressure inspection unit, a multi-dimensional comprehensive judgment of grafting quality is achieved, and the inspection results are uploaded to the collaborative control system in real time;

[0031] S8 Finished Product Unloading Station: Based on the quality inspection results, qualified seedlings and unqualified seedlings are sorted and transferred through the sorting unit.

[0032] Furthermore, the distributed multi-station collaborative control system achieves coordination between global scheduling and local execution through a three-level control architecture, which consists of a master control module, a slave control module, and an execution control module.

[0033] The main control module is responsible for global system decision-making, timing planning, and multi-station collaborative scheduling. It receives operational status data from each module and issues global control commands. The slave control module receives commands from the main control module, drives the execution unit of the corresponding station to complete specific process actions, collects sensor data of the station and uploads it to the main control module. The execution control module implements precise execution and closed-loop feedback of actions according to the commands of the slave control module. Each control level achieves real-time data interaction through a high-speed communication link to ensure the synchronization and accuracy of control commands.

[0034] Furthermore, the hybrid communication network ensures the real-time performance and reliability of data interaction between modules through a layered communication architecture. The main control layer and the software layer use industrial Ethernet to issue commands and upload status. The main control layer and the workstation control layer use a real-time bus to achieve synchronous transmission of control signals from multiple workstations. The workstation control layer and the execution sensing unit use an industrial fieldbus and analog signal link to achieve real-time data interaction. The entire communication network integrates communication status monitoring and redundancy switching functions to ensure the continuity of system communication and provide communication support for multi-module collaboration.

[0035] Furthermore, the visual guidance and intelligent detection module achieves full-process visual support through multi-vision unit collaboration and data fusion. The feeding vision unit collects seedling image data to provide visual basis for seedling identification, diameter measurement, and posture correction. The grafting and bonding vision unit collects three-dimensional data of the cutting surface to provide position guidance for the precise bonding of the rootstock and scion. The quality detection vision unit collects the appearance and three-dimensional bonding data after grafting to provide detection basis for quality judgment. After the data from each vision unit is analyzed and processed by the vision processing module, control guidance signals and quality judgment results are output and transmitted to the distributed collaborative control system to realize the collaboration between visual guidance and system control.

[0036] Furthermore, the deep learning algorithm and 3D vision detection algorithm provide algorithmic support for the visual guidance and intelligent detection module. The deep learning algorithm enables accurate identification of seedling varieties and accurate calculation of posture deviations, outputting posture adjustment commands. The 3D vision detection algorithm enables point cloud data processing of grafting and bonding areas, calculation of bonding degree and maximum gap, and outputting quality judgment results. The algorithm module interacts in real time with the visual acquisition module and the distributed collaborative control system to achieve closed-loop collaboration of image acquisition, algorithm processing and control feedback, ensuring the accuracy of visual guidance and quality detection.

[0037] The beneficial effects of this invention are:

[0038] 1. This invention designs a three-finger gripping structure based on biomimetic principles. Combining pressure threshold adaptation and fuzzy PID composite control algorithms, a closed-loop control system of perception, decision-making, and actuation is constructed. The fiber optic pressure sensor array senses the contact state between the gripper and the seedling in real time and feeds the pressure signal back to the control module. The fuzzy PID algorithm dynamically adjusts the output power of the memory alloy drive unit, so that the gripping force is adaptively matched with the characteristics of the seedling stem. At the same time, the composite structure of the flexible contact layer and the carbon fiber inner layer ensures both the buffering of the gripping and the structural support rigidity. With the gripping structure that can be quickly adapted, it can adapt to the gripping needs of different types of seedlings without complicated debugging. This design avoids the inherent defects of rigid gripping in principle, realizes non-destructive and broad-spectrum adaptability of seedling gripping, and lays the foundation for the accuracy of subsequent grafting processes.

[0039] 2. In this invention, the main controller is responsible for global timing planning and collaborative decision-making. Through timing planning, conflict detection, and dynamic compensation scheduling algorithms, it predicts the timing relationship of actions at each workstation in advance, avoiding the risk of multi-workstation interference. The workstations are responsible for receiving instructions from the controller and driving the local execution units to perform precise actions, ensuring the execution accuracy of each process. The hybrid communication network, through a layered architecture design, meets different communication needs such as instruction issuance, synchronous control, and sensor data interaction, ensuring the real-time performance and reliability of data transmission. This architecture enables multiple grafting processes to proceed in parallel and in an orderly manner, significantly improving overall work efficiency. At the same time, through dynamic compensation and conflict detection mechanisms, it improves the stability and security of system operation, breaking through the efficiency and stability bottlenecks of traditional equipment.

[0040] 3. In this invention, the feeding vision unit uses deep learning algorithms to identify seedling varieties and detect their posture, providing a precise basis for clamping adaptation and posture correction, ensuring feeding positioning accuracy from the source; the grafting and bonding vision unit uses 3D vision detection technology to collect three-dimensional data of the cutting surface, providing position guidance for the precise bonding of the rootstock and scion, ensuring the tightness of the bonding; the quality detection vision unit combines 2D appearance detection and 3D bonding accuracy detection, achieving multi-dimensional objective judgment of grafting quality through algorithm fusion. The entire vision system interacts in real time with the distributed collaborative control system, forming a closed-loop logic of image acquisition, algorithm processing, and control feedback, fundamentally replacing the traditional extensive control mode of mechanical limit and manual judgment, realizing precise control of the entire grafting process, and effectively improving the stability and consistency of grafting quality. Attached Figure Description

[0041] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0042] Figure 1This is a schematic diagram of the overall framework structure of a flexible seedling clamping and multi-station parallel grafting collaborative control system proposed in this invention.

[0043] Figure 2 This is a schematic diagram of the distributed collaborative control process of a seedling flexible clamping and multi-station parallel grafting collaborative control system proposed in this invention. Detailed Implementation

[0044] To make the technical means and objectives and effects of the present invention easier to understand, the embodiments of the present invention will be described in detail below with reference to specific illustrations.

[0045] like Figure 1-2 As shown, this invention discloses a flexible seedling clamping and multi-station parallel grafting collaborative control system, including a biomimetic adaptive flexible seedling clamping module, a multi-station parallel grafting execution module, a distributed multi-station collaborative control system, and a visual guidance and intelligent detection module. Each module realizes data interaction and collaborative work through a high-speed communication link.

[0046] The biomimetic adaptive flexible clamping and seedling loading module adopts a three-finger biomimetic structure, integrating a shape memory alloy driving unit, an elastic buffer unit, and a fiber optic pressure sensor array. It achieves damage-free clamping and precise seedling loading through a "pressure threshold adaptive + fuzzy PID" composite control algorithm. The fuzzy PID composite control algorithm satisfies the following formula:

[0047]

[0048]

[0049]

[0050] in, , , These are the initial proportional coefficient, integral coefficient, and derivative coefficient of the PID controller. , , These are the proportional coefficient correction, integral coefficient correction, and derivative coefficient correction values ​​output by the fuzzy controller, respectively. , , The input quantity is pressure deviation. Change rate of pressure deviation , , , To preset the clamping pressure threshold, Real-time clamping pressure collected by fiber optic pressure sensor array;

[0051] The distributed multi-workstation collaborative control system adopts a three-level control architecture consisting of a master controller, workstation slave controllers, and execution units. It introduces a time-sensitive network and an EtherCAT bus to construct a hybrid communication network. Through timing planning, conflict detection, and dynamic compensation collaborative scheduling algorithms, it achieves multi-workstation action synchronization and interference avoidance. The dynamic compensation algorithm satisfies the following formula:

[0052]

[0053] in, This is for the timing compensation of subsequent workstation actions. Delay time for the preceding workstation's actions. For compensation coefficient, This refers to the positioning deviation of the preceding workstation.

[0054] The visual guidance and intelligent detection module adopts a multi-camera distributed layout, uses deep learning algorithms to identify seedling varieties and detect postures, and combines 3D visual detection to detect grafting fit accuracy. All modules work together to achieve fully automated operation of the entire process of seedling feeding, cutting, grafting, curing, detection, and unloading.

[0055] Specifically, such as Figure 1 As shown, the four core modules of the seedling flexible clamping and multi-station parallel grafting collaborative control system of the present invention form a closed-loop collaborative architecture through a high-speed communication link. The hardware selection and connection relationship of each module are as follows:

[0056] The biomimetic adaptive flexible clamping and seedling loading module, the multi-station parallel grafting execution module, and the vision guidance and intelligent detection module all establish data connections with the distributed multi-station collaborative control system through communication interfaces. The main controller of the distributed multi-station collaborative control system is a Siemens S7-1516F series industrial-grade PLC, which serves as the core decision-making unit of the system and is responsible for coordinating the timing of actions and data interaction of each module.

[0057] The specific execution process of the pressure threshold adaptive and fuzzy PID composite control algorithm is as follows:

[0058] The first step involves acquiring images of the seedlings through the loading vision unit of the visual guidance and intelligent detection module. After identifying the seedling variety, the corresponding seedlings are matched from the system's preset parameter library. For example, tomato seedlings The range is 0.2-0.3N, corresponding to cucumber seedlings. The range is 0.3-0.4N;

[0059] In the second step, the gripper unit of the biomimetic adaptive flexible clamping seedling module closes at a low speed, and the fiber optic pressure sensor array collects data in real time. ,when When the value is ≥0.05N, it is determined that the gripper is in contact with the seedling, triggering the fuzzy PID control process;

[0060] The third step is to calculate the pressure deviation. and the rate of change of deviation ,Will and As inputs to the fuzzy controller, the universe of discourse for all inputs is set to [-6, 6], and the output ( , , The domain is set to [-3,3]. Using preset fuzzy rules, such as when... For the sake of righteousness, When it is upright, Take the upright and righteous, Take 0, Take the larger negative value, and then deduce... , , Then substitute into the formula The adjusted PID parameters are obtained, where , , The initial values ​​are obtained through the system's self-tuning function, and the initial ranges are as follows: ∈[5,15]、 ∈[0.1,1]、 ∈[1,5];

[0061] The fourth step is to control the output current of the shape memory alloy drive unit according to the adjusted PID parameters to achieve closed-loop regulation of the clamping pressure. Stable at Within the range of ±0.01N, it enters the pressure holding stage to dynamically compensate for the creep error of the shape memory alloy.

[0062] In the timing planning stage, a timing table for each workstation is developed based on the grafting process, specifying the start time, duration, and completion conditions for each action. For example, the duration of the rootstock loading action is set to 0.8 seconds, and the completion condition is that the rootstock positioning error is ≤ ±0.1 mm. The main controller sends a synchronization clock signal to the slave controllers of each workstation via the TSN communication module to ensure that the actions of each workstation start according to the preset timing. In the conflict detection stage, a system kinematic model is established to calculate the motion trajectory of the actuators at each workstation in real time. A safe distance threshold of 5 mm is set for the actuators at adjacent workstations. When the distance between two actuators is ≤ 5 mm, an interference risk is determined, and the system automatically adjusts the start time of the action at the next workstation. For example, the start time of the scion cutting workstation is delayed by 0.2 seconds. In the dynamic compensation stage, the position encoders at each workstation collect the action completion time and positioning deviation. When a delay occurs at a certain workstation When, substitute into the formula Calculate the timing compensation amount for subsequent workstations. The compensation coefficient k is set according to the station type; for cutting stations... =1.2, material loading station =1.0, for example, the delay of the rootstock cutting station. =0.2s, positioning deviation =0.03mm, then the compensation amount at the subsequent scion feeding station. The main controller commands the scaffold feeding station to start with a 0.23s delay, and at the same time adjusts the rotation sequence of the circular guide rail to ensure the continuity of the process.

[0063] In the biomimetic adaptive flexible clamping seedling module, the gripper unit has adaptive clamping and rapid adaptation functions. It can achieve damage-free clamping of seedlings through a flexible contact layer, and adapt to seedlings of different diameters through a quick-change clamping adaptation structure. The gripper unit works in conjunction with the shape memory alloy driving module and the elastic buffer module. Through the pressure sensing feedback of the flexible contact layer, combined with the precise driving adjustment of the shape memory alloy driving module, the clamping force is dynamically adapted to meet the clamping requirements of different varieties of seedlings.

[0064] Specifically, the flexible contact layer of the gripper unit is made of medical-grade silicone. The outer silicone layer is integrally molded with a diamond-shaped grid of micron-level anti-slip texture. The texture grid has a side length of 100μm and a depth of 20μm, which can enhance the friction with the seedling and avoid rigid contact that could damage the seedling stem. The inner side of the flexible contact layer is fixed with a carbon fiber inner layer through an embedded injection molding process. The carbon fiber inner layer is made of T700 grade carbon fiber composite material with a thickness of 4mm. The inner layer has a reserved channel for memory alloy wires and an elastic buffer unit mounting cavity to ensure the structural rigidity of the gripper unit.

[0065] The quick-change clamping adapter structure is as follows: a standardized quick-release interface is set at the root of the gripper unit. This interface adopts a conical surface fit + elastic pin locking structure. The taper of the conical surface is 1:10, the elastic pin is made of stainless steel with a diameter of 3mm, and a positioning pin hole is set at the quick-release interface to ensure that the coaxiality error of the gripper after replacement is ≤±0.02mm.

[0066] The system is equipped with three clamp specifications to adapt to different seedling diameters, corresponding to seedlings with diameters of 2-4mm, 4-6mm, and 6-8mm respectively. When changing, the operator presses the unlock button of the quick-release interface, the elastic pin retracts, and the current clamp can be removed. After changing to the target specification clamp, the elastic pin automatically pops out and locks. The entire replacement process takes ≤10 seconds.

[0067] The collaborative working process of the gripper unit, the shape memory alloy drive module, and the elastic buffer module is as follows: After the visual guidance and intelligent detection module identifies the variety and diameter of the seedling, the main controller sends an initial drive command to the shape memory alloy drive module. The shape memory alloy drive module outputs a corresponding current to drive the gripper unit to initially close. At the same time, the distributed pressure sensor array of the elastic buffer module collects the contact pressure data between the flexible contact layer and the seedling in real time and transmits the pressure signal to the clamping control module. The clamping control module sends an adjustment command to the shape memory alloy drive module based on the deviation between the pressure data and the preset threshold, dynamically adjusting the drive current, thereby adjusting the opening and closing degree of the gripper to achieve dynamic adaptation of the clamping force. For example, for thin-stemmed seedlings with a diameter of 2-4mm, the initial value of the drive current is set to 0.3A, which can be adjusted to the range of 0.2-0.4A according to the pressure feedback to ensure that the clamping force is between 0.1-0.3N.

[0068] The shape memory alloy drive unit provides precise driving power for the biomimetic adaptive flexible clamping seedling module. It can receive PWM drive commands from the clamping control module and control the extension and retraction of the drive components through precise current adjustment, thereby driving the gripper unit to complete the opening and closing action. The drive unit integrates a friction suppression structure and a stable transmission mechanism to ensure the smoothness and repeatability of the drive action and ensure the precise synchronization of the clamping action.

[0069] Specifically, the shape memory alloy drive unit includes a nickel-titanium shape memory alloy wire, a ceramic guide sleeve, a high-precision constant current source, and a drive control circuit board. The connection and operation of each component are as follows:

[0070] Each gripper finger corresponds to two nickel-titanium shape memory alloy wires, which are the clamping drive wire and the releasing drive wire, respectively. The two shape memory alloy wires are symmetrically distributed on both sides of the gripper finger. One end of the wire is fixed in the connecting groove at the end of the gripper finger by a threaded pressure block. The threaded pressure block is made of aluminum alloy and is locked with an M3 screw to ensure that the shape memory alloy wire is firmly fixed. The other end passes through the through-channel of the inner layer of carbon fiber and is fixed to the execution terminal of the drive control circuit board.

[0071] The friction suppression structure uses a ceramic guide sleeve, which is made of alumina ceramic with an inner diameter of 0.55 mm and an outer diameter of 2 mm. It is fixed to the channel of the inner layer of carbon fiber through a snap-fit ​​structure. The inner wall of the ceramic guide sleeve is coated with a polytetrafluoroethylene lubricating coating with a thickness of 6 μm, which can reduce the coefficient of friction of the shape memory alloy wire to below 0.01 during movement, thus preventing the shape memory alloy wire from breaking due to long-term friction.

[0072] A high-precision constant current source is integrated on the drive control circuit board, and a linear voltage regulator circuit design is adopted. The core chip supports PWM signal control with a frequency of 0-10kHz.

[0073] The PWM drive command issued by the clamping control module is processed by the signal conditioning circuit of the drive control circuit board and converted into a constant current source current regulation signal. By adjusting the duty cycle of the PWM signal, the output current is precisely regulated, thereby controlling the extension and retraction of the shape memory alloy wire. The extension and retraction of the shape memory alloy wire is linearly related to the input current, and its linear equation is: ,in Expansion / contraction amount, in mm; The input current is measured in amperes (A). This means that for every 0.1A change in the input current, the extension / retraction amount changes by 0.002mm, ensuring precise control of the opening and closing action of the gripper unit.

[0074] The stable transmission mechanism is achieved through the coordinated drive of two shape memory alloy wires. When the gripper needs to clamp, the clamping control module sends an instruction to increase the current to the constant current source corresponding to the clamping drive wire, and at the same time sends an instruction to decrease the current to the constant current source corresponding to the releasing drive wire. The clamping drive wire expands when heated and the releasing drive wire contracts when cooled, jointly driving the gripper fingers to close. When the gripper needs to release, the instruction is reversed to ensure the smoothness and repeatability of the gripper opening and closing action. The action repeatability error under the same clamping force is ≤±0.01mm.

[0075] The elastic buffer unit is used to achieve active buffering and pressure feedback during the clamping process. It can offset the rigid impact during the clamping process through an adjustable buffering mechanism to avoid damage to the seedling. At the same time, it integrates a distributed pressure sensing module to collect the pressure data of the contact between the gripper and the seedling in real time. After converting the pressure signal into a standard control signal, it is transmitted to the clamping control module to provide real-time pressure feedback data for the fuzzy PID composite control algorithm. Together with the buffering guidance mechanism, it ensures the stability of the clamping posture.

[0076] Specifically, the adjustable buffer mechanism of the elastic buffer unit consists of disc springs, a limiting platform, and adjusting bolts. Two disc springs, made of 619 series stainless steel, are symmetrically arranged in the mounting cavity at the base of the gripper fingers. The upper end of each disc spring abuts against the pressure plate at the base of the gripper fingers, and the lower end abuts against the limiting platform at the bottom of the mounting cavity. The limiting platform is threaded to the inner wall of the mounting cavity. An adjustment groove is provided on the side wall of the limiting platform. The operator can rotate the limiting platform using an adjustment tool to change its height within the mounting cavity, thereby adjusting the pre-compression of the disc springs and achieving adjustable elasticity. The elasticity adjustment range is 0.5-2 N / mm. When clamping seedlings with larger diameters and harder stems, adjusting the limiting platform increases the pre-compression, increasing the elasticity to 1.5-2 N / mm; when clamping seedlings with smaller diameters and softer stems, decreasing the pre-compression adjusts the elasticity to 0.5-1 N / mm, thus offsetting the rigid impact during clamping.

[0077] The distributed pressure sensing module consists of four fiber optic pressure sensors, a fiber optic signal conditioner, and transmission fibers. The four fiber optic pressure sensors are evenly distributed in a ring arrangement inside the flexible contact layer of the gripper unit, with a ring diameter of 10mm. The sensing surface of each sensor is flush with the flexible contact layer and is fixed to the mounting holes of the pressure sensor mounting base using epoxy resin. The pressure signal collected by the fiber optic pressure sensors is a light intensity signal, which is transmitted through the transmission fiber to the fiber optic signal conditioner. The signal conditioner converts the light intensity signal into a standard 4-20mA current signal with a conversion accuracy of 16 bits. This signal is then transmitted through a shielded cable to the AD acquisition interface of the clamping control module, with an acquisition frequency of 100Hz, providing real-time and accurate pressure feedback data for the fuzzy PID composite control algorithm.

[0078] The multi-station parallel grafting execution module has the functions of orderly flow of multiple stations and rapid process adaptation. It realizes the intermittent and precise flow of each station carrier through the circular flow module, and works with the station positioning module to ensure the positioning accuracy of each station. The execution module integrates modular functional units, and each functional unit can be quickly switched according to the grafting process requirements. Through signal interaction with the distributed collaborative control system, the coordinated linkage of each functional unit and station flow is realized, ensuring the orderly progress of the entire grafting process.

[0079] Specifically, the annular transfer module consists of an annular guide rail body, sliders, workstation trays, and a torque motor drive unit. The sliders and the annular guide rail body are in rolling friction engagement. Each slider is equipped with 8 precision balls, and each slider is connected to a workstation tray. The sliders and workstation trays are fixedly connected by 4 M5 bolts. The workstation trays are made of aluminum alloy and have seedling positioning clamps on their surface. The clamping range of the positioning clamps can be adaptively adjusted to accommodate seedlings of different diameters.

[0080] The torque motor drive unit includes a torque motor, a harmonic reducer, and an absolute position encoder. The harmonic reducer has a reduction ratio of 100:1, which can amplify the output torque of the torque motor to 1000 N·m to meet the driving requirements of the ring guide rail. The absolute position encoder can acquire the rotation angle of the ring guide rail in real time with an acquisition accuracy of 0.005°. The position encoder signal is transmitted to the main controller via EtherCAT bus. The main controller controls the start, stop, and speed of the torque motor according to the position signal to realize the intermittent rotation of the workstation tray. The rotation angle is 45° / step, which corresponds to the even distribution of 8 functional workstations. The rotation time of each step is ≤0.5s, and the positioning error after rotation is ≤±0.02mm.

[0081] The workstation positioning module adopts a dual positioning method using electromagnetic limiters and photoelectric sensors. Each workstation is initially positioned with an electromagnetic limiter and a photoelectric sensor. When the workstation tray rotates to the target workstation, the photoelectric sensor first detects the positioning protrusion of the workstation tray and sends a proximity signal to the main controller. The main controller controls the torque motor to brake and decelerate. When the electromagnetic limiter adheres to the limiting groove of the workstation tray, precise positioning is achieved with a positioning accuracy of ≤±0.01mm. The dual positioning method ensures the positioning reliability of each workstation.

[0082] The modular functional units include a cutting module, a bonding module, a curing module, a feeding module, a testing module, and a unloading module. Each module adopts a standardized mechanical interface, which uses a double positioning and locking structure of positioning pins and bolts. The positioning pins have a diameter of 8mm and a clearance of 0.005-0.01mm to ensure the positioning accuracy after the module is installed.

[0083] When switching grafting processes, such as from splitting to bonding, operators only need to disassemble the original cutting module and replace it with a cutting module adapted for the bonding process. Simultaneously, they select the corresponding process parameters through the human-machine interface. The main controller automatically loads the corresponding control timing and parameters, enabling rapid process switching with a switching time of ≤5 minutes. Each modular functional unit communicates with the workstation controller of the distributed collaborative control system via the EtherCAT bus, receiving control commands in real time and providing feedback on its working status, achieving coordinated operation with the workstation workflow.

[0084] like Figure 2 As shown, each functional station of the multi-station parallel grafting execution module realizes the corresponding grafting process function through the collaboration of functional units and sensing units. The time synchronization between stations is achieved through a distributed collaborative control system. The specific steps are as follows:

[0085] S1, Rootstock Loading Station: Through the collaboration of the flexible clamping unit and the posture pre-correction unit, the rootstock seedlings are accurately grasped and positioned for loading, and the loading signal is fed back to the collaborative control system.

[0086] Specifically, the flexible clamping unit adopts a biomimetic flexible gripper, and the attitude pre-correction unit consists of two micro servo motors and a correction platform. The correction platform is controlled to achieve rotational adjustment in the XY axis direction. During feeding, the rootstock feeding device transports the rootstock seedling to the gripping position. The feeding vision unit collects images of the rootstock seedling, identifies its attitude deviation, and transmits the deviation data to the station slave controller. The station slave controller controls the attitude pre-correction unit to adjust the angle of the correction platform, and at the same time controls the flexible clamping unit to grip the rootstock seedling and accurately place it on the rootstock positioning fixture on the station tray. The positioning error is ≤±0.1mm. After feeding is completed, the station slave controller sends a feeding completion signal to the main controller to trigger the preparation of the next station action.

[0087] S2, Rootstock Cutting Station: Through the collaboration of the cutting control unit and the rootstock positioning unit, precise cutting of the rootstock is achieved, and the cutting parameters can be adaptively adjusted according to the seedling type;

[0088] Specifically, the cutting control unit consists of high-speed pneumatic shears and a cutting drive cylinder. The cutter head is made of carbide and rotates at 6000 rpm. The rootstock positioning unit uses a pneumatic clamping fixture with adjustable clamping pressure ranging from 0.3 to 0.8 N. During cutting, the workstation tray transports the rootstock to the cutting position, and the pneumatic clamping fixture clamps the rootstock. The workstation controller, based on the seedling type identified by the vision guidance and intelligent detection module, calls the corresponding cutting parameters. For example, the cutting angle for cleft grafting is 30°, and the cutting angle for grafting is 45°. The controller then controls the cutting drive cylinder to move the high-speed pneumatic shears to the cutting position, completing the precise cutting of the rootstock tip. After cutting, the pneumatic clamping fixture releases and sends a cutting completion signal to the main controller.

[0089] S3, Scion feeding station: The biomimetic adaptive flexible clamping and feeding module enables precise feeding and posture correction of scion seedlings, ensuring alignment with the axis of the rootstock.

[0090] Specifically, the biomimetic adaptive flexible clamping and seedling loading module is invoked. The scion feeding device transports the scion seedling to the gripping area. The loading vision unit collects images of the scion seedling, identifies the scion variety and diameter, and matches the corresponding clamping pressure threshold. After the gripper unit of the biomimetic adaptive flexible clamping and seedling loading module grips the scion seedling, the scion posture is corrected by the posture adjustment platform and transported to the cutting position of the scion cutting station to ensure that the central axis of the scion seedling is aligned with the central axis of the rootstock seedling, with an alignment error of ≤±0.05mm. After loading is completed, a feedback signal is sent to the main controller.

[0091] S4, Scion Cutting Station: Through the collaboration of the cutting drive unit and the attitude adjustment unit, the scion cutting surface and the rootstock cutting surface are precisely matched.

[0092] Specifically, the cutting drive unit uses a precision cutting tool driven by a servo motor, and the attitude adjustment unit has the same structure as the attitude pre-correction unit at the rootstock loading station, enabling multi-directional attitude adjustment of the scion. During cutting, the station receives rootstock cutting angle data from the main controller and controls the attitude adjustment unit to adjust the scion's attitude so that the angle between the scion's cutting surface and the rootstock's cutting surface is consistent. Subsequently, the cutting drive unit drives the cutting tool to complete the scion cutting, ensuring that the flatness error of the cutting surface is ≤0.01mm and the cutting surface fit is ≥95%.

[0093] S5, Grafting and Fitting Station: Through the collaboration of dual flexible clamping and coordination units and precision displacement control unit, combined with pressure feedback, the precise fitting of rootstock and scion is achieved;

[0094] Specifically, the dual flexible clamping collaborative unit consists of two flexible grippers, which respectively clamp the rootstock and scion; the precision displacement control unit is an XYZ three-axis electric slide, with a stroke of 60mm on each axis and a positioning accuracy of ≤±0.01mm; pressure feedback is achieved by fiber optic pressure sensors mounted on the flexible grippers. During bonding, a 3D vision camera acquires three-dimensional data of the cutting surfaces of the rootstock and scion, transmits it to the vision processor, calculates the bonding alignment position, and transmits the alignment position data to the main controller. The main controller controls the precision displacement control unit to drive the dual flexible clamping collaborative unit to move, so that the cutting surfaces of the rootstock and scion are precisely bonded. The bonding pressure is collected in real time during the bonding process to ensure that the bonding pressure is stable at 0.1-0.2N. After bonding is completed, a bonding completion signal is fed back.

[0095] S6. Fixed curing station: Through the collaboration of the fixed execution unit and the constant temperature curing unit, the grafting parts are firmly fixed and cured;

[0096] Specifically, the fixing unit employs an automatic grafting clip installation mechanism, including a reel-type feeder, a pneumatic pusher, and a servo pressing mechanism. The reel-type feeder can hold 500 grafting clips with a feeding speed of 10 clips / minute; the servo pressing mechanism has a pressing pressure of 0.5-1N; the constant temperature curing unit uses a carbon fiber infrared heating tube with a heating temperature adjustment range of 35-40℃ and a temperature control accuracy of ±1℃. During fixing and curing, the reel-type feeder delivers the grafting clips to the installation position, the pneumatic pusher pushes the grafting clips to the grafting bonding area, the servo pressing mechanism completes the pressing and installation of the grafting clips, and then the infrared heating tube is activated to perform constant temperature heating and curing of the grafting area. The heating time is set to 5-8 seconds depending on the seedling type. After curing is completed, a feedback signal is sent to the main controller.

[0097] S7, Quality Inspection Station: Through the collaboration of the vision inspection unit and the pressure inspection unit, a multi-dimensional comprehensive judgment of grafting quality is achieved, and the inspection results are uploaded to the collaborative control system in real time;

[0098] Specifically, the vision inspection unit consists of a 2D industrial camera and a 3D vision camera. The 2D industrial camera detects the installation position of the grafting clips, whether there are any omissions or misalignments, and whether there is any damage to the seedling. The 3D vision camera detects the grafting fit and maximum gap. The pressure detection unit uses a miniature pressure sensor to detect the compressive strength of the grafting site, with a detection pressure of 0.5N. During inspection, the 2D and 3D vision cameras acquire images and 3D data of the grafted seedlings, while the pressure sensor acquires compressive strength data. All detection data is transmitted to the vision processor and the main controller. The main controller comprehensively judges the grafting quality according to preset judgment criteria, generates a pass / fail signal, and uploads it to the system database in real time.

[0099] S8 Finished Product Unloading Station: Based on the quality inspection results, qualified seedlings and unqualified seedlings are sorted and transferred through the sorting unit.

[0100] Specifically, the sorting unit consists of pneumatic grippers, a finished product conveyor line, and a waste sorting mechanism. The clamping pressure is adjustable from 0.2 to 0.5 N. The finished product conveyor line uses a belt conveyor with a conveying speed of 0.5 m / s. The waste sorting mechanism uses a pneumatic pusher plate. During unloading, the workstation pallet transports the grafted seedlings to the unloading position. The workstation receives the quality inspection results from the main controller. If the seedling is qualified, the pneumatic gripper grabs the grafted seedling and places it on the finished product conveyor line, which then transports it to the finished product box. If the seedling is unqualified, the pneumatic pusher plate of the waste sorting mechanism pushes it to the waste box. After unloading is completed, the workstation pallet returns to its initial position, ready for the next round of operation.

[0101] The main controller of the distributed collaborative control system achieves time synchronization between each workstation. The main controller receives the completion signals of each workstation in real time through the EtherCAT bus and issues the action instructions of the next workstation according to the preset timing table. This ensures that the entire grafting process proceeds in an orderly manner in the order of "rootstock loading - rootstock cutting - scion loading - scion cutting - grafting and bonding - fixing and curing - quality inspection - finished product unloading". The action connection time of each workstation is ≤0.1s, realizing parallel and efficient operation.

[0102] The distributed multi-workstation collaborative control system achieves coordinated global scheduling and local execution through a three-level control architecture. The main control module is responsible for global system decision-making, timing planning, and multi-workstation collaborative scheduling. It receives operational status data from each module and issues global control commands. The workstation slave control modules receive commands from the main control module, drive the corresponding workstation execution units to complete specific process actions, collect sensor data from their workstations, and upload it to the main control module. The execution control module, based on the commands from the slave control modules, achieves precise execution of actions and closed-loop feedback. All control levels achieve real-time data interaction through high-speed communication links, ensuring the synchronization and accuracy of control commands.

[0103] Specifically, the hardware configuration and coordination logic of the three-level control architecture are as follows:

[0104] Main Control Module: The core functions of the main control module include receiving user operation commands from the software control layer, such as start, stop, and parameter configuration; formulating a global timing plan table based on the grafting process flow; receiving operating status data uploaded from each workstation by the control module, such as action completion signals, sensor data, and fault alarms; processing data through a collaborative scheduling algorithm and issuing global control commands, such as workstation flow commands and process parameter commands; and storing system operating data, such as grafting quantity, pass rate, and fault information.

[0105] Workstation Slave Control Module: Each functional workstation is equipped with a microcontroller as a slave controller. The slave controller integrates an EtherCAT slave controller, a CANopen controller, and a 16-bit AD acquisition module. The EtherCAT slave controller supports the CoE protocol and communicates with the master control module via the EtherCAT bus with a communication cycle of ≤1ms, enabling the reception of control commands and the uploading of status data. The CANopen controller supports the DS301 protocol with a communication rate of 500kbps, used to connect to the intelligent execution units of this workstation, such as servo motors and pneumatic actuators. The 16-bit AD acquisition module has 8 acquisition channels and a sampling frequency of 100Hz, used to acquire analog signals from sensors of this workstation, such as pressure sensors and temperature sensors.

[0106] The specific working logic of the control module at the workstation is as follows: it receives targeted control commands from the main control module, such as the cutting angle and clamping pressure of a certain workstation; it converts the control commands into drive signals that the execution unit can recognize, such as PWM signals and analog signals; it collects the action feedback signals and sensor data of the execution unit at this workstation and performs preprocessing such as filtering and calibration; it uploads the preprocessed status data to the main control module; when a fault is detected at this workstation, such as abnormal sensor signals or failure of the execution unit to act, it immediately sends a fault alarm signal to the main control module and performs an emergency braking action.

[0107] The execution control module consists of the execution unit drive circuit and feedback signal processing circuit for each workstation. The execution unit includes servo motor driver, pneumatic solenoid valve, shape memory alloy driver, heating controller, etc.; the feedback signal processing circuit includes signal filtering circuit, signal amplification circuit, and A / D conversion circuit, if not integrated into the slave controller.

[0108] The execution control module works as follows: it receives drive signals from the workstation control module and drives the corresponding execution unit to complete specific actions, such as servo motor rotation, cylinder extension / retraction, and heating tube heating; it collects data on the action execution status through feedback elements equipped in the execution unit, such as encoders, position sensors, and temperature sensors; the feedback data is transmitted to the feedback signal processing circuit, where it undergoes noise reduction, amplification, and other processing before being transmitted to the workstation slave control module, forming a closed-loop control of the action. For example, after receiving the position command from the slave controller, the servo motor driver drives the servo motor to rotate. The encoder collects the motor rotation angle data in real time and feeds it back to the driver and slave controller. The slave controller adjusts the drive command based on the feedback data to ensure the positioning accuracy of the servo motor.

[0109] Real-time data interaction is achieved across all control levels via high-speed communication links: the main control module communicates with the workstation slave control module via an EtherCAT bus at a transmission rate of 100Mbps; the workstation slave control module communicates with the execution control module via a CANopen bus or analog link; and the main control module communicates with the software control layer via industrial Ethernet. The communication links utilize shielded cables, providing electromagnetic interference resistance and ensuring the synchronization of control commands and the accuracy of data transmission in industrial environments. Command transmission latency is ≤1ms, and data transmission success rate is ≥99.99%.

[0110] The hybrid communication network ensures the real-time and reliable data interaction of each module through a layered communication architecture. The main control layer and the software layer use industrial Ethernet to issue commands and upload status; the main control layer and the workstation control layer use a real-time bus to achieve synchronous transmission of control signals from multiple workstations; the workstation control layer and the execution / sensing unit use an industrial fieldbus and analog link to achieve real-time data interaction; the entire communication network integrates communication status monitoring and redundancy switching functions to ensure the continuity of system communication and provide communication support for multi-module collaboration.

[0111] Specifically, the layered architecture, communication protocols, and security mechanisms of hybrid communication networks are as follows:

[0112] Layer 1: Communication between the main control layer and the software layer. The main control layer is centered on the main controller, while the software layer includes the human-machine interface and the host computer monitoring system. The two communicate via industrial Ethernet, using Category 6 shielded network cable as the transmission medium, with a transmission rate of 1Gbps.

[0113] The communication adopts a client-server model, with the main controller acting as the server, with the IP address set to 192.168.0.1 and the port number 502; the software layer's human-computer interaction interface and the host computer act as clients, with the IP addresses set to 192.168.0.2 and 192.168.0.3 respectively.

[0114] The communication protocol uses ModbusTCP, which is a standard communication protocol in the industrial field and has good compatibility and reliability.

[0115] The specific communication content includes the software layer sending user operation commands and parameter modification commands downwards; and the main control layer sending system operating status data, fault alarm data, and real-time sensor data upwards. The communication latency at this level is ≤10ms, which meets the transmission requirements of non-real-time commands and status data.

[0116] The second layer: Communication between the main control layer and the workstation control layer. The main control layer is still centered on the main controller, while the workstation control layer consists of 8 workstation slave controllers. The two communicate via an EtherCAT real-time bus, using shielded twisted-pair network cable as the transmission medium, with a transmission rate of 100Mbps.

[0117] The EtherCAT bus adopts a master-slave communication architecture. The master controller acts as the EtherCAT master station, and the eight slave controllers are numbered sequentially (1#-8#) as slave stations, with slave addresses set to 0x01-0x08 respectively. The master station uses the APWR command to write synchronous data to all slave stations, including control commands and process parameters for each station; and uses the APRD command to read synchronous data from all slave stations, including the operating status and sensor data for each station.

[0118] The bus cycle of this level is ≤1ms and the synchronization accuracy is ≤100ns, which can ensure the synchronous transmission of control signals from multiple workstations and the real-time upload of status data, meeting the real-time requirements of multi-workstation collaborative control.

[0119] The third layer: the workstation control layer communicates with the execution / sensing units. This layer uses both CANopen bus and analog link communication in parallel. The CANopen bus is used to connect intelligent execution units, such as servo motors, pneumatic actuators, and automatic mounting mechanisms, etc. The transmission medium is shielded twisted-pair network cable, and the communication rate is 500kbps.

[0120] The workstation controller acts as the CANopen master station, and each intelligent actuator acts as a slave station. Real-time data interaction is achieved using PDOs (Power Directional Analysis), with a transmission cycle of ≤1ms. PDO data includes control commands and status feedback from the actuators. Non-real-time parameter configuration, such as the electronic gear ratio of a servo motor and the stroke parameters of a pneumatic actuator, is implemented using SDOs. Analog links connect analog sensors, such as fiber optic pressure sensors, temperature sensors, and displacement sensors, employing differential input with an input signal range of 4-20mA. Signal conversion is performed by the workstation controller's 16-bit AD acquisition module, with a sampling frequency of 100Hz. The acquired data is then digitally filtered before being used for control decisions.

[0121] Communication status monitoring and redundancy switching function: The entire hybrid communication network integrates a network management module, which consists of a communication status monitoring chip and a redundancy switching circuit, and is integrated into the main controller and the slave controllers of each workstation.

[0122] The communication status monitoring chip monitors the signal strength, transmission rate, and bit error rate of each communication link in real time. When a fault is detected in a communication link, a fault signal is immediately sent to the main controller. The redundancy switching circuit sets up redundant links for critical communication links. When the main link fails, the redundancy switching circuit automatically switches to the redundant link within 10ms to ensure the continuity of communication.

[0123] Meanwhile, the network management module has a fault diagnosis function, which can locate the specific location of the faulty link and upload the fault information to the main controller. The main controller displays the fault type and handling suggestions through the human-machine interface, which facilitates timely troubleshooting by operators.

[0124] The visual guidance and intelligent detection module provides full-process visual support through multi-vision unit collaboration and data fusion. The feeding vision unit collects seedling image data, providing visual basis for seedling identification, diameter measurement, and posture correction; the grafting and bonding vision unit collects three-dimensional data of the cutting surface, providing positional guidance for the precise bonding of rootstock and scion; the quality detection vision unit collects post-grafting appearance and three-dimensional bonding data, providing detection basis for quality judgment; after the data from each vision unit is analyzed and processed by the vision processing module, control guidance signals and quality judgment results are output and transmitted to the distributed collaborative control system, realizing the synergy between visual guidance and system control.

[0125] Specifically, the hardware configuration, installation layout, and data processing flow of the multi-vision unit are as follows:

[0126] Visual unit for loading seedlings: One set is set up at each of the rootstock loading station and the scion loading station. Each set includes a 2D industrial camera, a fixed-focus industrial lens, a ring light source, and a lens cleaning mechanism. The fixed-focus industrial lens clearly captures the stem area of ​​the seedling. The ring light source is a white LED light source with adjustable light intensity. The light source is coaxially arranged with the camera and installed 15cm below the camera to ensure the clarity and contrast of the seedling image. The lens cleaning mechanism adopts a high-pressure air blowing method, equipped with a miniature air pump and a solenoid valve. It cleans once every 100 acquisition actions, with a cleaning time of 0.1s, to avoid lens contamination affecting image quality.

[0127] Installation position of the feeding vision unit: The camera is fixed above the feeding station, with the lens pointing vertically downwards at the seedling grasping area, and the shooting range covers the feeding positioning area of ​​the seedling stem.

[0128] The data processing flow is as follows: After the camera acquires images of the seedlings, it transmits them to the vision processor via gigabit Ethernet; the vision processor preprocesses the images; it extracts the contour features of the seedling stems for seedling variety identification and diameter measurement; it extracts the stem centerline for posture deviation calculation; and it transmits the identification results, diameter data, and posture deviation data to the main controller of the distributed collaborative control system to provide visual basis for feeding positioning and posture correction.

[0129] Grafting and bonding vision unit: Set on the side of the grafting and bonding station, including a 3D vision camera, a laser light source and a protective housing. The 3D vision camera accurately collects the three-dimensional data of the cutting surfaces of the rootstock and scion; the laser light source is a red laser light source supporting the camera, with a power of 20 mW, and the light intensity can be adjusted adaptively according to the ambient light intensity; the protective housing is made of aluminum alloy, with dimensions of 300 mm × 200 mm × 150 mm, equipped with a transparent acrylic observation window, and a dust removal fan and a heating and defogging device are provided inside. The air volume of the dust removal fan is 5 CFM, and the heating power of the heating and defogging device is 50 W, ensuring that the 3D vision camera works normally in the dust and water vapor environment generated during the grafting operation.

[0130] Installation position of the grafting and bonding vision unit: Fixed on the bracket of the grafting and bonding station, the lens forms a 45° angle with the cutting surfaces of the rootstock and scion, and the shooting range covers the bonding area of the two cutting surfaces.

[0131] Its data processing flow is as follows: The 3D vision camera collects the three-dimensional point cloud data of the cutting surface and transmits it to the vision processor; the vision processor preprocesses the three-dimensional point cloud data; extracts the contour point clouds of the two cutting surfaces, calculates the target position for alignment and bonding; transmits the target position data to the main controller, and the main controller controls the precision displacement control unit to drive the double flexible clamping and collaborative unit to move, realizing the precise bonding of the rootstock and scion.

[0132] Quality inspection vision unit: Set above and on the side of the quality inspection station, including a 2D industrial camera and a 3D vision camera.

[0133] The parameters of the 2D industrial camera are the same as those of the loading vision unit. It is installed above the quality inspection station, and the lens is vertically downward aiming at the grafted seedling, used to collect the image of the installation position of the grafting clip and the appearance image of the seedling body; the parameters of the 3D vision camera are the same as those of the grafting and bonding vision unit. It is installed on the side of the quality inspection station, and the lens forms a 45° angle with the grafting and bonding part, used to collect the three-dimensional data of the grafting and bonding part.

[0134] Its data processing flow is as follows: After the 2D industrial camera collects the image, the vision processor analyzes the installation position of the grafting clip, judges whether there is missing installation or misinstallation, and at the same time detects whether there is damage to the stem or leaves of the seedling body; after the 3D vision camera collects the three-dimensional point cloud data, the vision processor calculates the grafting bonding degree and the maximum gap; fuses the 2D detection result and the 3D detection result, comprehensively judges the grafting quality; transmits the quality judgment result to the main controller, providing a basis for the sorting action of the finished product unloading station.

[0135] Vision processing module: An industrial-grade computer is used as the vision processor. The vision processor is connected to the cameras of the three vision units via Gigabit Ethernet, with a data transmission rate of ≥1Gbps. It is also connected to the main controller of the distributed collaborative control system via industrial Ethernet to realize the real-time uploading of detection results and the interaction of control commands, with an image processing latency of ≤100ms.

[0136] Deep learning algorithms and 3D vision detection algorithms provide algorithmic support for the visual guidance and intelligent detection modules. Deep learning algorithms enable accurate identification of seedling varieties and precise calculation of posture deviations, outputting posture adjustment commands. 3D vision detection algorithms process point cloud data of grafting and bonding areas, calculate bonding degree and maximum gap, and output quality judgment results. The algorithm modules interact in real time with the visual acquisition module and the distributed collaborative control system to achieve closed-loop collaboration of "image acquisition - algorithm processing - control feedback," ensuring the accuracy of visual guidance and quality detection.

[0137] Specifically, the detailed implementation process, formulas, and parameter explanations of deep learning algorithms and 3D vision detection algorithms are as follows:

[0138] The deep learning algorithms include seedling variety recognition algorithms and seedling posture deviation calculation algorithms, both of which are based on seedling image data collected by the feeding vision unit of the visual guidance and intelligent detection module.

[0139] The seedling variety identification algorithm is implemented using a lightweight MobileNetV3 CNN model, suitable for real-time identification needs in industrial settings. The specific process is as follows:

[0140] Step 1: Image Preprocessing. First, the acquired color seedling images are converted to grayscale using a weighted average grayscale conversion formula:

[0141]

[0142] in, Grayscale value , , These represent the pixel values ​​of the red, green, and blue channels of the color image, respectively. This formula effectively preserves the texture features of the seedling stem. Subsequently, Gaussian filtering is performed for noise reduction. The convolution kernel size for the Gaussian filter is set to 3×3, and the Gaussian function is:

[0143]

[0144] in, These are the pixel coordinates within the convolution kernel. Standard deviation is set here. =1.0, which can effectively remove Gaussian noise in images.

[0145] Finally, adaptive threshold segmentation is performed. The OTSU algorithm is used to automatically calculate the segmentation threshold, separating the seedling stem region from the background to obtain a binary image. The segmentation formula is as follows:

[0146]

[0147]

[0148] in, For pixel values ​​in a binary image, The segmentation threshold is calculated using the OTSU algorithm. This step allows for the precise extraction of the seedling stem region.

[0149] Step 2: Feature Extraction. The preprocessed binary image is scaled to 224×224 pixels and input into the feature extraction network of the MobileNetV3 lightweight CNN model.

[0150] The structure of the feature extraction network is as follows:

[0151] Input layer (224×224×1) → Convolutional layer 1 (3×3 kernels, stride 2, 16 output channels) → Batch normalization layer 1 → ReLU activation function layer 1 → Bottleneck layer 1 (16 layers, using inverted residual structure, expansion factor 1) → Batch normalization layer 2 → ReLU activation function layer 2 → Bottleneck layer 2 (24 layers, expansion factor 6) → Batch normalization layer 3 → ReLU activation function layer 3 → Bottleneck layer 3 (40 layers, expansion factor 6) → Batch normalization layer 4 → ReLU activation function layer 4 → Bottleneck layer 4 (48 layers, expansion factor 6) → Batch normalization layer 5 → ReLU activation function layer 5 → Bottleneck layer 5 (96 layers, expansion factor 6) → Batch normalization layer 6 → ReLU activation function layer 6 → Convolutional layer 2 (1×1 kernel, stride 1, 1280 output channels) → Global average pooling layer.

[0152] Through the feature extraction network, a 1×1×1280 feature vector is finally obtained, which contains key information such as the contour features and texture features of the seedling stem.

[0153] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A flexible seedling clamping and multi-station parallel grafting collaborative control system, characterized in that, It includes a biomimetic adaptive flexible clamping and seedling loading module, a multi-station parallel grafting execution module, a distributed multi-station collaborative control system, and a vision guidance and intelligent detection module. Each module achieves data interaction and collaborative work through a high-speed communication link. The biomimetic adaptive flexible clamping and seedling loading module adopts a three-finger biomimetic structure, integrating a shape memory alloy driving unit, an elastic buffer unit, and a fiber optic pressure sensor array. It achieves damage-free clamping and precise seedling loading through a pressure threshold adaptive and fuzzy PID composite control algorithm. The formula for the fuzzy PID composite control algorithm is as follows: in, , , These are the initial proportional coefficient, integral coefficient, and derivative coefficient of the PID controller. , , These are the proportional coefficient correction, integral coefficient correction, and derivative coefficient correction values ​​output by the fuzzy controller, respectively. , , The input quantity is pressure deviation. Change rate of pressure deviation , , , To preset the clamping pressure threshold, Real-time clamping pressure collected by fiber optic pressure sensor array; The distributed multi-station collaborative control system adopts a three-level control architecture consisting of a master controller, station slave controllers, and execution units. It incorporates a hybrid communication network built using a time-sensitive network and an EtherCAT bus. Through timing planning, conflict detection, and dynamic compensation collaborative scheduling algorithms, it achieves multi-station action synchronization and interference avoidance. The dynamic compensation algorithm formula is as follows: in, This is for the timing compensation of subsequent workstation actions. Delay time for the preceding workstation's actions. For compensation coefficient, This refers to the positioning deviation of the preceding workstation. The visual guidance and intelligent detection module adopts a multi-camera distributed layout, uses deep learning algorithms to identify seedling varieties and detect postures, and combines 3D visual detection algorithms to detect grafting fit accuracy.

2. The seedling flexible clamping and multi-station parallel grafting collaborative control system according to claim 1, characterized in that, The gripper unit of the biomimetic adaptive flexible clamping seedling module has adaptive clamping and rapid adaptation functions. It can achieve non-damaging clamping of seedlings through a flexible contact layer and adapt to seedlings of different diameters through a quick-change clamping adaptation structure. The gripper unit works in conjunction with the shape memory alloy drive module and the elastic buffer module. Through the pressure sensing feedback of the flexible contact layer, and in conjunction with the precise drive adjustment of the shape memory alloy drive module, the gripping force is dynamically adapted to meet the gripping needs of different varieties of seedlings.

3. The seedling flexible clamping and multi-station parallel grafting collaborative control system according to claim 1, characterized in that, The shape memory alloy drive unit provides precise driving power for the biomimetic adaptive flexible clamping seedling module. It can receive PWM drive commands from the clamping control module and control the extension and retraction of the drive components through precise current adjustment, thereby driving the gripper unit to complete the opening and closing action. The drive unit integrates a friction suppression structure and a stable transmission mechanism to ensure the smoothness and repeatability of the drive action and ensure the precise synchronization of the clamping action.

4. The seedling flexible clamping and multi-station parallel grafting collaborative control system according to claim 1, characterized in that, The elastic buffer unit is used to achieve active buffering and pressure feedback during the clamping process. It can offset the rigid impact during the clamping process through an adjustable buffering mechanism to avoid damage to the seedling. At the same time, it integrates a distributed pressure sensing module to collect the pressure data of the contact between the clamp and the seedling in real time. After converting the pressure signal into a standard control signal, it is transmitted to the clamping control module to provide real-time pressure feedback data for the fuzzy PID composite control algorithm. Together with the buffering guidance mechanism, it ensures the stability of the clamping posture.

5. The seedling flexible clamping and multi-station parallel grafting collaborative control system according to claim 1, characterized in that, The multi-station parallel grafting execution module has the functions of orderly flow of multiple stations and rapid process adaptation. It realizes the intermittent and precise flow of each station carrier through the circular flow module, and ensures the positioning accuracy of each station in conjunction with the station positioning module. The execution module integrates modular functional units, and each functional unit can be quickly switched according to the grafting process requirements. Through signal interaction with the distributed collaborative control system, the coordinated linkage of each functional unit and station flow is realized, ensuring the orderly progress of the entire grafting process.

6. The seedling flexible clamping and multi-station parallel grafting collaborative control system according to claim 5, characterized in that, Each functional workstation achieves its corresponding grafting process function through the collaboration of functional units and sensing units. The workstations are synchronized in time through a distributed collaborative control system. The specific steps are as follows: S1, Rootstock Loading Station: Through the collaboration of the flexible clamping unit and the posture pre-correction unit, the rootstock seedlings are accurately grasped and positioned for loading, and the loading signal is fed back to the collaborative control system. S2, Rootstock Cutting Station: Through the collaboration of the cutting control unit and the rootstock positioning unit, precise cutting of the rootstock is achieved, and the cutting parameters can be adaptively adjusted according to the seedling type; S3, Scion feeding station: The biomimetic adaptive flexible clamping and feeding module enables precise feeding and posture correction of scion seedlings, ensuring alignment with the axis of the rootstock. S4, Scion Cutting Station: Through the collaboration of the cutting drive unit and the attitude adjustment unit, the scion cutting surface and the rootstock cutting surface are precisely matched. S5, Grafting and Fitting Station: Through the collaboration of dual flexible clamping and coordination units and precision displacement control unit, combined with pressure feedback, the precise fitting of rootstock and scion is achieved; S6. Fixed curing station: Through the collaboration of the fixed execution unit and the constant temperature curing unit, the grafting parts are firmly fixed and cured; S7, Quality Inspection Station: Through the collaboration of the vision inspection unit and the pressure inspection unit, a multi-dimensional comprehensive judgment of grafting quality is achieved, and the inspection results are uploaded to the collaborative control system in real time; S8 Finished Product Unloading Station: Based on the quality inspection results, qualified seedlings and unqualified seedlings are sorted and transferred through the sorting unit.

7. The seedling flexible clamping and multi-station parallel grafting collaborative control system according to claim 1, characterized in that, The distributed multi-station collaborative control system achieves the coordination of global scheduling and local execution through a three-level control architecture, which consists of a master control module, a slave control module, and an execution control module. The main control module is responsible for global system decision-making, timing planning, and multi-station collaborative scheduling. It receives operational status data from each module and issues global control commands. The slave control module receives commands from the main control module, drives the execution unit of the corresponding station to complete specific process actions, collects sensor data of the station and uploads it to the main control module. The execution control module implements precise execution and closed-loop feedback of actions according to the commands of the slave control module. Each control level achieves real-time data interaction through a high-speed communication link to ensure the synchronization and accuracy of control commands.

8. The seedling flexible clamping and multi-station parallel grafting collaborative control system according to claim 1, characterized in that, The hybrid communication network ensures the real-time and reliable data interaction of each module through a layered communication architecture. The main control layer and the software layer use industrial Ethernet to issue commands and upload status. The main control layer and the workstation control layer use a real-time bus to achieve synchronous transmission of control signals from multiple workstations. The workstation control layer and the execution sensing unit use an industrial fieldbus and analog signal link to achieve real-time data interaction. The entire communication network integrates communication status monitoring and redundancy switching functions to ensure the continuity of system communication and provide communication support for multi-module collaboration.

9. The seedling flexible clamping and multi-station parallel grafting collaborative control system according to claim 1, characterized in that, The visual guidance and intelligent detection module achieves full-process visual support through multi-vision unit collaboration and data fusion. The feeding vision unit collects seedling image data to provide visual basis for seedling identification, diameter measurement, and posture correction. The grafting and bonding vision unit collects three-dimensional data of the cutting surface to provide position guidance for the precise bonding of rootstock and scion. The quality detection vision unit collects the appearance and three-dimensional bonding data after grafting to provide detection basis for quality judgment. After the data from each vision unit is analyzed and processed by the vision processing module, control guidance signals and quality judgment results are output and transmitted to the distributed collaborative control system to realize the collaboration between visual guidance and system control.

10. The flexible seedling clamping and multi-station parallel grafting collaborative control system according to claim 1, characterized in that, The deep learning algorithm and 3D vision detection algorithm provide algorithmic support for the visual guidance and intelligent detection module. The deep learning algorithm enables accurate identification of seedling varieties and accurate calculation of posture deviations, and outputs posture adjustment commands. The 3D vision detection algorithm processes point cloud data of the grafting and bonding parts, calculates the bonding degree and maximum gap, and outputs quality judgment results. The algorithm module interacts in real time with the visual acquisition module and the distributed collaborative control system to achieve closed-loop collaboration of image acquisition, algorithm processing and control feedback, ensuring the accuracy of visual guidance and quality detection.

Images