Intelligent granular mixture pressing process parameter decision-making method based on digital twinborn model

Abstract

The invention discloses an intelligent granular mixture pressing process parameter decision-making method based on a digital twinborn model, and relates to the field of granular mixture pressing process parameter decision-making. According to the method, a granular mixture pressing digital twinning model is constructed based on historical process parameter data and process parameter data acquired in real time, and a digital twinning visual decision platform is constructed. Through a case-based reasoning algorithm, case samples are retrieved by fully utilizing previous historical experience; and if the case-based reasoning algorithm does not generate an effective result, performing process parameter data prediction by using a digital twinning algorithm model based on a BCWOA-BP neural network, sending the output process parameter data to a physical entity for suppression, storing the process parameter data meeting the quality requirement in a process database for updating the digital twinning model, and performing data processing on the digital twinning model. The performance and accuracy of the granular mixture pressed digital twinning model are enhanced, virtual-real interaction is achieved, good process parameter data can be obtained visually and rapidly, and the problems of low efficiency and high cost are solved.

Description

Technical Field

[0001] The present invention relates to the field of bulk body pressing process parameter decision-making, and in particular to a bulk body pressing process parameter intelligent decision-making method based on a digital twin model. Background Art

[0002] The bulk body pressing process is currently the most widely used material forming process. The bulk body pressing process is to compress the bulk body in the mold into a compact with a certain size and shape by applying a pressing force. In the bulk body pressing process, the pressing process parameters such as initial bulk density, pressing rate, pressing force and holding time will directly affect the forming quality. Excessive pressing force will cause a large rebound of the formed grain, which is prone to defects such as falling blocks or cracks; the longer the holding time is conducive to improving the relative density of the forming, but there is a corresponding problem of low efficiency; the greater the initial bulk density, the greater the forming density during pressing. The traditional bulk body pressing process requires a lot of experimental experience and trial and error methods to determine the pressing process parameters and obtain high-quality grains, which is easy to cause a lot of material waste, high cost and low efficiency. Many scholars have proposed improvement methods and measures to address the above problems.

[0003] Chinese invention CN202210960287.7 provides a method for optimizing process parameters of powder pressing process based on digital twins, constructs a digital twin model of powder pressing process and a twin data generation module and generates twin data, and expands the twin data through the twin data expansion network; and based on the expanded data, trains the quality parameter and energy consumption parameter prediction model to construct a process parameter optimization module, and finally optimizes the process parameters through the particle swarm optimization algorithm; this method solves the problem of lack of sample data caused by experimental difficulties in the powder pressing process, and reduces production energy consumption. However, this method does not use historical process data, and the digital twin model may not be accurate enough in long-term prediction and optimization.

[0004] Chinese invention CN202210561679.6 provides a method for quantitatively predicting the pressing process and density of press-packed mixed explosives based on the AdaBoost algorithm. This method uses machine learning methods to establish the relationship between the process parameters of press-packed mixed explosives and the charge density with high precision and automation, which has guiding significance for the selection of process parameters. However, this invention does not combine digital twin technology, and the accuracy of process parameters still needs to be verified through traditional experiments.

[0005] Chinese invention CN202310654954.3 provides a method and device for optimizing the drug column compression molding process, which collects the production factor information of each device in the drug column compression molding process, inputs it into a pre-built digital twin model, outputs the optimization parameters in the drug column compression molding process, and sends it to the drug column compression molding equipment to achieve simulation optimization of the drug column compression molding process. This method is conducive to improving the quality and efficiency of drug column production, improving the degree of intelligent control of the drug column compression molding process, and improving the safety of the production process, but the invention does not propose a specific process parameter optimization method. Summary of the invention

[0006] In view of the shortcomings of the prior art, the present invention provides an intelligent decision-making method for granular pressing process parameters based on a digital twin model to solve the problem that when pressing a fixed material product, if the process information is changed, in order to obtain a high-quality drug column, it is usually necessary to re-perform a large number of granular pressing experiments to obtain a set of excellent process parameter data.

[0007] The technical solution of the present invention is:

[0008] The intelligent decision-making method for bulk compaction process parameters based on the digital twin model includes the following steps:

[0009] Step 1: Under the set process information, the process parameter data is collected in real time through the sensors arranged on the granular pressing equipment. After the collected process parameter data is cleaned, it is transmitted to the preset process database as a process case sample together with the corresponding process information;

[0010] Furthermore, the process information includes mold size, molding relative density, temperature and bulk density; the process parameter data includes the pressing force, holding time and pressing rate generated during the pressing process;

[0011] Step 2: Obtain the granular body pressing historical data and pressing equipment data;

[0012] The granular body pressing history data is the historical process parameter data stored in the process database; the pressing equipment data includes the geometric parameters, position parameters, die size and material property parameters of the granular body pressing equipment; the geometric parameters of the granular body pressing equipment are the external dimensions of the granular body pressing equipment; the position parameters are the relative position between the die and the upper die punch of the granular body pressing equipment, the die position and the sensor position;

[0013] Step 3: construct a digital twin model of granular body pressing according to the historical data of granular body pressing, the process parameter data collected in real time and the pressing equipment data; the digital twin model of granular body pressing includes a digital twin geometric model of granular body pressing, a digital twin physical model of granular body pressing, a behavior model and a digital twin algorithm model based on BCCOA-BP neural network;

[0014] Step 3.1: Construct a digital twin geometric model of granular pressing based on pressing equipment data;

[0015] Specifically, according to the geometric parameters, mold size, and position parameters of the granular pressing equipment, a three-dimensional model including the granular pressing equipment, the mold, and the sensor is drawn using a three-dimensional modeling software to obtain a digital twin geometric model of the granular pressing;

[0016] Step 3.2: Add the material property parameters, boundary conditions and initial conditions of the granular body in the finite element simulation software, set the contact properties and contact relationships of the granular body pressing equipment, mold and sensor in the constructed granular body pressing geometric model, and thus construct a digital twin physical model of granular body pressing;

[0017] Step 3.3: Combined with the granular body pressing process, the granular body pressing digital twin physical model is used in the finite element simulation software to simulate the behavior states of granular body pressing at different stages, and the behavior models of the preparation stage, the pressurization stage, the pressure holding stage, and the pressure relief and demolding stage are established respectively. The behavior models of the preparation stage, the pressurization stage, the pressure holding stage, and the pressure relief and demolding stage are calibrated through the process parameter data collected in real time by the sensor;

[0018] Step 3.4: Construct a BP neural network, and use the whale optimization algorithm based on quasi-reverse learning and improved crossover strategy to optimize the weights and thresholds of the BP neural network to obtain a digital twin algorithm model based on the BCCOA-BP neural network;

[0019] The input data of the BP neural network is process information, and the output data is process parameter data;

[0020] Step 3.4.1: According to the input data and output data, determine the number of input layers, output layers and hidden layer nodes, initialize the weights and thresholds of the BP neural network, and construct the BP neural network;

[0021] Step 3.4.2: Set the maximum number of iterations T max , the current number of iterations is 0;

[0022] Step 3.4.3: Set the population size N and the search space dimension d, and randomly generate an initial humpback whale population, wherein the initial humpback whale population includes a number of humpback whale individuals, each of which is a d-dimensional vector representing a set of weights and thresholds of the BP neural network;

[0023] Step 3.4.4: Calculate the fitness value of the humpback whale individuals in the current population, determine and record the current best humpback whale individual according to the fitness value; the fitness value is the root mean square error of the BP neural network; the best humpback whale individual is the humpback whale individual with the smallest fitness value;

[0024] Step 3.4.5: Determine whether the current number of iterations has reached the maximum number of iterations. If so, output the current best humpback whale individual and execute step 3.4.10. Otherwise, execute step 3.4.6.

[0025] Step 3.4.6: Use the quasi-reverse learning strategy to generate a quasi-reverse solution for each humpback whale individual, and update the position of each humpback whale individual according to the greedy strategy;

[0026] Specifically: The position of the i-th humpback whale individual is i is the number of the humpback whale individual, t is the current iteration number, j is the number of the dimension, is the value of the jth dimension in the position of the i-th humpback whale individual;

[0027] First generate the inverse solution for each individual humpback whale:

[0028]

[0029] In the formula, is the value of the jth dimension in the reverse solution, l b and u b are the upper and lower bounds of the search space respectively;

[0030] Then, according to the reverse solution of each humpback whale individual, the quasi-reverse solution of each humpback whale individual is obtained:

[0031]

[0032] In the formula, is the value of the jth dimension in the quasi-inverse solution, rand is a random number between [0,1];

[0033] The greedy strategy is incorporated as the evaluation rule, and the position of the humpback whale individual is updated using formula (4):

[0034]

[0035] In the formula, is the updated position of the i-th humpback whale individual obtained by the greedy strategy, is the quasi-reverse solution, f(·) is the fitness function;

[0036] Step 3.4.7: Calculate iteration variables And according to the coefficient vector of humpback whale individuals and The update formula updates the coefficient vector of humpback whale individuals and Generate a random number p between [0,1];

[0037] Step 3.4.8: When p < 0.5, if Then the humpback whale individual updates its position according to the way of surrounding the prey, otherwise it adopts the random search strategy to update its position. When p≥0.5, the dynamic threshold is calculated and a random number is generated. If the random number is less than the dynamic threshold, the spiral update position strategy is adopted to update the position, otherwise the horizontal cross strategy is adopted to update the position.

[0038] The dynamic threshold is:

[0039] T v =0.2×rand(N,1) (8)

[0040] Where: T v is a dynamic threshold, rand(N,1) represents a random number;

[0041] The horizontal crossover strategy updates the position of individual humpback whales based on the parity of the dimension:

[0042]

[0043] In the formula, is the value of the jth dimension in the updated position of the i-th humpback whale individual, α 1 and α 2 is a random number in (0,1), e 1 and e 2 is a random operator in (-1,1) used to enhance the search phase; is the value of the j-th dimension in the position of the i-1th humpback whale individual, The position of the i+1th humpback whale individual in the jth dimension, The location of the best humpback whale individuals at present;

[0044] The greedy strategy shown in formula (10) is used as the evaluation criterion to decide whether to update the position of the humpback whale individual:

[0045]

[0046] Step 3.4.9: Increase the current iteration number by 1 and return to step 3.4.4;

[0047] Step 3.4.10: Input the weights and thresholds corresponding to the current best humpback whale individual into the BP neural network, use the process case samples in the process database to train the BP neural network using the error back propagation mechanism, update the weights and thresholds, until the output error of the BP neural network is less than the preset error or the training reaches the maximum number of iterations, the training ends, and the digital twin algorithm model based on the BCCOA-BP neural network is obtained;

[0048] Step 4: Make decisions and solve process parameter data based on case-based reasoning algorithm and granular compaction digital twin model;

[0049] Step 4.1: Use the case-based reasoning algorithm to retrieve case samples based on the process information input by the user to obtain output case samples;

[0050] Step 4.1.1: The user inputs the set process information;

[0051] Step 4.1.2: Use the expert evaluation method to set the weight of each process information according to the degree of influence of each process information on the pressing quality during the granular pressing process. Based on the weight of each process information, calculate the overall similarity between the process information input by the user and the process information of each case sample in the process database;

[0052] First, the Hamming distance formula is used to calculate the similarity between each process information input by the user and the corresponding process information in the case sample, that is, the local similarity of the case sample, as shown in formula (11):

[0053]

[0054] In the formula, is the similarity between a process information input by the user and the corresponding process information in the case sample, and A process information entered by the user. is the corresponding process information in the case sample, r is the number of the process information, and The upper and lower limits of the process information;

[0055] According to the similarity between each process information input by the user and the corresponding process information in the case sample, the Kolodner formula is used to calculate the overall similarity between the process information input by the user and the process information in the case sample, as shown in formula (12):

[0056]

[0057] In the formula, sim(H,G) is the overall similarity between the process information input by the user and the process information in the case sample, q is the amount of process information, and w r is the weight of the rth process information and satisfies

[0058] Step 4.1.3: Set a similarity threshold, and find case samples whose overall similarity with the process information input by the user is greater than the similarity threshold as output case samples;

[0059] Step 4.2: If the number of output case samples is greater than 0, rank the output case samples according to the overall similarity and select the case sample with the highest overall similarity as the optimal solution, and then obtain the process parameter data in the optimal solution and execute step 4.4, otherwise execute step 4.3;

[0060] Step 4.3: Input the process information input by the user into the digital twin algorithm model based on the BCCOA-BP neural network to predict the process parameter data;

[0061] Step 4.4: Based on the obtained process parameter data, the behavior model is used to simulate the granular pressing and a granular pressing test is performed to verify and analyze the process parameter data to determine whether the process parameter data meets the pressing quality requirements. If so, the set of process parameter data and the corresponding process information are stored as a case sample in the process database, and the granular pressing digital twin model is updated. If not, return to step 4.1.

[0062] Compared with the prior art, the beneficial effects of the present invention are:

[0063] The present invention constructs a digital twin model of granular compaction based on historical process parameter data and real-time collected process parameter data, and builds a digital twin visualization decision-making platform. Digital twin technology can realize the virtual-real combination of granular compaction, and combined with the visualization decision-making platform, it is more conducive to intuitive monitoring and operation by operators.

[0064] Through the case-based reasoning algorithm (CBR algorithm), under the condition of certain process case samples as support, make full use of past historical experience to retrieve case samples; if the case-based reasoning algorithm does not produce effective results, the digital twin algorithm model based on the BCCOA-BP neural network is used to predict the process parameter data. The output process parameter data is sent to the physical entity for pressing, and the process parameter data that meets the quality requirements is stored in the process database for updating the digital twin model. The digital twin model of granular pressing is continuously updated, which enhances the performance and accuracy of the digital twin model of granular pressing, realizes virtual-real interaction, and is more conducive to operators to intuitively and quickly obtain excellent process parameter data, solving the problems of low efficiency and high cost.

[0065] The present invention integrates case reasoning and neural network algorithms, making full use of past historical experience, while also overcoming limitations when facing new processing conditions, and comprehensively considering the coordination of multiple goals in the processing process, thereby improving the efficiency, rationality and accuracy of process parameter decision-making. The proposed whale optimization algorithm (BCWOA) based on quasi-reverse learning and improved crossover strategy has faster convergence speed, better optimization accuracy and local search capability. The BCCOA algorithm is used to optimize the weights and thresholds of the BP neural network, effectively improving the prediction accuracy of the traditional BP neural network algorithm. Compared with the traditional method of obtaining excellent process parameters through compression experiments and trial and error and the unimproved BP neural network algorithm, the present invention can more accurately predict process parameter data. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 It is a flowchart of an intelligent decision-making method for granular pressing process parameters based on a digital twin model in an embodiment of the present invention;

[0067] Figure 2 This is a flow chart of the BCCOA algorithm optimizing the BP neural network in an embodiment of the present invention;

[0068] Figure 3 This is a comparison diagram of the convergence curves of the BCCOA algorithm for four standard test functions in an embodiment of the present invention;

[0069] Among them, (a) is the convergence curve of the BCCOA algorithm for the first test function selected in the CEC2005 standard test function; (b) is the convergence curve of the BCCOA algorithm for the second test function selected in the CEC2005 standard test function; (c) is the convergence curve of the BCCOA algorithm for the third test function selected in the CEC2005 standard test function; (d) is the convergence curve of the BCCOA algorithm for the fourth test function selected in the CEC2005 standard test function;

[0070] Figure 4 Flow chart of case-based reasoning algorithm in an embodiment of the present invention;

[0071] Figure 5 This is a diagram of a data management interface in an embodiment of the present invention;

[0072] Figure 6 This is a diagram of the process parameter data prediction interface in an embodiment of the present invention. DETAILED DESCRIPTION

[0073] In order to make those skilled in the art more clearly understand the objectives, technical solutions and advantages of the present invention, the present invention is further described in detail with reference to the accompanying drawings. It should be understood that the examples described herein are only used to illustrate and explain the present invention, and are not intended to limit the present invention.

[0074] For ease of understanding, the granular material used in the embodiment of the present invention is composed of 68% polyetheretherketone, 30% aluminum powder and 2% fluororubber, the fluororubber is a binder, and its specification is BDF25P. The hydraulic equipment model used is Instron8801, and the pressure range that can be applied is ±100kN.

[0075] like Figure 1 As shown in the figure, the intelligent decision-making method for bulk compacting process parameters based on the digital twin model includes the following steps:

[0076] Step 1: Under the set process information, the process parameter data is collected in real time through the sensors arranged on the granular pressing equipment. After the collected process parameter data is cleaned, it is transmitted to the preset process database as a process case sample together with the corresponding process information;

[0077] The process information includes mold size, molding relative density, temperature and bulk density; the sensor includes a pressure sensor and a displacement sensor; the process parameter data includes the pressing force, holding time and pressing rate generated during the pressing process; the pre-designed process database is established using SQL Sever, and the process parameter data and process information after data cleaning are sent to the process database through the serial port;

[0078] Step 2: Obtain the granular body pressing historical data and pressing equipment data;

[0079] The granular body pressing history data is the historical process parameter data stored in the process database; the pressing equipment data includes the geometric parameters, position parameters, die size and material property parameters of the granular body pressing equipment (hydraulic equipment); the geometric parameters of the granular body pressing equipment are the external dimensions of the granular body pressing equipment; the position parameters are the relative position between the die and the upper die punch of the granular body pressing equipment, the die position and the sensor position;

[0080] Step 3: construct a digital twin model of granular body pressing according to the historical data of granular body pressing, the process parameter data collected in real time and the pressing equipment data; the digital twin model of granular body pressing includes a digital twin geometric model of granular body pressing, a digital twin physical model of granular body pressing, a behavior model and a digital twin algorithm model based on BCCOA-BP neural network;

[0081] Step 3.1: Construct a digital twin geometric model of granular pressing based on pressing equipment data;

[0082] Specifically, the geometric model is an intuitive mapping of the physical entity. According to the geometric parameters, mold size, and position parameters of the granular pressing equipment, the Catia 3D modeling software is used to draw a 3D model including the granular pressing equipment, mold, and sensor. The established 3D model is imported into the Unity3D software to obtain the digital twin geometric model of granular pressing.

[0083] Step 3.2: The physical model is mainly used to express the basic characteristics and inherent properties of the physical entity. The material property parameters, boundary conditions and initial conditions of the granular body are added in the ANSYS finite element simulation software, and the contact properties and contact relationships of the granular body pressing equipment, mold and sensor in the constructed granular body pressing geometric model are set to construct a digital twin physical model of granular body pressing.

[0084] Step 3.3: Combined with the granular body pressing process, the granular body pressing digital twin physical model is used in the ANSYS finite element simulation software to simulate the behavior states of different stages of granular body pressing, and the behavior models of the preparation stage, the pressurization stage, the pressure holding stage, and the pressure relief and demolding stage are established respectively. The behavior models of the preparation stage, the pressurization stage, the pressure holding stage, and the pressure relief and demolding stage are calibrated through the process parameter data collected in real time by the sensor to ensure that the behavior of the behavior model is consistent with the actual granular body pressing behavior;

[0085] Step 3.4: Construct BP neural network and use Figure 2 The whale optimization algorithm based on quasi-reverse learning and improved crossover strategy shown in the figure optimizes the weights and thresholds of the BP neural network to obtain a digital twin algorithm model based on the BCCOA-BP neural network;

[0086] The input data of the BP neural network is process information (mold size, molding relative density, temperature and bulk density), and the output data is process parameter data (pressing force, holding time and pressing rate);

[0087] Step 3.4.1: In order to avoid the problem of model non-convergence and error fluctuation caused by the difference in order of magnitude, the input data and output data of the BP neural network are normalized, and the number of input layers, output layers and hidden layer nodes are determined according to the normalized input data and output data, the weights and thresholds of the BP neural network are initialized, and the BP neural network is constructed;

[0088] Set the input data of BP neural network to x 1 ,x 2 ,…,x n , the output data is y 1 ,y 2 ,…,ym Formula (1) is used to determine the number of hidden layer nodes:

[0089]

[0090] Where L is the number of hidden layer nodes, n is the number of input layers, m is the number of output layers, and g is an integer between 1 and 10;

[0091] The BP neural network has four input layers, namely, mold size, molding relative density, temperature and bulk density; and three output layers, namely, pressing force, holding time and pressing rate.

[0092] Step 3.4.2: Set the maximum number of iterations T max , the current number of iterations is 0;

[0093] Step 3.4.3: Set the population size N and the search space dimension d, and randomly generate an initial humpback whale population, wherein the initial humpback whale population includes a number of humpback whale individuals, each of which is a d-dimensional vector representing a set of weights and thresholds of the BP neural network;

[0094] In this implementation, at the beginning of the Whale Algorithm (WOA), a random initialization method is used to generate an initial humpback whale population, which has high randomness and cannot guarantee the diversity of the initial humpback whale population, which may cause the algorithm to converge prematurely and fall into a local optimal solution.

[0095] Step 3.4.4: Calculate the fitness value of the humpback whale individuals in the current population, determine and record the current best humpback whale individual according to the fitness value; the fitness value is the root mean square error of the BP neural network; the best humpback whale individual is the humpback whale individual with the smallest fitness value;

[0096] Step 3.4.5: Determine whether the current number of iterations has reached the maximum number of iterations. If so, output the current best humpback whale individual and execute step 3.4.10. Otherwise, execute step 3.4.6.

[0097] Step 3.4.6: Use the quasi-reverse learning strategy to generate a quasi-reverse solution for each humpback whale individual, and update the position of each humpback whale individual according to the greedy strategy;

[0098] Specifically: The position of the i-th humpback whale individual is i is the number of the humpback whale individual, t is the current iteration number, j is the number of the dimension, is the value of the jth dimension in the position of the i-th humpback whale individual;

[0099] First generate the inverse solution for each individual humpback whale:

[0100]

[0101] In the formula, is the value of the jth dimension in the reverse solution, l b and u b are the upper and lower bounds of the search space respectively;

[0102] Then, according to the reverse solution of each humpback whale individual, the quasi-reverse solution of each humpback whale individual is obtained:

[0103]

[0104] In the formula, is the value of the jth dimension in the quasi-inverse solution, rand is a random number between [0,1];

[0105] To ensure that each position update is meaningful, a greedy strategy is incorporated as the evaluation rule, and the position of the humpback whale individual is updated using formula (4):

[0106]

[0107] In the formula, is the updated position of the i-th humpback whale individual obtained by the greedy strategy, is the quasi-reverse solution, f(·) is the fitness function;

[0108] Step 3.4.7: Calculate iteration variables And according to the coefficient vector of humpback whale individuals and The update formula updates the coefficient vector of humpback whale individuals and Generate a random number p between [0,1];

[0109]

[0110]

[0111]

[0112] In the formula, is the iterative variable, which decreases linearly with the iteration process, gradually decreasing from the initial value 2 to 0. and is a random vector between [0,1];

[0113] Step 3.4.8: When p < 0.5, if Then the humpback whale individual updates its position according to the way of surrounding the prey, otherwise it adopts the random search strategy to update its position. When p≥0.5, the dynamic threshold is calculated and a random number is generated. If the random number is less than the dynamic threshold, the spiral update position strategy is adopted to update the position, otherwise the horizontal cross strategy is adopted to update the position.

[0114] In order to avoid the problem of the traditional WOA algorithm falling into the local optimum during the solution, which leads to the premature algorithm and low convergence accuracy, a dynamic threshold strategy is proposed for the spiral update position strategy and the horizontal cross strategy, so that the local optimization ability of the BCCOA algorithm is improved.

[0115] The dynamic threshold is:

[0116] T v =0.2×rand(N,1) (8)

[0117] Where: T v is a dynamic threshold, rand(N,1) represents a random number;

[0118] In order to solve the problem that the WOA algorithm is prone to fall into local optimality, a horizontal crossover strategy is used to update the position of humpback whale individuals according to the parity of the dimension:

[0119]

[0120] In the formula, is the value of the jth dimension in the updated position of the i-th humpback whale individual, α 1 and α 2 is a random number in (0,1), e 1 and e 2 is a random operator in (-1,1) used to enhance the search phase; is the value of the j-th dimension in the position of the i-1th humpback whale individual, The position of the i+1th humpback whale individual in the jth dimension, The location of the best humpback whale individuals at present;

[0121] Use the greedy strategy of formula (10) as the evaluation criterion to decide whether to update the position of the whale:

[0122]

[0123] In order to verify the effectiveness and advantages of the BCCOA algorithm, four test functions were selected from the CEC2005 standard test function to test the algorithm. The test results are shown in the figure. Figure 3 As shown;

[0124] Step 3.4.9: Increase the current iteration number by 1 and return to step 3.4.4;

[0125] Step 3.4.10: Input the weights and thresholds corresponding to the current best humpback whale individual into the BP neural network, use the process case samples in the process database to train the BP neural network using the error back propagation mechanism, update the weights and thresholds, and further optimize the performance of the BP neural network until the output error of the BP neural network is less than the preset error or the training reaches the maximum number of iterations. The training ends and a digital twin algorithm model based on the BCCOA-BP neural network is obtained;

[0126] Step 4: Make decisions and solve process parameter data based on case-based reasoning algorithm and granular compaction digital twin model;

[0127] Step 4.1: Use Figure 4 The case-based reasoning algorithm (CBR) shown performs case sample retrieval based on the process information input by the user;

[0128] Step 4.1.1: The user inputs the set process information;

[0129] Step 4.1.2: Use the expert evaluation method to set the weight of each process information according to the degree of influence of each process information on the pressing quality during the granular pressing process. Based on the weight of each process information, calculate the overall similarity between the process information input by the user and the process information of each case sample in the process database;

[0130] The Hamming distance formula is used to calculate the similarity between each process information input by the user and the corresponding process information in the case sample, that is, the local similarity of the case sample, as shown in formula (11):

[0131]

[0132] In the formula, is the similarity between a process information input by the user and the corresponding process information in the case sample, and A process information entered by the user. is the corresponding process information in the case sample, r is the number of the process information, and The upper and lower limits of the process information;

[0133] According to the similarity between each process information input by the user and the corresponding process information in the case sample (local similarity of the case sample), the Kolodner formula is used to calculate the overall similarity between the process information input by the user and the process information in the case sample (overall similarity of the case sample), as shown in formula (12):

[0134]

[0135] In the formula, sim(H,G) is the overall similarity between the process information input by the user and the process information in the case sample, q is the amount of process information, and w r is the weight of the rth process information and satisfies

[0136] Step 4.1.3: Set a similarity threshold based on experience and requirements, and find case samples whose overall similarity with the process information input by the user is greater than the similarity threshold as output case samples;

[0137] Step 4.2: If the number of output case samples is greater than 0, rank the output case samples according to the overall similarity and select the case sample with the highest overall similarity as the optimal solution, and then obtain the process parameter data in the optimal solution and execute step 4.4, otherwise execute step 4.3;

[0138] In this implementation, the overall similarity threshold is set according to experience and requirements, and similarity matching is performed. The overall similarity threshold is set to 90%, that is, case samples with a similarity greater than or equal to 90% are output;

[0139] Step 4.3: Input the process information input by the user into the digital twin algorithm model based on the BCCOA-BP neural network to predict the process parameter data;

[0140] Step 4.4: Based on the obtained process parameter data, the behavior model is used to simulate the granular pressing and the granular pressing test is performed to verify and analyze the process parameter data to determine whether the process parameter data meets the pressing quality requirements. If so, the set of process parameter data and the corresponding process information are stored as a case sample in the process database, and the granular pressing digital twin model is updated. If not, return to step 4.1;

[0141] This embodiment provides a digital twin visualization decision-making platform based on a process database and a granular pressing digital twin model.

[0142] The digital twin visualization decision-making platform is built using Visual Studio software and integrated with the granular compaction digital twin model and process database. The platform includes a data management interface (such as Figure 5 ), case-based reasoning interface, process parameter prediction interface (such as Figure 6). The platform can be actually operated and controlled by the user, and can realize the functions of database storage, data transmission, and addition, deletion, modification and query. The real-time decision of the process parameter data of the granular pressing process is realized through the granular pressing digital twin model. According to the process parameter data after the decision, the relevant pressing information is fed back to the physical entity, thereby realizing data interaction.

[0143] The data management interface is used to add, search, modify, delete and store case samples in the process database;

[0144] The case-based reasoning interface is used to retrieve case samples in the process database according to the process information input by the user to obtain process parameter data;

[0145] The process parameter prediction interface is used to obtain predicted process parameter data based on the process information input by the user using a digital twin algorithm model based on the BCCOA-BP neural network.

Claims

1. An intelligent decision-making method for bulk compaction process parameters based on a digital twin model, characterized in that: The following steps are involved: Step 1: Under the set process information, the process parameter data is collected in real time through the sensors arranged on the granular pressing equipment. After the collected process parameter data is cleaned, it is transmitted to the preset process database as a process case sample together with the corresponding process information; Step 2: Obtain the granular body pressing historical data and pressing equipment data; Step 3: construct a digital twin model of granular body pressing according to the historical data of granular body pressing, the process parameter data collected in real time and the pressing equipment data; the digital twin model of granular body pressing includes a digital twin geometric model of granular body pressing, a digital twin physical model of granular body pressing, a behavior model and a digital twin algorithm model based on BCCOA-BP neural network; Step 4: Make decisions and solve the process parameter data based on the case-based reasoning algorithm and the granular compaction digital twin model.

2. According to claim 1, the method for intelligent decision-making of bulk compacting process parameters based on digital twin model is characterized in that: The process information in step 1 includes mold size, molding relative density, temperature and bulk density; the process parameter data includes the pressing force, holding time and pressing rate generated during the pressing process.

3. The method for intelligent decision-making of bulk compacting process parameters based on digital twin model according to claim 2 is characterized in that: The granular body pressing history data in step 2 is the historical process parameter data stored in the process database; the pressing equipment data includes the geometric parameters, position parameters, mold dimensions and material property parameters of the granular body pressing equipment; the geometric parameters of the granular body pressing equipment are the external dimensions of the granular body pressing equipment; the position parameters are the relative position between the mold and the upper punch of the granular body pressing equipment, the mold position and the sensor position.

4. The method for intelligent decision-making of bulk compacting process parameters based on digital twin model according to claim 3 is characterized in that: Step 3 specifically includes: Step 3.1: Construct a digital twin geometric model of granular pressing based on pressing equipment data; Specifically, according to the geometric parameters, mold size, and position parameters of the granular pressing equipment, a three-dimensional model including the granular pressing equipment, the mold, and the sensor is drawn using a three-dimensional modeling software to obtain a digital twin geometric model of the granular pressing; Step 3.2: Add the material property parameters, boundary conditions and initial conditions of the granular body in the finite element simulation software, set the contact properties and contact relationships of the granular body pressing equipment, mold and sensor in the constructed granular body pressing geometric model, and thus construct a digital twin physical model of granular body pressing; Step 3.3: Combined with the granular body pressing process, the granular body pressing digital twin physical model is used in the finite element simulation software to simulate the behavior states of granular body pressing at different stages, and the behavior models of the preparation stage, the pressurization stage, the pressure holding stage, and the pressure relief and demolding stage are established respectively. The behavior models of the preparation stage, the pressurization stage, the pressure holding stage, and the pressure relief and demolding stage are calibrated through the process parameter data collected in real time by the sensor; Step 3.4: Construct a BP neural network, and use the whale optimization algorithm based on quasi-reverse learning and improved crossover strategy to optimize the weights and thresholds of the BP neural network to obtain a digital twin algorithm model based on the BCCOA-BP neural network; The input data of the BP neural network is process information, and the output data is process parameter data.

5. The method for intelligent decision-making of bulk compacting process parameters based on digital twin model according to claim 4 is characterized in that: Step 3.4 specifically includes: Step 3.4.1: According to the input data and output data, determine the number of input layers, output layers and hidden layer nodes, initialize the weights and thresholds of the BP neural network, and construct the BP neural network; Step 3.4.2: Set the maximum number of iterations T max , the current number of iterations is 0; Step 3.4.3: Set the population size N and the search space dimension d, and randomly generate an initial humpback whale population, wherein the initial humpback whale population includes a number of humpback whale individuals, each of which is a d-dimensional vector representing a set of weights and thresholds of the BP neural network; Step 3.4.4: Calculate the fitness value of the humpback whale individuals in the current population, determine and record the current best humpback whale individual according to the fitness value; the fitness value is the root mean square error of the BP neural network; the best humpback whale individual is the humpback whale individual with the smallest fitness value; Step 3.4.5: Determine whether the current number of iterations has reached the maximum number of iterations. If so, output the current best humpback whale individual and execute step 3.4.

10. Otherwise, execute step 3.4.

6. Step 3.4.6: Use the quasi-reverse learning strategy to generate a quasi-reverse solution for each humpback whale individual, and update the position of each humpback whale individual according to the greedy strategy; Step 3.4.7: Calculate iteration variables And according to the coefficient vector of humpback whale individuals and The update formula updates the coefficient vector of humpback whale individuals and Generate a random number p between [0,1]; Step 3.4.8: When p < 0.5, if Then the humpback whale individual updates its position according to the way of surrounding the prey, otherwise it adopts the random search strategy to update its position. When p≥0.5, the dynamic threshold is calculated and a random number is generated. If the random number is less than the dynamic threshold, the spiral update position strategy is adopted to update the position, otherwise the horizontal cross strategy is adopted to update the position. The dynamic threshold is: T v =0.2×rand(N,1) (8) Where: T v is a dynamic threshold, rand(N,1) represents a random number; Step 3.4.9: Increase the current iteration number by 1 and return to step 3.4.4; Step 3.4.10: Input the weights and thresholds corresponding to the current best humpback whale individual into the BP neural network, use the process case samples in the process database to train the BP neural network using the error back propagation mechanism, update the weights and thresholds, until the output error of the BP neural network is less than the preset error or the training reaches the maximum number of iterations, the training ends, and the digital twin algorithm model based on the BCCOA-BP neural network is obtained.

6. The method for intelligent decision-making of bulk compacting process parameters based on digital twin model according to claim 5 is characterized in that: Step 3.4.6 is as follows: The position of the i-th humpback whale individual is i is the number of the humpback whale individual, t is the current iteration number, j is the number of the dimension, is the value of the jth dimension in the position of the i-th humpback whale individual; First generate the inverse solution for each individual humpback whale: In the formula, is the value of the jth dimension in the reverse solution, l b and u b are the upper and lower bounds of the search space respectively; Then, according to the reverse solution of each humpback whale individual, the quasi-reverse solution of each humpback whale individual is obtained: In the formula, is the value of the jth dimension in the quasi-inverse solution, rand is a random number between [0,1]; The greedy strategy is incorporated as the evaluation rule, and the position of the humpback whale individual is updated using formula (4): In the formula, is the updated position of the i-th humpback whale individual obtained by the greedy strategy, is the quasi-inverse solution, and f(·) is the fitness function.

7. The method for intelligent decision-making of bulk compacting process parameters based on digital twin model according to claim 6 is characterized in that: The horizontal crossover strategy described in step 3.4.8 updates the position of individual humpback whales based on the parity of the dimension: In the formula, is the value of the jth dimension in the updated position of the i-th humpback whale individual, α1 and α2 are random numbers in (0,1), and e1 and e2 are random operators in (-1,1) used to enhance the search phase; is the value of the j-th dimension in the position of the i-1th humpback whale individual, The position of the i+1th humpback whale individual in the jth dimension, The location of the best humpback whale individuals at present; The greedy strategy shown in formula (10) is used as the evaluation criterion to decide whether to update the position of the humpback whale individual:

8. The method for intelligent decision-making of bulk compacting process parameters based on digital twin model according to claim 3 is characterized in that: Step 4 specifically includes: Step 4.1: Use the case-based reasoning algorithm to retrieve case samples based on the process information input by the user to obtain output case samples; Step 4.2: If the number of output case samples is greater than 0, rank the output case samples according to the overall similarity and select the case sample with the highest overall similarity as the optimal solution, and then obtain the process parameter data in the optimal solution and execute step 4.4, otherwise execute step 4.3; Step 4.3: Input the process information input by the user into the digital twin algorithm model based on the BCCOA-BP neural network to predict the process parameter data; Step 4.4: Based on the obtained process parameter data, the behavior model is used to simulate the granular pressing and a granular pressing test is performed to verify and analyze the process parameter data to determine whether the process parameter data meets the pressing quality requirements. If so, the set of process parameter data and the corresponding process information are stored as a case sample in the process database, and the granular pressing digital twin model is updated. If not, return to step 4.

1.

9. The method for intelligent decision-making of bulk compacting process parameters based on digital twin model according to claim 8 is characterized in that: Step 4.1 specifically includes: Step 4.1.1: The user inputs the set process information; Step 4.1.2: Use the expert evaluation method to set the weight of each process information according to the degree of influence of each process information on the pressing quality during the granular pressing process. Based on the weight of each process information, calculate the overall similarity between the process information input by the user and the process information of each case sample in the process database; First, the Hamming distance formula is used to calculate the similarity between each process information input by the user and the corresponding process information in the case sample, that is, the local similarity of the case sample, as shown in formula (11): In the formula, is the similarity between a process information input by the user and the corresponding process information in the case sample, and A process information entered by the user. is the corresponding process information in the case sample, r is the number of the process information, and The upper and lower limits of the process information; According to the similarity between each process information input by the user and the corresponding process information in the case sample, the Kolodner formula is used to calculate the overall similarity between the process information input by the user and the process information in the case sample, as shown in formula (12): In the formula, sim(H,G) is the overall similarity between the process information input by the user and the process information in the case sample, q is the amount of process information, and w r is the weight of the rth process information and satisfies Step 4.1.3: Set a similarity threshold and find case samples whose overall similarity with the process information input by the user is greater than the similarity threshold as output case samples.

Images