Adhesive production monitoring method, system, readable storage medium and computer

Abstract

This invention provides a method, system, readable storage medium, and computer for monitoring adhesive production. The method includes: acquiring reaction data in real time using data sensors and constructing a feature sample vector based on the reaction data; constructing a feature regression model using the feature sample vector and mixer parameters as inputs, and optimizing the mixer parameters using the feature regression model to obtain an optimal parameter vector; predicting viscosity based on the optimal parameter vector to obtain a basic predicted viscosity, and estimating the uncertainty by weighted residuals of the basic predicted viscosity; performing data analysis on the reaction variable set using a preset data analysis algorithm, and introducing variational autoencoders and Wasserstein distance distributions to obtain a multi-output dataset; constructing a fusion decision model based on an evidence algorithm, inputting the basic predicted viscosity, uncertainty, and multi-output dataset into the fusion decision model to generate corresponding monitoring indicators, and using the monitoring indicators to achieve production monitoring.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method, system, readable storage medium, and computer for monitoring adhesive production. Background Technology

[0002] The polymerization process of adhesives (such as polyurethane, epoxy resin, and acrylate) is a key step that determines the final properties of the product, such as bond strength, weather resistance, and viscosity stability. This reaction is typically carried out in a reactor equipped with stirring and a heating / cooling jacket, and is a typical highly nonlinear, time-varying exothermic process. During the reaction, the viscosity of the product increases sharply with the degree of polymerization. Accurately monitoring viscosity changes and precisely determining the reaction endpoint are crucial for ensuring product quality, avoiding "gelling" accidents, and reducing energy consumption.

[0003] Currently, the traditional methods used in industry for monitoring adhesive production processes are mainly divided into the following two types, but both have certain limitations: (1) Soft measurement methods based on a single mechanism model: This type of method indirectly estimates viscosity by establishing empirical or semi-empirical formulas (such as the coupling of the Arrhenius equation and the rheological model) between viscosity and process variables such as temperature and stirring torque. However, in actual polymerization, the rheological parameters of the material (such as consistency coefficient and rheological index) will undergo significant nonlinear time-varying changes with molecular chain growth, branching and cross-linking. Traditional methods usually assume that these parameters are fixed constants, which leads to a serious deviation between the model prediction value and the actual viscosity in the later stage of the reaction, which can easily cause misjudgment of the endpoint.

[0004] (2) Threshold monitoring method based on a single statistic: Some processes rely on monitoring the univariate trend of the stirring motor torque or current to determine the endpoint. The reaction stops when the torque value or its rate of change reaches a certain empirical threshold. Although this method is simple, the threshold setting is highly dependent on operational experience and cannot distinguish between torque changes caused by normal viscosity increases and torque fluctuations caused by mechanical failures (such as bearing wear) or mixing abnormalities (such as the rod climbing effect). Once an unexpected torque spike occurs, it is very easy to trigger false alarms or cause premature shutdown. Summary of the Invention

[0005] Based on this, the purpose of the present invention is to provide an adhesive production monitoring method, system, readable storage medium, and computer to at least address the shortcomings of the aforementioned technologies.

[0006] This invention proposes a method for monitoring adhesive production, comprising: The reaction data is collected in real time by a data sensor installed inside the reactor, and a feature sample vector for the current moment is constructed based on the reaction data. The reaction data includes reaction temperature, stirrer parameters, and reactor parameters. Using the feature sample vector and the mixer parameters as input, a feature regression model is constructed, and the feature regression model is used to optimize the mixer parameters to obtain the optimal parameter vector; Viscosity is predicted based on the optimal parameter vector to obtain the basic predicted viscosity at the current moment. Weighted residual estimation is performed on the basic predicted viscosity to obtain the corresponding uncertainty. A set of response variables is constructed, and a preset data analysis algorithm is used to analyze the data of the set of response variables. A variational autoencoder and Wasserstein distance distribution are introduced to process the data-analyzed set of response variables to obtain a multi-output dataset. A fusion decision model is constructed based on an evidence algorithm. The basic prediction viscosity, the uncertainty, and the multi-output dataset are input into the fusion decision model to generate corresponding monitoring indicators, and the monitoring indicators are used to realize production monitoring.

[0007] Furthermore, the step of acquiring reaction data in real time using data sensors installed inside the reactor, and constructing a feature sample vector for the current moment based on the reaction data, includes: The low-frequency signal is upsampled to a preset sampling frequency using linear interpolation, and the data of the reactor is sampled according to the preset sampling frequency. The collected data is then filtered by moving average to obtain the reaction temperature, agitator parameters, and reactor parameters at the current moment. Based on the reaction temperature, the stirrer parameters, and the reactor parameters, a feature sample vector is constructed for the current moment. The size of the sliding window is defined, and the feature sample vector is statistically analyzed within the sliding window to obtain the corresponding local data matrix.

[0008] Furthermore, the steps of constructing a feature regression model using the feature sample vector and the mixer parameters as input, and optimizing the mixer parameters using the feature regression model to obtain the optimal parameter vector include: Based on the time series of the mixer parameters, discretization calculations are performed using the GL definition to extract the corresponding discrete variables; The weight coefficients of the discrete variables are recursively processed using the gamma function, and a weighted sum is calculated to obtain the final eigenvalues. The feature sample vector and the final feature value are used as inputs to construct a feature regression model. The feature regression model is then solved using smoothness constraints, flow velocity constraints, and regression error to obtain the optimal parameter vector.

[0009] Furthermore, the steps of predicting viscosity based on the optimal parameter vector to obtain the basic predicted viscosity at the current time, and estimating the weighted residual of the basic predicted viscosity to obtain the corresponding uncertainty include: The viscosity is predicted by using the feature regression model to obtain the basic predicted viscosity at the current time. For each historical sample at the current moment, viscosity prediction is performed according to the optimal parameter vector to obtain the viscosity prediction value of each historical sample; Each predicted viscosity value is compared with the basic predicted viscosity, and the weighted residual variance of the comparison result is calculated. The corresponding uncertainty is then calculated based on the weighted residual variance and the basic predicted viscosity.

[0010] Furthermore, the steps of constructing a set of response variables, performing data analysis on the set of response variables using a preset data analysis algorithm, and introducing a variational autoencoder and Wasserstein distance distribution to process the analyzed set of response variables to obtain a multi-output dataset include: Construct a set of reaction variables, wherein the set of reaction variables includes a set of physical variables and a set of torque variables; The torque variable set is processed using the independent component analysis algorithm to extract Gaussian residual vectors and calculate non-Gaussian statistical features; The physical variable set and the Gaussian residual vector are concatenated into a shared information matrix, and recursive principal component analysis is performed on it to extract the residual vector and calculate the Gaussian statistical features. A variational autoencoder reconstruction error is introduced, and the Wasserstein distance distribution offset index is used to construct the corresponding multi-output dataset.

[0011] This invention also proposes an adhesive production monitoring system, comprising: The data acquisition module is used to collect reaction data in real time through data sensors installed in the reactor, and to construct a feature sample vector at the current moment based on the reaction data. The reaction data includes reaction temperature, stirrer parameters and reactor parameters. The parameter optimization module is used to construct a feature regression model with the feature sample vector and the mixer parameters as input, and to optimize the mixer parameters using the feature regression model to obtain the optimal parameter vector. The viscosity prediction module is used to predict viscosity based on the optimal parameter vector to obtain the basic predicted viscosity at the current moment, and to perform weighted residual estimation on the basic predicted viscosity to obtain the corresponding uncertainty. The data processing module is used to construct a set of response variables, perform data analysis on the set of response variables using a preset data analysis algorithm, and introduce variational autoencoder and Wasserstein distance distribution to process the data-analyzed set of response variables to obtain a multi-output dataset. The production monitoring module is used to construct a fusion decision model based on an evidence algorithm. The basic prediction viscosity, the uncertainty, and the multi-output dataset are input into the fusion decision model to generate corresponding monitoring indicators, and the monitoring indicators are used to realize production monitoring.

[0012] Furthermore, the data acquisition module is specifically used for: The low-frequency signal is upsampled to a preset sampling frequency using linear interpolation, and the data of the reactor is sampled according to the preset sampling frequency. The collected data is then filtered by moving average to obtain the reaction temperature, agitator parameters, and reactor parameters at the current moment. Based on the reaction temperature, the stirrer parameters, and the reactor parameters, a feature sample vector is constructed for the current moment. The size of the sliding window is defined, and the feature sample vector is statistically analyzed within the sliding window to obtain the corresponding local data matrix.

[0013] Furthermore, the parameter optimization module is specifically used for: Based on the time series of the mixer parameters, discretization calculations are performed using the GL definition to extract the corresponding discrete variables; The weight coefficients of the discrete variables are recursively processed using the gamma function, and a weighted sum is calculated to obtain the final eigenvalues. The feature sample vector and the final feature value are used as inputs to construct a feature regression model. The feature regression model is then solved using smoothness constraints, flow velocity constraints, and regression error to obtain the optimal parameter vector.

[0014] Furthermore, the viscosity prediction module is specifically used for: The viscosity is predicted by using the feature regression model to obtain the basic predicted viscosity at the current time. For each historical sample at the current moment, viscosity prediction is performed according to the optimal parameter vector to obtain the viscosity prediction value of each historical sample; Each predicted viscosity value is compared with the basic predicted viscosity, and the weighted residual variance of the comparison result is calculated. The corresponding uncertainty is then calculated based on the weighted residual variance and the basic predicted viscosity.

[0015] Furthermore, the data processing module is specifically used for: Construct a set of reaction variables, wherein the set of reaction variables includes a set of physical variables and a set of torque variables; The torque variable set is processed using the independent component analysis algorithm to extract Gaussian residual vectors and calculate non-Gaussian statistical features; The physical variable set and the Gaussian residual vector are concatenated into a shared information matrix, and recursive principal component analysis is performed on it to extract the residual vector and calculate the Gaussian statistical features. A variational autoencoder reconstruction error is introduced, and the Wasserstein distance distribution offset index is used to construct the corresponding multi-output dataset.

[0016] The present invention also proposes a storage medium storing a computer program that, when executed by a processor, implements the above-described adhesive production monitoring method.

[0017] The present invention also proposes a computer, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described adhesive production monitoring method.

[0018] The adhesive production monitoring method, system, readable storage medium, and computer of this invention utilize the construction of corresponding feature sample vectors from reaction data, and construct a feature regression model using the feature sample vectors and mixer parameters. The model is then optimized to obtain the optimal parameter vector, which can quantify the long-range memory effect of polymer melt during shearing. This effectively compensates for the deficiency of traditional integer derivative models in describing the historical dependence of viscoelasticity, enabling the regression model to perceive richer information on changes in the microstructure of materials and significantly improving the resolution of complex rheological behavior of non-Newtonian fluids. Furthermore, the use of a collaborative feature regression model to solve the parameter vectors in the viscosity coupling prediction equation in real time allows the prediction model to learn and track the nonlinear migration of rheological properties caused by the dramatic increase in crosslinking density in the later stages of the polymerization reaction, thereby maintaining extremely high viscosity prediction accuracy. Attached Figure Description

[0019] Figure 1 This is a flowchart of the adhesive production monitoring method in the first embodiment of the present invention; Figure 2 This is a structural block diagram of the adhesive production monitoring system in the second embodiment of the present invention; Figure 3 This is a structural block diagram of the computer in the third embodiment of the present invention.

[0020] The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation

[0021] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0022] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0023] Example 1 Please see Figure 1 The figure shows a method for monitoring adhesive production in the first embodiment of the present invention, the method specifically including steps S101 to S104: S101, real-time reaction data is collected by a data sensor installed inside the reactor, and a feature sample vector at the current moment is constructed based on the reaction data. The reaction data includes reaction temperature, stirrer parameters, and reactor parameters. Furthermore, step S101 specifically includes steps S1011 to S1012: S1011, the low-frequency signal is upsampled to a preset sampling frequency using linear interpolation, and the data of the reactor is sampled according to the preset sampling frequency. The collected data is then filtered by moving average to obtain the reaction temperature, agitator parameters and reactor parameters at the current moment. S1012, construct the feature sample vector at the current moment based on the reaction temperature, the stirrer parameters and the reactor parameters, define the sliding window size, and perform statistics on the feature sample vector within the sliding window to obtain the corresponding local data matrix.

[0024] In practical implementation, the following online sensors are installed on the adhesive polymerization reactor, such as temperature sensor, torque sensor, pressure sensor and speed sensor. Since the sampling frequencies of each sensor are different, timestamp alignment and resampling are required. Using a preset sampling frequency (10Hz in this embodiment) as the reference time axis, low-frequency signals such as temperature and pressure are upsampled to the preset sampling frequency using linear interpolation. Data is sampled from the reactor using this sampling frequency, and the sampled data is filtered by moving average with a window width of 1 second (i.e., 10 sampling points) to eliminate electromagnetic interference from the motor and high-frequency noise caused by the passing frequency of the stirring blade.

[0025] Furthermore, a feature sample vector for the current moment is constructed based on the reaction temperature, stirrer parameters, and reactor parameters:

[0026] In the formula, Indicates the rate of change of torque; Indicates the current time The absolute temperature below; This represents the estimated degree of response at the previous sampling time. Indicates by the rate of change of torque Obtained by second-order polynomial fitting and differentiation, with a window width of 11 points; It represents the fractional-order torque characteristic.

[0027] Furthermore, a sliding window size is defined (in this embodiment, the sliding window size is 50), and the obtained feature sample vectors are statistically analyzed within the sliding window to obtain the corresponding local data matrix.

[0028] S102, using the feature sample vector and the mixer parameters as input, construct a feature regression model, and use the feature regression model to optimize the mixer parameters to obtain the optimal parameter vector; Furthermore, step S102 specifically includes steps S1021 to S1023: S1021, Based on the time series of the mixer parameters, discretization calculation is performed using the GL definition to extract the corresponding discrete variables; S1022, The weight coefficients of the discrete variables are recursively processed and a weighted sum is calculated using the gamma function to obtain the final eigenvalues; S1023, the feature sample vector and the final feature value are used as input to construct a feature regression model, and the feature regression model is solved by vector calculation using smoothing constraints, flow velocity constraints and regression error to obtain the optimal parameter vector.

[0029] In practical implementation, the aforementioned fractional-order torque characteristics are addressed. The time series of mixer parameters is discretized using the GL definition, and the weight coefficients of the discretization result are recursively processed and weighted by the gamma function to obtain the result. Furthermore, using the feature sample vector and the mixer parameters as input, a feature regression model is constructed, and the parameter vector is defined as follows: ,in, Indicates the sampling time The viscosity coefficient at that point, Indicates the sampling time rheological index under the following conditions Indicates the sampling time Temperature sensitivity coefficient and Indicates the sampling time The degree of reaction correction factor, in the process of adhesive polymerization, the quantitative relationship between viscosity and torque, temperature, and degree of reaction is represented by the following viscosity coupling prediction function:

[0030] In the formula, For the first Viscosity coupling prediction function for each sample. To calculate the extent of the reaction using the Arrhenius equation, This indicates absolute temperature.

[0031] Specifically, smoothing constraints, manifold constraints, and regression errors are constructed, where the smoothing constraint is:

[0032] In the formula, This represents the parameter vector to be optimized. Indicates the previous sampling time The optimal parameter vector; The manifold constraint is:

[0033] In the formula, Indicates the size of the sliding window. , These are the indices of two different samples within the sliding window. Indicates sample and samples manifold adjacency weights between them Indicates sample eigenvectors, Indicates sample eigenvectors, Indicates sample The predicted viscosity, Indicates sample The predicted viscosity.

[0034] The regression error is:

[0035] In the formula, This represents the sample index within the sliding window. Indicates the first Feature vectors of historical samples Indicates the first Predicted viscosity of a historical sample, Indicates the first Reference viscosity values ​​for historical samples; Construct the objective function using smoothness constraints, manifold constraints, and regression error:

[0036] In the formula, To smooth out the constraint regularization coefficients, is the manifold constraint regularization coefficient.

[0037] The optimal parameter vector is obtained by using the dynamic parameter vector in the real-time feature regression model with the objective function.

[0038] S103, perform viscosity prediction based on the optimal parameter vector to obtain the basic predicted viscosity at the current moment, and perform weighted residual estimation on the basic predicted viscosity to obtain the corresponding uncertainty; Furthermore, step S103 specifically includes steps S1031 to S1033: S1031, The viscosity of the optimal parameter vector is predicted using the feature regression model to obtain the basic predicted viscosity at the current moment; S1032, perform viscosity prediction on each historical sample at the current time according to the optimal parameter vector to obtain the viscosity prediction value of each historical sample; S1033, compare each of the predicted viscosity values ​​with the basic predicted viscosity, calculate the weighted residual variance of the comparison results, and calculate the corresponding uncertainty based on the weighted residual variance and the basic predicted viscosity.

[0039] In practical implementation, the obtained feature regression model is used to predict viscosity using the optimal parameter vector to obtain the base predicted viscosity at the current time. To evaluate the reliability of the prediction results, viscosity prediction is performed for each historical sample at the current time using the obtained optimal parameter vector. The obtained viscosity prediction values ​​are compared with the base predicted viscosity, and the weighted residual variance within the calculation sliding window of the comparison results is calculated. The corresponding uncertainty is then calculated using the weighted residual variance and the base predicted viscosity.

[0040] In the formula, This represents the weighted residual variance. This represents the basic predicted viscosity.

[0041] S104, construct a set of response variables, perform data analysis on the set of response variables using a preset data analysis algorithm, and introduce variational autoencoder and Wasserstein distance distribution to process the data-analyzed set of response variables to obtain a multi-output dataset; Furthermore, step S104 specifically includes steps S1041 to S1044: S1041, Construct a set of reaction variables, wherein the set of reaction variables includes a set of physical variables and a set of torque variables; S1042, The torque variable set is processed using the independent component analysis algorithm to extract the Gaussian residual vector and calculate the non-Gaussian statistical features; S1043, the physical variable set and the Gaussian residual vector are concatenated into a shared information matrix, and recursive principal component analysis is performed on it to extract the residual vector and calculate the Gaussian statistical features; S1044 introduces the variational autoencoder reconstruction error and utilizes the Wasserstein distance distribution offset index to construct the corresponding multi-output dataset.

[0042] In practice, the physical variables in the reaction process are constructed as a set of physical variables, and the torque variables in the reaction process are constructed as a set of torque variables. Data of normal batches are collected according to the data in the set of physical variables and the set of torque variables. The set of physical variables includes temperature, pressure, and reaction degree, and the set of torque variables includes torque change rate and fractional torque characteristics. Specifically, the torque variable set is processed using the independent component analysis algorithm to extract Gaussian residual vectors and calculate non-Gaussian statistical features. The physical variable set and the obtained Gaussian residual vectors are concatenated into a shared information matrix, and recursive principal component analysis is performed on it to extract residual vectors and calculate Gaussian statistical features. Furthermore, a variational autoencoder is constructed, consisting of an encoder and a decoder. The encoder compresses the input feature vector into a 2-dimensional distribution of latent vectors, and the decoder samples a latent vector from the latent distribution to reconstruct the original feature vector. The mean of the output of each decoder is calculated to obtain the error in feature reconstruction, and the reconstruction probability is used as an indicator to measure how well the sample conforms to the normal data distribution. The Wasserstein distance distribution offset index is used to detect the slow drift of the process data distribution in order to construct the corresponding multi-output dataset.

[0043] S105, construct a fusion decision model based on the evidence algorithm, input the basic prediction viscosity, the uncertainty and the multi-output dataset into the fusion decision model to generate corresponding monitoring indicators, and use the monitoring indicators to realize production monitoring.

[0044] In practical implementation, a fusion decision model is constructed based on an evidence algorithm, specifically the Dempster-Shafer evidence algorithm. This is achieved by defining an identification framework. ,in, This indicates that the reaction has reached its endpoint. Indicating that the reaction has not reached its endpoint, a basic probability assignment function is constructed for each source of evidence, including viscosity evidence, reconstruction probability evidence, and Wasserstein distance evidence. A basic probability assignment function is constructed for each of the three independent sources of evidence, denoted as follows: , , .

[0045] About functions Based on the predicted viscosity and target viscosity relative deviation And use linear interpolation to map the bias ratio to the support in the [0,1] interval:

[0046]

[0047] In the formula, This indicates the deviation compared to a threshold value, which is 0.1 in this embodiment. This represents the lower threshold of the deviation ratio; in this embodiment, this value is 0.03. Used to represent, based on viscosity deviation, the proposition The higher the support level, the stronger the support. Used to represent, based on viscosity deviation, the proposition The higher the support level, the more support is for the project not yet finished. Introducing the uncertainty obtained above This reflects the unreliability of current predictions. Allocating some trust to the "unknown" is necessary. :

[0048]

[0049]

[0050] About functions The reconstruction rate reflects the degree of consistency between the current data pattern and the normal operating condition pattern; the lower the probability, the more abnormal the pattern. First, the reconstruction probability is converted to a logarithmic scale. In the formula, Indicates the reconstruction probability. Representing extremely small positive numbers, in this embodiment, it is selected as... Set the maximum threshold for logarithms Minimum threshold for sum and logarithm To linearly map logarithmic values ​​to logarithmic values Support level:

[0051]

[0052]

[0053] This indicates the level of trust assigned to evidence that is "not yet concluded." The lower the probability of reconstruction (…), the lower the trust level. The smaller the value, the larger this item becomes.

[0054] About functions The Wasserstein distance measures the difference between the current data distribution and the normal baseline distribution. The larger the distance, the further the process deviates.

[0055] Calculate the relative distance. ;in, Indicates the current time The distance to Wasserstein below, This indicates the maximum reference distance under normal operating conditions; the relative distance is directly used as the reference distance. Support level:

[0056]

[0057]

[0058] The Dempster combination rule is used to fuse the aforementioned sources of evidence into a comprehensive BPA function. This comprehensive BPA function is then used to calculate the validity of the proposition. The comprehensive reliability function determines that the response has reached its endpoint and issues a termination command when the following conditions are met: 1. The overall reliability function is greater than the preset overall reliability threshold (in this embodiment, the threshold is set to 0.85), that is, based on all the current evidence, there is more than 85% confidence that the response has ended; 2. The first derivative of the predicted viscosity value is approximately equal to 0; 3. Change in the parameter vector Less than the parameter convergence tolerance.

[0059] In summary, the adhesive production monitoring method in the above embodiments of the present invention utilizes the construction of corresponding feature sample vectors from reaction data, and constructs a feature regression model using the feature sample vectors and mixer parameters. The optimal parameter vector is obtained through model optimization, which can quantify the long-range memory effect of polymer melt during shearing. This effectively compensates for the deficiency of traditional integer derivative models in describing the historical dependence of viscoelasticity, enabling the regression model to perceive richer information on changes in the microstructure of materials and significantly improving the resolution of complex rheological behavior of non-Newtonian fluids. Furthermore, the use of a collaborative feature regression model to solve the parameter vectors in the viscosity coupling prediction equation in real time allows the prediction model to learn and track the nonlinear migration of rheological properties caused by the dramatic increase in crosslinking density in the later stages of the polymerization reaction, thereby maintaining extremely high viscosity prediction accuracy.

[0060] Example 2 In another aspect, this invention also proposes an adhesive production monitoring system, please refer to [link / reference needed]. Figure 2 The figure shows an adhesive production monitoring system according to a second embodiment of the present invention, the system comprising: The data acquisition module 11 is used to acquire reaction data in real time through data sensors installed in the reactor, and construct a feature sample vector at the current moment based on the reaction data. The reaction data includes reaction temperature, stirrer parameters and reactor parameters. The parameter optimization module 12 is used to construct a feature regression model with the feature sample vector and the mixer parameters as input, and to optimize the mixer parameters using the feature regression model to obtain the optimal parameter vector. The viscosity prediction module 13 is used to predict viscosity based on the optimal parameter vector to obtain the basic predicted viscosity at the current moment, and to perform weighted residual estimation on the basic predicted viscosity to obtain the corresponding uncertainty. Data processing module 14 is used to construct a set of response variables, perform data analysis on the set of response variables using a preset data analysis algorithm, and introduce variational autoencoder and Wasserstein distance distribution to process the data-analyzed set of response variables to obtain a multi-output dataset. The production monitoring module 15 is used to construct a fusion decision model based on the evidence algorithm, input the basic prediction viscosity, the uncertainty and the multi-output dataset into the fusion decision model to generate corresponding monitoring indicators, and use the monitoring indicators to realize production monitoring.

[0061] Furthermore, the data acquisition module 11 is specifically used for: The low-frequency signal is upsampled to a preset sampling frequency using linear interpolation, and the data of the reactor is sampled according to the preset sampling frequency. The collected data is then filtered by moving average to obtain the reaction temperature, agitator parameters, and reactor parameters at the current moment. Based on the reaction temperature, the stirrer parameters, and the reactor parameters, a feature sample vector is constructed for the current moment. The size of the sliding window is defined, and the feature sample vector is statistically analyzed within the sliding window to obtain the corresponding local data matrix.

[0062] Furthermore, the parameter optimization module 12 is specifically used for: Based on the time series of the mixer parameters, discretization calculations are performed using the GL definition to extract the corresponding discrete variables; The weight coefficients of the discrete variables are recursively processed using the gamma function, and a weighted sum is calculated to obtain the final eigenvalues. The feature sample vector and the final feature value are used as inputs to construct a feature regression model. The feature regression model is then solved using smoothness constraints, flow velocity constraints, and regression error to obtain the optimal parameter vector.

[0063] Furthermore, the viscosity prediction module 13 is specifically used for: The viscosity is predicted by using the feature regression model to obtain the basic predicted viscosity at the current time. For each historical sample at the current moment, viscosity prediction is performed according to the optimal parameter vector to obtain the viscosity prediction value of each historical sample; Each predicted viscosity value is compared with the basic predicted viscosity, and the weighted residual variance of the comparison result is calculated. The corresponding uncertainty is then calculated based on the weighted residual variance and the basic predicted viscosity.

[0064] Furthermore, the data processing module 14 is specifically used for: Construct a set of reaction variables, wherein the set of reaction variables includes a set of physical variables and a set of torque variables; The torque variable set is processed using the independent component analysis algorithm to extract Gaussian residual vectors and calculate non-Gaussian statistical features; The physical variable set and the Gaussian residual vector are concatenated into a shared information matrix, and recursive principal component analysis is performed on it to extract the residual vector and calculate the Gaussian statistical features. A variational autoencoder reconstruction error is introduced, and the Wasserstein distance distribution offset index is used to construct the corresponding multi-output dataset.

[0065] The functions or operation steps implemented by the above modules and units are largely the same as those in the above method embodiments, and will not be repeated here.

[0066] The adhesive production monitoring system provided in this embodiment of the invention has the same implementation principle and technical effect as the aforementioned method embodiment. For the sake of brevity, any parts not mentioned in the system embodiment can be referred to the corresponding content in the aforementioned method embodiment.

[0067] Example 3 This invention also proposes a computer, please refer to [link / reference]. Figure 3 The computer shown in the third embodiment of the present invention includes a memory 10, a processor 20, and a computer program 30 stored in the memory 10 and executable on the processor 20. When the processor 20 executes the computer program 30, it implements the above-described adhesive production monitoring method.

[0068] The memory 10 includes at least one type of storage medium, such as flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 10 can be an internal storage unit of a computer, such as the computer's hard disk. In other embodiments, the memory 10 can be an external storage device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. Furthermore, the memory 10 can include both internal and external storage units of the computer. The memory 10 can be used not only to store application software and various types of data installed on the computer, but also to temporarily store data that has been output or will be output.

[0069] In some embodiments, the processor 20 may be an electronic control unit (ECU), a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip, used to run program code stored in the memory 10 or process data, such as executing access restriction programs.

[0070] It should be pointed out that, Figure 3 The structure shown does not constitute a limitation on the computer. In other embodiments, the computer may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0071] This invention also proposes a storage medium storing a computer program that, when executed by a processor, implements the adhesive production monitoring method described above.

[0072] Those skilled in the art will understand that the logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can mean any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0073] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0074] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0075] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0076] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for monitoring adhesive production, characterized in that, include: The reaction data is collected in real time by a data sensor installed inside the reactor, and a feature sample vector for the current moment is constructed based on the reaction data. The reaction data includes reaction temperature, stirrer parameters, and reactor parameters. Using the feature sample vector and the mixer parameters as input, a feature regression model is constructed, and the feature regression model is used to optimize the mixer parameters to obtain the optimal parameter vector; Viscosity is predicted based on the optimal parameter vector to obtain the basic predicted viscosity at the current moment. Weighted residual estimation is performed on the basic predicted viscosity to obtain the corresponding uncertainty. A set of response variables is constructed, and a preset data analysis algorithm is used to analyze the data of the set of response variables. A variational autoencoder and Wasserstein distance distribution are introduced to process the data-analyzed set of response variables to obtain a multi-output dataset. A fusion decision model is constructed based on an evidence algorithm. The basic prediction viscosity, the uncertainty, and the multi-output dataset are input into the fusion decision model to generate corresponding monitoring indicators, and the monitoring indicators are used to realize production monitoring.

2. The method for monitoring adhesive production according to claim 1, characterized in that, The steps of acquiring reaction data in real time using data sensors installed inside the reactor, and constructing a feature sample vector for the current moment based on the reaction data, include: The low-frequency signal is upsampled to a preset sampling frequency using linear interpolation, and the data of the reactor is sampled according to the preset sampling frequency. The collected data is then filtered by moving average to obtain the reaction temperature, agitator parameters, and reactor parameters at the current moment. Based on the reaction temperature, the stirrer parameters, and the reactor parameters, a feature sample vector is constructed for the current moment. The size of the sliding window is defined, and the feature sample vector is statistically analyzed within the sliding window to obtain the corresponding local data matrix.

3. The method for monitoring adhesive production according to claim 1, characterized in that, The steps of constructing a feature regression model using the feature sample vector and the mixer parameters as input, and then using the feature regression model to optimize the mixer parameters to obtain the optimal parameter vector include: Based on the time series of the mixer parameters, discretization calculations are performed using the GL definition to extract the corresponding discrete variables; The weight coefficients of the discrete variables are recursively processed using the gamma function, and a weighted sum is calculated to obtain the final eigenvalues. The feature sample vector and the final feature value are used as inputs to construct a feature regression model. The feature regression model is then solved using smoothness constraints, flow velocity constraints, and regression error to obtain the optimal parameter vector.

4. The method for monitoring adhesive production according to claim 1, characterized in that, The steps of predicting viscosity based on the optimal parameter vector to obtain the basic predicted viscosity at the current time, and estimating the weighted residual of the basic predicted viscosity to obtain the corresponding uncertainty include: The viscosity is predicted by using the feature regression model to obtain the basic predicted viscosity at the current time. For each historical sample at the current moment, viscosity prediction is performed according to the optimal parameter vector to obtain the viscosity prediction value of each historical sample; Each predicted viscosity value is compared with the basic predicted viscosity, and the weighted residual variance of the comparison result is calculated. The corresponding uncertainty is then calculated based on the weighted residual variance and the basic predicted viscosity.

5. The method for monitoring adhesive production according to claim 1, characterized in that, The steps of constructing a set of response variables, performing data analysis on the set of response variables using a preset data analysis algorithm, and introducing a variational autoencoder and Wasserstein distance distribution to process the analyzed set of response variables to obtain a multi-output dataset include: Construct a set of reaction variables, wherein the set of reaction variables includes a set of physical variables and a set of torque variables; The torque variable set is processed using the independent component analysis algorithm to extract Gaussian residual vectors and calculate non-Gaussian statistical features; The physical variable set and the Gaussian residual vector are concatenated into a shared information matrix, and recursive principal component analysis is performed on it to extract the residual vector and calculate the Gaussian statistical features. A variational autoencoder reconstruction error is introduced, and the Wasserstein distance distribution offset index is used to construct the corresponding multi-output dataset.

6. A monitoring system for adhesive production, characterized in that, include: The data acquisition module is used to collect reaction data in real time through data sensors installed in the reactor, and to construct a feature sample vector at the current moment based on the reaction data. The reaction data includes reaction temperature, stirrer parameters and reactor parameters. The parameter optimization module is used to construct a feature regression model with the feature sample vector and the mixer parameters as input, and to optimize the mixer parameters using the feature regression model to obtain the optimal parameter vector. The viscosity prediction module is used to predict viscosity based on the optimal parameter vector to obtain the basic predicted viscosity at the current moment, and to perform weighted residual estimation on the basic predicted viscosity to obtain the corresponding uncertainty. The data processing module is used to construct a set of response variables, perform data analysis on the set of response variables using a preset data analysis algorithm, and introduce variational autoencoder and Wasserstein distance distribution to process the data-analyzed set of response variables to obtain a multi-output dataset. The production monitoring module is used to construct a fusion decision model based on an evidence algorithm. The basic prediction viscosity, the uncertainty, and the multi-output dataset are input into the fusion decision model to generate corresponding monitoring indicators, and the monitoring indicators are used to realize production monitoring.

7. The adhesive production monitoring system according to claim 6, characterized in that, The data acquisition module is specifically used for: The low-frequency signal is upsampled to a preset sampling frequency using linear interpolation, and the data of the reactor is sampled according to the preset sampling frequency. The collected data is then filtered by moving average to obtain the reaction temperature, agitator parameters, and reactor parameters at the current moment. Based on the reaction temperature, the stirrer parameters, and the reactor parameters, a feature sample vector is constructed for the current moment. The size of the sliding window is defined, and the feature sample vector is statistically analyzed within the sliding window to obtain the corresponding local data matrix.

8. The adhesive production monitoring system according to claim 6, characterized in that, The parameter optimization module is specifically used for: Based on the time series of the mixer parameters, discretization calculations are performed using the GL definition to extract the corresponding discrete variables; The weight coefficients of the discrete variables are recursively processed using the gamma function, and a weighted sum is calculated to obtain the final eigenvalues. The feature sample vector and the final feature value are used as inputs to construct a feature regression model. The feature regression model is then solved using smoothness constraints, flow velocity constraints, and regression error to obtain the optimal parameter vector.

9. A readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the adhesive production monitoring method as described in any one of claims 1 to 5.

10. A computer comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the adhesive production monitoring method as described in any one of claims 1 to 5.

Images