Production raw material mixing control method of UV curing adhesive, electronic equipment and medium

Abstract

The invention belongs to the technical field of production control, and provides a production raw material mixing control method of a UV curing adhesive, electronic equipment and a medium. The method comprises the following steps: pre-training a deep optimization model and a state prediction model by using training data; determining first attribute information and first component information of a plurality of production raw materials of the to-be-mixed UV curing adhesive, and processing by using a deep optimization model to obtain a first stirring path; the real-time temperature in the mixing barrel is monitored, and when the real-time temperature is higher than a temperature threshold value, second attribute information and second component information of all the production raw materials in the mixing barrel are predicted through the state prediction model; and performing secondary treatment by using the deep optimization model to obtain a second stirring path, and continuing stirring according to the second stirring path until a stirring ending condition is met. According to the stirring device, the stirring effect on the UV curing adhesive raw materials can be remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of production control, and more particularly, to a method for controlling the mixing of raw materials for producing UV-curable adhesives, an electronic device, and a medium. Background Art

[0002] Due to its characteristics such as rapid curing, environmental friendliness, and high efficiency, UV-curable adhesives are widely used in many fields such as electronics, optics, and packaging. With the continuous improvement of the market's requirements for the performance of UV-curable adhesives, the accuracy and stability of raw material mixing during its production process have become the key factors determining product quality.

[0003] Currently, the mixing of raw materials for producing UV-curable adhesives mostly adopts traditional stirring equipment and manual control methods. This method has many drawbacks. On the one hand, it is difficult for manual control to accurately grasp the raw material ratio and mixing time, which easily leads to uneven mixing and affects the core properties such as the curing speed and bonding strength of the curable adhesive. On the other hand, the rotation speed and stirring path of traditional stirring equipment are fixed and cannot be flexibly adjusted according to the characteristics of different raw materials and formula requirements, resulting in low mixing efficiency and difficulty in meeting the needs of large-scale production.

[0004] Therefore, it is urgent to develop a new method for controlling the mixing of raw materials for producing UV-curable adhesives to solve the problems existing in the prior art and improve the production quality and efficiency of UV-curable adhesives. Summary of the Invention

[0005] In response to this, the present invention provides a method for controlling the mixing of raw materials for producing UV-curable adhesives, an electronic device, a medium, and a computer program product to solve at least one of the above technical problems.

[0006] The present invention provides a method for controlling the mixing of raw materials for producing UV-curable adhesives, including the following method steps:

[0007] Pre-train the deep optimization model and the state prediction model using training data;

[0008] Determine the first attribute information and the first component information of various raw materials for the UV-curable adhesive to be mixed, and use the deep optimization model to process the first attribute information and the first component information to obtain the first stirring path;

[0009] During the process of stirring and mixing the various raw materials in the mixing tank according to the first stirring path, monitor the real-time temperature in the mixing tank, and use the state prediction model to predict the second attribute information and the second component information of the various raw materials in the mixing tank when the real-time temperature is higher than the temperature threshold;

[0010] The second attribute information and the second component information are processed twice using a deep optimization model to obtain a second stirring path, and stirring is continued according to the second stirring path until a stirring end condition is reached.

[0011] The present invention also provides an electronic device, which includes: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the computer program implements any of the aforementioned methods when executed by the processor.

[0012] The present invention also provides a medium storing a computer program that can be executed by a processor to implement any of the methods described above.

[0013] The present invention also provides a computer program product, which includes a computer program that can be executed by a processor to implement any of the methods described above.

[0014] The beneficial effects of the present invention are:

[0015] The UV curing adhesive production raw material mixing control method of the present invention accurately grasps the relationship between raw material characteristics and stirring path and temperature changes through pre-training deep optimization and state prediction models, and generates an optimal stirring path; at the same time, during mixing, the stirring strategy can also be dynamically adjusted according to temperature changes, solving the problems of traditional uneven mixing and low efficiency, significantly improving mixing uniformity and stability, increasing product qualification rate, reducing labor costs, and adapting to large-scale production. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0017] Figure 1 The present invention discloses a flow chart of a method for controlling raw material mixing in the production of UV curing adhesive. DETAILED DESCRIPTION

[0018] The following specific embodiments illustrate the implementation of this application. Those familiar with the art can easily understand the other advantages and functions of this application from the content disclosed in this specification. Obviously, the described embodiments are part of the embodiments of this application, but not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0019] In addition, the technical features involved in different embodiments of the present application described below can be combined with each other as long as they do not conflict with each other.

[0020] As Figure 1 shown, an embodiment of the present invention discloses a method for controlling the mixing of raw materials for producing UV curable glue, including the following method steps:

[0021] S10. Use training data to pre-train the deep optimization model and the state prediction model.

[0022] The present invention pre-constructs a deep optimization model and a state prediction model. The deep optimization model is used to analyze and obtain the optimal stirring path of the stirring paddle in the mixing barrel, and the state prediction model is used to predict the actual state of each raw material in the mixing barrel when the temperature in the stirring barrel reaches the temperature threshold based on the actual stirring situation.

[0023] Deep optimization model: Constructed based on deep learning algorithms. The training data covers a large amount of data such as the attributes and component information of raw materials for producing UV curable glue, and the corresponding ideal stirring paths. Through training, the deep optimization model can learn the mapping relationship between raw material attributes, components, and stirring paths, so as to output a suitable stirring path according to the input raw material information in actual production to meet the mixing requirements of different raw materials.

[0024] State prediction model: Also using deep learning algorithms, the training data includes data on attribute changes and component changes of raw materials under different temperatures and stirring stages. By learning these data, the state prediction model can predict the changes in the attributes and components of raw materials when the temperature is higher than the threshold based on the currently monitored temperature in the mixing barrel, providing a basis for subsequent adjustment of the stirring path.

[0025] It should be noted that neural network algorithms such as DNN, RNN, and CNN can be used for deep learning algorithms, but variational autoencoders (VAEs) are preferably used. A VAE is a generative model composed of an encoder and a decoder. The encoder maps the input data to a distribution in a latent space, and the decoder samples from the latent space and generates reconstructed data. By minimizing the reconstruction error and the regularization term in the latent space, the model can learn the latent representation of the data and generate different stirring paths through sampling in the latent space, providing more possibilities for optimizing the mixing process. In addition, multiple VAEs can be used to construct a model group, and each VAE in the model group is used to predict the stirring path and the state of raw materials, and then the prediction results of each are integrated (such as weighted calculation) and other processing to improve the mixing effect.

[0026] S20. Determine the first attribute information and the first component information of various production raw materials of the UV curable glue to be mixed, and use a deep optimization model to process the first attribute information and the first component information to obtain a first stirring path.

[0027] The first attribute information includes the type of raw materials and the physical and chemical properties such as the viscosity, density, chemical activity, and thermal stability of the raw materials; the first component information refers to the specific ratio or dosage of each raw material in the mixing system. The first attribute information can be obtained from the production management system, which stores the specific formula information of the UV curable glue to be produced this time, including the above attribute information and component information of each production raw material.

[0028] The raw materials of UV curable glue usually include prepolymers, reactive diluents, photoinitiators, additives, etc., which are specifically described as follows:

[0029] 1. Prepolymers

[0030] Acrylate prepolymers: They are the most commonly used prepolymers in UV curable glue, such as epoxy acrylate, polyurethane acrylate, and polyester acrylate.

[0031] 2. Reactive diluents

[0032] Mono-functional diluents: Such as methyl acrylate, butyl acrylate, etc. They can reduce the viscosity of the system and improve the flexibility of the curable glue.

[0033] Multi-functional diluents: Such as trimethylolpropane triacrylate (TMPTA), pentaerythritol triacrylate (PETA), etc. They can increase the curing speed and crosslinking density, thereby improving the hardness, strength, and chemical resistance of the curable glue, but will increase the viscosity of the system and decrease the flexibility.

[0034] 3. Photoinitiators

[0035] Free radical photoinitiators: Include benzoin and its ethers, acetophenones, etc. Benzoin dimethyl ether is a commonly used free radical photoinitiator, which can rapidly generate free radicals under UV light irradiation and initiate the polymerization reaction of prepolymers and reactive diluents.

[0036] Cationic photoinitiators: Such as iodonium salts, sulfonium salts, etc. They are mainly used in cationic polymerization UV curing systems and have good thermal stability and storage stability.

[0037] 4. Additives

[0038] Plasticizers: Such as dibutyl phthalate (DBP), dioctyl phthalate (DOP), etc. They can increase the flexibility and toughness of the curable glue and improve its low-temperature performance.

[0039] Fillers: Commonly used ones include silica, calcium carbonate, talc, etc. Adding fillers can reduce costs and improve the hardness, wear resistance and heat resistance of the cured adhesive.

[0040] Coupling agent: such as silane coupling agent, which can improve the bonding strength between the cured adhesive and the surface of the adherend, and enhance the water resistance and resistance to wet and hot aging.

[0041] Antioxidants: Hindered phenolic antioxidants can prevent UV-curing adhesives from oxidative degradation during storage and use, thereby improving their stability and service life.

[0042] The first attribute information and first component information obtained above are input into a pre-trained deep optimization model. The deep optimization model outputs a first stirring path for the current raw material combination based on the mapping relationship learned during training. This first stirring path can ensure that the raw materials are dispersed quickly and evenly during the initial mixing stage.

[0043] S30, in the process of stirring and mixing the various production raw materials in the mixing barrel according to the first stirring path, monitor the real-time temperature in the mixing barrel, and use the state prediction model to predict the second attribute information and second component information of each production raw material in the mixing barrel when the real-time temperature is higher than the temperature threshold.

[0044] When raw materials are mixed, their temperature may rise due to factors such as stirring work and chemical reactions. This temperature change can have a significant impact on the properties of the raw materials and the mixing process. Therefore, the present invention provides for real-time monitoring of the temperature of the raw materials within the mixing barrel during stirring according to the first stirring path. Specifically, multiple temperature sensors can be installed on the wall of the mixing barrel or on the paddles of the stirring paddle to achieve real-time detection of the temperature of the raw materials being stirred within the mixing barrel.

[0045] When the temperature of the raw materials being stirred in the mixing barrel exceeds a temperature threshold, the mixing process has entered a new phase, indicating that the state and properties of the raw materials may have changed. At this point, the mixing path needs to be changed for subsequent mixing. The temperature threshold can be pre-determined based on actual production experience and corresponds to different raw material formulations. The specifics are not detailed here.

[0046] Specifically, the state prediction model is used to predict the new attribute information, i.e., the second attribute information, and the new component information, i.e., the second component information, of each production raw material in the mixing barrel at this time based on the current mixing situation, temperature changes, and previous raw material information.

[0047] It should be noted that some raw materials may undergo chemical reactions during the mixing process to generate new "raw materials", which will cause changes in the type and amount of raw materials in the mixing barrel.

[0048] S40: Perform secondary processing on the second attribute information and the second component information using a deep optimization model to obtain a second stirring path, and continue stirring according to the second stirring path until a stirring end condition is reached.

[0049] The second attribute information and second component information output by the state prediction model are re-entered into the deep optimization model. Since the raw material state has changed, the deep optimization model re-analyzes and recalculates based on the newly input data and outputs a second mixing path that adapts to the current raw material state. This path further optimizes the mixing effect based on the changed raw material properties, ensuring mixing uniformity and product quality.

[0050] Preset stirring termination conditions include, for example, reaching a set mixing time, meeting mixing uniformity standards (e.g., by measuring whether the variance of the raw material concentration distribution is less than a certain threshold), and achieving target values for the physical and chemical properties of the mixed system (e.g., viscosity, density, etc.). When these conditions are met, stirring is stopped, completing the entire mixing process for the raw materials used in UV curing adhesive production.

[0051] The UV curing adhesive production raw material mixing control method of the present invention accurately grasps the relationship between raw material characteristics and stirring path and temperature changes through pre-training deep optimization and state prediction models, and generates an optimal stirring path; at the same time, during mixing, the stirring strategy can also be dynamically adjusted according to temperature changes, solving the problems of traditional uneven mixing and low efficiency, significantly improving mixing uniformity and stability, increasing product qualification rate, reducing labor costs, and adapting to large-scale production.

[0052] It should be noted that the above-mentioned solution of the present invention also has a preset premise, that is, the structural parameters of the mixing barrel are fixed. Correspondingly, the training data used for the deep optimization model and the state prediction model training and the prediction of the mixing path are all based on the same above-mentioned structural parameters.

[0053] As an example, the use of training data to pre-train the depth optimization model and the state prediction model includes:

[0054] Training a depth optimization model using the first training dataset, and training a state prediction model using the second training dataset;

[0055] During the pre-training process, the training effect is evaluated a specified number of times to obtain a first effect evaluation set corresponding to the depth optimization model and a second effect evaluation set corresponding to the state prediction model;

[0056] If the first effect evaluation set and / or the second effect evaluation set does not meet the mid-training condition, using a generative adversarial network to update and expand the first training data set and / or the second training data set to obtain a first updated training data set and / or a second updated training data set;

[0057] Continuously train the deep optimization model and the state prediction model using the first updated training dataset and / or the second updated training dataset;

[0058] Repeat the above process until the deep optimization model and the state prediction model reach the training goal.

[0059] In this embodiment, a certain amount of historical stirring data corresponding to the same stirring mechanism (i.e., the structural parameters of the mixing barrel are the same as those of the mixing barrel for this stirring) is collected in advance. These historical stirring data contain corresponding stirring effect labels, and based on this, training data is constructed respectively, so as to integrate and obtain the first training dataset and the second training dataset. Among them, the first training dataset and the second training dataset are used to train the deep optimization model and the state prediction model respectively, because the functions of the two models are different. The deep optimization model is used to generate a stirring path according to the raw material attributes and component information, and the state prediction model is used to predict the change of the raw material state according to information such as the real-time temperature, raw material formula, cumulative stirring duration, and actual stirring speed.

[0060] Since the historical stirring data collected is for the same stirring mechanism, it limits the quantity and diversity index of the training data, which is not conducive to the training effect of the model. Therefore, the present invention is set to evaluate the training effect of the model according to a preset number of times during the training process, so as to evaluate the actual performance of the training data used. If it does not meet the expectations, the training data will be optimized and adjusted again.

[0061] Among them, for the deep optimization model, the evaluation results form the first effect evaluation set; for the state prediction model, the evaluation results form the second effect evaluation set. These evaluation sets record the performance of the model in the previous training stage, such as indicators such as prediction accuracy and error. The prediction accuracy index refers to the proportion of correct prediction results, and the error index can be calculated using the mean square error (MSE) or the mean absolute error (MAE), etc.

[0062] When the first effect evaluation set and the second effect evaluation set show that the training effect of the model does not meet the conditions set in the middle stage of training (such as too large prediction error, too low accuracy, etc.), it indicates that the data diversity of the training data collected in the early stage is insufficient, resulting in the model being prone to falling into a local optimum. To solve this technical problem, the present invention is set to use the generative adversarial network technology to expand and update the training data of the initially determined first training dataset and second training dataset, and then use the new training dataset with better diversity index to train the model subsequently, which can ensure that the model does not fall into a local optimum and is also conducive to improving the training efficiency of the model.

[0063] Specifically, the generative adversarial network consists of a generator and a discriminator, and generates new training data similar to the original data through the adversarial game between the two, so as to expand and update the first training data set and the second training data set to obtain a new training data set, that is, the first updated training data set and the second updated training data set.

[0064] In the process of using the new training data set to conduct subsequent training on the model, the foregoing methods can also be repeatedly adopted for processes such as evaluation of training effects, update and expansion of training data sets, and retraining until the deep optimization model and the state prediction model reach the training objectives (that is, the prediction accuracy of the model, the error range of predicted values, etc. all meet the standards). At this time, it is considered that the model training is completed and can be put into practical application.

[0065] As an example, the determination method of the specified number of times is as follows:

[0066] Determine the historical stirring data corresponding to each training data in the first training data set and the second training data set, and extract production raw material information and stirring parameter information from the historical stirring data;

[0067] Statistically obtain the type quantity of the production raw material information, calculate the parameter range, parameter distribution information, and parameter entropy value according to the stirring parameter information; use a clustering algorithm to divide each stirring parameter information into different clusters, and statistically obtain the number of clusters;

[0068] Construct a diversity feature according to the type quantity, parameter range, parameter distribution information, parameter entropy value, and number of clusters, and use a classifier to classify the diversity feature to obtain a diversity index value;

[0069] Match the specified number of times according to the diversity index value and the positive correlation.

[0070] In this embodiment, multiple evaluations of the training effect can improve the accuracy of the evaluation of the training effect as much as possible, but too many evaluations will lead to a decrease in training efficiency. When there is a lot of training data to be used, this decrease in training efficiency is even unacceptable. To solve this problem, the present invention evaluates the diversity of the training data in the preliminarily constructed training data sets (that is, the first training data set and the second training data set), and dynamically adjusts the number of evaluations in the training process according to the evaluation results, so as to achieve a balance between evaluation accuracy and training efficiency.

[0071] Specifically, the production raw material information for stirring and the specific stirring parameter information used are respectively obtained from each piece of historical stirring data. Among them, the production raw materials used for different types of UV curable adhesives are different. By statistically calculating the raw material formulas in each piece of historical stirring data, the type quantity of the production raw material information, that is, how many raw material formulas there are in total, can be obtained. And the stirring parameter information used in the stirring process includes the stirring speed, stirring duration, and the variation range of the stirring blade angle, etc. Based on this, the parameter range, parameter distribution information, and parameter entropy value corresponding to each piece of historical stirring data can be analyzed as follows:

[0072] Parameter range: Examine the value range of each dimension in the historical stirring parameters. For example, the value range of the stirring speed, the length range of the stirring time, the variation range of the stirring blade angle, etc. The wider the value range, the higher the diversity of these historical stirring data in this dimension.

[0073] Parameter distribution information: Obtain the distribution form of the stirring parameters within their value range by plotting a histogram or probability density function. If the data is evenly distributed, it indicates that there is relatively rich data in each value interval, and the diversity of these historical stirring parameters is good; if the data is concentrated in certain specific intervals and sparse in other intervals, the diversity is poor.

[0074] Entropy value: Entropy is an index to measure the uncertainty or chaos degree of data. Calculate the joint entropy of all historical stirring data. The higher the entropy value, the greater the uncertainty of the data and the higher the diversity. For example, when the value types of the stirring speed are more and the distribution is relatively uniform, its entropy value will be larger, reflecting better diversity of this parameter.

[0075] Among them, the stirring speed is, for example, 100 r / min, 200 r / min, 300 r / min, 400 r / min, 500 r / min, and its entropy value calculation formula is:

[0076]

[0077] Among them, is the entropy value of the stirring speed, is the stirring speed array, is the stirring speed corresponding probability.

[0078] The entropy value calculation formula of the stirring duration is:

[0079]

[0080] The stirring duration follows a normal distribution , and its probability density function is , then the entropy value of the stirring time is:

[0081]

[0082] Among them, is the mean value, is the standard deviation.

[0083] The entropy value calculation formula for the change range of the stirring paddle blade angle is:

[0084] The value range of the stirring paddle blade angle is divided into n intervals, namely A1, A2, ⋯, An. Through the statistics of historical stirring parameter data, the frequencies f1, f2, ⋯, fn of the stirring paddle blade angle appearing in each interval are obtained, and the frequencies are regarded as probabilities , then the entropy value H calculation formula for the change range of the stirring paddle blade angle is

[0085] .

[0086] Based on the above three calculated entropy values, namely , , the joint entropy can be integrated and calculated.

[0087] Cluster analysis: Use the clustering algorithm to divide the historical stirring parameter data into different clusters. If the number of clusters in the clustering result is large, and the data within each cluster is relatively compact, and the difference between clusters is large, it indicates that these historical stirring data can be clearly divided into multiple different categories, reflecting that the data has high diversity; on the contrary, if only a few clusters can be divided, or the data within the clusters is relatively scattered, and the difference between clusters is not obvious, it indicates that the diversity of these historical stirring data is low.

[0088] Integrate the above data such as the type number of the production raw material information, the parameter range of the stirring parameter information, the parameter distribution information, the parameter entropy value, and the number of clusters obtained previously to form a diversity feature set. Input the constructed diversity features into a pre-trained classifier, and the classifier will classify the diversity degree of the data according to these features and output a diversity index value. This index value can quantitatively represent the diversity level of the training data, and the higher the value, the better the diversity of the data. The classifier can be constructed based on random forests, decision trees, etc., which will not be elaborated here.

[0089] A positive correlation is established in advance between the diversity index value and the specified number of times, that is, the higher the diversity index value, the more corresponding specified number of times. According to this positive correlation, after obtaining the diversity index value, the specified number of times for evaluating the training effect during the pre-training process can be matched. If the diversity of the training data is high, it is expected that the training effect of the model is good, then the model needs to be evaluated more times during the training process to ensure the training effect, so as to timely discover problems and make adjustments (such as data augmentation operations); conversely, if the data diversity is low, it is expected that the training effect of the model is insufficient, and at this time, a relatively small number of evaluation times may be sufficient to obtain an accurate evaluation result. The positive correlation can be represented by a look-up table or a fitting function, and specific forms are not limited.

[0090] As an example, the stirring parameter information includes the stirring speed, the stirring duration, and the variation range of the stirring paddle angle. The parameter entropy value is obtained by integrating the entropy values of the stirring speed, the stirring duration, and the variation range of the stirring paddle angle. Specifically:

[0091] Calculate the entropy values of the variation ranges of the stirring speed, the stirring duration, and the stirring paddle angle respectively, calculate the mutual information between the entropy values pairwise, and approximately calculate the joint entropy based on the calculated mutual information.

[0092] In this embodiment, during the actual stirring process, parameters such as the stirring speed and the stirring paddle angle are affected by various factors. The actual implemented stirring speed and stirring paddle angle may deviate greatly from the set values, resulting in the relationship between parameters not being a simple deterministic function relationship, making it difficult to directly and accurately calculate the joint entropy. Therefore, the present invention does not directly calculate the joint entropy among the variation ranges of the stirring speed, the stirring duration, and the stirring paddle angle, that is, does not adopt the following calculation formula:

[0093]

[0094] Count the number of occurrences of each combination of the stirring speed, the stirring duration, and the stirring paddle angle. Assume that a total of N stirring operations are performed, and a certain combination The number of occurrences is , is the joint probability of each combination.

[0095] Instead, first calculate the entropy values of the stirring speed, the stirring duration, and the stirring paddle angle respectively, and then approximately calculate these entropy values to obtain the joint entropy. Specifically as follows:

[0096] Stirring speed and stirring duration The mutual information between them:

[0097]

[0098] in, yes and The joint probability when .

[0099] Stirring speed The range of variation of the stirring blade angle Mutual information between:

[0100]

[0101] in, yes and The joint probability when .

[0102] Stirring time The range of variation of the stirring blade angle Mutual information between:

[0103]

[0104] in, yes and The joint probability when .

[0105] The calculation formula of approximate joint entropy is as follows:

[0106]

[0107] The above-mentioned integration method of the present invention takes into account the mutual relationship between the three parameters and avoids repeated calculation of information by subtracting the mutual information between any two parameters, thereby obtaining a relatively more accurate joint entropy approximation.

[0108] As an example, using a deep optimization model to perform secondary processing on the second attribute information and the second component information to obtain a second stirring path includes:

[0109] Retrieving a first average time consumed by the state prediction model for a prediction and a second average time consumed by the depth optimization model for a prediction, obtaining a compensation coefficient based on the sum of the first average time consumed and the second average time consumed, and multiplying the compensation coefficient by a temperature threshold to obtain a predicted temperature;

[0110] A second stirring path is obtained by performing secondary processing on the second attribute information, the second component information and the predicted temperature using a deep optimization model.

[0111] In this embodiment, when the temperature of the raw materials in the mixing barrel reaches the temperature threshold, it takes a certain amount of time for the state prediction model and the depth optimization model to make corresponding predictions. During this prediction process, the temperature of the raw materials in the mixing barrel may further increase. If the second stirring path is predicted based on the temperature threshold, the obtained second stirring path is obviously not optimal.

[0112] To address this, the present invention first obtains the prediction history of the state prediction model and the depth optimization model, and statistically calculates the average time consumption for each prediction of the two. Then, the sum value of the two average time consumptions is calculated, and this sum value represents the total prediction time from when the real-time temperature in the mixing barrel is higher than the temperature threshold until the second stirring path is obtained.

[0113] The control relationship between the sum value of the average prediction time consumption and the compensation coefficient is preset. The following table shows an example of this control relationship. By querying this control relationship, a suitable compensation coefficient can be obtained.

[0114] Sum range (First average time + Second average time) Compensation coefficient 0-5 0.8 5-10 1.05 10-15 1.10 15-20 1.15 20-25 1.20 Above 25 1.25

[0115] Note: The unit of the sum value in the above table is the unit time length, and the unit time length can be, for example, 2s, 5s, 10s, etc.

[0116] Multiply the obtained compensation coefficient by the preset temperature threshold to calculate the predicted temperature. This predicted temperature is closer to the temperature of the raw materials in the mixing barrel when the second stirring path is executed compared to the temperature threshold.

[0117] Input the second attribute information, the second component information of each production raw material in the mixing barrel, and the calculated predicted temperature into the depth optimization model together. The depth optimization model will comprehensively consider these input information and perform secondary processing on these data.

[0118] It can be understood that when training the depth optimization model, the training data also includes the temperatures of different types of raw materials in the barrel. As an alternative solution, the depth optimization model here can also be a model specifically used for secondary processing. When training it, the training data includes the temperatures of different types of raw materials in the barrel, while for the depth optimization model for primary processing, the temperatures of different types of raw materials in the barrel included in its training data can be fixed values.

[0119] As an example, the change of the stirring path is achieved by adjusting the relevant structure of the stirring blades.

[0120] In this embodiment, the angle of the blades on the stirring paddle is adjustable, and the change of the stirring path can be achieved by adjusting its angle.

[0121] For example, vertical mixers feature an adjustment mechanism to change the angle of the mixing blades relative to the horizontal plane. This can be accomplished by adjusting the cylinder, extending the piston rod and pushing the adjustment seat along the mixing shaft. This in turn rotates the mixing blades around the axis, changing their angle relative to the horizontal plane. This in turn propels the mixture along a different path, improving mixing efficiency and uniformity. Alternatively, some mixing mechanisms feature a mixing blade that rotates in conjunction with a fixed shaft, allowing the blade's inclination to be adjusted during mixing, thus altering the mixing path.

[0122] Correspondingly, the stirring path obtained above is a specific blade angle / angle combination, and the angle combination means that the angle of the blade changes during the stirring process of one stage.

[0123] As an example, the stirring path is changed by adjusting the combined structure of multiple stirring components.

[0124] In this embodiment, for a mixing barrel equipped with multiple stirring paddles, the stirring path can be changed by optimizing the blade angles of the multiple stirring paddles. Correspondingly, the stirring path obtained above is the set of blade angles / angle combinations of the multiple stirring paddles.

[0125] An embodiment of the present invention further provides an electronic device comprising: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the computer program implements any of the aforementioned methods when executed by the processor.

[0126] An embodiment of the present invention further provides a medium storing a computer program that can be executed by a processor to implement any of the methods described above.

[0127] An embodiment of the present invention further provides a computer program product, which includes a computer program that can be executed by a processor to implement any of the methods described above.

[0128] Those of ordinary skill in the art will appreciate 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 terms of function in the above description. 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.

[0129] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A method for controlling the mixing of raw materials for producing a UV-curable adhesive, characterized in that: It includes the following method steps: Pre-train the deep optimization model and the state prediction model using training data; Determine the first attribute information and the first component information of various production raw materials of the UV curable glue to be mixed, and use the deep optimization model to process the first attribute information and the first component information to obtain the first stirring path; During the process of stirring and mixing each production raw material in the mixing barrel according to the first stirring path, monitor the real-time temperature in the mixing barrel, and use the state prediction model to predict that when the real-time temperature is higher than the temperature threshold, the second attribute information and the second component information of each production raw material in the mixing barrel; Use the deep optimization model to perform secondary processing on the second attribute information and the second component information to obtain the second stirring path, and continue stirring according to the second stirring path until the stirring end condition is reached.

2. The production raw material mixing control method of a UV curable adhesive according to claim 1, characterized in that: The pre-training of the deep optimization model and the state prediction model using training data includes: Train the deep optimization model using the first training data set, and train the state prediction model using the second training data set; During the pre-training process, evaluate the training effect a specified number of times to obtain the first effect evaluation set corresponding to the deep optimization model and the second effect evaluation set corresponding to the state prediction model; If the first effect evaluation set and / or the second effect evaluation set do not meet the mid-training conditions, use the generative adversarial network to update and expand the training data of the first training data set and / or the second training data set to obtain the first updated training data set and / or the second updated training data set; Use the first updated training data set and / or the second updated training data set to continue training the deep optimization model and the state prediction model; Repeat the above process until the deep optimization model and the state prediction model reach the training goal.

3. The method for controlling the mixing of raw materials for producing a UV curable adhesive according to claim 2, characterized in that: The determination method of the specified number of times is: Determine the historical stirring data corresponding to each training data in the first training data set and the second training data set, and extract the production raw material information and the stirring parameter information from the historical stirring data; Statistically obtain the type number of the production raw material information; calculate the parameter range, the parameter distribution information, and the parameter entropy value according to the stirring parameter information; Use the clustering algorithm to divide each stirring parameter information into different clusters, and statistically obtain the number of clusters; Construct a diversity feature according to the type number, the parameter range, the parameter distribution information, the parameter entropy value, and the number of clusters, and use a classifier to classify the diversity feature to obtain a diversity index value; Match the specified number of times according to the diversity index value and the positive correlation relationship.

4. A method for controlling the mixing of raw materials for producing a UV curable adhesive according to claim 3, characterized in that: The stirring parameter information includes the change ranges of the stirring speed, the stirring duration, and the stirring paddle angle. The parameter entropy value is obtained by integrating the entropy values of the change ranges of the stirring speed, the stirring duration, and the stirring paddle angle. Specifically: Calculate the entropy values of the change ranges of the stirring speed, the stirring duration, and the stirring paddle angle respectively, calculate the mutual information between the entropy values pairwise, and approximately calculate the joint entropy based on the calculated mutual information.

5. A method for controlling the mixing of raw materials for producing a UV-curable adhesive according to claim 1, characterized in that: Performing secondary processing on the second attribute information and the second component information using a depth optimization model to obtain a second stirring path, including: Retrieving the first average time taken for a single prediction by the state prediction model and the second average time taken for a single prediction by the depth optimization model, matching a compensation coefficient based on the sum of the first average time and the second average time, and multiplying the compensation coefficient by the temperature threshold to obtain a predicted temperature; Performing secondary processing on the second attribute information, the second component information, and the predicted temperature using the depth optimization model to obtain a second stirring path.

6. The production raw material mixing control method of a UV curable adhesive according to claim 1, characterized in that: Changing the stirring path by adjusting the relevant structure of the stirring blades.

7. A method for controlling the mixing of raw materials for producing a UV-curing adhesive according to claim 1, characterized in that: Changing the stirring path by adjusting the combined structure of multiple stirring components.

8. An electronic device, characterized in that: The electronic device includes: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, and the computer program, when executed by the processor, implements the method according to any one of claims 1-7.

9. A medium, characterized in that: The medium stores a computer program executable by a processor to implement the method according to any one of claims 1-7.

10. A computer program product, characterized in that: The computer program product includes a computer program executable by a processor to implement the method according to any one of claims 1-7.

Images