Intelligent production adjusting method, device and system for meat flaking

Through data collection and intelligent algorithms, the extrusion angle of the slab machine is dynamically adjusted, which solves the problem of uneven thickness of the meat slices in the meat slice machine, and achieves accurate control of the thickness of the meat slices and improves the production efficiency.

CN118915825AInactive Publication Date: 2024-11-08YIYANG ZIYANG DISTRICT XIANGSHAN ANIMAL HUSBANDRY DEVELOPMENT CO LTD
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202410960362.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-17
Publication Date
2024-11-08
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

It is difficult for existing meat flake machines to achieve uniform control of the thickness of meat flakes during the flake process. It is mainly due to the irregular shape of the raw materials that cause changes in the cutting depth and angle of the blade, and the existing technology cannot achieve real-time prediction and active adjustment.

Method used

Data collection, support vector machine (SVM) model regression, backtracking and prediction compensation mechanism, and PID controller algorithm are used to dynamically adjust the extrusion angle of the chip slab through historical data and intelligent algorithms to achieve accurate prediction and control of the thickness of the meat slices.

Benefits of technology

The stability and consistency of the thickness of the meat slices is achieved, production efficiency is improved, manual intervention and defective rate is reduced, and product quality is significantly improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118915825A_ABST
    Figure CN118915825A_ABST
Patent Text Reader

Abstract

The invention discloses an intelligent production adjusting method, device and system for meat flaking. The invention relates to the technical field of food engineering. The independent variable x comprises the thickness difference fluctuation value s of the previous multiple rounds of slicing and the target sliced meat thickness T; inputting the independent variable x into a regression function f (x), executing mapping based on an RBF kernel function, and outputting a prediction dependent variable y '; executing a prediction compensation mechanism to obtain an extrusion angle EA (t) required at the current moment t so as to eliminate hysteresis; the extrusion angle EA (t) is substituted into a PID controller algorithm, an electric signal e (t) is output, and then the extrusion degree of the raw materials is controlled to accept flaking operation; according to the method, the independent variables (the historical slicing thickness fluctuation value and the target sliced meat thickness) and the dependent variables (such as the extrusion angle) are collected, regression prediction is carried out through the support vector machine (SVM), the extrusion angle needed for achieving the expected sliced meat thickness can be intelligently predicted, and therefore the control precision of the sliced meat thickness is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of food engineering, in particular to the technical field of flaking meat raw materials, and in particular to an intelligent production and adjustment method, device and system for meat flaking. Background Art

[0002] Meat slicer is a kind of equipment specially used to cut frozen meat or room temperature meat into slices of different thicknesses. It can easily handle the fast slices of finished meat such as pork, fish, chicken breast, etc. by using the powerful rotation of the blade, while reducing the loss of nutrients and protein in the meat.

[0003] Generally speaking, consistent meat slice thickness ensures uniform taste and texture after cooking. Uniform meat slices are easier to control cooking time and heat. However, in actual operation, the slicer easily leads to uneven slice thickness for the same raw material. This is mainly because the raw meat block is irregular in shape, with convexities or concavities with curvature characteristics, while the position of the slicer blade is relatively unchanged. Therefore, the slicing process will affect the cutting depth and angle of the blade, resulting in uneven thickness of the meat slices.

[0004] Therefore, in order to make the slicer maintain uniform thickness for the processed meat slices, some existing technologies have studied and improved this problem, such as:

[0005] A prior art [1] A meat slice detection device is disclosed, which can accurately measure the thickness of meat slices through the cooperation of grating and CCD camera. However, this prior art can only determine the thickness of meat slices after the meat slices have been formed, and process the meat slices with poor thickness according to the existing information, and provide debugging information to the flaking machine; it is not only inefficient, but also wastes resources, and cannot be directly adjusted and applied to the flaking machine body;

[0006] Several other existing technologies [2-4] All of them disclose slicers with adjustable thickness of meat slices, which can adjust the thickness of meat slices through a unique gear mechanism; however, meat slices of poor thickness may still be produced; because the user needs to visually observe the difference in thickness of the processed meat slices and make corresponding gear adjustments, there is a certain lag and subjectivity, and the efficiency is not high;

[0007] Another existing technology [5]The invention discloses a meat slice thickness detection and slicing technology, which can quantitatively detect the fat and meat pixels of the cut meat slices, thereby determining the quality of the meat slices in different batches and classifying them. However, this method still relies on the difference in thickness of the processed meat slices to implement the automatic sorting function. Although the thickness of different types of meat slices after sorting is consistent, it cannot fundamentally solve the problem that the thickness of the slicer is inconsistent during the meat slice processing process and cannot be adjusted intelligently and actively;

[0008] To this end, the present invention proposes a meat slicing intelligent production adjustment method, device and system.

[0009] The citations in the present invention are as follows:

[0010] [1] Ma Xiaoping, Ma Xiaojun. A thickness detection module for beef and mutton slices [P]. CN202321982182.8: 2024-04-19.

[0011] [2] Liu Chaojian, Han Kai. A meat planer with thickness adjustment function for pet food processing [P]. CN202220926959.8: 2022-12-16.

[0012] [3] Liu Yiying, Jin Lexin, Duan Xinyu, et al. A manual meat slicer [P]. CN201720843203.6: 2018-03-13.

[0013] [4] Gao Ding, Gao Yuan. Thickness adjustment component and intelligent slicer [P]. CN201721180537.6: 2018-08-21.

[0014] [5] Zhang Wei, Wang Chunxiao, Hu Mojie, et al. Beef processing system based on intelligent grading[P]. CN202310378066.3: 2023-09-19. Summary of the invention

[0015] In view of this, the embodiments of the present invention hope to provide a meat slicing intelligent production adjustment method, device and system to solve or alleviate the technical problems existing in the prior art, that is, how to establish a closed-loop feedback and machine control mechanism to enable the slicing machine to predict its thickness during the slicing process and actively perform corresponding production adjustments; the technical solution of the present invention is implemented as follows:

[0016] First, the intelligent production and adjustment method of meat slicers:

[0017] 1. Overview:

[0018] This solution aims to solve the above technical problems; through data collection, SVM model regression, backtracking and prediction compensation mechanism, and PID controller algorithm, etc., the thickness of the meat slices during the slicing process of the slicing machine is accurately predicted and controlled. By using historical data and intelligent algorithms to dynamically adjust the extrusion angle, the influence of the irregular shape of the raw meat block on the slicing effect is eliminated, ensuring the consistency of the thickness of the produced meat slices.

[0019] (II) Technical solution:

[0020] To achieve the above technical objectives, the following steps S1 to S4 need to be performed.

[0021] 2.1 Step S1, Data Engineering:

[0022] Collect independent variables x (input features) and dependent variables y (target variables), where:

[0023] 2.1.1 Independent variable x:

[0024] Including the thickness difference fluctuation value s of the previous multiple rounds of slicing and the target meat slice thickness T;

[0025] 1) Difference fluctuation value s: obtained by calculating the coefficient of variation (CV) of the thickness of the slices in several consecutive rounds of slicing to reflect the stability of the thickness:

[0026]

[0027] Where ω is the standard deviation of the thickness of the meat slices, and μ is the mean thickness of the meat slices.

[0028] 2) Target meat slice thickness T: an expected value set by the user, indicating the desired meat slice thickness.

[0029] 2.1.2 Dependent variable y:

[0030] The extrusion angle EA of the first multiple rounds of flaking is a parameter that needs to be adjusted during the flaking process to affect the thickness of the meat slices. In data collection, this value is recorded as a dependent variable.

[0031] 2.2 Step S2, execute SVM regression model:

[0032] Input the independent variable x into the regression function f(x), perform mapping based on the RBF kernel function, and output the predicted dependent variable y';

[0033] 2.2.1 Step S200, executing regression function f(x):

[0034]

[0035] Where: N is the number of support vectors. α i and α i * is the Lagrange multiplier obtained through the SVM training process; K(x,x i ) is the RBF kernel function, x i is a support vector in the training data set, and b is a bias term;

[0036] 2.2.2 Step S201, perform RBF kernel function mapping:

[0037]

[0038] Among them, σ is the width parameter, which controls the radial range of the function. Υ is the hyperparameter of the RBF kernel function, which controls how the similarity of data points decays with their distance in the original space.

[0039] The process of regression prediction is represented as based on the support vector x i , the corresponding Lagrange multiplier α i and α i * and the bias term b, and the predicted dependent variable y' (i.e. the predicted extrusion angle EA') is calculated through the regression function f(x). In this process, the dot product between the support vector and the input data is calculated and weighted, and the bias term is added.

[0040] 2.3SVM model training:

[0041] Before executing steps S1 to S2, the SVM regression model needs to be trained in advance before deploying and executing step S2; the steps include:

[0042] 2.3.1 The first step is to read the data set D:

[0043] Collect the independent variables x and dependent variables y in the historical data to form a data set D; and perform conventional data preprocessing operations (such as normalization and standardization).

[0044] 2.3.2 The second step is to construct and solve the optimization problem:

[0045] 2.3.2.1 Setting the optimization problem:

[0046] Finding an optimal hyperplane to divide the data is equivalent to dealing with a convex optimization problem:

[0047]

[0048] Where W is the normal vector of the hyperplane. C is the penalty coefficient, which is used to control the trade-off between model complexity and fitting data. i is the support vector x i The corresponding true value.i and ξi i * is a slack variable.

[0049] in, is to x i The function mapped to the high-dimensional space implicitly implements the dot product operation of the high-dimensional space through the RBF kernel function K(x,xi) (the so-called "kernel technique"). The dot product of the RBF kernel function to K(x,xi) is expressed as:

[0050] K(x i ,x j )=φ(x i )·φ(x j );

[0051] Among them, the dot product operation represents the similarity between two vectors in high-dimensional space.

[0052] Among them, ε is a parameter in the insensitive loss function that defines the error range allowed for regression, that is, the error range is constrained based on the following insensitive loss function:

[0053]

[0054] Where y is the true value (i.e. the corresponding dependent variable). ε is the radius of the insensitive zone, which is a user-defined parameter used to control the model's tolerance for prediction errors. When the absolute value of the difference between the predicted value and the true value is less than or equal to ε, the prediction is considered accurate and the loss is 0. Once the absolute value of the difference between the predicted value and the true value exceeds this range, the loss accumulates linearly.

[0055] 2.3.2.2 Solving the optimization problem:

[0056] Use the quadratic programming (QP) algorithm to find a set of Lagrange multipliers α i and α i *, which correspond to the data points and find the location of the hyperplane W.

[0057] First set the Lagrangian function L and introduce a set of Lagrangian multipliers in

[0058]

[0059] About L Taking the partial derivative and setting it to 0, we can get the following relationship:

[0060]

[0061] Substituting these relations into the Lagrangian function, we can obtain a function that only contains α i and α i * is the dual problem of the original problem. Then the quadratic programming (QP) algorithm is used to solve this dual problem. The QP algorithm is an efficient mathematical optimization method, especially suitable for solving optimization problems with quadratic objective functions and linear constraints. The QP algorithm can find the optimal α i and α i * values, and then they can be used to determine the position of the hyperplane W. Specifically, the normal vector W of the hyperplane can be determined by α i and α i * and the corresponding data point x i At the same time, the value of the bias term b can also be calculated through some support vectors.

[0062] 2.3.3 The third step is to use the dataset D combined with k-fold cross validation to train the SVM regression model, including the following steps:

[0063] 1) Randomly divide the data set D into k subsets of equal size, namely D1, D2, ..., D k .

[0064] 2) For i from 1 to k:

[0065] 2.1) Select subset D i As the test set D test , and the remaining subsets are combined as training set D train .

[0066] 2.2) Using training set D test Train the SVM model (solve the optimization problem) and apply the RBF kernel function K(x,x') to it:

[0067] 2.3) Using subset D i Test the trained SVM model and record the performance metric (recall).

[0068] 3) Calculate the average of k test results as the final evaluation of model performance.

[0069] 4) Repeat the above steps and use grid search to try different combinations of penalty coefficients C and hyperparameters Υ to find the parameter combination that optimizes model performance.

[0070] 2.4 Step S3, execute prediction compensation mechanism:

[0071] Based on the prediction dependent variable y' obtained in step S2, the prediction compensation mechanism is executed to obtain the required extrusion angle EA(t) at the current time t to eliminate the hysteresis;

[0072] 2.4.1 Step S300, obtaining error feedback coefficient β i :

[0073] Error feedback coefficient β i It is used to control the degree of compensation of the prediction error at each moment in the past for the current prediction value. It is determined by the least squares method to minimize the prediction error after compensation, that is, to minimize the following residual sum of squares:

[0074]

[0075] Where M is the total number of observations. This expression measures the difference between the model predictions and the actual observations. The goal of the least squares method is to find a set of β i The partial derivatives are calculated and set to zero, thus forming a linear system of equations that minimizes the residual sum of squares S. Solving this system of equations yields the β that minimizes S. i value.

[0076] 2.4.2 Step S301, prediction compensation mechanism:

[0077]

[0078] Where: y' t is the extrusion angle directly predicted by the SVM regression model at time t. t-i is the prediction error at time ti: e t-i =y t-i -y' t-i ; n is the number of historical error periods considered, that is, how many periods of historical prediction errors we look back to make compensation.

[0079] Its mechanism is that at each time t, the predicted value y' of the SVM regression model is recorded t and the actual observed value yt, and calculate the error e between them t By considering the forecast errors of the past n periods, the error feedback coefficient β i These errors are weighted and summed to obtain a total error compensation value. Then the extrusion angle y' directly predicted by the SVM regression model is t Added to the total error compensation value, the required extrusion angle EA(t) adjusted by the compensation mechanism is obtained. The adjusted extrusion angle EA(t) is then applied to the meat shavings production process to eliminate the hysteresis of the system and improve production efficiency and product quality.

[0080] 2.5 Step S4, execute PID controller algorithm:

[0081] The extrusion angle EA(t) is introduced into the PID controller algorithm, and the electrical signal e(t) is output, thereby controlling the degree of extrusion of the raw material to accept the flaking operation;

[0082]

[0083] Where: K p K is the proportional gain, which is used to amplify or reduce the error signal. i is the integral gain, used to eliminate steady-state errors. K d is the derivative gain, which is used to predict future deviations and react in advance to reduce overshoot and improve system stability.

[0084] error(t) is the deviation at time t, calculated as the set value minus the actual value: e(t) =

[0085] r(t)-y(t); wherein r(t) is the set value (the current extrusion angle EA), and y(t) is the actual value (the extrusion angle EA(t) obtained in step S3).

[0086] Error(τ) represents the deviation value of a certain time variable τ, which refers to the difference between the set value (expected value) and the actual value at the time point corresponding to the time variable τ.

[0087] The above PID controller algorithm can also be discretized and implemented:

[0088]

[0089] Where: t is the discrete time step. error(t) is the deviation at time step t. error(j) is the deviation at time step j.

[0090] In practice, step S4 follows the following steps:

[0091] S400, at each time t, calculate the deviation error(t) between the set value (desired squeezing angle) and the actual value (current squeezing angle EA(t)).

[0092] S401, PID calculation: Using the PID controller algorithm, the output electrical signal e(t) is calculated according to the deviation, the accumulation of the deviation and the rate of change of the deviation. The calculated electrical signal e(t) is output to the actuator (i.e., the second conveyor belt assembly described later) to control the degree of extrusion of the raw material.

[0093] S402, read the dictionary table: In order to quickly respond to different extrusion angles EA(t), a dictionary table can be established in advance for different extrusion angles, and the corresponding electrical signal values ​​can be stored in the table. In this way, in the actual control process, the approximate electrical signal value can be directly obtained by looking up the table, reducing the calculation time. At the same time, the size parameters of the raw material extrusion part corresponding to different extrusion angles EA(t) and the size parameters of the shavings should also be preset in the dictionary table;

[0094] S403, determining the difference of the electric signal: determining a small adjustment amount by calculating the difference of the electric signal e(t) to achieve a smoother control process.

[0095] (III) Principles for solving technical problems:

[0096] 3.1SVM model regression:

[0097] The VM model can learn from historical data and predict the extrusion angle that should be used in the current situation to achieve the target slice thickness T, allowing the system to intelligently adjust the extrusion angle based on past experience and current goals.

[0098] 3.2 Retrospective and forecast compensation mechanism:

[0099] The predicted value of the extrusion angle is further optimized through the backtracking and prediction compensation mechanism, so that the predicted value can be fine-tuned to ensure a more precise control effect.

[0100] 3.3PID controller algorithm execution:

[0101] The optimized extrusion angle EA(t) is executed by the PID controller algorithm. The PID controller can dynamically adjust the extrusion angle according to the real-time feedback data to ensure the stability and consistency of the meat slice thickness. This closed-loop control system can effectively deal with various interference factors in the production process and ensure the precise control of the meat slice thickness.

[0102] Second, intelligent production and adjustment device for meat slicers:

[0103] It includes a CCD camera, a feeding mechanism and a shaving mechanism installed on the frame.

[0104] The feeding mechanism conveys the raw material and forms an extrusion portion facing the flaking mechanism. When the extrusion portion is cut by the flaking mechanism according to its size, a meat slice having a thickness corresponding to the size of the extrusion portion is formed;

[0105] The CCD camera detects the thickness of the meat slices. After executing the intelligent production adjustment method as described above, the electrical signal e(t) is input into the PLC controller. The PLC controller controls the feeding mechanism and adjusts the extrusion degree of the extrusion part so that meat slices of target thickness can be formed when the meat slices are shaving by the rotation.

[0106] Thirdly, the intelligent production and adjustment system for meat slicers:

[0107] like Figure 6 As shown, the system includes a processor and a memory connected to the processor, wherein the memory stores program instructions. When the program instructions are executed by the processor, the processor controls the intelligent production regulation device as described above to execute the intelligent production regulation method as described above.

[0108] Compared with the prior art, the present invention has the following beneficial effects:

[0109] 1. Intelligent prediction and decision-making: The present invention collects independent variables (historical slicing thickness fluctuation values, target meat slice thickness) and dependent variables (such as extrusion angle), and uses support vector machine (SVM) for regression prediction. It can intelligently predict the extrusion angle required to achieve the expected meat slice thickness, thereby improving the control accuracy of meat slice thickness.

[0110] 2. Dynamic adjustment and optimization: The present invention includes a backtracking and prediction compensation mechanism that can be dynamically adjusted based on previous prediction results and actual feedback. It effectively reduces prediction errors and enables the system to quickly adapt to changes in the production process. In addition, the PID controller algorithm can monitor the thickness of the meat slices in real time and adjust the extrusion angle based on the difference between the actual thickness and the expected thickness. This real-time feedback mechanism ensures the stability of the production process and the consistency of the thickness of the meat slices.

[0111] 3. Double improvement of efficiency and quality: The present invention reduces the downtime adjustment time caused by the thickness of the meat slices through the intelligent prediction and adjustment of the algorithm. It directly improves the production efficiency and makes the slicing process smoother and more efficient. At the same time, the algorithm significantly reduces the defective rate caused by uneven thickness by accurately controlling the thickness of the meat slices. It not only saves costs, but also improves the overall product quality.

[0112] 4. Reduce manual intervention: The technical solution of the present invention eliminates the need for operators to frequently make manual adjustments, which reduces work intensity and improves the convenience of operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0113] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0114] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0115] Figure 2 It is an overall three-dimensional schematic diagram of the device of the present invention;

[0116] Figure 3 is a three-dimensional schematic diagram of a flaking mechanism in the device of the present invention;

[0117] Figure 4 is a three-dimensional schematic diagram of a feeding mechanism in the device of the present invention;

[0118] Figure 5 Schematic diagram of the training method of the SVM regression model of the present invention;

[0119] Figure 6 It is a schematic diagram of the system composition of the present invention;

[0120] Figure 7 It is a line graph of training effect evaluation in Example 2;

[0121] Figure 8 It is a line chart of the comparative effect of the test example.

[0122] Figure numerals: 1. frame shape; 2. CCD camera; 3. feeding mechanism; 301. lower frame; 302. upper frame; 303. first motor; 304. first transmission belt assembly; 305. first conveyor belt assembly; 306. connecting frame; 307. second conveyor belt assembly; 308. second motor; 4. flaking mechanism; 401. knife disc; 402. third motor; 403. second transmission belt assembly; 5. PLC controller; 6. display. DETAILED DESCRIPTION

[0123] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below;

[0124] It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same or similar parts between the various embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0125] Example 1: Please refer to Figure 1 , this embodiment discloses an intelligent production and adjustment method for meat slices; the method is mainly used in the meat processing industry, especially when using a slicer to cut meat slices, to ensure the stability and consistency of the thickness of the meat slices through intelligent prediction and adjustment. This method needs to be performed on a slicer, and must be implemented based on historical data after the slicer has cut at least one slice of meat. In addition, in order to maintain the continuity of the production process and the consistency of the quality of the meat slices, the set target meat slice thickness T needs to be kept unchanged during each subsequent slice process. If the set value of the target meat slice thickness T changes, the entire adjustment method needs to be restarted. The method includes the following steps S1 to S4.

[0126] In this embodiment, regarding step S1, data engineering: collect independent variables x (input features) and dependent variables y (target variables), where:

[0127] 1) The independent variable x includes the thickness difference fluctuation value s of the previous rounds of slicing and the target slice thickness T;

[0128] 1.1) Difference fluctuation value s: obtained by calculating the coefficient of variation CV of the thickness of the slices in consecutive rounds of slicing to reflect the stability of the thickness; the coefficient of variation is a standardized measure used to reflect the degree of dispersion or stability of a set of data. Here, it is used to quantify the stability of the thickness of the slices:

[0129]

[0130] Among them, ω represents the standard deviation of the meat slice thickness, which is used to measure the degree of dispersion of the meat slice thickness data; μ represents the average value of the meat slice thickness, which is used to indicate the central trend of the meat slice thickness. By calculating the coefficient of variation, a standardized fluctuation value can be obtained. The larger this value is, the worse the stability of the meat slice thickness is; conversely, the stability of the meat slice thickness is better.

[0131] 1.2) Target slice thickness T: Target slice thickness T is an expected value set by the user, indicating the slice thickness that is expected to be achieved during the slicing process. This value is fixed and is used to guide the cutting depth of the slicing machine to ensure that the thickness of the slices produced meets the expectations.

[0132] 2) Dependent variable y: The dependent variable y is the extrusion angle EA of the previous round or rounds of flaking. The extrusion angle is an important parameter that needs to be adjusted during the flaking process (see the solution in Example 4 for details), which directly affects the thickness of the meat slices. During the data collection stage, this value will be recorded as the dependent variable for subsequent analysis of its relationship with the thickness of the meat slices.

[0133] Furthermore, the python execution program of step S1 is as follows:

[0134]

[0135]

[0136] Among them, the MeatSlicingData class manages the collection and storage of slice thickness data and extrusion angle data. It has three main attributes: target_thickness (target slice thickness), slice_thicknesses (a list recording historical slice thickness), and extrusion_angles (a list recording historical extrusion angles). The update_data method is used to update the historical data of slice thickness and extrusion angle. After each slicing, this method is called to record new data points. The calculate_variability method calculates the coefficient of variation (CV) of slice thickness, that is, the ratio of the standard deviation to the mean. This method requires thickness data of at least two slices of meat to calculate. The get_input_features and get_target_variable methods are used to obtain the independent variables x (difference fluctuation value s and target slice thickness T) and the dependent variable y (the most recent extrusion angle EA), respectively.

[0137] In this embodiment, regarding step S2, the SVM regression model is executed: the independent variable x is input into the regression function f(x), mapping is performed based on the RBF kernel function, and the predicted dependent variable y' is output; it includes the following steps S200-S201.

[0138] Specifically, step S200, executing the regression function f(x):

[0139]

[0140] Among them: only a part of the training data points (support vector x i ) make a real contribution to the model. These support vectors are expressed by the Lagrange multiplier α i and α i * Play a role in the model. These multipliers are obtained through the training process of SVM. RBF kernel function K(x,x i) represents the Radial Basis Function (RBF) kernel function, also known as the Gaussian kernel function. The RBF kernel function can map data from the original space to a higher-dimensional feature space, thereby making the data that was originally linearly inseparable in the original space linearly separable. This mapping can enhance the expressiveness of the model and enable it to handle more complex nonlinear relationships. b is the bias term, which is used to adjust the output of the model to ensure that the model can fit the data more accurately.

[0141] Specifically, step S201 executes the RBF kernel function K(x,x i ) mapping:

[0142]

[0143] Among them, x is the input independent variable (including the difference fluctuation value s and the target meat slice thickness T), x i is a support vector in the training data set, σ is the width parameter, which controls the radial range of the function, that is, determines the influence range of the function. Υ is a hyperparameter of the RBF kernel function, which controls how the similarity of data points decays with their distance in the original space.

[0144] It can be understood that the process of regression prediction is represented as based on the support vector x i , the corresponding Lagrange multiplier α i and α i * and the bias term b, and the predicted dependent variable y' (i.e. the predicted extrusion angle EA') is calculated through the regression function f(x). This process involves the dot product operation between the support vector and the input data, the weighted summation, and the addition of the bias term.

[0145] Furthermore, step S201 constrains the error range based on the following insensitive loss function:

[0146]

[0147] Where y is the true value (i.e. the corresponding dependent variable). ε is the radius of the insensitive zone, which is a user-defined parameter used to control the model's tolerance for prediction errors. When the absolute value of the difference between the predicted value and the true value is less than or equal to ε, the prediction is considered accurate and the loss is 0. Once the absolute value of the difference between the predicted value and the true value exceeds this range, the loss accumulates linearly.

[0148] It can be understood that through the RBF kernel function K(x,x i) maps the data to a high-dimensional space, and the model is able to capture more complex nonlinear relationships, thereby improving prediction accuracy. This helps to control the thickness of the meat slices more accurately. The use of an insensitive loss function allows the model to have a certain tolerance for prediction errors, which helps to reduce the risk of overfitting and improve the generalization ability of the model. This means that the model can still maintain good prediction performance when faced with new, unseen data. By adjusting the parameters of the RBF kernel function (such as σ and Υ) and the radius ε of the insensitive loss function, the performance of the model can be optimized according to specific needs. This flexibility and adjustability allows the method to adapt to different production environments and meat slice characteristics.

[0149] Furthermore, the Python execution program of step S2 is as follows:

[0150] from sklearn.svm import SVR

[0151] import joblib

[0152] import numpy as np

[0153] #A trained SVM regression model is saved as 'svm_regression_model.pkl'

[0154] model_path='svm_regression_model.pkl'

[0155] svm_model = joblib.load(model_path) # Load the pre-trained model

[0156] #A new set of input data x_new, which is a two-dimensional numpy array

[0157] #Each row represents a sample, and each column represents a feature (difference fluctuation value and target meat slice thickness)

[0158] x_new=np.array([[0.1,1.5],[0.2,1.6],[0.15,1.4]])

[0159] #Use the loaded model to make predictions

[0160] y_pred=svm_model.predict(x_new)

[0161] # Output the predicted extrusion angle

[0162] print("Predicted extrusion angle:",y_pred)

[0163] In the above program, the scikit-learn library is used to load the class of the SVM model; then the joblib.load function is used to load a pre-trained SVM regression model from the disk. This model has learned how to predict the output (extrusion angle) based on input features (such as difference fluctuation values ​​and target meat slice thickness) during the training phase. By inputting a new set of input data x_new, the new input data is predicted using the predict method of the loaded SVM regression model. This method will execute the calculation steps of the SVM regression model internally, including executing the regression function f(x), applying the RBF kernel function for mapping, and performing weighted summation based on the support vector and the corresponding Lagrange multiplier. Finally, the predicted extrusion angle y_pred is output. These predicted values ​​can be directly used to guide the real-time adjustment of the meat slice production line to achieve the desired meat slice thickness. The specific architecture and program of the SVM model will be disclosed by Example 2.

[0164] In this embodiment, regarding step S3, a prediction compensation mechanism is executed: based on the prediction dependent variable y' obtained in step S2, a prediction compensation mechanism is executed to obtain the required extrusion angle EA(t) at the current time t to eliminate hysteresis; it includes the following steps S300~S301.

[0165] Specifically, step S300, obtaining the error feedback coefficient β i : This coefficient controls the degree to which the prediction errors at each moment in the past compensate for the current prediction value. In short, the error feedback coefficient β i It can help the system "remember" and "correct" past forecast errors, thereby improving the accuracy of current forecasts.

[0166] In order to determine these error feedback coefficients β i , the least squares method is used to find the best function matching the data by minimizing the residual sum of squares S. Its goal is to find a set of β i value, so that the prediction error after compensation is minimized.

[0167] First, define a residual sum of squares function S(β1,β2,…,β n ), which measures the difference between the model prediction and the actual observed value. This function considers all data points from time n+1 to the total number of observed data M. For each time t, the difference between the extrusion angle EA(t) and the predicted value after error compensation is calculated, and the squares of these differences are accumulated. In order to find the β that minimizes S i value, for S with respect to each β i Taking partial derivatives and setting them equal to zero results in a system of linear equations:

[0168]

[0169] By solving this system of equations, we can obtain a set of β that minimizes S. i The benefit of this method is that it not only considers the current predicted value y't, but also takes into account the past prediction errors, thereby optimizing the prediction results more comprehensively. In this way, the algorithm of this embodiment can more accurately control the thickness of the meat slices, improve production efficiency and product quality. At the same time, this data-based dynamic adjustment strategy also makes the system more adaptable and flexible, and can cope with changes in different production environments and product requirements.

[0170] Specifically, step S301, prediction compensation mechanism: It is based on the extrusion angle y't directly predicted by the SVM regression model at time t and the prediction error e in the past n periods t-i Specifically, the mechanism is adjusted by the following formula:

[0171]

[0172] Where: y' t is the extrusion angle directly predicted by the SVM regression model at time t. t-i is the prediction error at time ti: e t-i =y t-i -y' t-i ; n is the number of historical error periods considered, that is, how many periods of historical prediction errors we look back to make compensation.

[0173] The mechanism works as follows: At each time t, the system records the predicted value y't of the SVM regression model and the actual observed value y t , and calculate the error e between them t Then, the system considers the forecast errors of the past n periods and uses the error feedback coefficient β previously determined by the least squares method i These errors are weighted and summed to obtain a total error compensation value, which reflects the impact of past forecast errors on current forecasts and helps correct possible systematic deviations.

[0174] Finally, the extrusion angle y' t Adding the total error compensation value, we get the required squeeze angle EA(t) adjusted by the compensation mechanism. This adjusted squeeze angle is more accurate because it not only takes into account the current prediction value, but also integrates the past prediction errors, thus effectively eliminating the lag of the system.

[0175] Furthermore, applying the adjusted extrusion angle EA(t) to the meat slice production process can significantly improve production efficiency and product quality. This is because precise extrusion angle control can ensure the uniformity of meat slice thickness, reduce the production of unqualified products, and improve the stability and efficiency of the production line. This predictive compensation mechanism makes the entire production system more intelligent and adaptive, and can better cope with changes in various production environments and conditions.

[0176] Furthermore, the Python execution program of step S3 is as follows:

[0177]

[0178]

[0179]

[0180] In the above program, two empty lists ea_list and error_list are first initialized to store the compensated squeeze angle and prediction error respectively. At the same time, in order to calculate the prediction error of the previous n periods, n zeros are first filled in error_list. Then the prediction value list of the SVM regression model is traversed. At each time t, the prediction error of the current moment is calculated and added to error_list. Then the total error compensation value is calculated using the error feedback coefficient βi and the prediction error of the past n periods. This value is obtained by multiplying the prediction error of each period with its corresponding error feedback coefficient and adding the results. Finally, the total error compensation value is added to the prediction value of the SVM regression model to obtain the compensated squeeze angle and add it to ea_list. In order to keep the length of error_list always n, its length is checked at the end of each loop. If the length exceeds n, the first element in the list is deleted. Finally, the list of compensated squeeze angles is output.

[0181] In this embodiment, regarding step S4, a PID controller algorithm is executed: the extrusion angle EA(t) is brought into the PID controller algorithm, an electrical signal e(t) is output, and the degree of extrusion of the raw material is controlled to receive the flaking operation;

[0182]

[0183] Where: K p K is the proportional gain, which is used to amplify or reduce the error signal. i is the integral gain, used to eliminate steady-state errors. K d is the derivative gain, which is used to predict future deviations and react in advance to reduce overshoot and improve system stability.

[0184] error(t) is the deviation at time t, calculated as the set value minus the actual value:

[0185] e(t)=r(t)-y(t);

[0186] Wherein r(t) is the set value (the current extrusion angle EA), and y(t) is the actual value (the extrusion angle EA(t) obtained in step S3).

[0187] Error(τ) represents the deviation value of a certain time variable τ, which refers to the difference between the set value (expected value) and the actual value at the time point corresponding to the time variable τ.

[0188] It is understandable that in order to adapt to digital control systems, the above PID controller algorithm can also be discretized:

[0189]

[0190] Where: t represents the discrete time step. error(t) is the deviation at time step t. error(j) is the deviation at time step j. By accumulating past deviation values ​​(integral term) and calculating the change in deviation between adjacent time steps (differential term), the discrete PID control output can be obtained.

[0191] Substituting the extrusion angle EA(t) into the PID controller algorithm, the output electrical signal e(t) will be used to control the degree of extrusion of the raw material. In this way, we can accurately adjust the state of the raw material before the flaking operation, thereby improving the quality and production efficiency of the meat slices. This intelligent adjustment method not only improves the automation level of the production process, but also enhances the system's ability to adapt to raw material changes and external interference.

[0192] Furthermore, the python execution program of step S4 and its linkage with the PLC controller 5 in the fourth embodiment is as follows:

[0193]

[0194]

[0195]

[0196] In the above program, the parameters of the PID controller (Kp, Ki, Kd) are first set, and the variables used for PID calculation (previous_error and integral) are initialized. At the same time, a serial port connection is established for communication with the PLC controller. The discretized PID controller algorithm is implemented in the pid_controller function. This function receives the set value (expected value), actual value, time step and current time step as input and calculates the electrical signal e(t). The global variables previous_error and integral are used in the calculation process to track the error and integral of the previous time step. The pid_controller function is called at each time step to calculate the electrical signal e(t), which is then converted into a string format and sent to the PLC controller through the serial port. The serial port connection is closed after all time steps are calculated.

[0197] Example 2: Please refer to Figure 1 , 5 Although the first embodiment uses an SVM model, the SVM model needs to be trained in advance before the corresponding step S2 can be executed. This embodiment will disclose the training method of the SVM model used in the first embodiment, and its steps include the following P1 to P3:

[0198] In this embodiment, regarding step P1, read the data set D: collect the independent variable x and its corresponding dependent variable y in the historical data to form the data set D: D = [(x1, y1), (x2, y2), ..., (x m ,y m )];

[0199] Among them, (x i ,y i ) is an arbitrary sample, i is the sample index, and m is the number of samples.

[0200] Then, the conventional normalization and standardization are performed. Normalization is to adjust the range of the data to [0, 1] or [-1, 1] to improve the training efficiency and performance of the model. Standardization (also known as z-score standardization) is to convert the data into a distribution with a mean of 0 and a standard deviation of 1.

[0201] Furthermore, the Python execution program of step P1 is as follows:

[0202]

[0203]

[0204] In the above program, the read_csv function of the pandas library is used to read the dataset file in CSV format. This function can easily process the data in the CSV file and convert it into a DataFrame object, which is a two-dimensional, resizable, labeled data structure. The data of the independent variables (features) and dependent variables (targets) are extracted from the DataFrame.

[0205] sklearn.preprocessing.StandardScaler performs data standardization. The fit_transform method of StandardScaler first calculates the mean and standard deviation of the data (fit), and then uses these parameters to transform the data (transform) to achieve standardization.

[0206] In this embodiment, regarding step P2, an optimization problem is constructed and solved:

[0207] (1) First, set the optimization problem: find an optimal hyperplane to divide the data, which is equivalent to dealing with a convex optimization problem:

[0208]

[0209] Where W is the normal vector of the hyperplane, which determines the direction of the hyperplane. C is the penalty coefficient, which is used to control the trade-off between model complexity and fitting data. This parameter will be determined in the subsequent training process. i and ξi i * is a slack variable used to handle noise and outliers in the data and make the model more robust.

[0210] Specifically, The purpose of this is to map the data from the original space into a higher dimensional space so that a linear hyperplane can be found to partition the data. However, directly computing the dot products in this high dimensional space may be computationally infeasible. Therefore, a kernel trick is used to implicitly compute these dot products:

[0211] K(x i ,x j )=φ(x i )·φ(x j );

[0212] The dot product operation represents the similarity between two vectors in a high-dimensional space. The RBF kernel function can measure the similarity between two samples even if they are not linearly separable in the original space.

[0213] Among them, ε is a parameter in the insensitive loss function that defines the error range allowed for regression, that is, the error range is constrained based on the following insensitive loss function:

[0214]

[0215] Where y is the true value (i.e. the corresponding dependent variable). ε is the radius of the insensitive zone, which is a user-defined parameter used to control the model's tolerance for prediction errors. When the absolute value of the difference between the predicted value and the true value is less than or equal to ε, the prediction is considered accurate and the loss is 0. Once the absolute value of the difference between the predicted value and the true value exceeds this range, the loss accumulates linearly.

[0216] It is understandable that the optimization problem constructed above expresses our desire to:

[0217] 1.1) Finding the optimal hyperplane: This hyperplane should be able to best separate or regress the data while maintaining a certain generalization ability and avoiding overfitting.

[0218] 1.2) Balance model complexity and data fitting: By minimizing |w| 2 (i.e., the square norm of the hyperplane normal vector), find a simple model (i.e., the weight vector is as small as possible) to prevent the model from being too complicated. At the same time, by introducing the penalty coefficient C and the slack variable ξ i and i *, the model can tolerate classification or regression errors to a certain extent to achieve better data fitting. This reflects finding a balance between model complexity and data fitting.

[0219] 3) Handling nonlinear and noisy data: through mapping function The use of kernel techniques, especially the RBF kernel function, enables the optimization problem to handle nonlinearly separable data in the original space. In addition, the existence of slack variables enables the model to handle noise and outliers in the data, improving the robustness of the model.

[0220] 4) Control the tolerance of prediction error: The introduction of the insensitive loss function and parameter ε allows us to control the tolerance of the model to prediction errors. The model's requirements for prediction accuracy can be set according to the specific application scenario, and the acceptable prediction error can be defined by adjusting ε.

[0221] In summary, this optimization problem aims to find a hyperplane that is both simple (avoiding overfitting) and can effectively fit the data (including nonlinear data and noisy data), while allowing us to adjust the tolerance to prediction errors according to actual needs.

[0222] (2) Then solve this optimization problem: find a set of Lagrange multipliers α through the quadratic programming algorithm i and α i*, which correspond to the data points and find the location of the hyperplane W.

[0223] First set the Lagrangian function L and introduce a set of Lagrangian multipliers in

[0224]

[0225] About L Find the partial derivative and set it to 0. This will give us a series of relationships that convert the original variables With Lagrange multipliers The specific relationship is as follows:

[0226]

[0227] Substituting these relations into the Lagrangian function, we can obtain a function that only contains α i and α i * is the dual problem of the original problem. Then the quadratic programming (QP) algorithm is used to solve this dual problem. The QP algorithm is an efficient mathematical optimization method, especially suitable for solving optimization problems with quadratic objective functions and linear constraints. Through the QP algorithm, we can find the optimal α i and α i * values, and then they can be used to determine the position of the hyperplane W. Specifically, the normal vector W of the hyperplane can be determined by α i and α i * and the corresponding data point x i At the same time, we can also calculate the value of the bias term b through some support vectors.

[0228] It is understandable that in step P2, by transforming the dual problem, the original complex optimization problem is simplified to a quadratic programming problem involving only Lagrange multipliers, thereby reducing the difficulty of solving. By solving the dual problem, the data points (i.e., support vectors) that have a decisive influence on the position of the hyperplane can be determined, which is of great significance for understanding the data distribution and model decision boundary. Moreover, by optimizing the regularization terms and slack variables in the problem, the complexity of the model can be controlled and noisy data can be processed, thereby improving the generalization ability of the model on unseen data.

[0229] Furthermore, the Python execution program of the above step P2 (which relies on the cvxopt library to implement the QP algorithm) is as follows:

[0230]

[0231]

[0232]

[0233]

[0234] In the above program, the fit method is used to train the model, which constructs and solves the quadratic programming problem. The predict method is used to make predictions for new data. The kernel function is passed as a parameter to the constructor of SVMOptimizer. The other functions are built-in programs of the cvxopt library and will not be described in detail.

[0235] In this embodiment, regarding step P3, the SVM regression model is trained using the data set D in combination with k-fold cross validation, including the following steps:

[0236] P3.1) Randomly divide the data set D into k subsets of equal size, namely D1, D2, ..., D k .

[0237] P3.2) For i from 1 to k:

[0238] P3.2.1) Select subset D i As the test set D test , and the remaining subsets are combined as training set D train This practice ensures that every data point has a chance to be used as test data, providing a more accurate assessment of model performance.

[0239] P3.2.2) Use training set D test Train the SVM model (solve the optimization problem) and apply the RBF kernel function K(x,x') to it; it helps the model better handle nonlinear relationships in high-dimensional space and helps solve complex problems such as predicting the thickness of meat slices.

[0240] P3.2.3) Use subset D i Test the trained SVM model and record the performance metric (recall).

[0241] P3.3) Calculate the average of k test results as the final evaluation of model performance. After completing k rounds of cross-validation, calculate the average of all rounds of test results, and use this average as the final evaluation indicator of model performance. In this way, a more robust and reliable model performance evaluation result can be obtained. At the same time, record the accuracy and recall rate during the training process and form the following Figure 7 Comparison chart shown.

[0242] P3.4) Repeat the above steps and use grid search to try different combinations of penalty coefficients C and hyperparameters Υ to find the parameter combination that makes the model perform best.

[0243] P3.5) When the model converges or reaches the predetermined number of iterations, the best combination of penalty coefficient C and hyperparameter Υ is selected, and the bias term b and slack variable ξi updated in the last round of training are replaced by i and ξi i *、Lagrange multiplier α i and α i *, the normal vector W of the hyperplane and several support vectors x i as the optimal model parameters.

[0244] Furthermore, the Python execution program of step P3 above is as follows (the program relies on the implementation of SVM provided by the cikit-learn library as well as the grid search and cross-validation tools):

[0245]

[0246]

[0247]

[0248] In the above program, SVR is the implementation of the support vector machine regression model in scikit-learn, and the Gaussian function of the Euclidean distance between data points is used to measure similarity. KFold divides the data set into k consecutive folds, using k-1 folds as training data each time and the remaining fold as test data. This process is repeated k times, and each fold is used as a test set, thereby providing k performance indicators. The GridSearchCV class implements a grid search of parameters. It tries all parameter combinations and uses cross-validation to evaluate the performance of each set of parameters. In the end, it selects the best performing parameter combination. After training, joblib.dump is used to save the best model to a file for future use. The SVM model trained by the entire program is encapsulated in 'svm_regression_model.pkl', which can be used by the program of step S2 in Example 1.

[0249] Example 3: Please refer to Figure 1 In order to make the PID controller algorithm of step S4 in embodiment 1 more intelligent, step S4 should follow the following steps:

[0250] S400: Calculate the deviation: At each time t, first calculate the deviation error(t) between the set value (i.e. the expected extrusion angle) and the actual value (the current extrusion angle EA): error(t) = set value - EA(t). This deviation is the basis for the PID controller to adjust, reflecting the gap between the current system state and the expected state.

[0251] S401: PID calculation and output: Use the PID controller algorithm to process the deviation. The PID controller calculates the output electrical signal e(t) based on the deviation, the accumulation of the deviation (integral term) and the rate of change of the deviation (differential term). This electrical signal is the key to controlling the degree of extrusion of the raw material.

[0252] S402: Read the dictionary table. In order to improve the response speed and efficiency of the control process, a dictionary table is established in advance for different extrusion angles. The dictionary table stores the electrical signal values ​​corresponding to different extrusion angles. In the actual control process, when the extrusion angle needs to be adjusted, the processor can directly obtain the approximate electrical signal value by looking up the table, thereby greatly reducing the time for real-time calculation. In addition, the dictionary table also presets the size parameters of the raw material extrusion part corresponding to different extrusion angles EA(t) and the size parameters expected to be shaved, which provides an important reference for subsequent production adjustments.

[0253] S403: Determine the difference of the electrical signal In order to achieve a smoother control process, it is necessary to calculate the difference of the electrical signal e(t). This difference reflects the small adjustment of the electrical signal, which helps the controller to make fine adjustments. By continuously calculating and adjusting this difference, it can be ensured that the degree of extrusion of the raw material gradually approaches the set value, thereby achieving accurate control of the thickness of the meat slice.

[0254] It can be understood that this embodiment can achieve precise control of the degree of raw material extrusion through precise calculation and adjustment of the PID controller and the assistance of the dictionary table, thereby improving the uniformity and consistency of the thickness of the meat slices. The introduction of the dictionary table greatly reduces the time of real-time calculation, enables the system to respond more quickly to changes in the extrusion angle, and improves production efficiency. Moreover, by calculating the difference of the electrical signal and making fine adjustments, the smoothness of the control process can be ensured, avoiding excessive fluctuations or overshoot, thereby protecting the production equipment and improving product quality.

[0255] Example 4: Please refer to Figure 2-4 This embodiment discloses an intelligent production and adjustment device for meat slicing; it includes a CCD camera 2, a feeding mechanism 3 and a slicing mechanism 4 installed on a frame 1;

[0256] In this embodiment, the composition and principle of the above mechanism will be gradually introduced according to the execution process of the device. First, the feeding mechanism 3: please refer to Figure 2 , 4 Its general shape includes a lower frame 301 and an upper frame 302 installed on the upper part thereof.

[0257] Specifically, the lower frame 301 is provided with a first conveyor belt assembly 305, which is consistent with a conventional synchronous toothed conveyor belt (timing belt), including a belt body with a toothed cross-section, a driving roller meshed in the middle of the belt body, and two driven rollers located on both sides of the belt body (because it is a prior art, no number is shown in the figure), and when the driving roller rotates, the belt body can be driven to move;

[0258] Specifically, the upper frame 302 is provided with a second conveyor belt assembly 307 similar to the first conveyor belt assembly 305, and its mechanical principle is consistent with the first conveyor belt assembly 305 (but the functional purpose is different, which will be introduced later), and will not be repeated here; however, it should be particularly pointed out that the belt body of the first conveyor belt assembly 305 is horizontally arranged, while the belt body of the second conveyor belt assembly 307 is inclined compared with the belt body of the first conveyor belt assembly 305;

[0259] Furthermore, in order to drive the belt bodies of the first conveyor belt assembly 305 and the second conveyor belt assembly 307 to move synchronously, the feeding mechanism 3 further includes a first motor 303 and a first transmission belt assembly 304, the first transmission belt assembly 304 includes a driving wheel (fixedly connected to the output shaft of the first motor 303), and two driven wheels, and the two driven wheels establish a transmission relationship with the driving wheel through two transmission belts; wherein:

[0260] 1) A driven wheel is fixedly connected to the driving roller of the first conveyor belt assembly 305, so that when the first motor 303 is started, the belt body of the first transmission belt assembly 304 can roll accordingly; in this scheme, when the meat raw material to be sliced ​​(hereinafter referred to as "raw material") is placed on the above-mentioned belt body, the belt body can transport the raw material according to the above-mentioned principle to the flaking mechanism 4 to receive flaking;

[0261] 2) Another driven wheel is fixedly connected to the driving roller of the second conveyor belt assembly 307; when the raw material moves on the belt body of the first transmission belt assembly 304 to the second conveyor belt assembly 307, due to the above-mentioned inclination, the second conveyor belt assembly 307 forms an extrusion portion with the portion of the raw material facing the flaking mechanism 4, and the flaking mechanism 4 performs rotational flaking on the extrusion portion and outputs the meat slices;

[0262] It should be pointed out that if Figure 4 As shown, when the second conveyor belt assembly 307 performs extrusion, its belt body is not in full contact with the raw material, but the belt body on the driven roller at the flaking mechanism 4 performs extrusion, thereby controlling the forming orientation of the extrusion portion.

[0263] It is understandable that the extruded portion formed by the extruded raw material has a denser texture, which is convenient for the flaking mechanism 4 to cut and flake. However, this will affect the thickness of the flakes, because when the density of the raw material is high, its ability to resist cutting is enhanced, resulting in an increase in the resistance encountered by the flaking mechanism 4 during the cutting process. Therefore, under the same cutting conditions, raw materials with high density produce thinner flakes, because it is more difficult for the flaking mechanism 4 to cut into the raw material. On the contrary, raw materials with lower density provide less resistance during cutting, and the flaking mechanism 4 is easier to cut into, thereby producing thicker flakes.

[0264] It can be further understood that the conventional flaking machine is prone to cause the thickness of the sliced ​​meat from the same raw material to be non-uniform in actual operation, mainly because the raw meat block is irregular in shape and has convexities or concave areas with curvature characteristics, which will affect the cutting depth and angle of the blade during the flaking process, thereby causing the thickness of the meat slices to be non-uniform; however, when the above raw material is squeezed, the convexities or concave areas of the curved surface are eliminated due to the flexible characteristics of the meat, so that the flaking mechanism 4 is not easily affected by the above convexities or concave areas during the cutting process. The principle is very simple common sense: when we cut meat manually, we manually press the meat cutting position to cut the meat, and the thickness is easier to control than directly cutting the meat. In the technical solution of this embodiment, the second conveyor belt assembly 307 is essentially used as the function of manually pressing the meat cutting position. That is, the feeding mechanism 3 conveys the raw material and forms an extrusion portion facing the flaking mechanism 4. When the extrusion portion is cut by the flaking mechanism 4 according to its size, a meat slice of thickness corresponding to the size of the extrusion portion will be formed.

[0265] Therefore, in the technical solution provided in this embodiment, in order to realize dynamic adjustment of the extrusion part, so as to provide a basis for controlling the thickness of the shavings, please continue to refer to Figure 4 : The feeding mechanism 3 further introduces a connecting frame 306, which is rotatably matched in the upper frame body 302 through a bearing, and the driving roller and the driven roller of the second conveyor belt assembly 307 are installed in the connecting frame 306; at the same time, a second motor 308 is fixedly arranged outside the upper frame body 302, and its output shaft is fixedly connected to the connecting frame 306. When the output shaft of the second motor 308 rotates, it can drive the connecting frame 306 and its second conveyor belt assembly 307 to adjust the inclination angle facing the first conveyor belt assembly 305, so that the second conveyor belt assembly 307 can control the size of the extrusion part when extruding the raw material;

[0266] It should be noted that in order to avoid the phenomenon of mechanism interference, when the second motor 308 is started, the first motor 303 must be stopped. That is, the first motor 303 also needs to be controlled by the PLC controller 5;

[0267] It is understandable that, although the second conveyor belt assembly 307 squeezes the raw materials, it still essentially performs the function of a conveyor belt, and thus will not interfere with the transportation of the raw materials on the first conveyor belt assembly 305 .

[0268] It should be pointed out that in order to make the whole machine output meat slices of target thickness, on the basis of the above functions, the electrical signal e(t) determined by the meat slicing intelligent production adjustment method described in the above embodiments one to three is handed over to the PLC controller 5, and the PLC controller 5 controls the second motor 308 to execute the corresponding angle output according to the principle of the PID controller algorithm, that is, controls the size of the extrusion portion to be formed; and then the slicing is performed by the flaking mechanism 4; therefore, in fact, the dictionary table in step S4 of the above embodiments one to three can be further preset with different extrusion angles EA(t) corresponding to the size of the raw material extrusion portion and its flaking size; because the position of the flaking mechanism 4 is relatively unchanged, when different extrusion angles EA(t) correspond to different sizes of the raw material extrusion portion, the flaking size can also be directly determined.

[0269] It can be understood that, based on the above mechanical principles, the first conveyor belt assembly 305 slowly feeds the material, the second conveyor belt assembly 307 extrude and continuously forms the extrusion portion of our target calculated by Examples 1 to 3, and the flaking mechanism 4 continuously flaks the extrusion portion until the raw material is flaked.

[0270] In this example, see Figures 2-3 The flaking mechanism 4 includes a cutter disc 401 rotatably matched with the frame 1, a third motor 402 and a second transmission belt assembly 403. The second transmission belt assembly 403 also includes a driving wheel, a driven wheel and a transmission belt. The driving wheel is driven by the output shaft of the third motor 402, and the driven wheel is fixedly connected to the cutter disc 401. When the third motor 402 rotates, the cutter disc 401 can be controlled to rotate and flak. The third motor 402 is controlled by a PLC controller 5.

[0271] In this example, see Figure 2 A CCD camera 2 is also mounted on the frame 1, and the lens of the CCD camera 2 is facing the flaking position of the raw material extrusion part and the flaking mechanism 4 ( Figure 3A area); when the flaking mechanism 4 cuts the extrusion part and forms the meat slices, especially when the meat slices have not yet separated from the extrusion part, the CCD camera 2 takes pictures and detects the thickness of the meat slices (the graphic file can be captured and displayed by the display 6), determines the actual thickness of the meat slices in this round of flaking process (i.e., the basic data of the difference fluctuation value s established in step S1), and determines the electrical signal e(t) based on the meat flaking intelligent production adjustment method described in the above embodiments 1 to 3; therefore, the above raw materials are squeezed by the second conveyor belt assembly 307 and formed into the "electrical signal e(t)" required by the second motor 308 in the extrusion part, which is actually calculated from the previous moment t-1. However, no matter at which moment, when the CCD camera 2 detects the thickness of the meat slices and executes the above-mentioned intelligent production adjustment method, the electrical signal e(t) is input into the PLC controller 5, and the PLC controller 5 controls the feeding mechanism 3 to adjust the extrusion degree of the extrusion part so that it can form meat slices of target thickness when it is rotated for flaking.

[0272] It can be understood that the CCD camera 2, based on its photoelectric element, realizes the thickness of the object or the micro-displacement change according to the displacement change of the image point imaged on the CCD and then judges the thickness of the meat slice, which determines the independent variable x in the meat slicing intelligent production adjustment method described in Examples 1 to 3; its specific thickness detection algorithm can be implemented using many existing image detection and processing technologies, such as the meat thickness detection algorithm based on (CCD) imaging graphics disclosed in CN201510024277.2, CN202321982182.8, CN202210376515.6, CN202111208086.3 or CN201811330725.1, or the meat thickness detection algorithm disclosed in the following documents:

[0273] Liu Jin, Yang Haima, Zhang Jing. Research on online thickness measurement method based on CCD[J]. Chinese Journal of Scientific Instrument, 2006(S2):1217-1218.DOI:10.19650 / j.cnki.cjsi.2006.s2.080;

[0274] This embodiment further introduces an intelligent production and regulation system for meat slicers: Figure 6 As shown, it includes a processor and a memory connected to the processor, wherein program instructions are stored in the memory;

[0275] At the same time, the CCD camera 2 also stores the historically detected meat slice thickness data in the memory, and the memory also stores the historical pressing angle EA data; at the same time, the memory also has program instructions for implementing the meat slicer intelligent production adjustment method described in the above-mentioned embodiments 1 to 3;

[0276] The user sets the target meat slice thickness T, and when the program instructions are executed by the processor, the system forms an electrical signal e(t) at the current moment t, and then the processor transmits it to the PLC controller 5, so that the PLC controller 5 controls the first motor 303, the second motor 308 and the third motor 402 to perform corresponding operations, that is, controls the intelligent production adjustment device of this embodiment.

[0277] Test example: This example is used to verify whether the flaking machine disclosed in Example 4 can make autonomous decisions and adjustments on the thickness processing of meat slices when combined with the intelligent adjustment methods provided in Examples 1 to 3.

[0278] 1. Overview:

[0279] In this case, the performance of the experimental group (using the intelligent adjustment method) and the control group (using the traditional fixed extrusion angle method) during the slicing process was compared to analyze the difference between the two groups in controlling the thickness of the meat slices.

[0280] 2. Materials and methods

[0281] 2.1 Experimental Materials

[0282] Experimental group and control group: two flaking devices as provided in Example 4;

[0283] From a beef cattle, take two tenderloins, each weighing 2.8 kg after removing the bones.

[0284] 2.2 Experimental methods

[0285] Experimental group: a flaking device using an intelligent adjustment method, performing steps S1 to S4 as disclosed in the first embodiment;

[0286] Control group: the same type of flaking device was used, and its extrusion angle EA was set to a fixed value;

[0287] Set the slicing thickness to 1.5 mm and slice the two beef tenderloins separately;

[0288] Both groups performed 35 rounds of flaking, and the thickness of the meat slices was recorded by CCD camera 2 in each round;

[0289] Compare the meat slice thickness data of the two groups and draw a line chart for comparison.

[0290] (III) Results and Discussion

[0291] 3.1 Experimental results:

[0292] like Figure 8As shown in the figure, the deviation of the thickness of the meat slices in the experimental group was controlled at about ±0.03mm, showing a high stability. In the control group, the thickness of the meat slices gradually exceeded 1.5mm after the 9th slice, approached 2mm at the 27th slice, and then dropped sharply to 1.2mm at the 28th to 29th slices, and then the thickness increased sharply again. The error was about ±0.48mm;

[0293] 3.2 Analysis:

[0294] The instability of the thickness of the meat slices in the control group may be related to the changes in the shape and texture of the meat. In particular, during the 28th to 29th slicing, the cutting site was located in the hip bone area of ​​the cow, which has a large concave feature, affecting the cutting depth and angle of the blade, resulting in uneven thickness of the meat slices. In contrast, the experimental group effectively overcame this problem by adjusting the extrusion angle in real time through an intelligent adjustment method, maintaining the stability of the thickness of the meat slices.

[0295] (IV) Conclusion:

[0296] This example experimentally verifies the effectiveness of the meat slicer intelligent production adjustment method in maintaining the thickness stability of meat slices. Compared with the traditional method of fixed extrusion angle, the intelligent adjustment method can significantly improve the quality and production efficiency of meat slices, providing a new technical solution for the meat processing industry.

[0297] For those skilled in the art, it can be further appreciated that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0298] At the same time, those skilled in the art can understand that all or part of the processes in all the above-mentioned embodiments can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided in this application and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0299] The above embodiments only express the implementation methods of the relevant practical applications of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. An intelligent production and adjustment method for meat slicers, characterized in that: The steps include: S1, data engineering: collect the thickness difference fluctuation value s of the previous multiple rounds of slicing, the target meat slice thickness T, and form the independent variable x; collect the extrusion angle EA of the previous multiple rounds of slicing, and form the dependent variable y; S2, execute the SVM regression model: input the independent variable x into the regression function f(x), perform mapping based on the RBF kernel function, and output the predicted dependent variable y'; S3, executing the prediction compensation mechanism: executing the prediction compensation mechanism on the prediction dependent variable y' to obtain the required extrusion angle EA(t) at the current time t; S4, executing the PID controller algorithm: bringing the extrusion angle EA(t) into the PID controller algorithm, outputting the electrical signal e(t), and further controlling the degree of extrusion of the raw material to accept the flaking operation.

2. The intelligent production adjustment method according to claim 1, characterized in that: In S1, the difference fluctuation value s is obtained based on the coefficient of variation CV: Where ω is the standard deviation of the thickness of the meat slices, and μ is the mean thickness of the meat slices.

3. The intelligent production adjustment method according to claim 1, characterized in that: In S2, the regression function f(x) and the mapping are respectively performed by the following two steps: S200: Where N is the number of support vectors; α i and α i * is the Lagrange multiplier; K(x,x i ) is the RBF kernel function, x i is a support vector, b is a bias term; S201: Among them, σ is the width parameter and Υ is the hyperparameter of the RBF kernel function.

4. The intelligent production adjustment method according to claim 3 is characterized in that: The SVM regression model should be trained before executing S1 to S2, the steps are as follows: The first step is to read the data set D: collect the independent variables x and dependent variables y in the historical data, form the data set D and divide it into the training set D train and the test set D test ; In the second step, we set the optimization problem to find an optimal hyperplane to divide the data: Where W is the normal vector of the hyperplane; C is the penalty coefficient; ξi i and ξi i * is a slack variable; x i The support vector x i Function mapped to high-dimensional space; y i is the support vector x i The corresponding true value; A set of the Lagrange multipliers α is found by a quadratic programming algorithm. i and α i *, and find the position of the hyperplane W and the bias term b; The third step is to use the data set D in combination with k-fold cross validation to train the SVM regression model to obtain the penalty coefficient C and the hyperparameter Υ.

5. The intelligent production adjustment method according to claim 4, characterized in that: In S2, the RBF kernel function K(x,x i ) When performing the mapping, the error range is constrained based on the following insensitive loss function: Where ε is the radius of the insensitive area.

6. The intelligent production adjustment method according to claim 3, characterized in that: In S3, the method for calculating the extrusion angle EA(t) is: Where: y' t is the extrusion angle directly predicted by the SVM regression model at time t; e t-i is the forecast error at time ti; n is the number of historical error periods considered; β i It is the error feedback coefficient used to control the degree of compensation of the prediction errors at each moment in the past for the current prediction; The error feedback coefficient β i The method to obtain is: Where M is the total number of observations and S is the residual sum of squares.

7. The intelligent production adjustment method according to any one of claims 1 to 6, characterized in that: In S4, the PID controller algorithm is: Where: K p is the proportional gain, K i is the integral gain, K d is the differential gain, error(τ) represents the deviation value of the variable τ at a certain time; error(t) is the deviation at time t: e(t)=r(t)-y(t); Wherein r(t) is the current extrusion angle EA, and y(t) is the extrusion angle EA(t) obtained by S3.

8. The intelligent production adjustment method according to claim 7, characterized in that: The S4 follows the following steps: S400, at each time t, calculating the deviation error(t) between the set value and the actual value; S401, obtaining the electrical signal e(t); S402, read the dictionary table: find the preset electric signal value corresponding to the electric signal e(t), and the size parameters of the corresponding raw material extrusion part, and the size parameters to be shaved.

9. Meat slicing intelligent production and adjustment device, characterized by: It comprises a CCD camera (2), a feeding mechanism (3) and a shaving mechanism (4); The feeding mechanism (3) conveys the raw material and forms an extrusion portion facing the flaking mechanism (4). When the extrusion portion is cut by the flaking mechanism (4) according to its size, meat slices of a thickness corresponding to the size of the extrusion portion are formed; The CCD camera (2) detects the thickness of the meat slices; After executing the intelligent production adjustment method as described in any one of claims 1 to 8, the electrical signal e(t) is input into the PLC controller (5), and the PLC controller (5) controls the feeding mechanism (3) to adjust the extrusion degree of the extrusion part so that meat slices of target thickness can be formed when the meat is shaving by the rotation.

10. Meat shavings intelligent production and regulation system, characterized by: It includes a processor and a memory connected to the processor, wherein program instructions are stored in the memory, and when the program instructions are executed by the processor, the processor controls the intelligent production regulation device as described in claim 9 to execute the intelligent production regulation method as described in any one of claims 1-8.

Citation Information

Cited By

  • Intelligent control method and system for toilet paper roll core forming based on data driving

    CN119937434A

  • Data-driven intelligent control method and system for the forming of toilet paper cores

    CN119937434B