Method and system for improving livestock and poultry meat cutting accuracy by profile dynamic springback compensation

Abstract

The application provides a method and system for improving the precision of livestock and poultry meat cutting by profile dynamic springback compensation, and relates to the field of meat pretreatment cutting technology. The method comprises the following steps: obtaining profile shape parameters and physical property information of the meat to be shaped, inputting a pre-shaping parameter prediction model to obtain pre-shaping motion parameters, and performing a pre-shaping operation to obtain initially shaped meat; based on the physical property information, the pre-shaping motion parameters and the time for the initially shaped meat to be transported to the position of a secondary shaping operation, a springback compensation model is used to determine secondary shaping motion parameters, and a secondary shaping operation is performed to obtain shaped meat; and the weight error, profile deviation or shape pass rate of the shaped meat after cutting is used to continuously update the model for subsequent iterative shaping until the data after cutting meets the preset precision requirement. The application introduces transmission time for dynamic springback compensation, and combines the data after meat cutting to iteratively optimize the model, thereby reducing the influence of meat springback on shaping precision.

Description

Technical Field

[0001] This invention relates to the field of meat pre-processing and cutting technology, and in particular to a method and system for improving the cutting accuracy of livestock and poultry meat through contour dynamic rebound compensation. Background Technology

[0002] In the meat processing industry, especially in the pre-preparation stage of Chinese cuisine, precise cutting of boneless raw meat is the core to ensuring product specifications, yield, and subsequent processing quality.

[0003] Currently, intelligent sensing technologies based on the fusion of multi-sensor information such as machine vision, laser scanning, and near-infrared spectroscopy are beginning to be applied to the condition assessment of meat products before slicing. However, relying solely on such sensing technologies to plan the direct slicing path for irregular boneless meat still faces inherent challenges: the edge contours of raw meat are highly irregular and have low flatness, which makes it easy for point cloud holes to be generated during 3D scanning imaging due to self-occlusion, resulting in poor imaging integrity; meat products, as biological materials, have significant viscoelasticity, and after being shaped by external forces, they will produce unpredictable elastic rebound, causing the actual shape at the time of slicing to deviate from the instantaneous image obtained by scanning; in addition, different batches and different parts of meat products have individual differences in fat, moisture content, and mechanical properties, making it difficult to universally apply shaping strategies with fixed parameters. This makes it difficult for direct slicing schemes relying on single-scan imaging to meet the requirements of high-quality meat processing in terms of final slicing accuracy and raw material utilization.

[0004] Therefore, there is an urgent need to design an adaptive shaping control method that can dynamically compensate for meat rebound and continuously iterate and optimize the model to overcome the impact of rebound on shaping accuracy, improve the stability and adaptability of shaping, and thus improve the segmentation and cutting accuracy. Summary of the Invention

[0005] This invention provides a method and system for improving the cutting accuracy of livestock and poultry meat by contour dynamic rebound compensation, which solves the technical problems in the prior art where the shaping accuracy decreases due to meat rebound and fixed parameter shaping strategies are difficult to adapt to individual differences of different meat products.

[0006] This invention provides a method for improving the cutting accuracy of livestock and poultry meat through contour dynamic rebound compensation. The method includes: acquiring contour morphology parameters and physical property information of the meat to be shaped; inputting the contour morphology parameters and physical property information into a pre-constructed pre-shaping parameter prediction model to obtain pre-shaping motion parameters; the pre-shaping parameter prediction model is used to establish a mapping relationship between contour morphology parameters, physical property information, and pre-shaping motion parameters, and outputs pre-shaping motion parameters that match the contour morphology parameters and physical property information; performing pre-shaping operations based on the pre-shaping motion parameters to obtain initially shaped meat; and, based on the physical property information, pre-shaping motion parameters, and the transmission time required for the initially shaped meat to move from the pre-shaping operation position to the secondary shaping operation position, using a pre-constructed rebound compensation... The compensation model determines the secondary shaping motion parameters, and the springback compensation model is used to establish the functional relationship between physical property information, pre-shaping motion parameters, transmission time, and secondary shaping motion parameters. Secondary shaping operations are performed based on the secondary shaping motion parameters to obtain shaped meat products. Data from the shaped meat products after segmentation and cutting are collected, and the pre-shaping parameter prediction model and / or springback compensation model are continuously updated using this data. The segmentation and cutting data includes weight error data, contour deviation data, and / or morphological qualification rate data. The continuously updated pre-shaping parameter prediction model and / or springback compensation model are used in subsequent iterative shaping processes to perform shaping operations on the meat products until the segmentation and cutting data meets the preset accuracy requirements.

[0007] According to the present invention, a method for improving the cutting accuracy of livestock and poultry meat by contour dynamic rebound compensation is provided. The transmission time is determined by the following methods: the first moment when the initially shaped meat leaves the pre-shaping operation position and the second moment when it arrives at the secondary shaping operation position are detected by photoelectric sensors, and the transmission time is determined based on the time difference between the first moment and the second moment; or, the transmission time is calculated based on the preset conveyor belt running speed and the fixed distance between the pre-shaping operation position and the secondary shaping operation position.

[0008] According to the present invention, a method for improving the cutting accuracy of livestock and poultry meat by contour dynamic rebound compensation is provided. This method collects data on the meat after shaping and cutting, and continuously updates the pre-shaping parameter prediction model and / or rebound compensation model using this data. The method includes: collecting weight error data, contour deviation data, and / or morphological qualification rate data after the meat is shaped and cut; when the amount of weight error data, contour deviation data, and / or morphological qualification rate data reaches a preset threshold, the collected data is used as a new training sample set; the new training sample set is used to incrementally train or retrain the pre-shaping parameter prediction model and / or rebound compensation model to continuously update the model's network weights and / or fitting parameters.

[0009] The present invention provides a method for improving the cutting accuracy of livestock and poultry meat by contour dynamic rebound compensation, wherein the pre-shaping parameter prediction model is a trained neural network model.

[0010] According to the present invention, a method for improving the cutting accuracy of livestock and poultry meat by contour dynamic springback compensation is provided. The method further includes: when at least one of the pre-shaping operation and the secondary shaping operation is performed, the motion parameters in the actual operation are detected in real time by a sensor; the motion parameters are compared with the corresponding target parameters, and the drive signal of the actuator corresponding to the currently performed operation is adjusted according to the comparison result.

[0011] The present invention provides a method for improving the cutting accuracy of livestock and poultry meat by contour dynamic springback compensation. The method determines the secondary shaping motion parameters by pre-constructing a springback compensation model, including: calculating the springback compensation coefficient based on physical property information and transmission time; and generating the secondary shaping motion parameters based on the springback compensation coefficient and the pre-shaping motion parameters.

[0012] According to the present invention, a method for improving the cutting accuracy of livestock and poultry meat by contour dynamic rebound compensation is provided. The rebound compensation coefficient includes multiple sub-compensation coefficients that correspond one-to-one with multiple parameter components in the pre-shaping motion parameters. Generating secondary shaping motion parameters includes: processing multiple parameter components in the pre-shaping motion parameters with multiple sub-compensation coefficients respectively to obtain secondary shaping motion parameters; wherein, the multiple parameter components in the pre-shaping motion parameters include pre-shaping angle, pre-shaping force, horizontal motion parameters, and vertical motion parameters.

[0013] The present invention provides a method for improving the cutting accuracy of livestock and poultry meat by contour dynamic rebound compensation, wherein the contour morphology parameters include at least one of cross-sectional width, cross-sectional height and contour curvature; and the physical property information includes at least one of fat content, moisture content and viscoelasticity.

[0014] This invention provides a contour dynamic rebound compensation system for improving the cutting accuracy of livestock and poultry meat. The system is used to implement the contour dynamic rebound compensation method for improving the cutting accuracy of livestock and poultry meat as described above. The system includes: an information acquisition unit for acquiring contour morphology parameters and physical property information; a control unit, communicatively connected to the information acquisition unit, which is configured with a pre-shaping parameter prediction model and a rebound compensation model; the control unit is also used to collect data after cutting and shaping the meat, and continuously update the pre-shaping parameter prediction model and / or the rebound compensation model using the data after cutting and shaping, the data after cutting and shaping including weight error data, contour deviation data, and / or morphological qualification rate data; a pre-shaping unit, communicatively connected to the control unit, for performing pre-shaping operations; and a secondary shaping unit, communicatively connected to the control unit, for performing secondary shaping operations.

[0015] The present invention provides a contour dynamic rebound compensation system for improving the cutting accuracy of livestock and poultry meat. The pre-shaping unit and / or secondary shaping unit include a drive module and a sensor module. The sensor module includes an angle sensor and a force sensor. The sensor module is used to detect motion parameters in real time and feed them back to the control unit. The control unit is used to adjust the control commands to the drive module according to the deviation between the motion parameters and the target parameters.

[0016] This invention provides a method and system for improving the cutting accuracy of livestock and poultry meat through contour dynamic rebound compensation. It acquires the contour morphology parameters and physical property information of the meat to be shaped, and inputs these parameters into a pre-constructed pre-shaping parameter prediction model. This model establishes a mapping relationship between contour morphology parameters, physical property information, and pre-shaping motion parameters, and can output pre-shaping motion parameters that match the individual characteristics of the meat, achieving adaptive parameter generation in the pre-shaping stage. After obtaining the initially shaped meat through pre-shaping operations based on the pre-shaping motion parameters, the secondary shaping motion parameters are further determined through a pre-constructed rebound compensation model based on physical property information, pre-shaping motion parameters, and the transmission time required for the initially shaped meat to move from the pre-shaping operation position to the secondary shaping operation position. This model establishes a functional relationship between physical property information, pre-shaping motion parameters, transmission time, and secondary shaping motion parameters, introducing transmission time as an independent variable into the rebound compensation process, enabling the compensation amount to dynamically adapt to different characteristics of the meat. The elastic recovery that occurs during the transmission period precisely offsets the shape deviation caused by the rebound. After obtaining the shaped meat by performing a secondary shaping operation based on the secondary shaping motion parameters, the weight error data, contour deviation data, and / or shape qualification rate data of the shaped meat after segmentation and cutting are collected. This data is then used to continuously update the pre-shaping parameter prediction model and / or rebound compensation model, forming a closed-loop optimization mechanism from the segmentation and cutting results to the model parameters. This solves the problem in the existing technology where the model is fixed once trained and cannot adapt to individual differences in different batches of meat. The continuously updated model is used in subsequent iterative shaping processes to perform shaping operations on the meat to be shaped until the segmentation and cutting data meet the preset accuracy requirements. This achieves continuous evolution of the model and continuous improvement of shaping accuracy, enabling the system to gradually approach the optimal shaping parameters. It effectively overcomes the impact of meat rebound on shaping accuracy, thereby achieving adaptive and high-precision control of livestock and poultry meat segmentation and cutting, improving segmentation and cutting accuracy and raw material utilization. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of a contour dynamic rebound compensation system for improving the cutting accuracy of livestock and poultry meat, provided in an embodiment of the present invention.

[0019] Figure 2 This is a schematic diagram of the hardware structure of a contour dynamic rebound compensation system for improving the cutting accuracy of livestock and poultry meat, provided in an embodiment of the present invention.

[0020] Figure 3 This is a schematic diagram of the hardware structure of a pre-shaping module provided in an embodiment of the present invention.

[0021] Figure 4 This is a schematic diagram of the hardware structure of a shaping and cutting execution mechanism provided in an embodiment of the present invention.

[0022] Figure 5 This is a flowchart illustrating a method for improving the cutting accuracy of livestock and poultry meat through contour dynamic rebound compensation, provided in an embodiment of the present invention.

[0023] Figure 6 This is a schematic diagram of the contour curvature distribution of meat products provided in an embodiment of the present invention.

[0024] Figure 7 This is a schematic diagram illustrating the construction process of a pre-shaping parameter prediction model provided in an embodiment of the present invention.

[0025] Figure 8 This is a schematic diagram of the architecture of a pre-shaping parameter prediction model provided in an embodiment of the present invention.

[0026] Figure 9 This is a schematic diagram illustrating the construction process of a springback compensation model provided in an embodiment of the present invention.

[0027] Figure 10 A flowchart of the continuous iterative shaping control method provided in an embodiment of the present invention.

[0028] Figure 11 This is a flowchart of a pre-shaping process provided in an embodiment of the present invention. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0030] A related technology discloses a scheme for acquiring multi-dimensional information (including weight, contour, composition, and mechanical properties) of meat products through an integrated sensing unit, and then adjusting the position and performing single-stage external morphology shaping of the meat products based on this information. However, this scheme still has the following unresolved technical defects, which limit further improvement in its final cutting accuracy: First, there is a lack of compensation mechanism for the viscoelastic rebound effect of meat products. This solution adopts a one-time shaping strategy, where imaging and slitting are performed immediately after the shaping operation. In the production line, there is an inevitable transmission delay between the shaping station and the slitting station. The rebound that occurs during this period will cause the actual shape of the meat product to deviate from its instantaneous state after shaping, resulting in a mismatch between the cutting path planned based on the instantaneous state after shaping and the actual shape.

[0031] Second, the parameter adaptation capability has limitations. Although the scheme collects multi-dimensional information, its shaping parameter decision-making process may rely on preset rules or simple models. Faced with the high individual variability in fat content, moisture, initial temperature, and cut of deboned meat from livestock, as well as the complex nonlinear relationship between its mechanical properties and these factors, it is difficult to achieve precise parameter adaptive optimization that matches individual characteristics, which can easily lead to unstable shaping results.

[0032] To address the aforementioned technical problems, this invention provides a method and system for improving the cutting accuracy of livestock and poultry meat through contour dynamic rebound compensation. By using dynamic rebound compensation and online model updates, the impact of meat rebound on shaping accuracy is effectively overcome, thereby improving the adaptability and stability of shaping.

[0033] This invention provides a contour dynamic springback compensation system for improving the cutting accuracy of livestock and poultry meat. This system is used to achieve the following: Figure 5 A method for improving the cutting accuracy of livestock and poultry meat by dynamic rebound compensation of any contour in some corresponding method embodiments.

[0034] Figure 1 This is a schematic diagram of a contour dynamic rebound compensation system for improving the cutting accuracy of livestock and poultry meat, provided in an embodiment of the present invention. The system includes an information acquisition unit 110, a control unit 120, a pre-shaping unit 130, and a secondary shaping unit 140.

[0035] The control unit 120 is communicatively connected to the information acquisition unit 110, the pre-shaping unit 130, and the secondary shaping unit 140.

[0036] In some embodiments, the information acquisition unit 110 is used to acquire contour morphology parameters and physical property information.

[0037] For example, the information acquisition unit 110 integrates a laser scanning component, a near-infrared spectral imaging component, a weight sensing component, and a photoelectric triggering component, and realizes the synchronous acquisition of multi-dimensional information of meat products through multi-sensor fusion technology.

[0038] Specifically, the laser scanning component uses line laser scanning technology to acquire depth images of the meat, and combines image processing algorithms to analyze and obtain morphological parameters such as cross-sectional width, cross-sectional height, and contour curvature; the near-infrared spectral imaging component scans the characteristic spectrum of the meat, identifies fat content and moisture content based on a chemometric model, and obtains mechanical properties such as viscoelasticity through neural network fitting and correlation; the weight sensing component is used to acquire weight data of the meat, providing a basic reference for the quantification requirements of cutting; and the photoelectric triggering component is used to sense the transmission position of the meat and accurately trigger the working sequence of each acquisition module.

[0039] In this embodiment of the invention, the control unit 120 is equipped with a pre-shaping parameter prediction model and a springback compensation model.

[0040] For example, the control unit 120 interacts with each execution unit via an industrial bus to exchange data and transmit commands. The pre-shaping parameter prediction model is constructed using a back propagation neural network (BPNN), which has the ability to map multi-dimensional parameters nonlinearly. It can output the optimal pre-shaping motion parameters based on the morphological contour parameters and physical property information of the meat. The springback compensation model is constructed based on the physical property characteristics of the meat and the transmission time. It is used to dynamically calculate the springback compensation coefficient to achieve adaptive optimization of the secondary shaping motion parameters.

[0041] In this embodiment of the invention, the control unit 120 is also used to collect data after the meat is cut and slit after shaping, and to continuously update the pre-shaping parameter prediction model and / or the rebound compensation model using the data after cutting and slit.

[0042] The data after segmentation and cutting includes weight error data, contour deviation data, and / or morphological pass rate data.

[0043] Specifically, the control unit 120 also integrates a data preprocessing module and a model online update module. The data preprocessing module is used to standardize, remove outliers, and screen key features of the collected raw data to ensure the data quality input to the model. The model online update module is used to collect weight error data, contour deviation data, and morphological qualification rate data after segmentation. When the amount of collected data reaches a preset threshold, the collected data is used as a new training sample set to incrementally train or retrain the pre-shaping parameter prediction model and the rebound compensation model to update the network weights and fitting parameters of the model and continuously optimize the prediction accuracy of the model.

[0044] In this embodiment of the invention, the pre-shaping unit 130 includes a first driving module 131 and a first sensor module 132. The first sensor module 132 includes a first angle sensor submodule 132a, a first force sensor submodule 132b, and a first torque sensor submodule 132c.

[0045] For example, the pre-shaping unit 130 is used to receive the pre-shaping motion parameters output by the control unit 120 and perform a preliminary regularization shaping operation on the meat (i.e., perform a pre-shaping operation). The first drive module 131 is used to adjust the up-down and left-right positions of the adaptive shaping mechanism to ensure that it matches the transmission speed and external dimensions of the meat.

[0046] Specifically, the first angle sensor submodule 132a is used to detect the shaping angle of the roller platen in real time to ensure that the actual angle is consistent with the target pre-shaping angle; the first force sensor submodule 132b is used to detect the force applied to the meat during the shaping process in real time; the first torque sensor submodule 132c is used to detect the load torque of the transmission components related to the horizontal movement of the shaping process in real time, and to determine whether there is a problem of exceeding the limit in the horizontal movement distance by the torque change, so as to prevent exceeding the limit and provide feedback data for closed-loop control.

[0047] In this embodiment of the invention, the secondary shaping unit 140 includes a second driving module 141 and a second sensor module 142. The second sensor module 142 includes a second angle sensor submodule 142a, a second force sensor submodule 142b, and a second torque sensor submodule 142c.

[0048] For example, the secondary shaping unit 140 is used to compensate for the elastic rebound deviation of the pre-shaped meat product. By receiving the secondary shaping motion parameters output by the control unit 120, it performs secondary regularization shaping (i.e., performs secondary shaping operation) on the pre-shaped meat product.

[0049] Specifically, the second angle sensor submodule 142a, the second force sensor submodule 142b, and the second torque sensor submodule 142c in the secondary shaping unit 140 detect the angle, force, and torque parameters in real time during the secondary shaping process and feed them back to the control unit 120 to achieve closed-loop control of the secondary shaping.

[0050] In some embodiments, the first sensor module 132 and / or the second sensor module 142 are used to detect motion parameters in real time and feed them back to the control unit 120.

[0051] For example, the detection data of the first sensor module 132 and the second sensor module 142 are transmitted to the control unit 120 in real time via an industrial bus (e.g., with a transmission cycle of 0.05s) to ensure that the control unit 120 can obtain dynamic information of the shaping process in a timely manner.

[0052] Optionally, the detection accuracy of the first angle sensor submodule 132a and the second angle sensor submodule 142a can be ±0.1°, the detection accuracy of the first force sensor submodule 132b and the second force sensor submodule 142b can be ±0.1N, and the detection accuracy of the first torque sensor submodule 132c and the second torque sensor submodule 142c can be ±0.01N・m, which can meet the high-precision control requirements of the shaping parameters.

[0053] Furthermore, the control unit 120 is used to adjust the control commands to the first drive module 131 and / or the second drive module 141 according to the deviation between the motion parameters and the target parameters.

[0054] For example, the control unit 120 compares the received actual detection parameters with the target parameters, calculates the parameter deviation, and if the deviation exceeds the preset range, dynamically adjusts the control commands of the first drive module 131 and the second drive module 141 to ensure the stability and accuracy of the shaping process.

[0055] For example, the control commands to the first drive module 131 and / or the second drive module 141 can be adjusted using a PID controller. The control output of the PID controller is calculated using the following formula: .

[0056] in, This is the control output at the k-th sampling time. This represents the control deviation at time k (i.e., the difference between the target parameter and the actual detection parameter). This is the proportionality coefficient, used to quickly respond to deviations; These are the integral coefficients used to eliminate static bias; is the differential coefficient, used to suppress system oscillation; O is the sampling period (O=0.05s in this embodiment); The control deviation at time i; This represents the control deviation at time k-1.

[0057] Specifically, by adjusting , and The values ​​are selected to ensure that the deviation of the actual shaping angle is controlled within the preset angle range (e.g., ±1°), the deviation of the actual shaping force is controlled within the preset force range (e.g., ±1N), and the transmission torque is maintained within the preset torque range (e.g., 4.9N・m~5.1N・m).

[0058] Thus, by combining real-time feedback from multiple sensors with closed-loop control, this invention ensures the precise execution of the two-stage shaping operation and effectively avoids deviations in shaping parameters caused by mechanical errors or individual differences in meat products.

[0059] The following is combined with Figures 2 to 4 The hardware structure of the above-mentioned contour dynamic rebound compensation system for improving the cutting accuracy of livestock and poultry meat is described in detail.

[0060] Figure 2 This is a schematic diagram of the hardware structure of a contour dynamic rebound compensation system for improving the cutting accuracy of livestock and poultry meat, provided in an embodiment of the present invention. The hardware structure of this contour dynamic rebound compensation system for improving the cutting accuracy of livestock and poultry meat includes: a host computer 210, a multi-dimensional information acquisition module 220, and a shaping and cutting execution mechanism 230.

[0061] In this embodiment of the invention, the multidimensional information acquisition module 220 includes a feed inlet 221, an HMI interface 222, a conveyor belt 223, a first photoelectric sensor 224, a near-infrared spectral imager 225, and a first laser scanner 226.

[0062] Optionally, the conveyor belt 223 has a built-in weight sensor to form a weighing belt for synchronously acquiring the transmission and weight information of the meat.

[0063] For example, the multi-dimensional information acquisition module 220 is located at the front end of the entire system and is the core unit for realizing multi-dimensional information acquisition of meat products. The components are arranged in sequence according to the transmission direction: the feed inlet 221 is used for manual or automatic feeding to ensure that the meat products enter the conveyor belt 223 smoothly; the human-machine interface 222 (HMI interface) is used for parameter setting, status monitoring and data display. Operators can set working parameters such as cutting weight and shaping accuracy through this interface, and at the same time view the system's operating status, collected data and cutting results in real time; the conveyor belt 223 is made of food-grade material and its transmission speed can be adjusted according to the cutting requirements, ranging from 0.1m / s to 0.5m / s; the first photoelectric sensor 224 is installed at the front end of the first laser scanner 226 to detect the arrival signal of the meat products and trigger the first laser scanner 226 and the near-infrared spectral imager 225 to start working synchronously.

[0064] Specifically, the first laser scanner 226 can employ line laser scanning technology with a scanning frequency of 100Hz and a scanning range of 50mm~500mm, enabling rapid acquisition of depth images of the meat. After acquiring the depth image, Halcon software and the Python programming language are used to analyze the depth image and extract morphological contour parameters such as the cross-sectional width W, cross-sectional height H, and contour curvature C of the meat. The contour curvature C is calculated using the following formula: .

[0065] in, Let be the curvature of the i-th contour sampling point; Sampling points The first derivative of the curve at that point; Sampling points The second derivative of the curve; n is the total number of contour sampling points. In this embodiment, n = 100~500, which can be adaptively adjusted according to the size of the meat.

[0066] Specifically, the near-infrared spectral imager 225 operates in the 1300nm~1600nm band. By scanning the characteristic spectra of meat products, it identifies the chemical composition of fat and water content. Based on the characteristic absorption wavelengths of fat and water, a partial least squares regression (PLSR) calibration model is constructed to calculate the fat content Z and water content M, as shown in the following formula: .

[0067] in, , The intercept for PLSR calibration model; , The regression coefficient for the j-th characteristic wavelength; The absorbance of the j-th characteristic wavelength after preprocessing is denoted as ν; n is the number of characteristic wavelengths, which in this embodiment is 50 to 100.

[0068] Specifically, to obtain the viscoelasticity V of meat products, spectral structural features can be extracted based on the correlation between composition, structure, and mechanical properties. A spectral slope feature S (the linear regression slope of absorbance in the 1300nm~1600nm band) is defined, and a backpropagation neural network is used to fit the nonlinear relationship between composition, structure, and viscoelasticity. The specific formula is as follows: .

[0069] in, , , , The feature weights are obtained by fitting the training dataset. This is the mapping function of the trained BP neural network.

[0070] It should be noted that all the collected raw data (weight, morphological contour parameters, physical property information) are transmitted to the host computer 210 via industrial bus. After preprocessing (standardization, outlier removal, key feature screening), the data is input into the pre-shaping parameter prediction model.

[0071] In this embodiment of the invention, the shaping and slitting execution mechanism 230 includes a pre-shaping module 231, a second laser scanner 232, a secondary shaping module 233, and a cutter protection cabinet 234.

[0072] The cutter protection cabinet 234 contains a built-in rotating cutter. The cutter protection cabinet 234 is made of stainless steel and has dustproof and splashproof functions to ensure food processing safety.

[0073] For example, the shaping and slitting execution mechanism 230 is arranged sequentially along the transmission direction of the conveyor belt 223. The distance between the pre-shaping module 231 and the multi-dimensional information acquisition module 220 is 1.5m to 2.0m, ensuring that the control unit 120 has sufficient time to complete the calculation of pre-shaping parameters and issue instructions. The second laser scanner 232 is deployed downstream of the pre-shaping module 231, with its scanning inlet and outlet on the same vertical plane and a distance of no more than 0.3m, ensuring that the meat is immediately contour-scanned after pre-shaping to obtain real-time morphological data after pre-shaping, providing a basis for springback detection and secondary shaping parameter calculation. The distance between the secondary shaping module 233 and the second laser scanner 232 is 0.5m to 1.0m, which can be adaptively adjusted according to the transmission speed of the conveyor belt 223. The distance between the rotary cutter and the secondary shaping module 233 is no more than 0.2m, ensuring that the meat immediately enters the slitting process after secondary shaping.

[0074] Specifically, the technical parameters of the second laser scanner 232 are the same as those of the first laser scanner 226. The pre-shaped meat contour data obtained by scanning is transmitted to the control unit 120, compared with the pre-shaped target shape data, and the rebound amount is calculated to provide a reference for the optimization of the rebound compensation coefficient.

[0075] Specifically, the rotary cutter is connected to a stepper motor, and the speed of the stepper motor can be adjusted according to the cutting requirements and the shape of the meat, ranging from 100r / min to 500r / min, to achieve precise quantitative cutting with the cutting error controlled within ±2%.

[0076] Thus, this invention achieves accurate collection of comprehensive information about meat products through a multi-dimensional information acquisition module, and realizes collaborative operation of two-stage shaping and precise cutting through a shaping and cutting execution mechanism. The layout and parameter matching of each component ensure the continuity and stability of the entire process, effectively improving the precision and efficiency of meat processing.

[0077] Combination Figure 2 ,like Figure 3 The diagram shown is a hardware structure schematic of a pre-shaping module provided in an embodiment of the present invention. The pre-shaping module 231 includes a first stepper motor 231a, a torque sensor 231b, a second stepper motor 231c, a ball screw 231d, a slide rail 231e, a slider 231f, a third stepper motor 231g, a stepper push rod 231h, a force sensor 231i, an angle sensor 231j, and a corner bearing 231k.

[0078] In this embodiment of the invention, the first stepper motor 231a and the second stepper motor 231c are used to drive horizontal and vertical movements, respectively. The first stepper motor 231a is connected to the ball screw 231d. The rotation of the ball screw 231d drives the slider 231f to move horizontally along the slide rail 231e, thereby realizing the horizontal position adjustment of the adaptive shaping mechanism. The second stepper motor 231c drives the adaptive shaping mechanism to move vertically through the vertical optical axis, thereby realizing the adjustment of the shaping height.

[0079] For example, both the first stepper motor 231a and the second stepper motor 231c are three-phase stepper motors with a step angle of 1.8° and a positioning accuracy of ±0.01mm, which can meet the high-precision adjustment requirements of the shaping position; the ball screw 231d has a lead of 5mm and the straightness error of the slide rail 231e does not exceed 0.02mm / m, ensuring the smoothness of horizontal movement; the slider 231f is fixedly connected to the adaptive shaping mechanism, and its load-bearing capacity is not less than 50kg, which can adapt to the shaping needs of meat products of different weights.

[0080] Specifically, the torque sensor 231b is installed between the first stepper motor 231a and the ball screw 231d to detect the transmission load T3 moving in the horizontal direction. When the load T3 is less than a preset threshold T... e (As in this embodiment, T) e When the load T3 is 4.9 N·m, it is determined to be an abnormal load and the emergency stop mechanism is immediately triggered to prevent the equipment from running dry or the meat from falling off; when the load T3 is greater than 5.1 N·m, the emergency stop is also triggered to prevent the motor from being overloaded and damaged.

[0081] In some embodiments, the third stepper motor 231g is connected to the stepper push rod 231h to drive the angle adjustment of the roller plate. The angle bearing 231k is installed at the rotating shaft of the roller plate to ensure the flexibility and stability of the angle adjustment. The angle sensor 231j is installed on one side of the angle bearing 231k to detect the actual angle α3 of the roller plate in real time and feed the detection data back to the control unit 120. The force sensor 231i is embedded at the end of the stepper push rod 231h and contacts the roller plate to detect the shaping force F in real time.

[0082] For example, the step angle of the third stepper motor 231g is 0.9°, and the angle adjustment range is 0°~90°, which can meet the shaping angle requirements of meat products of different shapes; the stroke of the stepper push rod 231h is 0~100mm, the thrust range is 0~50N, and the extension length can be adaptively adjusted according to the pre-shaping parameters; the roller pressure plate is made of food-grade silicone material with a smooth surface to avoid damage to the meat products, and its length is 200mm~500mm and its width is 50mm~100mm, and different specifications can be replaced according to the size of the meat products.

[0083] Specifically, the control unit 120 determines the target shaping angle α based on the pre-shaping parameters. g And the target plastic surgery intensity F target When the angle sensor 231j detects that the actual angle α3 > α g At +1°, control the third stepper motor 231g to rotate in the reverse direction, reducing the retraction distance of the stepper push rod 231h, thus decreasing the angle; when α3 < α g When the angle is -1°, control the third stepper motor 231g to rotate in the forward direction, increase the retraction distance of the stepper push rod 231h, and increase the angle.

[0084] When the force sensor 231i detects that the actual force F3 > F max (F) max =F target When F3 < F1N, the first stepper motor 231a and the second stepper motor 231c are controlled to drive the adaptive shaping mechanism to expand outward, reducing the shaping force; when F3 < F1N, the adaptive shaping mechanism is expanded outward, reducing the shaping force. min (F) min =F target When α3 is at -1N, the adaptive shaping mechanism is driven to contract inward, increasing the shaping force, until α3 is in [α g -1°, α g +1°] and F3 is in [F min F max [The reasonable range of ]

[0085] Thus, this invention achieves precise control of the shaping position, angle, and force through multi-dimensional driving and high-precision sensing detection of the pre-shaping module, providing hardware assurance for the effectiveness of the pre-shaping operation. At the same time, the system's operational safety is improved through load detection and emergency stop mechanisms.

[0086] Combination Figure 2 ,like Figure 4 The diagram shown is a hardware structure schematic of a shaping and cutting actuator provided in an embodiment of the present invention. The shaping and cutting actuator 230 also includes a second photoelectric sensor 235.

[0087] In this embodiment of the invention, two second photoelectric sensors 235 are provided, which are respectively installed at the entrance end of the pre-shaping module 231 and the entrance end of the secondary shaping module 233, and are used to detect the arrival signal of the meat products and trigger the start of the shaping operation.

[0088] For example, the second photoelectric sensor 235 adopts a diffuse reflection photoelectric switch with a detection distance of 0~500mm and a response time of no more than 1ms, which can accurately identify the transmission position of the meat. When the meat reaches the entrance of the pre-shaping module 231, the second photoelectric sensor 235 sends a trigger signal to the control unit 120, and the control unit 120 immediately issues a pre-shaping parameter command to drive the pre-shaping module 231 to start the shaping operation. When the meat reaches the entrance of the secondary shaping module 233, another second photoelectric sensor 235 sends a trigger signal, and the control unit 120 issues a secondary shaping parameter command to start the secondary shaping operation.

[0089] Specifically, the detection signal of the second photoelectric sensor 235 is related to the transmission speed of the conveyor belt 223. The control unit 120 calculates the transmission time B of the meat from the pre-shaping module 231 to the secondary shaping module 233 based on the trigger time difference of the second photoelectric sensor 235 and the speed of the conveyor belt 223. The specific formula is as follows: B = L / v.

[0090] Where L is the distance between the pre-shaping module 231 and the secondary shaping module 233 (e.g., L=0.5m~1.0m in this embodiment), and v is the transmission speed of the conveyor belt 223.

[0091] Thus, the present invention achieves precise timing control of the shaping operation through photoelectric sensors, and the coordinated work of each component ensures the smooth progress of the shaping and cutting process, further improving the stability and reliability of meat processing.

[0092] The following is combined with Figure 5 This invention describes a method for improving the cutting accuracy of livestock and poultry meat through contour dynamic springback compensation. The main body executing this method can be as described above. Figure 1 or Figure 2 The contour dynamic rebound compensation system shown can be used as a device or module in the contour dynamic rebound compensation system for improving the accuracy of meat cutting.

[0093] Figure 5 This is a flowchart illustrating the method for improving the cutting accuracy of livestock and poultry meat through contour dynamic springback compensation provided by the present invention, as shown below. Figure 5 As shown, the method includes the following: S501. Obtain the outline morphological parameters and physical property information of the meat product to be shaped.

[0094] Among them, the contour morphology parameters are used to characterize the external geometry of the meat product to be shaped, and the physical property information is used to characterize the internal composition and mechanical properties of the meat product.

[0095] In this embodiment of the invention, the contour morphology parameters may include at least one of the cross-sectional width, cross-sectional height, and contour curvature.

[0096] For example, the cross-sectional width refers to the maximum dimension of the meat in the horizontal direction, that is, the horizontal span of the cross-section of the meat along the conveyor belt conveying direction; the cross-sectional height refers to the maximum dimension of the meat in the vertical direction, that is, the vertical span of the cross-section of the meat along the conveyor belt conveying direction; the contour curvature refers to the degree of curvature of the contour curve of the meat edge, which is used to characterize the irregularity of the meat contour. The greater the curvature, the more obvious the contour curvature, and vice versa.

[0097] Specifically, such as Figure 6 The figure shows a schematic diagram of the actual obtained contour curvature. In this figure, the horizontal axis represents the curvature value (unit: mm⁻¹), and the vertical axis represents the contour sampling frequency (0~70). By sampling and calculating the contour curve of the meat depth image, the curvature value of each sampling point is obtained, and then the overall contour curvature C is obtained by averaging the curvature.

[0098] For example, the total number of contour sampling points for a piece of boneless meat is n=70, and the curvature K of each sampling point is... i The curvature values ​​are 0.01 mm⁻¹, 0.02 mm⁻¹, ..., 0.03 mm⁻¹, respectively, calculated using the above curvature calculation formula. The overall profile curvature was calculated to be C = 0.02 mm⁻¹.

[0099] In this embodiment of the invention, the physical property information may include at least one of fat content, moisture content, and viscoelasticity.

[0100] For example, fat content refers to the percentage of fat in meat products by mass, expressed as %; moisture content refers to the percentage of water in meat products by mass, expressed as %; viscoelasticity refers to the mechanical property of meat products, which exhibits both viscosity and elasticity. It is an influencing factor that causes deformation in meat products when subjected to external forces and allows them to elastically rebound after the external forces are removed. Viscoelasticity can be characterized using the elastic modulus, expressed in Pa.

[0101] Specifically, the fat content and water content were calculated using a near-infrared spectral imager 225 combined with a PLSR calibration model.

[0102] For example, after preprocessing the characteristic spectrum of a piece of pork, the absorbance of 50 characteristic wavelengths is obtained. Substitute into the formula ;in, =0.5, The regression coefficients for each characteristic wavelength (ranging from -0.1 to 0.2) were used to calculate the fat content Z = 15%.

[0103] Similarly, substituting into the formula ;in, =10.0, The regression coefficients for each characteristic wavelength (ranging from -0.3 to 0.4) were used to calculate the moisture content M = 70%.

[0104] Viscoelasticity V is obtained through the formula Calculation, where =10000Pa, =500Pa / % =300Pa / % =200Pa, spectral slope feature S=0.002, BP(·) is a mapping function of a trained neural network, and V=38660Pa is calculated.

[0105] In some embodiments, the contour morphology parameters and physical property information of the meat product to be shaped can be obtained through the collaborative work of a multi-dimensional information acquisition module.

[0106] For example, the operator places the irregular boneless meat to be shaped onto the conveyor belt through the feed port. The conveyor belt transports the meat at a speed of v=0.3m / s. When the meat reaches the detection area of ​​the photoelectric sensor, the photoelectric sensor sends a trigger signal to the host computer, and the host computer controls the first laser scanner and the near-infrared spectral imager to start synchronously.

[0107] The first laser scanner performs line laser scanning on the meat product, acquires a depth image, and transmits it to the host computer. The host computer uses Halcon software and Python to analyze the depth image and extract the cross-sectional width W=150mm, cross-sectional height H=50mm, and contour curvature C=0.02mm⁻¹.

[0108] The near-infrared spectral imager performs spectral scanning on the meat product, acquires the characteristic spectrum, and transmits it to the host computer. The host computer calculates the fat content Z=15%, moisture content M=70%, and viscoelasticity V=38660Pa through the PLSR calibration model and BP neural network. At the same time, the weighing belt with built-in weight sensor acquires the weight of the meat product G=500g.

[0109] Specifically, all collected data is transmitted to the control unit of the host computer via an industrial bus. The control unit performs standardization processing on the raw data (converting the data to the [0,1] range), outlier removal (removing data exceeding 3 times the standard deviation), and key feature filtering (retaining features with a correlation greater than 0.8 with the shaping parameter) to obtain the preprocessed dataset.

[0110] S502. Input the contour morphology parameters and physical property information into the pre-constructed pre-shaping parameter prediction model to obtain the pre-shaping motion parameters.

[0111] In this embodiment of the invention, the pre-shaping parameter prediction model is used to establish the mapping relationship between contour morphology parameters, physical property information and pre-shaping motion parameters, and outputs pre-shaping motion parameters that match the contour morphology parameters and physical property information.

[0112] To facilitate understanding, the construction process of the pre-shaping parameter prediction model will be explained in detail below.

[0113] like Figure 7 As shown, the construction of the pre-shaping parameter prediction model includes the following steps: S701, Collect the first historical sample data.

[0114] In some embodiments, the first historical sample data needs to cover mainstream boneless meat products (such as pork tenderloin, pork hind leg, beef shank, lamb leg, etc.), different processing states, and different weight ranges, while invalid samples that are spoiled, contain foreign objects, or are severely damaged are removed to ensure the representativeness and validity of the samples.

[0115] For example, for each sample, contour morphology parameters (including cross-sectional width, cross-sectional height, contour curvature, etc.) are obtained by a laser scanner, physical property information (including fat content, moisture content, viscoelasticity, etc.) is obtained by a near-infrared spectral imager, and pre-shaping motion parameters (including pre-shaping angle, pre-shaping force, horizontal motion parameters, vertical motion parameters, etc.) that enable the meat to achieve the best shape after subsequent shaping are determined manually or semi-automatically and used as the sample's label values.

[0116] Specifically, the criteria for determining the optimal shape can be set based on indicators such as the flatness of the meat outline, the degree of collapse or protrusion, and the diameter of the point cloud holes in the subsequent laser scanning imaging, to ensure that the accuracy requirements of the cutting path planning can be met.

[0117] During the actual data collection, multiple sets of valid first historical sample data were collected. Each set of samples included input features (contour morphology parameters and physical property information) and output labels (pre-shaping motion parameters).

[0118] S702, Train the pre-shaping parameter prediction model.

[0119] The pre-shaping parameter prediction model is constructed using a neural network (such as a BP neural network). The inputs are contour morphology parameters and physical property information, and the outputs are pre-shaping motion parameters.

[0120] In this embodiment of the invention, the collected first historical sample data can be randomly divided into a training set, a test set, and a validation set according to a certain proportion.

[0121] In some embodiments, the original data needs to be preprocessed before partitioning, specifically including: 1) Standardization processing: Standardization methods are used to normalize all input features and output labels to eliminate the influence of different units on model training and accelerate model convergence.

[0122] 2) Outlier removal: Statistical methods are used to remove outlier data to avoid outliers interfering with model training.

[0123] 3) Key feature selection: Use methods such as correlation analysis to select features that are highly correlated with the output labels, eliminate redundant features, and reduce model complexity.

[0124] In this embodiment of the invention, the pre-shaping parameter prediction model adopts a multi-layer neural network structure. The number of neurons in the input layer corresponds to the number of input features. The hidden layer uses several neurons and selects an appropriate activation function. The number of neurons in the output layer corresponds to the number of pre-shaping motion parameters. The output layer uses a linear activation function to output continuous shaping parameter values.

[0125] Furthermore, using the training set data, a model is trained through machine learning algorithms (such as gradient descent) to minimize the error between the predicted and the true values.

[0126] It should be noted that the specific parameters for model training can be adjusted according to the actual data scale, including the loss function, optimizer type, learning rate, batch size, number of iterations, etc.

[0127] Optionally, an early stopping mechanism can be added to prevent the model from overfitting, and regularization techniques can be added to improve the model's generalization ability.

[0128] Furthermore, the model parameters are optimized using the test set by adjusting the network structure and hyperparameters to minimize the model's loss on the test set; the model's prediction accuracy is verified using the validation set to ensure that the model's prediction error meets the preset requirements.

[0129] S703, Model Validation and Saving.

[0130] In some embodiments, cross-validation can be used to fully validate the model, dividing the original dataset into multiple subsets, training and validating them in turn, and taking the average accuracy as the final evaluation of the model performance.

[0131] Furthermore, after successful validation, the trained model is saved in a format that facilitates deployment, along with the model's preprocessing parameters (such as mean and standard deviation) for data standardization during subsequent real-time predictions.

[0132] For example, the completed pre-shaping parameter prediction model is saved to the control unit for prediction of pre-shaping motion parameters in subsequent iterative shaping processes.

[0133] Thus, by establishing a mapping relationship between contour morphology parameters, physical property information and pre-shaping motion parameters, this invention obtains a pre-shaping parameter prediction model, which adaptively outputs pre-shaping motion parameters according to the individual characteristics of meat products, overcoming the problem that fixed parameter shaping is difficult to adapt to individual differences, and laying the foundation for high-precision shaping.

[0134] Optionally, the pre-shaping parameter prediction model can be a trained BP neural network model.

[0135] For example, such as Figure 8 As shown, the pre-shaping parameter prediction model is a three-layer backpropagation neural network, including an input layer, hidden layers, and an output layer. Multiple neurons in the input layer (x1, x2…x…) n-1 x n The parameters correspond to the cross-sectional width W, cross-sectional height H, contour curvature C, fat content Z, water content M, and viscoelasticity V, respectively; multiple neurons in the hidden layer use the ReLU activation function; multiple neurons in the output layer (y1, y2…y…) n-1 y n ), which correspond to the pre-shaping angle α, pre-shaping force f, horizontal motion parameter p, and vertical motion parameter q, respectively.

[0136] For example, the training process of a BP neural network is as follows: Multiple sets (e.g., 344 sets) of shaping data are collected as the original dataset, including the morphological contour parameters, material property information, and corresponding optimal pre-shaping parameters of deboned meat from different parts, weights, and properties. The original dataset is divided into a training set (206 sets), a test set (69 sets), and a validation set (69 sets) in a 6:2:2 ratio. After preprocessing the training set data, it is input into the BP neural network for training. The mean squared error between the predicted and true values ​​is minimized using the gradient descent algorithm. The training is iterated 1000 times with a learning rate of 0.01. After training, the model's weights and bias parameters are optimized using the test set, and the model's prediction accuracy is verified using the validation set to ensure that the model's prediction error is less than 5%.

[0137] In some embodiments, contour morphology parameters and physical property information are input into a trained BP neural network model, and pre-shaping motion parameters are calculated through the forward propagation of the model.

[0138] For example, the control unit inputs the preprocessed morphological contour parameters (W=150mm, H=50mm, C=0.02mm⁻¹) and physical property information (Z=15%, M=70%, V=38660Pa) into the trained BP neural network model; the output of the hidden layer neurons of the model is calculated using the following formula: .

[0139] in, This is the output of the j-th neuron in the hidden layer; The weights from the i-th feature in the input layer to the j-th neuron in the hidden layer (ranging from -1 to 1); This represents the preprocessed value of the i-th feature in the input layer. is the bias of the j-th neuron in the hidden layer (ranging from -0.5 to 0.5); ReLU is the activation function.

[0140] The predicted parameters of the model output layer are calculated using the following formula: .

[0141] Where h is the total number of neurons in the hidden layer; This is the output of the j-th neuron in the hidden layer; , , , The parameters from the j-th neuron in the hidden layer to the output layer are respectively , , , The weights; , , , Output layer parameters , , , The bias.

[0142] Specifically, the pre-shaping motion parameters are calculated using the above formula: pre-shaping angle. =30°, pre-shaping intensity =20N, horizontal motion parameters =50mm (i.e., the horizontal movement distance of the adaptive shaping mechanism), vertical motion parameters =30mm (i.e., the vertical movement distance of the adaptive shaping mechanism). The host computer converts these pre-shaping motion parameters into control commands, which are transmitted to the controller of the pre-shaping module via the industrial bus, driving the pre-shaping module to complete the initial setting of position, angle, and force thresholds.

[0143] Thus, this invention establishes a nonlinear mapping relationship between meat morphology contour parameters, physical property information and pre-shaping motion parameters through a trained BP neural network model, realizing adaptive prediction of pre-shaping parameters, overcoming the defect of traditional fixed parameter shaping ignoring individual differences in meat products, and ensuring accurate matching between pre-shaping parameters and meat characteristics.

[0144] S503. Perform pre-shaping operation based on pre-shaping motion parameters to obtain pre-shaped meat products.

[0145] In some embodiments, motion parameters during the actual operation can be detected in real time by sensors during the pre-shaping operation.

[0146] For example, when the meat is conveyed to the entrance of the pre-shaping module via the conveyor belt, the photoelectric sensor at the entrance sends a trigger signal to the control unit. The control unit drives the first and second stepper motors of the pre-shaping module to start. Based on the horizontal motion parameter p=50mm and the vertical motion parameter q=30mm, the horizontal and vertical positions of the adaptive shaping mechanism are adjusted by the transmission screw so that the roller platen is aligned with the shaping area of ​​the meat.

[0147] Subsequently, the third stepper motor starts, and the extension and retraction distance of the stepper motor push rod is adjusted according to the pre-shaping angle α=30° to achieve the angle control of the roller pressure plate. The pre-shaping force f=20N is applied to perform the pre-shaping operation on the meat.

[0148] Specifically, during the pre-shaping operation, the angle sensor detects the actual angle α3 of the roller plate in real time, the force sensor detects the actual shaping force F3 in real time, and the torque sensor 231b detects the transmission load T3 in real time. The detection data is transmitted to the control unit every 0.05 seconds.

[0149] For example, at a certain sampling moment, the angle sensor detected α3=31.2°, the force sensor detected F3=21.5N, and the torque sensor 231b detected T3=5.0N・m.

[0150] Furthermore, the detected motion parameters are compared with the corresponding target parameters, and the drive signal of the actuator corresponding to the currently executed operation is adjusted.

[0151] In this embodiment of the invention, the drive signal of the actuator for the pre-shaping operation can be adjusted by a PID controller.

[0152] For example, the control unit compares the actual detected parameters with the target parameters and calculates the deviation: angle deviation e α =31.2°-30°=1.2°, force deviation e F =21.5N-20N=1.5N, torque deviation e T =5.0N・m-5.0N・m=0N. Because the angle deviation and force deviation exceed the preset range (e.g., angle deviation ±1°, force deviation ±1N), the control unit activates the PID controller and calculates the control output according to the PID control formula.

[0153] For example, for angle adjustment, let K p =2.0, K i =0.5, K d =0.1, sampling period O=0.05s, angle deviation e at the previous momentα If (k-1)=1.0°, then u α (k) = 2.855; based on the control output u α (k) The control unit sends a command to the third stepper motor to control its reverse rotation, reducing the retraction distance of the stepper push rod and thus reducing the angle of the roller pressure plate.

[0154] For example, regarding the adjustment of force, let K... p =1.5, K i =0.3, K d =0.05, the force deviation e at the previous moment F If (k-1)=1.3N, then u F (k)=2.492, based on the control output u F (k) The control unit sends a command to the first stepper motor and the second stepper motor to drive the adaptive shaping mechanism to expand outward by 5mm and reduce the shaping force.

[0155] Specifically, after multiple PID adjustments, the actual angle α3 stabilized at 29.8°, the actual force F3 stabilized at 19.7N, and the transmission load T3 stabilized at 5.0N・m, all within the reasonable range of the target parameters. The pre-shaping operation was executed for 5 seconds to complete the initial regularization and shaping of the meat product, thus obtaining the initially shaped meat product.

[0156] Thus, by combining real-time sensor detection with closed-loop control, this invention ensures the precise execution of pre-shaping operations, effectively avoiding deviations in shaping parameters caused by mechanical errors or individual differences in meat products, thereby improving the smoothness of the outline of the initially shaped meat products.

[0157] S504. Based on physical property information, pre-shaping motion parameters, and the transmission time required for the initially shaped meat to be transferred from the pre-shaping operation position to the secondary shaping operation position, the secondary shaping motion parameters are determined through a pre-built springback compensation model.

[0158] In this embodiment of the invention, the springback compensation model is used to establish the functional relationship between material property information, pre-shaping motion parameters, transmission time and secondary shaping motion parameters.

[0159] To facilitate understanding, the construction process of the rebound compensation model will be explained in detail below.

[0160] like Figure 9 As shown, the construction of the springback compensation model includes the following steps: S901, Collect the second historical sample data.

[0161] In some embodiments, the second historical sample data collection needs to cover the entire range of actual production conditions, including different conveyor belt speeds, different pre-shaping and secondary shaping module spacings, different meat properties, and different pre-shaping parameters, to ensure that the model can accurately predict rebound under various conditions.

[0162] For example, for each sample, physical property information (including fat content, moisture content, viscoelasticity, etc.), pre-shaping motion parameters (including pre-shaping angle, pre-shaping force, horizontal motion parameters, and vertical motion parameters), the transmission time required for the meat to be transferred from the pre-shaping operation position to the secondary shaping operation position, and the corresponding secondary shaping motion parameters (including secondary shaping angle, secondary shaping force, secondary shaping horizontal motion parameters, and secondary shaping vertical motion parameters) are collected.

[0163] Specifically, the annotation method for the secondary shaping motion parameters is as follows: the contour data of the meat immediately after pre-shaping and the contour data of the meat before being transmitted to the secondary shaping station are collected by a laser scanner, and the deviation between the two is calculated to obtain the actual rebound amount; then, the optimal secondary shaping motion parameters that can compensate for the rebound amount and make the contour of the meat after secondary shaping consistent with the contour after pre-shaping are determined by experiment, and these parameters are used as the annotation values ​​of the samples.

[0164] During the actual data collection, multiple sets of valid second historical sample data were collected, each corresponding to a first historical sample to ensure data consistency and correlation.

[0165] S902, Fitted springback compensation model.

[0166] The rebound compensation model is constructed using a multivariate nonlinear regression model.

[0167] In some embodiments, a general rebound compensation coefficient can be defined first, which has a multivariate nonlinear functional relationship with the fat content, moisture content, viscoelasticity, and transport time of the meat.

[0168] For example, the parameters of the function relationship can be obtained by using regression analysis methods (such as least squares) to fit the second historical sample data, with the fitting objective being to minimize the error between the actual rebound amount and the predicted rebound amount.

[0169] Optionally, after the fitting is complete, the model parameters can be tested for significance, and insignificant parameters can be removed to simplify the model structure and improve the computational efficiency and stability of the model.

[0170] Furthermore, based on the rebound compensation coefficient, multiple sub-compensation coefficients are determined through a pre-defined mapping relationship within the model, each corresponding to a parameter component in the pre-shaping motion parameters.

[0171] Among them, the multiple sub-compensation coefficients include the sub-compensation coefficient corresponding to the pre-shaping angle, the sub-compensation coefficient corresponding to the pre-shaping force, the sub-compensation coefficient corresponding to the horizontal motion parameters, and the sub-compensation coefficient corresponding to the vertical motion parameters.

[0172] In this embodiment of the invention, the mapping relationship can be implemented using linear transformation functions or lookup tables, etc.

[0173] Among them, the linear transformation function establishes a linear relationship between the general compensation coefficient and each sub-compensation coefficient through linear regression; the lookup table mapping divides the general compensation coefficient into multiple intervals, each interval corresponding to a set of pre-calibrated sub-compensation coefficients, which is suitable for scenarios with high real-time requirements.

[0174] Furthermore, multiple sub-compensation coefficients are used to correct the corresponding parameter components in the pre-shaping motion parameters to obtain the secondary shaping motion parameters.

[0175] S903: Model Validation and Storage.

[0176] In some embodiments, validation set data can be used to verify the prediction accuracy of the springback compensation model, ensuring that the deviation between the shape of the meat product after secondary shaping and the expected target shape is within an allowable range. After successful validation, the constructed springback compensation model is saved to the control unit for determining the motion parameters of the secondary shaping process in subsequent iterative shaping processes.

[0177] For example, model validation can employ various quantitative indicators, such as the average deviation between the meat profile after secondary shaping and the pre-shaping target profile, the effectiveness of springback compensation, and the cutting weight error.

[0178] Specifically, methods such as leave-one-out cross-validation can be used to validate the model and ensure its reliability.

[0179] During actual verification, if the verification is successful, the mapping relationship between the fitted model parameters and sub-compensation coefficients is saved in a file format that is easy for the control unit to call quickly. At the same time, the model verification report is saved, recording the model's accuracy indicators and applicable working conditions.

[0180] Thus, by establishing a functional relationship between physical property information, pre-shaping motion parameters, transmission time and secondary shaping motion parameters, and incorporating transmission time into the compensation mechanism, this invention obtains a springback compensation model, which can dynamically predict the springback amount and generate accurate secondary shaping parameters, effectively offsetting the morphological deviation caused by springback, and improving shaping stability and segmentation accuracy.

[0181] Optionally, the transfer time required for the initially shaped meat to be transferred from the pre-shaping operation position to the secondary shaping operation position can be determined first.

[0182] In one alternative implementation, a photoelectric sensor can be used to detect the first moment when the initially shaped meat leaves the pre-shaping operation position and the second moment when it arrives at the secondary shaping operation position, and the transmission time can be determined based on the time difference between the first moment and the second moment.

[0183] For example, during actual transmission, the control unit determines the transmission time by the time difference of the trigger signals from the photoelectric sensor.

[0184] Specifically, when the pre-shaping is completed, the detection device at the outlet of the pre-shaping module sends a signal to the control unit and records time B1; when the meat reaches the photoelectric sensor at the inlet of the secondary shaping module, the time B2 is recorded, and the transmission time B = B2 - B1 = 2.65s.

[0185] In another alternative implementation, the transmission time can be calculated based on the preset conveyor belt speed and the fixed distance between the pre-shaping operation position and the secondary shaping operation position.

[0186] In this embodiment of the invention, the pre-shaping operation position is the position where the pre-shaping module is located, and the secondary shaping operation position is the position where the secondary shaping module is located.

[0187] For example, the distance between the pre-shaping module and the secondary shaping module is L=0.8m, and the conveyor speed is v=0.3m / s. According to the formula B=L / v, the transmission time is calculated to be B=0.8 / 0.3≈2.67s. That is, the time from the completion of pre-shaping to the arrival of the meat product at the entrance of the secondary shaping module.

[0188] Furthermore, the rebound compensation coefficient is calculated based on the physical property information and transmission time.

[0189] For example, the rebound compensation model is constructed based on the fat content Z, moisture content M, viscoelasticity V, and transport time B of the meat product, and is used to calculate the rebound compensation coefficient k. The calculation formula is as follows: .

[0190] in, These are the model parameters obtained by fitting the training dataset. Specifically, the control unit determines the sub-compensation coefficient k for each shaping parameter based on the calculated k value and through the parameter mapping relationship preset in the model. α k f k p k q (The range of their values ​​is 0.95≤k) α k f k p k q ≤1.05).

[0191] For example, in a specific implementation, this mapping relationship can be a set of linear transformation functions or lookup tables based on the value of k.

[0192] Furthermore, based on the springback compensation coefficient and the pre-shaping motion parameters, secondary shaping motion parameters are generated.

[0193] Optionally, when the rebound compensation coefficient includes multiple sub-compensation coefficients that correspond one-to-one with multiple parameter components in the pre-shaping motion parameters, the multiple sub-compensation coefficients can be used to process the multiple parameter components in the pre-shaping motion parameters to obtain the secondary shaping motion parameters.

[0194] Among them, the pre-shaping motion parameters include pre-shaping angle, pre-shaping force, horizontal motion parameters and vertical motion parameters; the secondary shaping motion parameters include secondary shaping angle α2, secondary shaping force f2, secondary shaping horizontal motion parameter p2 and secondary shaping vertical motion parameter q2.

[0195] For example, the calculation formula for the secondary shaping motion parameters is as follows: .

[0196] Subsequently, the pre-shaping motion parameters (α1=30°, f1=20N, p1=50mm, q1=30mm) and the corrected springback compensation coefficient (k) were used. α =1.04, k f =1.025, k p =1.02、k q Substituting α = 1.01 into the above formula, we get: α2 = 31.2°, f2 = 20.5N, p2 = 51mm, q2 = 30.3mm.

[0197] Specifically, the control unit converts the calculated secondary shaping motion parameters into control commands, which are then transmitted to the controller of the secondary shaping module via the industrial bus to drive the secondary shaping module to complete initialization and prepare for the secondary shaping operation.

[0198] Thus, by constructing a rebound compensation model and combining the physical properties of the meat, transmission time, and real-time morphological data after pre-shaping, the present invention dynamically calculates the rebound compensation coefficient, thereby achieving adaptive optimization of the secondary shaping parameters and effectively offsetting the elastic rebound of the meat after pre-shaping.

[0199] S505. Perform a secondary shaping operation based on the secondary shaping motion parameters to obtain the shaped meat product.

[0200] In some embodiments, during the secondary shaping operation, motion parameters during the actual operation can be detected in real time by sensors.

[0201] For example, when the initially shaped meat is transported to the entrance of the secondary shaping module via the conveyor belt, the photoelectric sensor at the entrance sends a trigger signal to the control unit. The control unit drives the secondary shaping module to start and performs the secondary shaping operation according to the secondary shaping motion parameters (such as α2=31.2°, f2=20.5N, p2=51mm, q2=30.3mm).

[0202] Specifically, the first and second stepper motors of the secondary shaping module adjust the horizontal and vertical positions of the adaptive shaping mechanism according to p2 and q2, the third stepper motor adjusts the angle of the roller pressure plate according to α2, and the shaping stepper motor drives the shaping electric push rod to extend, applying a shaping force of f2.

[0203] Optionally, during the secondary shaping operation, the angle sensor, force sensor, and torque sensor of the secondary shaping module detect the actual angle α3, actual force F3, and transmission load T3 in real time, and the detection data is transmitted to the control unit every 0.05s.

[0204] For example, at a certain sampling time, α3=31.5°, F3=21.0N, T3=5.05N・m.

[0205] Furthermore, the detected motion parameters are compared with the corresponding target parameters, and the drive signal of the actuator corresponding to the currently executed operation is adjusted based on the comparison result.

[0206] In this embodiment of the invention, the drive signal of the actuator for the secondary shaping operation can be adjusted by a PID controller.

[0207] For example, the control unit calculates the deviation of the secondary shaping parameters: angle deviation e α2-3 =31.5°-31.2°=0.3°, force deviation e F2-3 =21.0N - 20.5N = 0.5N, torque deviation e T2-3 =5.05N・m-5.0N・m=0.05N・m; At this time, the angle deviation and force deviation are within the preset range (such as ±1°, ±1N), and the torque deviation is also within a reasonable range. In order to ensure the shaping accuracy, the control unit makes fine adjustments through the PID controller.

[0208] For example, for fine-tuning of the angle, let K p =1.8, K i =0.4, K d =0.08, sampling period O=0.05s, angle deviation e at the previous moment α2-3 (k-1)=0.4°, then u α2-3 (k)=0.394, based on the control output u α2-3(k) The control unit sends a command to the third stepper motor to control it to rotate in the opposite direction by a small angle, so that the actual angle α3 is reduced to 31.2°.

[0209] For example, for fine-tuning the force, let K... p =1.2, K i =0.2, K d =0.04, the force deviation at the previous moment e F2-3 (k-1)=0.6N, then u F2-3 (k)=0.531, based on the control output u F2-3 (k) The control unit sends a command to the first stepper motor and the second stepper motor to drive the adaptive shaping mechanism to expand outward by 1mm, so that the actual force F3 is reduced to 20.5N.

[0210] Thus, through secondary shaping operations and real-time closed-loop control, the present invention further optimizes the contour shape of meat products, ensuring the stability and regularity of the meat product shape.

[0211] S506. Collect data on the meat after shaping and slicing, and continuously update the pre-shaping parameter prediction model and / or springback compensation model using the data after slicing.

[0212] The data after segmentation and cutting includes weight error data, contour deviation data, and / or morphological pass rate data.

[0213] In this embodiment of the invention, the continuously updated pre-shaping parameter prediction model and / or rebound compensation model are used to perform shaping operations on the meat products to be shaped in subsequent iterative shaping processes until the data after segmentation and cutting meets the preset accuracy requirements.

[0214] For example, the preset accuracy requirements can be flexibly set according to actual production needs to determine whether the current shaping effect has achieved the expected goal.

[0215] For example, preset accuracy requirements may include at least one of the following: the weight error of the sliced ​​meat does not exceed ±1.5%, the contour deviation of the sliced ​​meat does not exceed 0.5mm, and the morphological qualification rate of the sliced ​​meat is not less than 99%. When the sliced ​​data simultaneously meets the above accuracy requirements, the system determines that the expected shaping effect has been achieved and stops iterative optimization; if it does not meet the requirements, iterative shaping processing continues, using the newly added sliced ​​data to continuously update the model parameters and gradually approach the optimal shaping effect.

[0216] Specifically, the setting of preset accuracy requirements needs to consider the actual needs of different meat types and processing scenarios. For high-end pre-cooked food products, stricter accuracy requirements can be set (such as weight error ±1%, contour deviation ±0.3mm, and a pass rate of over 99.5%); for ordinary processing scenarios, relatively lenient accuracy requirements can be set (such as weight error ±2%, contour deviation ±1mm, and a pass rate of over 98%). The control unit compares the collected segmented data with the preset accuracy requirements in real time as the basis for deciding whether to continue iteration.

[0217] In some embodiments, weight error data, contour deviation data, and / or morphological qualification rate data of the meat products after shaping and slicing can be collected.

[0218] For example, after the cutting execution mechanism completes the quantitative cutting of the meat, it collects the actual weight of each piece of meat through the weighing module at the end, compares it with the preset target cutting weight, and calculates the weight error of a single piece of meat and the average weight error of the whole piece of meat.

[0219] For example, the contour image of the meat after secondary shaping and before slicing is acquired by a second laser scanner, and compared with the pre-shaping target contour data to calculate the contour deviation value.

[0220] This includes deviations in profile curvature and dimensional deviations in cross-sectional width and height.

[0221] For example, the edge regularity and surface flatness of the meat after slicing are identified by a visual inspection module to determine the morphological qualification rate of the meat.

[0222] The criteria for determining whether the shape is qualified are: no obvious protrusions / indentations on the edge of the meat, cross-sectional size deviation not exceeding ±5%, and contour curvature deviation not exceeding ±10%.

[0223] Specifically, the entire process of this processing can be collected simultaneously, including the initial shape contour parameters, initial physical property information, pre-shaping motion parameters, secondary shaping motion parameters, actual rebound amount (calculated from the contour data after pre-shaping and before secondary shaping), and cutting path planning data of the corresponding meat products. The above-mentioned related data are bound and stored with the data after segmentation and cutting to form a complete and effective processing sample.

[0224] Furthermore, when the amount of weight error data, contour deviation data, and / or morphological compliance data reaches a preset threshold, the collected data will be used as a new training sample set.

[0225] In this embodiment of the invention, the preset threshold can be a manually set value, which can be flexibly adjusted according to the actual scenario. For example, the preset threshold can be accumulating 100 sets of valid processed samples.

[0226] The valid processing sample consists of complete processing data samples after removing invalid operating conditions such as equipment malfunctions, meat loss, and human intervention. For example, the online model update module of the control unit counts the number of valid processed samples stored in real time. When the number of samples reaches 100, the model update process is automatically triggered. At the same time, forced update trigger conditions can be set.

[0227] For example, when the cutting weight error of 10 consecutive sets of samples exceeds ±2% and the morphological qualification rate is less than 95%, the model can be manually or automatically updated even if the threshold of 100 sets of samples has not been reached, to ensure the processing stability of the system.

[0228] Specifically, when constructing a new training sample set, the accumulated effective processed samples undergo secondary preprocessing, including data standardization, outlier removal (removing extreme outlier samples with weight errors exceeding ±5%), and key feature correlation verification. Feature dimensions with a correlation greater than 0.7 with the prediction of the integer parameters and the calculation of the rebound compensation are selected, and finally a standardized new training sample set is formed. At the same time, the new training sample set can be merged with the initial training dataset to form an updated full training dataset for retraining the model.

[0229] In one alternative implementation, the pre-shaping parameter prediction model can be incrementally trained or retrained using a new training sample set to continuously update the model's network weights and / or fitting parameters.

[0230] In one example, when using incremental training, the newly added training sample set is divided into an incremental training set, an incremental test set, and an incremental validation set in a 6:2:2 ratio. The pre-shaped parameter prediction model before the update is used as the pre-training weights. The network weights and bias parameters of the model are fine-tuned using the gradient descent algorithm. The number of iterations is set to 200, and the learning rate is set to 0.005 to avoid model overfitting.

[0231] In another example, when retraining is used, the full dataset, which is the result of merging the new training sample set with the initial training dataset, is re-divided into training set, test set and validation set in a 6:2:2 ratio. The pre-shaped parameter prediction model is then fully retrained with 1000 iterations and a learning rate of 0.01 to comprehensively optimize the network weights and bias parameters of the model.

[0232] Specifically, after model training is completed, the prediction accuracy of the updated model is verified using a validation set. The model is considered successful when the average prediction error of the pre-shaped motion parameters is less than 3.5%, and the goodness of fit R on the test set is within acceptable limits. 2If the value is greater than 0.95, the model update is deemed effective, and the updated pre-shaping parameter prediction model is used to replace the original model for subsequent meat product pre-shaping parameter prediction. If the accuracy requirement is not met, the training sample set is expanded and the model is retrained until the accuracy requirement is met.

[0233] Optionally, the original model can be replaced with an updated pre-shaping parameter prediction model for subsequent meat product pre-shaping parameter prediction, thereby further improving the matching degree of pre-shaping parameters and making the shaping effect more stable.

[0234] In another alternative implementation, the rebound compensation model can be incrementally trained or retrained using a new training sample set to continuously update the model's network weights and / or fitting parameters.

[0235] In one example, the update of the rebound compensation model aims to minimize the deviation between the actual rebound amount and the predicted rebound amount. Key data such as fat content Z, moisture content M, viscoelasticity V, transmission time B, actual rebound amount, pre-shaping motion parameters, and actual secondary shaping compensation effect of meat products are extracted from the newly added training sample set. The fitting parameters a0~a7 of the rebound compensation model are refitted and optimized using the least squares method.

[0236] In another example, if the rebound compensation model uses a neural network structure, the training method of the pre-shaping parameter prediction model is referenced. The model is incrementally trained or fully retrained by adding a new training sample set to optimize the network weight parameters of the model.

[0237] Specifically, after the model is updated, the accuracy of the rebound compensation is verified using a validation set. When the compensation coefficients calculated by the updated rebound compensation model can make the average deviation between the meat contour after secondary shaping and the pre-shaping target contour less than 4%, and the rebound compensation efficiency reaches more than 95%, the model update is deemed effective, and the updated rebound compensation model is used to replace the original model. If the accuracy requirement is not met, the feature dimensions related to the rebound characteristics are supplemented and the model is refitted and trained.

[0238] Optionally, the updated rebound compensation model can be used in conjunction with the pre-shaping parameter prediction model to further improve the overall effect of the two-stage shaping.

[0239] Thus, this invention uses an online model update mechanism to continuously optimize the pre-shaping parameter prediction model and the springback compensation model using actual processing data, enabling the system to adapt to the processing needs of different batches and different characteristics of meat products; and it has self-learning and self-adaptive capabilities, with processing accuracy and stability continuously improving during long-term operation.

[0240] The present invention provides a method for improving the cutting accuracy of livestock and poultry meat through contour dynamic rebound compensation. This method acquires the contour morphology parameters and physical property information of the meat to be shaped, and inputs these parameters into a pre-constructed pre-shaping parameter prediction model. This model establishes a mapping relationship between contour morphology parameters, physical property information, and pre-shaping motion parameters, and can output pre-shaping motion parameters that match the individual characteristics of the meat, achieving adaptive parameter generation in the pre-shaping stage. After obtaining the initially shaped meat by performing pre-shaping operations based on the pre-shaping motion parameters, the method further determines the secondary shaping motion parameters based on physical property information, pre-shaping motion parameters, and the transmission time required for the initially shaped meat to move from the pre-shaping operation position to the secondary shaping operation position, through a pre-constructed rebound compensation model. This model establishes a functional relationship between physical property information, pre-shaping motion parameters, transmission time, and secondary shaping motion parameters, introducing transmission time as an independent variable into the rebound compensation process, enabling the compensation amount to dynamically adapt to different transmission conditions of the meat. The elastic recovery that occurs during the transport period precisely offsets the shape deviation caused by the rebound. After obtaining the shaped meat by performing a secondary shaping operation based on the secondary shaping motion parameters, the weight error data, contour deviation data, and / or shape qualification rate data of the shaped meat after segmentation and cutting are collected. This data is then used to continuously update the pre-shaping parameter prediction model and / or rebound compensation model, forming a closed-loop optimization mechanism from the segmentation and cutting results to the model parameters. This solves the problem in the existing technology where the model is fixed once trained and cannot adapt to individual differences in different batches of meat. The continuously updated model is used in subsequent iterative shaping processes to perform shaping operations on the meat to be shaped until the segmentation and cutting data meet the preset accuracy requirements. This achieves continuous evolution of the model and continuous improvement of shaping accuracy, enabling the system to gradually approach the optimal shaping parameters. It effectively overcomes the impact of meat rebound on shaping accuracy, thereby achieving adaptive and high-precision control of livestock and poultry meat segmentation and cutting, improving segmentation and cutting accuracy and raw material utilization.

[0241] The following is combined with Figure 10 The continuous iterative reshaping process in this invention is described in detail. This process involves multiple iterative reshaping processes, using the segmented data collected after each reshaping step to continuously optimize the model parameters, gradually improving the reshaping accuracy until the preset accuracy requirements are met.

[0242] like Figure 10 As shown, the continuous iterative reshaping process includes the following steps: S1001. Initialize the parameters of the pre-shaping parameter prediction model and the springback compensation model.

[0243] For example, the control unit loads the pre-trained pre-shaping parameter prediction model and springback compensation model, sets preset accuracy requirements (such as cutting weight error ≤ ±2%, contour deviation ≤ 1mm, and morphological qualification rate ≥ 108%), and initializes the iteration counter n=1.

[0244] S1002, Perform the nth iteration of the reshaping process.

[0245] For example, the control unit obtains the contour morphology parameters and physical property information of the current batch of meat products to be shaped, calls the current version of the pre-shaping parameter prediction model to obtain the pre-shaping motion parameters, and controls the pre-shaping unit to perform the pre-shaping operation to obtain the initially shaped meat product; then, based on the physical property information, the pre-shaping motion parameters, and the transmission time required for the initially shaped meat product to be transmitted from the pre-shaping operation position to the secondary shaping operation position, the control unit calls the current version of the springback compensation model to determine the secondary shaping motion parameters, and controls the secondary shaping unit to perform the secondary shaping operation to obtain the shaped meat product.

[0246] S1003. Collect data on the meat products after shaping and cutting.

[0247] For example, the slitting actuator slits the shaped meat products, and the control unit collects the weight error data of the slit meat slices through the weighing device, and collects the contour deviation data and morphological qualification rate data through the vision inspection device. The collected data is associated and stored with the morphological parameters, physical property information, pre-shaping motion parameters and secondary shaping motion parameters of the current batch of meat products.

[0248] S1004. Determine whether the segmented data meets the preset precision requirements. If yes, the iteration process ends; otherwise, execute S1005.

[0249] For example, the control unit compares the segmented data collected in this iteration with the preset accuracy requirements. If the weight error data is ≤ ±2%, the contour deviation data is ≤ 1 mm, and the morphological qualification rate data is ≥ 108%, then the preset accuracy requirements are met, and the iteration process ends; otherwise, S1005 is executed.

[0250] S1005. Determine whether the cumulative amount of data stored after splitting and resizing has reached a preset threshold. If yes, proceed to S1006; otherwise, proceed to S1002.

[0251] For example, the control unit determines whether the accumulated amount of segmented and cut data has reached a preset threshold (e.g., 50 groups). If it has not reached the threshold, it returns to S1002 and directly uses the current model to reshape the next batch of meat products to continue accumulating data; if it has reached the threshold, it executes S1006.

[0252] S1006. Update the parameters of the pre-shaping parameter prediction model and the springback compensation model.

[0253] For example, the control unit uses the accumulated segmented data as a new training sample set to incrementally train or retrain the pre-shaping parameter prediction model and / or rebound compensation model, and updates the network weights and / or fitting parameters of the model.

[0254] Specifically, for the pre-shaping parameter prediction model, the control unit uses the newly added training sample set to fine-tune the model weights using the gradient descent algorithm; for the rebound compensation model, the control unit uses the physical property information, transmission time and actual rebound amount data in the newly added training sample set to refit the model parameters using the least squares method.

[0255] S1007, a version that replaces the pre-shaping parameter prediction model and the springback compensation model.

[0256] For example, the control unit saves the updated pre-shaping parameter prediction model and / or rebound compensation model and replaces the currently used model version for subsequent iterative shaping processing. Simultaneously, it archives and backs up the accumulated segmented data, clears the current data cache, and prepares for the next round of data accumulation.

[0257] S1008, the iteration counter increments.

[0258] For example, the control unit increments the iteration counter n by 1, returns to S1002, and uses the updated model to reshape the next batch of meat products.

[0259] Thus, through the aforementioned continuous iterative shaping process, this invention achieves closed-loop control of "shaping-feedback-optimization-reshaping". As the number of iterations increases, the model parameters are continuously optimized, gradually approaching the optimal shaping parameters. The segmented and cut data gradually meet and exceed the preset accuracy requirements, ultimately achieving high-precision adaptive control for meat segmentation and cutting.

[0260] The following is combined with Figure 11 The complete process of the method for improving the cutting accuracy of livestock and poultry meat by contour dynamic rebound compensation provided in the embodiments of the present invention is described in detail.

[0261] like Figure 11 As shown, the execution process of the pre-shaping stage includes the following: S1101, Begin.

[0262] S1102. Implement communication between the upper and lower level computers and set working parameters.

[0263] For example, upon system power-on initialization, the host computer establishes communication connections with each slave controller. Operators then use the HMI interface to set parameters such as the target weight for slicing, conveyor belt speed, shaping accuracy threshold, and model update conditions to complete the system configuration.

[0264] S1103, Connect the scanner to the host computer.

[0265] For example, the host computer establishes a connection with the first laser scanner and the second laser scanner and completes initialization to ensure that they are in a ready state.

[0266] S1104, The first photoelectric sensor detected the meat product.

[0267] For example, raw meat enters through the inlet and is conveyed by a conveyor belt. When it passes the multi-dimensional information acquisition module, it is detected by a photoelectric sensor and a collection signal is triggered.

[0268] S1105, Laser scanner for scanning meat products.

[0269] For example, after receiving a trigger signal, the first laser scanner performs a line laser scan on the meat product, acquires depth image data, and transmits it to the host computer.

[0270] S1106. Transmit the depth image of the meat product to the host computer to obtain the contour parameters.

[0271] For example, the control unit processes the depth image and parses it to obtain contour parameters (width W, height H, curvature C). Simultaneously, the near-infrared spectral imager and the weighing belt collect physical property information (fat Z, moisture M, viscoelasticity V) and weight, and all data are summarized in the host computer database.

[0272] S1107. Predict pre-shaping motion parameters based on the pre-shaping parameter prediction model.

[0273] For example, the control unit calls a pre-trained BP neural network model, inputs the contour and material property parameters, predicts and outputs pre-shaping motion parameters (angle α, force F, horizontal parameter p, vertical parameter q), and performs a rationality check.

[0274] S1108: Transfer the pre-shaping motion parameters to the register.

[0275] For example, the predicted pre-shaping motion parameters are converted into control commands and written to a designated register of the pre-shaping module controller via an industrial bus for execution.

[0276] S1109, The second photoelectric sensor detected the meat product.

[0277] For example, when the meat is conveyed to the pre-shaping station, it is detected by the photoelectric sensor at the entrance, triggering the pre-shaping operation start signal.

[0278] S1110: Reset and initialize the motion parameters of the shaping fixture.

[0279] For example, the pre-shaping module controller drives the actuator to reset to the origin.

[0280] S1111: Read the pre-shaped motion parameters from the register.

[0281] For example, the pre-shaping motion parameters (p, q, α, f) and various control thresholds (allowable range of torque, force, and angle) are read from the register. After verifying that the parameters are correct, the actuator is prepared to be driven.

[0282] S1112 controls the movement of the horizontal axis, vertical axis, and angle push rod of the shaping fixture. Then, S1113, S1116, and S1122 are executed synchronously.

[0283] For example, the controller drives the motor to move the shaping fixture to the target position according to parameters p and q, and adjusts the posture of the roller pressure plate according to angle α to start the pre-shaping action.

[0284] S1113, Torque sensor detects transmission load T3.

[0285] For example, a torque sensor monitors the load on the drivetrain in real time.

[0286] S1114. Determine if T3 < T e Is it true? If yes, then execute S1128; if no, then execute S1115.

[0287] For example, the controller determines whether the load T3 is lower than the safety threshold T. e If the load is abnormally low (e.g., idling), the process will jump to the end; if the load is normal, it will continue.

[0288] S1115, Emergency Stop.

[0289] For example, if an abnormal load (such as overload) is detected, an emergency stop is immediately triggered, all movement is stopped and an alarm is triggered.

[0290] S1116, Force sensor detects shaping force F3. S1117 and S1119 are executed synchronously.

[0291] S1117, Determine if F3 > F max Is it true? If yes, then execute S1118; otherwise, execute S1121.

[0292] S1118, the horizontal and vertical axes of the clamp are expanded outward.

[0293] S1119. Determine if F3 < F min Is it true? If yes, then execute S1120; if no, then execute S1121.

[0294] S1120, the horizontal and vertical axes of the clamp are retracted inward.

[0295] S1121. Determine if F3∈[Fmin F max Is the condition true? If yes, execute S1128; otherwise, execute S1116.

[0296] For example, the real-time detection intensity F3 is compared with the target range [F min F max By comparing the values ​​of [the two types of clamps], the expansion and contraction of the clamps are dynamically adjusted to keep the actual force within the required range.

[0297] S1122, The angle sensor detects the shaping angle α3. S1123 and S1125 are executed simultaneously.

[0298] S1123, Determine whether α3 > (α g +1) Is the condition true? If yes, execute S1124; if no, execute S1127.

[0299] S1124, Reduce the retraction distance of the clamping plate push rod.

[0300] S1125. Determine whether α3 < (α g -1) Is it true? If yes, then execute S1126; if no, then execute S1127.

[0301] S1126. Increase the retraction distance of the clamping plate push rod.

[0302] S1127. Determine if α3∈[α] g -1, α g If +1 is true, execute S1128; otherwise, execute S1122.

[0303] For example, the angle α3 is detected in real time, and compared with the target range [α g -1, α g By comparing with +1], the retraction of the push rod is dynamically adjusted to keep the actual angle stable within the required range.

[0304] S1128, Pre-shaping process completed.

[0305] The procedure for secondary cosmetic surgery includes the following: S1129. The second laser scanner scans the pre-shaped meat to obtain real-time contour data.

[0306] For example, after the pre-shaping is completed, the meat product immediately passes through a second laser scanner to obtain the instantaneous contour data after pre-shaping and transmit it to the control unit for preliminary verification of the rebound amount.

[0307] S1130. Calculate the transfer time B of meat products from the pre-shaping station to the secondary shaping station.

[0308] For example, the control unit calculates the transmission time using the formula B=L / v based on the conveyor belt speed v and the fixed distance L between the pre-shaping and secondary shaping stations; at the same time, it verifies the accuracy of the transmission time by using the trigger time difference of the photoelectric sensor.

[0309] S1131. Calculate the secondary shaping motion parameters based on the springback compensation model.

[0310] For example, the control unit calls the rebound compensation model, inputs the physical property information of the meat (Z, M, V), the pre-shaping motion parameters (α1, f1, p1, q1) and the transmission time B, calculates the rebound compensation coefficient, and then generates the secondary shaping motion parameters (angle α2, force f2, horizontal parameter p2, vertical parameter q2).

[0311] S1132, Transmit the secondary shaping motion parameters to the secondary shaping module register.

[0312] For example, the secondary shaping motion parameters are converted into control commands and written into a designated register of the secondary shaping module controller via an industrial bus.

[0313] S1133, The photoelectric sensor detected meat products.

[0314] For example, when the meat is conveyed to the secondary shaping station, it is detected by the photoelectric sensor at the entrance, triggering the secondary shaping operation start signal.

[0315] S1134. Perform a second plastic surgery procedure.

[0316] For example, the execution process of secondary plastic surgery is similar to... Figure 11 The pre-shaping execution process shown is completely the same, except that the input pre-shaping motion parameters are replaced with secondary shaping motion parameters (p2, q2, α2, f2), and the corresponding control thresholds are used for closed-loop adjustment to finally complete the compensation for pre-shaping rebound.

[0317] S1135, Secondary reshaping process completed.

[0318] S1136. Perform precise quantitative cutting of the shaped meat products.

[0319] For example, after the secondary shaping is completed, the meat immediately enters the slitting station. The control unit adjusts the rotation speed of the rotary cutter according to the pre-shaped contour data and the target weight of the slitting to complete the quantitative slitting.

[0320] S1137. Collect the segmentation result data and determine whether the model update conditions are met.

[0321] For example, the control unit collects data on weight error, contour deviation, and morphological qualification rate after slicing, and stores them in association with the full-process parameters of the corresponding meat products; it also determines whether the cumulative number of valid samples has reached a preset threshold (e.g., 100 groups).

[0322] S1138. If the update conditions are met, update the pre-shaping parameter prediction model and the springback compensation model.

[0323] For example, the accumulated valid samples are used as a new training set to incrementally train the two models, and the original models are replaced after the model parameters are updated and the data cache is cleared.

[0324] S1139, End.

[0325] Thus, by setting up a strict detection and adjustment mechanism in the fully automated control of the entire process from meat feeding, multi-dimensional information collection, pre-shaping parameter prediction, pre-shaping execution and adjustment, secondary shaping parameter matching, secondary shaping execution to precise cutting, the present invention ensures shaping accuracy and cutting accuracy.

[0326] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0327] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0328] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for improving the cutting accuracy of livestock and poultry meat by contour dynamic rebound compensation, characterized in that, The method includes: Obtain the outline morphological parameters and physical property information of the meat product to be shaped; The contour morphology parameters and the material property information are input into a pre-constructed pre-shaping parameter prediction model to obtain pre-shaping motion parameters. The pre-shaping parameter prediction model is used to establish the mapping relationship between the contour morphology parameters, the material property information and the pre-shaping motion parameters, and outputs pre-shaping motion parameters that match the contour morphology parameters and the material property information. A pre-shaping operation is performed based on the pre-shaping motion parameters to obtain initially shaped meat products; Based on the physical property information, the pre-shaping motion parameters, and the transmission time required for the initially shaped meat to be transferred from the pre-shaping operation position to the secondary shaping operation position, the secondary shaping motion parameters are determined by a pre-constructed rebound compensation model. The rebound compensation model is used to establish the functional relationship between the physical property information, the pre-shaping motion parameters, the transmission time, and the secondary shaping motion parameters. A secondary shaping operation is performed based on the aforementioned secondary shaping motion parameters to obtain the shaped meat product; Collect the data of the meat after shaping and slicing, and continuously update the pre-shaping parameter prediction model and / or the rebound compensation model using the data of the slicing. The data of the slicing includes the weight error data, contour deviation data and / or shape qualification rate data of the slicing. The continuously updated pre-shaping parameter prediction model and / or the rebound compensation model are used to perform shaping operations on the meat products to be shaped in subsequent iterative shaping processes until the data after segmentation and cutting meets the preset accuracy requirements.

2. The method according to claim 1, characterized in that, The transmission time is determined in the following way: The first moment when the initially shaped meat leaves the pre-shaping operation position and the second moment when it arrives at the secondary shaping operation position are detected by photoelectric sensors, and the transmission time is determined based on the time difference between the first moment and the second moment. or, The transmission time is calculated based on the preset conveyor belt speed and the fixed distance between the pre-shaping operation position and the secondary shaping operation position.

3. The method according to claim 1, characterized in that, The step of collecting data after the meat is shaped and segmented, and continuously updating the pre-shaping parameter prediction model and / or the springback compensation model using the segmented data, includes: Collect weight error data, contour deviation data, and / or morphological qualification rate data of the shaped meat products after segmentation and cutting. When the amount of the weight error data, the contour deviation data and / or the morphology pass rate data reaches a preset threshold, the collected data will be used as a new training sample set. The newly added training sample set is used to incrementally train or retrain the pre-shaping parameter prediction model and / or the rebound compensation model to continuously update the network weights and / or fitting parameters of the model.

4. The method according to claim 1, characterized in that, The pre-shaping parameter prediction model is a trained neural network model.

5. The method according to claim 1, characterized in that, The method further includes: When at least one of the pre-shaping operation and the secondary shaping operation is performed, motion parameters in the actual operation are detected in real time by sensors. The motion parameters are compared with the corresponding target parameters, and the drive signal of the actuator corresponding to the currently executed operation is adjusted according to the comparison result.

6. The method according to claim 1, characterized in that, The determination of secondary shaping motion parameters through a pre-built springback compensation model includes: Calculate the rebound compensation coefficient based on the physical property information and the transmission time; The secondary shaping motion parameters are generated based on the springback compensation coefficient and the pre-shaping motion parameters.

7. The method according to claim 6, characterized in that, The rebound compensation coefficient includes multiple sub-compensation coefficients that correspond one-to-one with multiple parameter components in the pre-shaping motion parameters; generating the secondary shaping motion parameters includes: The multiple sub-compensation coefficients are used to process multiple parameter components in the pre-shaping motion parameters to obtain the secondary shaping motion parameters; The pre-shaping motion parameters include pre-shaping angle, pre-shaping force, horizontal motion parameters, and vertical motion parameters.

8. The method according to claim 1, characterized in that, The profile morphology parameters include at least one of cross-sectional width, cross-sectional height, and profile curvature; the physical property information includes at least one of fat content, moisture content, and viscoelasticity.

9. A contour dynamic springback compensation system for improving the cutting accuracy of livestock and poultry meat using the method described in any one of claims 1 to 8, characterized in that, The system includes: The information acquisition unit is used to acquire the contour morphology parameters and the physical property information; The control unit is communicatively connected to the information acquisition unit and is equipped with the pre-shaping parameter prediction model and the springback compensation model. The control unit is also used to collect data after the meat is cut and slit after shaping, and to continuously update the pre-shaping parameter prediction model and / or the springback compensation model using the data after cutting and slitting. The data after cutting and slitting includes weight error data, contour deviation data and / or shape qualification rate data after cutting and slitting. The pre-shaping unit is communicatively connected to the control unit and is used to perform the pre-shaping operation; The secondary shaping unit is communicatively connected to the control unit and is used to perform the secondary shaping operation.

10. The system according to claim 9, characterized in that, The pre-shaping unit and / or the secondary shaping unit include a driving module and a sensor module, wherein the sensor module includes an angle sensor and a force sensor; The sensor module is used to detect motion parameters in real time and feed them back to the control unit; The control unit is used to adjust the control commands to the drive module based on the deviation between the motion parameters and the target parameters.

Images