An edge-computing-based intelligent meat cutting system cooperative control method and device

Abstract

The application discloses an intelligent meat cutting system based on edge computing, and a collaborative control method and device thereof, and belongs to the technical field of food processing equipment. The method comprises the following steps: three-dimensional point cloud data acquisition and preprocessing; distributed clock synchronization based on a time-sensitive network; modeling a multi-degree-of-freedom mechanical arm and a high-precision intelligent cutting machine as an intelligent agent, and generating a collaborative cutting strategy by solving Nash equilibrium through multi-agent game; triggering the dynamic reconstruction of cutting parameters by an impedance control model through high-frequency force feedback during the cutting process; and performing secondary scanning after the cutting is completed, and adopting a deep reinforcement learning closed-loop optimization game model. The device comprises a three-dimensional vision acquisition module, an edge computing and game decision module, a multi-actuator collaboration module, a high-frequency force feedback module, a secondary scanning and closed-loop optimization module and a control unit. The application realizes microsecond-level collaborative synchronization and millisecond-level dynamic reconstruction, significantly improves cutting precision, efficiency and raw material utilization, and promotes the intelligent upgrading of food processing.

Description

Technical Field

[0001] This invention belongs to the field of food processing equipment technology, specifically relating to a collaborative control method and device for an intelligent meat cutting system based on edge computing, which is particularly suitable for adaptive high-precision cutting of irregularly shaped and easily deformable meats such as chicken breast and fish. Background Technology

[0002] The meat processing industry is undergoing a transformation from labor-intensive to technology-intensive. Currently, meat cutting mainly relies on manual labor or semi-automated equipment, and faces the following technological bottlenecks:

[0003] Low efficiency: Manually positioning the direction of cutting meat is labor-intensive and difficult to operate continuously for a long time;

[0004] Poor precision: Manual operation is greatly affected by skill level, resulting in poor stability of cutting precision;

[0005] Poor adaptability: Traditional 2D vision-guided cutting cannot perceive the three-dimensional shape of food (such as thickness and curvature), which can easily cause over-cutting of irregularly shaped or easily deformable food, resulting in a high loss rate.

[0006] Poor coordination: In existing equipment, the robotic arm and the cutting machine adopt a master-slave control mode, which results in a large delay in coordination response and makes it difficult to achieve high-speed and high-precision cutting.

[0007] While some existing technologies incorporate machine vision and industrial robots, for example, patent document CN116897978A discloses an automatic meat cutting device based on dual collaborative robotic arms. The device uses a camera to locate the position of the meat pieces, and a controller controls two collaborative robotic arms to grab the meat pieces and cut them using a cutting machine.

[0008] However, the solution still has the following shortcomings: it uses 2D vision and cannot obtain the three-dimensional shape information of the meat; there is a lack of deep collaboration mechanism between the robotic arm and the cutting machine; there is no real-time force feedback and adaptive adjustment capability; and it cannot dynamically optimize the cutting strategy according to the individual differences of the meat.

[0009] Therefore, there is an urgent need for a high-precision, high-efficiency, and highly adaptable intelligent cutting control method and device. Summary of the Invention

[0010] The present invention aims to solve the following problems existing in the prior art: (1) 2D vision cannot perceive the three-dimensional shape of meat, resulting in limited cutting accuracy; (2) The collaborative control accuracy between the robotic arm and the cutting machine is insufficient, and the response delay is large; (3) There is a lack of adaptive ability to the time-varying characteristics of the stiffness of meat raw materials; (4) The cutting strategy is fixed and cannot be dynamically optimized according to individual differences, resulting in low raw material utilization.

[0011] To achieve the above objectives, this invention provides a collaborative control method for an intelligent meat cutting system based on edge computing, comprising the following steps:

[0012] Step S1: 3D point cloud data acquisition and preprocessing: The meat to be cut on the conveyor belt is scanned by the 3D vision module to obtain high-precision 3D point cloud data, which is then transmitted to the edge computing unit for noise reduction and feature extraction to generate a complete 3D model of the meat.

[0013] Specifically, the 3D vision module uses a high-resolution 3D laser camera (resolution not less than 0.1mm) in conjunction with an industrial camera. When meat enters the detection area, scanning is automatically triggered to collect the raw material's three-dimensional point cloud data. The edge computing unit incorporates a point cloud preprocessing model, including: a noise filtering module that uses statistical filtering algorithms to remove environmental interference points and outliers; a point cloud completion module that uses a generative adversarial network (GAN) to complete missing areas caused by occlusion or reflection; and a feature extraction module used to identify key features of the meat, such as edges, corners, and thickness variations.

[0014] Step S2: Distributed clock synchronization based on time-sensitive networking: A unified clock domain is established between the control unit, the robotic arm controller and the cutting machine controller through a time-sensitive networking (TSN) switch to achieve microsecond-level task synchronization between each actuator.

[0015] The TSN achieves clock synchronization between controllers based on the IEEE 802.1AS protocol (gPTP), with a synchronization accuracy of 1 microsecond. The communication cycle is configured through a Time-Aware Shaper (TAS), dividing the cycle into multiple time slots. These time slots are assigned deterministic transmission time slots for multi-degree-of-freedom robotic arm control commands, high-precision intelligent cutting machine status feedback, and visual data streams, ensuring coordinated consistency during high-speed movement.

[0016] Step S3: Multi-agent game-theoretic cutting planning: The multi-degree-of-freedom robotic arm and the high-precision intelligent cutting machine are modeled as intelligent agents with autonomous decision-making capabilities. A value function matrix based on game theory is constructed. Combined with preset cutting parameters and a three-dimensional model of meat, a collaborative strategy for the multi-degree-of-freedom robotic arm posture adjustment sequence and the high-precision intelligent cutting machine feed trajectory is generated by solving the Nash equilibrium.

[0017] The multi-agent game model includes: a first agent (a multi-degree-of-freedom robotic arm) whose value function aims to minimize attitude adjustment time and energy consumption; and a second agent (a high-precision intelligent cutting machine) whose value function aims to maximize cutting accuracy and cutting speed. A cross-sensing term is introduced into the value function matrix, enabling the multi-degree-of-freedom robotic arm to predict the actions of the high-precision intelligent cutting machine and respond in advance, while the high-precision intelligent cutting machine can also sense the attitude deviations of the multi-degree-of-freedom robotic arm and adjust its feed strategy. The Nash equilibrium solution process is completed in the edge computing unit using the Alternating Direction Multiplier Method (ADMM), with a solution time of less than 50ms, meeting real-time control requirements.

[0018] Step S4: Dynamic Reconstruction and Impedance Control Integration of Cutting Task: During the cutting process, force feedback data is collected in real time by a high-frequency force feedback module installed at the end of the multi-degree-of-freedom robotic arm and the cutter head of the high-precision intelligent cutting machine. When abnormal fluctuations in cutting resistance are detected, the impedance control model is triggered to dynamically reconstruct the cutting parameters, and the support stiffness of the multi-degree-of-freedom robotic arm and the feed speed of the high-precision intelligent cutting machine are adjusted in real time to achieve collaborative operation of multiple actuators and adaptive cutting.

[0019] The impedance control model includes: a stiffness adjustment module for dynamically adjusting the support stiffness of the multi-degree-of-freedom robotic arm end effector on the meat based on the cutting resistance; and a damping adjustment module for dynamically adjusting the feed speed of the high-precision intelligent cutter based on the cutting resistance. When the detected cutting resistance exceeds a preset threshold, the system reconstructs and sends the cutting parameters within 1ms. The impedance control model employs a neural network adaptive algorithm, using force error and its rate of change as input, to update stiffness and damping parameters online, achieving adaptive compensation for the time-varying stiffness characteristics of the meat raw material.

[0020] Step S5: Secondary scanning and closed-loop optimization after cutting: After cutting, the finished product is scanned a second time through the vision module, and the cutting deviation data is fed back to the edge computing unit to update the value function matrix of the game model, thereby realizing iterative optimization of system performance.

[0021] The closed-loop optimization employs a deep reinforcement learning algorithm, using cutting deviations (such as size errors and weight errors) as input to the reward function. The value function matrix of the game model is updated through the policy gradient method, enabling the system to continuously optimize the cutting strategy during continuous operation.

[0022] The cutting parameters include one or more combinations of fixed-width cutting, fixed-length cutting, fixed-weight cutting, equal-weight bisecting cutting, and surface-area similar cutting. Taking fixed-size cutting of chicken breast as an example, the target dimensions are a length range of Lmm (allowable range L1-L2) and a width of Dmm (allowable tolerance ±0.5mm), and the optimization objective is to maximize the number of qualified meat strips.

[0023] The present invention also provides a collaborative control device for an intelligent meat cutting system based on edge computing, comprising:

[0024] 3D Vision Acquisition Module: Includes a 3D laser camera and triggering unit, used to acquire 3D point cloud data of meat raw materials. The 3D laser camera has a resolution of no less than 0.1mm and a scanning speed of no less than 30 frames / second, which can meet the needs of high-speed production lines.

[0025] Edge computing and game decision-making module: This module integrates a time-sensitive network synchronization unit, a multi-agent game solver, and an impedance control model to generate collaborative cutting strategies. It also incorporates a built-in FPGA hardware acceleration unit to accelerate real-time computation of point cloud preprocessing, game solving, and the impedance control model, ensuring an overall processing latency of less than 50ms.

[0026] Multi-actuator collaboration module: This module includes at least one multi-degree-of-freedom robotic arm and at least one high-precision intelligent cutting machine. Connected to the edge computing module via a time-sensitive network switch, it receives and executes collaborative cutting commands. The multi-degree-of-freedom robotic arm adjusts the posture and cutting direction of the meat, while the high-precision intelligent cutting machine executes the cutting task according to the generated cutting path. The collaborative control latency between the two is less than 5ms.

[0027] High-frequency force feedback module: Installed at the end of the multi-degree-of-freedom robotic arm and at the cutter head of the high-precision intelligent cutting machine, it is used to collect force feedback data in real time during the cutting process. The force feedback module adopts a six-dimensional force sensor with a sampling frequency of not less than 10kHz and a force feedback accuracy of 0.1N, which can detect minute changes in cutting resistance.

[0028] The secondary scanning and closed-loop optimization module scans the finished product, collects cutting deviation data, and feeds it back to the game decision module for model iterative optimization. This module uses the same 3D vision acquisition hardware as step S1 to ensure data consistency.

[0029] Control unit: Used to coordinate the communication and timing of actions between modules through a time-sensitive network, and adopts a real-time industrial Ethernet communication protocol to ensure the deterministic response of the overall system.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] 1. High precision in collaborative control: Through TSN microsecond-level synchronization and multi-agent game collaboration, the collaborative control precision of the multi-degree-of-freedom robotic arm and high-precision intelligent cutting machine is improved to within ±0.3mm, which is superior to existing technologies;

[0032] 2. Fast response speed: The collaborative control latency is less than 5ms, supporting high-speed cutting of up to 20 times / second, which greatly improves production efficiency;

[0033] 3. High adaptability: It adopts three-dimensional vision and adaptive impedance control, which can effectively deal with individual differences of irregular and easily deformable meats such as chicken breast and fish, and supports rapid switching between multiple categories without the need for manual parameter adjustment;

[0034] 4. High raw material utilization rate: Through multi-agent game planning and closed-loop optimization, the cutting loss rate is reduced by more than 20%, and the raw material utilization rate is significantly improved, which greatly reduces costs for customers.

[0035] 5. Full-process automation: From 3D scanning, game theory planning, collaborative execution to closed-loop optimization, the entire process requires no manual intervention, saving a lot of labor and promoting the upgrade of food processing from experience-driven to data-driven. Attached Figure Description

[0036] The present application will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0037] Figure 1 This is a schematic diagram of the system structure according to an embodiment of the present invention;

[0038] Figure 2 This is a flowchart of a method according to an embodiment of the present invention;

[0039] Figure 3 This is a flowchart of point cloud preprocessing in an embodiment of the present invention;

[0040] Figure 4 This is a schematic diagram of a time-sensitive network synchronization architecture in an embodiment of the present invention;

[0041] Figure 5 This is a schematic diagram illustrating the principle of multi-agent game-theoretic collaborative control in an embodiment of the present invention.

[0042] Figure 6 This is a flowchart of the dynamic reconstruction of impedance control and cutting tasks in an embodiment of the present invention;

[0043] Figure 7 This is a schematic diagram of the closed-loop optimization principle in an embodiment of the present invention;

[0044] Figure 8 This is a schematic diagram illustrating an example of the application of the present invention to the fixed-size cutting of chicken breast.

[0045] The markings in the attached diagram are as follows:

[0046] 100. 3D Vision Acquisition Module; 101. 3D Laser Camera; 102. Trigger Unit; 200. Edge Computing and Game Theory Decision Module; 201. TSN Synchronization Unit; 202. Multi-Agent Game Solver; 203. Impedance Control Model; 204. FPGA Hardware Acceleration Unit; 300. Multi-Actuator Collaboration Module; 301. Multi-DOF Robotic Arm; 302. High-Precision Intelligent Cutting Machine; 400. High-Frequency Force Feedback Module; 401. Six-Dimensional Force Sensor; 402. Single-Dimensional Force Sensor; 500. Secondary Scanning and Closed-Loop Optimization Module; 600. Control Unit. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Those skilled in the art should understand that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0048] Example 1: System Overall Architecture:

[0049] like Figure 1 As shown, this embodiment provides a collaborative control device for an intelligent meat cutting system based on edge computing, including: a three-dimensional vision acquisition module 100, an edge computing and game decision-making module 200, a multi-actuator collaborative module 300, a high-frequency force feedback module 400, a secondary scanning and closed-loop optimization module 500, and a control unit 600.

[0050] The 3D vision acquisition module 100 includes a 3D laser camera 101 and a trigger unit 102, which are installed above the conveyor belt. When meat enters the detection area, it automatically triggers scanning to acquire the 3D point cloud data of the raw material.

[0051] The edge computing and game decision-making module 200 includes a TSN synchronization unit 201, a multi-agent game solver 202, an impedance control model 203, and an FPGA hardware acceleration unit 204. This module receives 3D point cloud data, performs point cloud preprocessing, game solving, and impedance control parameter generation, and connects to each actuator through a TSN switch.

[0052] The multi-actuator collaboration module 300 includes a multi-degree-of-freedom robotic arm 301, which in this embodiment is a six-axis collaborative multi-degree-of-freedom robotic arm 301, and also includes a high-precision intelligent cutting machine 302. The multi-actuator collaboration module 300 further integrates a high-performance servo motion control system for achieving high-precision tracking of the cutting path. This servo motion control system includes a servo driver, a high-resolution encoder, a motion controller, and a real-time communication interface. Its specific structure and working principle are as follows:

[0053] (1) Hardware components:

[0054] The servo drive uses a three-phase AC servo drive, supports the EtherCAT industrial Ethernet protocol, and is equipped with a three-closed-loop control structure consisting of a current loop, a speed loop, and a position loop. The position loop update frequency is 10kHz, the speed loop update frequency is 20kHz, and the current loop update frequency is 50kHz, ensuring a fast response in the control loop.

[0055] The high-resolution encoder employs a 23-bit absolute photoelectric encoder, mounted on the output shafts of each joint of the multi-degree-of-freedom robotic arm 301 and the feed shaft of the high-precision intelligent cutting machine 302. It achieves a resolution of 8,388,608 pulses per revolution, with position feedback accuracy of ±0.001° (joints) and ±0.001mm (linear axis). Encoder data is transmitted in real-time to the servo drive and motion controller via bidirectional serial communication.

[0056] The motion controller employs a high-performance motion control card based on an FPGA+DSP heterogeneous architecture, with built-in 512MB DDR3 memory and 4MB non-volatile storage, supporting synchronous control of up to 32 axes. This motion controller communicates in real-time with the edge computing and game decision-making module 200 via a TSN switch, receiving the desired trajectory data (including three-dimensional spatial position sequence, velocity sequence, and acceleration sequence) from the optimal cutting scheme.

[0057] (2) Control principle:

[0058] During the cutting process, the motion controller performs precise interpolation of the desired trajectory with a control cycle of 1ms, and uses a cubic spline interpolation algorithm to generate the position command sequence for each joint and tool axis. The interpolated commands are then sent to each servo driver via the EtherCAT bus using a distributed clock synchronization method, achieving a clock synchronization accuracy of 100ns.

[0059] To achieve high-precision tracking, the servo motion control system employs the following key technologies:

[0060] ① Feedforward + Feedback Composite Control: Velocity and acceleration feedforward based on a dynamic model are introduced into the forward channel to compensate for system inertial delay and friction effects. The velocity feedforward coefficient is calculated online based on the joint inertia, while the acceleration feedforward coefficient is dynamically adjusted according to the tool load. The feedback channel employs an improved PID controller, coupled with a second-order low-pass filter and a notch filter in series to suppress mechanical resonance and measurement noise, ensuring the position tracking error is controlled within ±0.03mm.

[0061] ② Contour Error Pre-compensation: For complex spatial cutting trajectories, the motion controller has a built-in contour error estimator that calculates the normal deviation based on the shortest distance between the real-time feedback position and the desired trajectory. A cross-coupled controller is used to decompose the contour error to each motion axis, dynamically adjusting the commands of each axis to control the contour error within ±0.05mm.

[0062] ③ Dynamic Friction Compensation: Based on the LuGre friction model, the frictional torque of the joint and guide rail is estimated online according to the current speed, acceleration, and temperature, and compensation is added to the feedforward channel. This mechanism effectively eliminates low-speed crawling and reversal spikes, ensuring a smooth and vibration-free cutting process.

[0063] ④ Load Adaptive Gain Scheduling: The servo motion control system monitors the current, torque, and speed feedback of each axis in real time. Combined with the force / torque data from the high-frequency force feedback module 400, it identifies load changes online (such as resistance fluctuations when cutting meat of different hardness). Based on the identification results, it automatically adjusts the PID gain parameters of the position loop and speed loop to ensure consistent dynamic response and tracking accuracy under different load conditions.

[0064] This embodiment uses a servo-driven band saw cutting machine. The end of the multi-degree-of-freedom robotic arm 301 is equipped with an electric gripper for gripping and adjusting the posture of the meat.

[0065] The high-frequency force feedback module 400 includes a six-dimensional force sensor 401 installed at the end of a multi-degree-of-freedom robotic arm 301 and a single-dimensional force sensor 402 installed on the cutter head of a high-precision intelligent cutting machine 302, with a sampling frequency of 10kHz and an accuracy of 0.1N.

[0066] The secondary scanning and closed-loop optimization module 500 adopts the same hardware configuration as the three-dimensional vision acquisition module 100, and is used to perform secondary scanning on the finished product after cutting and collect deviation data.

[0067] The control unit 600 serves as the central hub of the system, connecting to each module via a TSN switch and employing a real-time industrial Ethernet communication protocol to coordinate communication and action timing between modules.

[0068] Example 2: Detailed Method Flow:

[0069] like Figure 2 As shown, this embodiment provides a collaborative control method for an intelligent meat cutting system based on edge computing, using chicken breast cutting as an example for detailed explanation. The target dimensions are: length range 80-120mm, width 15mm±0.5mm, and the optimization objective is to maximize the number of qualified meat strips.

[0070] Step S1: 3D point cloud data acquisition and preprocessing:

[0071] When a piece of chicken breast enters the detection area via the conveyor belt, the trigger unit 102 activates the 3D laser camera 101 to scan and acquire raw point cloud data. The point cloud data is then uploaded to the edge computing and game theory decision-making module 200 and enters the point cloud preprocessing flow (e.g., ...). Figure 3 (as shown)

[0072] (1) Noise filtering: The statistical filtering algorithm is used to calculate the average distance between each point and its k nearest neighbors, and outliers with an average distance exceeding the threshold (μ±3σ in this embodiment) are removed to eliminate environmental interference.

[0073] (2) Point cloud completion: The filtered point cloud is input into the generative adversarial network (GAN) model, which has been pre-trained on a dataset containing 100,000 chicken breast point clouds. It can reasonably complete the missing areas caused by occlusion or reflection and generate a complete point cloud model.

[0074] (3) Feature extraction: Curvature analysis is used to identify key features of the meat, such as edges, inflection points, and thickness variations, providing input for subsequent game planning. The output of this step is a complete 3D model of the meat, containing approximately 500,000 3D points with uniform point cloud density and clear key features.

[0075] Step S2: Distributed clock synchronization based on time-sensitive networks:

[0076] like Figure 4 As shown, the control unit 600, the robotic arm controller, and the cutting machine controller are connected via a TSN switch, and clock synchronization is achieved based on the IEEE 802.1AS protocol. The specific synchronization process is as follows:

[0077] (1) Best master clock election: The control unit 600 with the highest accuracy is selected as the master clock through the BMCA algorithm.

[0078] (2) Link delay measurement: The peer delay mechanism is used to measure the link delay between the master clock and each slave clock.

[0079] (3) Clock calibration: PI closed-loop calibration is performed based on the measurement results to ensure that the clock deviation between each controller is less than 1 microsecond.

[0080] (4) Time-aware shaping: The communication cycle (1ms) is divided into 4 time slots: 0-200μs for transmitting multi-degree-of-freedom robotic arm control commands (priority 7), 200-400μs for transmitting high-precision intelligent cutting machine status feedback (priority 6), 400-600μs for transmitting visual data streams (priority 4), and 600-1000μs for transmitting best-effort traffic (priority 0). Guard bands are inserted between each time slot to prevent data packet collisions.

[0081] Step S3: Multi-agent game segmentation planning:

[0082] The multi-degree-of-freedom robotic arm 301 and the high-precision intelligent cutting machine 302 are modeled as two intelligent agents, and the game theory model is constructed as follows:

[0083] The value function of robotic arm agent A:

[0084]

[0085] The value function of the cutting machine agent B:

[0086]

[0087] The above two formulas are illustrated in Table 1 below;

[0088] Table 1: Value Functions of Robotic Arm Agent A and Cutting Machine Agent B (Symbol Explanation)

[0089]

[0090] Status tracking item:

[0091] The representations are shown in Table 2 below;

[0092] Table 2: Definitions of Status Tracking Item Symbols

[0093]

[0094] Physical meaning: This measure indicates whether a multi-degree-of-freedom robotic arm accurately follows a planned reference trajectory. If the multi-degree-of-freedom robotic arm deviates from the predetermined trajectory, this value increases, leading to increased costs.

[0095] Energy consumption items:

[0096] The representations are shown in Table 3 below;

[0097] Table 3: Explanation of symbols for energy consumption items:

[0098]

[0099] Physical meaning: This measure measures the control energy consumed by a multi-degree-of-freedom robotic arm. The more intense the control movements (faster speed, greater torque), the higher this value, and the higher the cost. This helps to avoid unnecessary violent movements and extend the life of the equipment.

[0100] Collaborative sensing items:

[0101] The representations are shown in Table 4 below;

[0102] Table 4: Explanation of symbols for collaborative sensing terms:

[0103]

[0104] Specific form:

[0105] The representations are shown in Table 5 below;

[0106] Table 5: Explanation of Cross-Sensing Function Symbols:

[0107]

[0108] Physical meaning: When a high-precision intelligent cutting machine accelerates or changes its motion... Increase Increased, due to the presence of the negative sign, the overall cost Reduce (equivalent to rewarding a multi-degree-of-freedom robotic arm).

[0109] This incentivizes the multi-degree-of-freedom robotic arm to proactively predict and coordinate with the movements of the high-precision intelligent cutting machine. For example, when the high-precision intelligent cutting machine is about to cut into a harder part, the multi-degree-of-freedom robotic arm increases its support stiffness in advance, instead of passively waiting for instructions.

[0110] V. Complete Technical Explanation:

[0111] In a meat-cutting scenario, this value function guides the actions of the multi-degree-of-freedom robotic arm 301:

[0112] 1. Precise tracking: The end effector of the multi-degree-of-freedom robotic arm 301 must accurately deliver the meat to the predetermined cutting position (minimize the first requirement);

[0113] 2. Energy-saving operation: While ensuring accuracy, minimize unnecessary vigorous movements (the second item is minimization).

[0114] 3. Active collaboration: The multi-degree-of-freedom robotic arm 301 should actively observe the movements of the high-precision intelligent cutting machine 302, predict its intentions, and cooperate in advance (maximizing the third item, i.e. minimizing costs).

[0115] Balancing the three elements: by adjusting the weight matrix Q A R A Sum of coefficients It can achieve a balance between tracking accuracy, energy consumption, and coordination.

[0116] In contrast, the value function of the cutting machine agent B is:

[0117]

[0118] The representations are shown in Table 6 below;

[0119] Table 6: Explanation of the symbols for the value function of the cutting machine agent B:

[0120]

[0121] Two-way collaborative mechanism: The value function of a multi-degree-of-freedom robotic arm contains This prompted them to actively cooperate with the high-precision intelligent cutting machine;

[0122] The value function of the cutting machine contains When the attitude deviation of the multi-degree-of-freedom robotic arm increases, the cost of the high-precision intelligent cutting machine increases, prompting it to adjust its feed strategy (such as deceleration) to wait for the multi-degree-of-freedom robotic arm 301 to correct.

[0123] This bidirectional coupling design enables true game-theoretic collaboration, rather than traditional one-way master-slave control.

[0124] Based on the 3D model generated in step S1 and the preset cutting parameters (width 15mm), the multi-agent game solver 202 uses the Alternating Direction Multiplier Method (ADMM) to solve for the Nash equilibrium. The specific steps are as follows:

[0125] Initialization: Set the initial posture u of the multi-DOF robotic arm 301. A (0) and the initial feed speed u of the high-precision intelligent cutting machine 302 B (0);

[0126] Iterative optimization: In the k-th iteration, fix u B (k-1) Optimization u A (k), then fix u A (k) Optimize u B (k);

[0127] Convergence criterion: When (In this embodiment, the value is 0.01) and when the number of iterations is ≤ Rmax (in this embodiment, the value is 20), it is considered that an approximate Nash equilibrium has been reached;

[0128] The definitions of the various symbols in the formula are shown in Table 7 below;

[0129] Table 7: Explanation of symbols in convergence judgment formulas:

[0130]

[0131] Explanation of the formula:

[0132] This formula is a convergence criterion in the solution of multi-agent games, used to determine whether the Alternating Direction Multiplier Method (ADMM) iteration has reached Nash equilibrium. The specific logic is as follows:

[0133] The numerator represents the change in the sum of the value functions of the two agents between the current iteration and the previous iteration. If the change is small, it indicates that the strategies of the two agents have stabilized and are no longer changing significantly.

[0134] Threshold judgment: When the absolute value of the change is less than the preset convergence threshold. When the iteration is considered to have converged, the iteration can be stopped and the current optimal control sequence can be output.

[0135] Physical meaning: In the meat cutting scenario, this criterion ensures that the collaborative strategy of the multi-degree-of-freedom robotic arm 301 and the high-precision intelligent cutting machine 302 has been stabilized, and subsequent iterations will no longer produce meaningful improvements, thereby saving computational resources while ensuring the accuracy of the solution.

[0136] Typical value: In embodiments of the present invention, the convergence threshold The value is 0.01.

[0137] Output: The optimal control sequence for the current iteration, with the first control variable taken as the real-time instruction.

[0138] The solution time is less than 50ms, and the output results include: the attitude adjustment sequence of the multi-degree-of-freedom robotic arm 301 (rotating the chicken breast to the optimal cutting direction so that its width direction is perpendicular to the feed direction of the high-precision intelligent cutting machine) and the feed trajectory of the high-precision intelligent cutting machine (including the cutting start point, end point, and feed speed curve).

[0139] Step S4: Integration of dynamic reconstruction and impedance control for the cutting task:

[0140] During the cutting process, the high-frequency force feedback module 400 collects force feedback data in real time. This embodiment adopts neural network adaptive impedance control. During the cutting process, force feedback data is collected in real time by a six-dimensional force sensor 401 installed at the end of the multi-degree-of-freedom robotic arm 301 and a single-dimensional force sensor 402 installed on the cutter head of the high-precision intelligent cutting machine 302. When abnormal fluctuations in cutting resistance are detected, the impedance control model is triggered to dynamically reconstruct the cutting parameters, and the support stiffness of the multi-degree-of-freedom robotic arm 301 and the feed speed of the high-precision intelligent cutting machine 302 are adjusted in real time to achieve collaborative operation of multiple actuators and adaptive cutting. The entire parameter reconstruction and distribution process is completed within 1ms, ensuring a smooth transition in the cutting process.

[0141] Step S5: Secondary scanning and loop closure optimization after cutting:

[0142] After cutting, the finished product enters the secondary scanning area via a conveyor belt. The secondary scanning and closed-loop optimization module 500 performs a 3D scan on the finished product and collects actual cutting size data. In this embodiment, the actual width of the cut meat strips is 15.2mm, and the lengths are 118mm, 105mm, and 92mm, all within the allowable range.

[0143] The cutting deviation data (width deviation + 0.2mm) is fed back to the edge computing and game decision-making module 200, and the value function matrix of the game model is updated using a deep reinforcement learning algorithm. Specifically:

[0144] (1) Define the reward function:

[0145] The definitions are as shown in Table 8.

[0146] Table 8: Explanation of Reward Function Symbols

[0147]

[0148] Explanation of the formula:

[0149] This reward function is used in deep reinforcement learning (such as Q-learning or PPO algorithms) to evaluate the quality of the cutting results. The specific logic is as follows:

[0150] The negative sign indicates a penalty mechanism; the greater the deviation, the greater the penalty (the smaller the reward value, the more negative it is).

[0151] Absolute value accumulation: Add up the absolute values ​​of all deviations, regardless of the direction of the deviation (whether it is too large or too small, as long as it deviates from the target, it will be included in the penalty).

[0152] Optimization objective: The reinforcement learning algorithm aims to maximize the cumulative reward. This enables the system to learn to minimize cutting deviations, thereby improving cutting accuracy and yield.

[0153] (2) The value function matrix of the game model is updated using the policy gradient method (the PPO algorithm is used in this embodiment);

[0154] (3) The updated model will automatically adjust the cutting strategy during the next cut to reduce similar deviations.

[0155] Through continuous operation and closed-loop optimization, the system performance has been continuously improved, and both cutting accuracy and raw material utilization have been enhanced.

[0156] Example 3: Application of multiple cutting modes:

[0157] The cutting parameters of this invention can be flexibly configured, supporting multiple cutting modes:

[0158] Mode A: Fixed-width cutting – suitable for chicken breast strip products. Target width 20mm, length unlimited. The game theory model aims to maximize the number of qualified meat strips by arranging as many 20mm wide strips as possible on a 3D model.

[0159] Mode B: Fixed-weight cutting – suitable for quantitatively packaged products. Target weight 100g ± 5g. The system estimates the volume of each part based on a 3D model, converts the weight using density, and plans a cutting scheme to ensure that each finished product meets the weight requirements.

[0160] Mode C: Equal Weight Division Cutting – Suitable for uniformly dividing large pieces of raw material. The system finds the optimal cutting position to divide the raw material into 2-4 pieces of equal weight, with no waste generated.

[0161] Mode D: Similar Surface Area Cutting – Suitable for breaded and deep-fried products. The system ensures that the surface areas of each piece are similar after cutting, guaranteeing consistent oil absorption during frying.

[0162] This embodiment uses the fixed-size cutting of chicken breast as an example to demonstrate the entire process from feeding, scanning, game-theoretic planning, collaborative execution to secondary optimization. Testing showed that the system's collaborative control accuracy reached ±0.3mm, cutting efficiency was increased by more than 30%, and raw material utilization was improved by 15%, fully achieving the invention's objectives.

[0163] Example 4: Hardware acceleration implementation:

[0164] To meet real-time requirements, the edge computing and game decision-making module 200 of this invention incorporates an FPGA hardware acceleration unit 204. This unit utilizes a Xilinx Zynq UltraScale+ series FPGA to achieve hardware acceleration for the following functions:

[0165] (1) Point cloud preprocessing: The noise filtering algorithm adopts a pipeline architecture, and the processing time for one frame of point cloud (500,000 points) is less than 10ms;

[0166] (2) Game theory solution: The matrix operations of the ADMM algorithm are implemented in parallel on the FPGA, and the solution time is less than 30ms;

[0167] (3) Impedance control: The neural network forward computation uses a customized IP core, and the single inference time is less than 0.1ms.

[0168] The overall processing delay is controlled within 50ms, meeting the needs of high-speed production line operation.

[0169] This invention provides a collaborative control method and device for an intelligent meat cutting system based on edge computing, which can be widely applied to intelligent cutting and processing production lines for various meat products such as chicken breast, fish, pork, and beef. The device has a compact structure and advanced control method, significantly improving cutting accuracy and efficiency while reducing labor costs and raw material waste, demonstrating good industrial applicability and economic benefits.

[0170] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A collaborative control method for an intelligent meat cutting system based on edge computing, characterized in that, Includes the following steps: Step S1: 3D point cloud data acquisition and preprocessing: The meat to be cut on the conveyor belt is scanned by the 3D vision module to obtain high-precision 3D point cloud data, and transmitted to the edge computing unit for noise reduction and feature extraction to generate a complete 3D model of the meat. Step S2: Distributed clock synchronization based on time-sensitive network: A unified clock domain is established between the control unit, the robotic arm controller and the cutting machine controller through a TSN switch to achieve microsecond-level task synchronization between each actuator; Step S3: Multi-agent game-theoretic cutting planning: The multi-degree-of-freedom robotic arm and the high-precision intelligent cutting machine are modeled as intelligent agents with autonomous decision-making capabilities. A value function matrix based on game theory is constructed. Combined with the preset cutting parameters and the three-dimensional model of meat, the collaborative strategy of the multi-degree-of-freedom robotic arm posture adjustment sequence and the high-precision intelligent cutting machine feed trajectory is generated by solving the Nash equilibrium. Step S4: Dynamic Reconstruction and Impedance Control Integration of Cutting Task: During the cutting process, force feedback data is collected in real time by a high-frequency force feedback module installed at the end of the multi-degree-of-freedom robotic arm and the cutter head of the high-precision intelligent cutting machine. When abnormal fluctuations in cutting resistance are detected, the impedance control model is triggered to dynamically reconstruct the cutting parameters and adjust the support stiffness of the multi-degree-of-freedom robotic arm and the feed speed of the high-precision intelligent cutting machine in real time to achieve collaborative operation of multiple actuators and adaptive cutting. Step S5: Secondary scanning and closed-loop optimization after cutting: After cutting, the finished product is scanned a second time through the vision module, and the cutting deviation data is fed back to the edge computing unit to update the value function matrix of the game model, thereby realizing iterative optimization of system performance.

2. The method according to claim 1, characterized in that, The point cloud preprocessing in step S1 includes: Noise filtering module, used to remove environmental interference points; The point cloud completion module completes missing regions based on generative adversarial networks. The feature extraction module is used to identify key features of meat, such as edges, corners, and thickness variations.

3. The method according to claim 1, characterized in that, The time-sensitive network in step S2 uses the IEEE 802.1AS protocol to achieve clock synchronization between controllers with a synchronization accuracy of 1 microsecond. The communication cycle is configured through a time-aware shaper to ensure the coordination and consistency between the multi-degree-of-freedom robotic arm and the high-precision intelligent cutting machine during high-speed movement.

4. The method according to claim 1, characterized in that, The multi-agent game model in step S3 includes: The first intelligent agent: a multi-degree-of-freedom robotic arm whose value function aims to minimize attitude adjustment time and energy consumption; The second intelligent agent is a high-precision intelligent cutting machine, whose value function aims to maximize cutting accuracy and cutting speed. The value function matrix introduces a cross-sensing term, enabling the multi-degree-of-freedom robotic arm to predict the actions of the high-precision intelligent cutting machine and respond in advance. The high-precision intelligent cutting machine can also sense the posture deviation of the multi-degree-of-freedom robotic arm and adjust the feeding strategy accordingly. The Nash equilibrium solution process is completed in the edge computing unit using the alternating direction multiplier method, with a solution time of less than 50ms, which meets the real-time control requirements.

5. The method according to claim 1, characterized in that, The impedance control model in step S4 includes: The stiffness adjustment module is used to dynamically adjust the support stiffness of the multi-degree-of-freedom robotic arm end to the meat according to the cutting resistance. The damping adjustment module is used to dynamically adjust the feed speed of the high-precision intelligent cutting machine according to the cutting resistance. When the detected cutting resistance exceeds the preset threshold, the system completes the reconstruction and distribution of cutting parameters within 1ms; The impedance control model employs a neural network adaptive algorithm, taking the force error and its rate of change as input, to update stiffness and damping parameters online, thereby achieving adaptive compensation for the time-varying stiffness characteristics of meat raw materials.

6. The method according to claim 1, characterized in that, The closed-loop optimization in step S5 employs a deep reinforcement learning algorithm, using the cutting bias as the input to the reward function, and updating the value function matrix of the game model through the policy gradient method to achieve continuous optimization of the cutting strategy.

7. The method according to claim 1, characterized in that, The cutting parameters include one or more combinations of fixed-width cutting, fixed-length cutting, fixed-weight cutting, equal-weight bisecting cutting, and surface area similar cutting.

8. A collaborative control device for an intelligent meat cutting system based on edge computing, characterized in that, include: 3D vision acquisition module: including a 3D laser camera and triggering unit, used to acquire 3D point cloud data of meat raw materials; Edge computing and game decision-making module: integrates a time-sensitive network synchronization unit, a multi-agent game solver, and an impedance control model to generate collaborative cutting strategies; Multi-actuator collaboration module: includes at least one multi-degree-of-freedom robotic arm and at least one high-precision intelligent cutting machine, connected to the edge computing module through a time-sensitive network switch, to receive and execute collaborative cutting instructions; High-frequency force feedback module: installed at the end of the multi-degree-of-freedom robotic arm and at the cutter head of the high-precision intelligent cutting machine, used to collect force feedback data in real time during the cutting process; Secondary scanning and closed-loop optimization module: used to scan the finished product, collect cutting deviation data and feed it back to the game decision module for model iterative optimization; Control unit: Used to coordinate the communication and timing of actions between modules via a time-sensitive network.

9. The apparatus according to claim 8, characterized in that, The edge computing and game decision-making module has a built-in FPGA hardware acceleration unit, which is used to accelerate the real-time calculation of point cloud preprocessing, game solving and impedance control model.

10. The apparatus according to claim 8, characterized in that, The multi-degree-of-freedom robotic arm in the multi-actuator collaboration module and the high-precision intelligent cutting machine achieve collaborative control through a time-sensitive network, with a collaborative control delay of less than 5ms.

11. The apparatus according to claim 8, characterized in that, The sampling frequency of the high-frequency force feedback module is not less than 10kHz, and the force feedback accuracy reaches 0.1N.

12. The apparatus according to claim 8, characterized in that, The multi-actuator collaborative module also includes a servo motion control system; the servo motion control system includes a servo driver, a high-resolution encoder, a motion controller and a real-time communication interface, and adopts a feedforward + feedback composite control, contour error pre-compensation, dynamic friction compensation and load adaptive gain scheduling mechanism to achieve high-precision tracking of the cutting path, with the position tracking error controlled within ±0.03mm.

Images