EVA hot melt adhesive production monitoring method and system

Abstract

The invention relates to the technical field of hot melt adhesives, in particular to an EVA hot melt adhesive production monitoring method and system.The method comprises the steps that firstly, environmental condition data and production process parameter data of an EVA hot melt adhesive production site are collected and processed, and EVA hot melt adhesive production data are obtained; then, carrying out relation learning on historical production data and EVA hot melt adhesive quality data, and constructing an EVA hot melt adhesive production stability monitoring model; thirdly, inputting EVA hot melt adhesive real-time production data into the stability monitoring model, and obtaining an EVA hot melt adhesive quality data prediction result through feature extraction and fusion; and finally, judging whether the current production condition is stable or not according to a prediction result, and monitoring and reminding. According to the method, a plurality of influence factors in the EVA hot melt adhesive preparation and processing link are comprehensively collected and analyzed, so that the stability of the EVA hot melt adhesive preparation and processing flow is improved, and the quality and consistency of the EVA hot melt adhesive product are further improved.

Description

Technical Field

[0001] The invention relates to the technical field of hot melt adhesives, and in particular to an EVA hot melt adhesive production monitoring method and system. Background Art

[0002] EVA hot melt adhesive is a hot melt adhesive with ethylene-vinyl acetate copolymer (EVA) as the main component. It is widely used in many fields for its low melting point, strong adhesion, flexibility and weather resistance, as well as its non-toxic and harmless environmental protection characteristics during use, meeting the needs of various industries and consumer products.

[0003] EVA hot melt adhesive is used for bonding purposes, such as edge banding and veneer bonding of furniture in furniture manufacturing, bonding and laminating of fabrics in textile processing, bonding and curing of shoe bodies in footwear manufacturing, fixing and insulating electronic components in the electronics and electrical industry, etc. In addition, EVA hot melt adhesive is also widely used in handicraft production and is suitable for the production and bonding of various handicrafts. The quality and safety of EVA hot melt adhesive products will affect the production and application of related manufacturing industries. Its bonding strength, heat resistance, weather resistance and safety will affect the quality and safety of related products, thereby limiting the market competitiveness of related products. Therefore, it is necessary to optimize the preparation and processing methods of EVA hot melt adhesive.

[0004] There are many optimization methods for the preparation and processing of EVA hot melt adhesive in the prior art. For example, a method for preparing and processing EVA hot melt adhesive provided by a Chinese patent with publication number CN113043565A proposes a solution of using an air cooling device and a shaping device to solve the problems of low cooling efficiency and poor straightness of EVA hot melt adhesive sticks. The hot melt adhesive sticks are initially shaped by air cooling to reduce insufficient cooling of the hot melt adhesive sticks and bending deformation under gravity. This method improves the cooling efficiency and enhances the straightness and overall quality of the hot melt adhesive sticks. A modified EVA hot melt adhesive and a preparation method provided by a Chinese patent with publication number CN116254073A first react part of the amino group of octaaminopropyl cage-type polysilsesquioxane with dimethyl phosphite, and then react the remaining amino group with acrylic acid-modified polyethylene wax to obtain a composite additive. By adding the composite additive, the EVA hot melt adhesive can be given good comprehensive properties. A Chinese patent with publication number CN112172072A provides an EVA hot melt adhesive preparation system. This patent addresses the problems of difficult adjustment of the diameter of the glue stick, low production efficiency and poor stability of the fixing device during the hot melt adhesive production process. It proposes adjusting the diameter of the glue stick and improving production efficiency through an adjustable extruder head and a cutting device, providing mounting parts to enhance the stability of the fixing frame, and controlling the diameter of the glue stick by connecting a driving airbag and providing a collecting device to improve operational convenience, thereby achieving the production of various types of glue sticks and improving the applicability and production efficiency of the device.

[0005] The existing technology has improved the comprehensive performance of EVA hot melt adhesive products to a certain extent, but has not fully considered the impact of real-time changes in environmental conditions and production process parameters during the preparation and processing of EVA hot melt adhesive on the melting of raw materials and the material extrusion process during the production process, resulting in the problem of low quality and consistency of EVA hot melt adhesive products.

[0006] Therefore, a method and system for monitoring the production of EVA hot melt adhesive are proposed. Summary of the invention

[0007] The purpose of the present invention is to provide an EVA hot melt adhesive production monitoring method and system. The present invention relates to the field of EVA hot melt adhesive production technology, specifically to an EVA hot melt adhesive production monitoring method and system. First, the environmental condition data and production process parameter data of the EVA hot melt adhesive production site are collected and processed to obtain the EVA hot melt adhesive production data. Then, by learning the relationship between historical production data and EVA hot melt adhesive quality data, an EVA hot melt adhesive production stability monitoring model is constructed; then, the real-time production data of the EVA hot melt adhesive is input into the stability monitoring model, and the prediction results of the EVA hot melt adhesive quality data are obtained through feature extraction and fusion; finally, through the prediction results, it is judged whether the current production conditions are stable and monitoring reminders are given.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] A method and system for monitoring the production of EVA hot melt adhesive, comprising:

[0010] Collect real-time environmental condition data and production process parameter data of the EVA hot melt adhesive production site; the environmental condition data include ambient temperature, relative air humidity, equipment vibration intensity and workshop light intensity; the production process parameter data include melting temperature, extrusion pressure, extrusion speed, cooling temperature;

[0011] Preprocessing environmental condition data and production process parameter data; the data preprocessing includes timestamp alignment, normalization processing and data integration; the timestamp alignment is used to synchronize the time of different data sources; the normalization processing scales different dimensional data to the same range;

[0012] Real-time monitoring of production stability, inputting the environmental condition data and production process parameter data into the production stability monitoring model, and obtaining the EVA hot melt adhesive quality data prediction result through the feature extraction layer, representation learning layer, feature fusion layer and output layer in sequence. The prediction result is used to judge whether the current production conditions are stable and issue a stability monitoring reminder; the EVA hot melt adhesive quality data includes bonding strength, flexibility and appearance quality scores.

[0013] The data preprocessing includes normalization, timestamp alignment and data integration;

[0014] Processing missing values, outliers and noise data of the EVA hot melt adhesive production data, and converting all numerical data into unified floating point data to obtain standardized data;

[0015] The standardized data are scaled to the same range through normalization processing to obtain data of the same dimension.

[0016] The training process of the production stability monitoring model includes:

[0017] Step 1: Collect historical production data, including historical environmental condition data, historical production process parameter data and historical EVA hot melt adhesive quality data;

[0018] Step 2: preprocessing the historical production data, and then dividing the training data according to the ratio to obtain training data and test data; the preprocessing operation includes normalization, timestamp alignment and data integration;

[0019] Step 3: Extracting environmental time series features and production parameter time series features; the extraction of environmental time series features and production parameter time series features includes:

[0020] Applying a sliding window to the historical environmental condition data and the historical EVA hot melt adhesive quality data in the training data to generate an environmental condition window sequence;

[0021] Input each environmental condition window sequence into the LSTM network, calculate the hidden state and output of the LSTM, and obtain the environmental time series characteristics;

[0022] Applying a sliding window to the historical production parameter data and the historical EVA hot melt adhesive quality data in the training data to generate a production parameter window sequence;

[0023] Input each production parameter window sequence into the LSTM network, calculate the hidden state and output of the LSTM, and obtain the sequence characteristics between production parameters;

[0024] Step 4: Input the historical environmental condition data and the historical EVA hot melt adhesive quality data into the convolution layer in combination with the environmental time series characteristics to obtain the environmental associated feature vector; input the historical production process parameter data and the historical EVA hot melt adhesive quality data into the convolution layer in combination with the production parameter time series characteristics to obtain the production parameter associated feature vector;

[0025] The process of obtaining the environment-related feature vector and the production parameter-related feature vector is as follows:

[0026] Environment-related feature vector acquisition:

[0027] Input the historical environmental condition data, historical EVA hot melt adhesive quality data and environmental time series features in the training data into a one-dimensional convolution layer with a convolution kernel size of 3×3 and a number of 12 to obtain the first layer of environmental data association features;

[0028] The first layer of environmental data features is input into a one-dimensional convolution layer with a convolution kernel size of 3×3 and 24 convolution kernels, and an attention mechanism is added to the convolution layer to obtain the second layer of environmental data association features;

[0029] The second layer of environmental data features is input into a one-dimensional convolution layer with a convolution kernel size of 3×3 and 48 convolution kernels, and a nonlinear transformation is performed through a ReLU activation function to obtain the third layer of environmental data association features;

[0030] The first-layer environment data association features, the second-layer environment data association features, and the third-layer environment data association features are integrated to obtain an environment association feature vector;

[0031] The calculation formula of the environment-related feature vector is:

[0032]

[0033] represents the environment-related feature vector output after fusion, F1, F2 and F3 represent the first-layer environment data-related features, the second-layer environment data-related features and the third-layer environment data-related features respectively; W1, W2 and W3 represent the weight matrices of the three convolutional layers respectively, b1, b2 and b3 are the bias items of the three convolutional layers, and σ is the activation function; and ω3 are fusion weight coefficients, and satisfy

[0034] Production parameter associated feature vector acquisition:

[0035] Input the production process parameter data, EVA hot melt adhesive quality data and production parameter time series features in the training data into a one-dimensional convolution layer with a convolution kernel size of 3×3 and a number of 12 to obtain the first layer of production parameter association features;

[0036] The first layer of production parameter association features is input into a one-dimensional convolution layer with a convolution kernel size of 3×3 and 24 convolution kernels, and an attention mechanism is added to the convolution layer to obtain the second layer of production parameter association features;

[0037] The second layer of production parameter association features is input into a one-dimensional convolution layer with a convolution kernel size of 3×3 and 48 convolution kernels, and a nonlinear transformation is performed through a ReLU activation function to obtain the third layer of production parameter association features;

[0038] The first-layer production parameter association features, the second-layer production parameter association features, and the third-layer production parameter association features are merged to obtain a production parameter association feature vector;

[0039] The calculation formula of the production parameter associated feature vector is:

[0040]

[0041] in, Represents the associated feature vector of production parameters output after fusion, F1 prod 、F2 prod and F3 prod Respectively represent the first-layer production parameter association features, the second-layer production parameter association features, and the third-layer production parameter association features; W1 prod 、W2 prod and W3 prod Represents the weight matrices of the three convolutional layers, b1 prod 、b2 prod and b3 prod are the bias terms of the three convolutional layers, σ is the activation function; λ1, λ2 and λ3 are the fusion weight coefficients, and they satisfy λ1+λ2+λ3=1;

[0042] Step 5: Input the environment-related feature vector and the production parameter-related feature vector into the feature fusion layer for fusion, and use the test data for evaluation and optimization;

[0043] Step 6: After evaluation and optimization, the best hyperparameter combination is used as the model hyperparameter of the production stability monitoring model;

[0044]

[0045] in, is the output of the feature fusion layer; W out is the weight matrix of the output layer, represents the environment-related feature vector, represents the production parameter associated feature vector, b out is the bias term of the output layer, φ is the activation function of the output layer; α and β are the weight coefficients of feature fusion, and α+β=1;

[0046] Define hyperparameters and their value ranges; evaluate the model on test data;

[0047] Combining the evaluation index with the hyperparameters to perform multiple fittings until convergence, to obtain a production stability monitoring model with the best hyperparameter combination;

[0048] Define the hyperparameters and their value ranges:

[0049]

[0050] Model Evaluation:

[0051]

[0052] Hyperparameter iterative optimization:

[0053]

[0054] Among them, Θ is defined as the set of hyperparameters, θ i represents the i-th hyperparameter, Θ i Defined as the value range of the i-th hyperparameter; represents the output of the initial training model, D test represents the standardized preparation data test set, M is defined as the evaluation indicator set, m j represents the jth evaluation indicator; Indicates that in the test set D test Calculate the jth evaluation index m j The value of θ * is the hyperparameter combination that finally minimizes the sum of the evaluation indicators on the test set.

[0055] Compared with the prior art, the present invention has the following beneficial effects:

[0056] 1. The present invention obtains input data for the production stability monitoring model by comprehensively collecting real-time production data of EVA hot melt adhesive. After multiple data preprocessing such as timestamp alignment, normalization processing and data integration, the consistency and availability of the data are ensured. By improving the quality of input data and reducing the interference of noise and outliers, the quality prediction of EVA hot melt adhesive is made more efficient and accurate, thereby further improving the accuracy and reliability of production monitoring, and helping to achieve accurate control and optimization of the production process.

[0057] 2. The present invention constructs an EVA hot melt adhesive production stability monitoring model by using the historical production data of EVA hot melt adhesive. The model obtains the time series characteristics of environmental condition data and production process parameter data through feature extraction, and then uses representation learning to generate production parameter-related feature vectors and environment-related feature vectors, and then fuses these features to finally output the prediction results of EVA quality data; among them, the iterative optimization process further improves the prediction effect of the model.

[0058] 3. The present invention collects and analyzes real-time production data of EVA hot melt adhesive in real time, and constructs a production stability monitoring model to predict and evaluate the impact of changes in production conditions on product quality. By timely monitoring and feedback of EVA hot melt adhesive production conditions, production personnel are helped to discover and predict potential problems, ensure the stability of the production process, reduce variability and uncertainty in production, thereby ensuring the consistency of product quality, and further improving production quality and efficiency, and enhancing the market competitiveness of enterprises. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 A flow chart of an implementation method of an EVA hot melt adhesive production monitoring method and system provided in an embodiment of the present invention;

[0060] Figure 2 A method and system structure diagram of EVA hot melt adhesive production monitoring provided by an embodiment of the present invention;

[0061] Figure 3 A production stability monitoring model training flow chart provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0062] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0063] The present invention provides an EVA hot melt adhesive production monitoring method, which is applied to an EVA hot melt adhesive production monitoring system. The specific method and system flow chart are shown in Figure 1 and Figure 2 .

[0064] Embodiment 1

[0065] This example collects and analyzes data for the EVA hot melt adhesive preparation and processing workshop of Factory A;

[0066] Reference Figure 1 S10 in the embodiment is applied to an EVA hot melt adhesive production monitoring method for collecting real-time environmental condition data and production process parameter data of an EVA hot melt adhesive production site;

[0067] The real-time environmental condition data collection method includes:

[0068] Ambient temperature: Temperature sensors are deployed at key locations in the production workshop to reflect the temperature of the entire workshop;

[0069] Relative humidity of air: Humidity sensors are deployed near temperature sensors to ensure coordinated collection of workshop ambient temperature and humidity;

[0070] Equipment vibration intensity: deployed on extruders and cooling equipment to ensure that the vibration of the equipment during operation can be monitored;

[0071] Workshop light intensity: Light sensors are deployed near the main light sources in the production workshop to ensure that they can accurately reflect the lighting conditions in the workshop;

[0072] In this embodiment, the real-time environmental condition data and production process parameter data of the EVA hot melt adhesive production site of Factory A are collected every 5 minutes;

[0073] An example of environmental condition data collection is shown in Table 1:

[0074] Table 1 Example of environmental data collection

[0075]

[0076] The real-time production process parameter data collection method includes:

[0077] Melt temperature: High temperature sensor, deployed in the heating zone of the extruder, ensures that the temperature in the melt state can be monitored;

[0078] Extrusion pressure: The pressure sensor is deployed at the discharge port of the extruder to ensure that the pressure during extrusion can be monitored;

[0079] Extrusion speed: The speed sensor is deployed at the discharge port of the extruder to ensure that the extrusion speed can be monitored;

[0080] Cooling temperature: Temperature sensors are deployed in the cooling water tank to ensure that the temperature of the cooling medium can be monitored;

[0081] An example of production process parameter data collection is shown in Table 2:

[0082] Table 2 Example of production process parameter data collection

[0083]

[0084] In this embodiment, the real-time environmental condition data and production process parameter data collected at the EVA hot melt adhesive production site comprehensively cover multiple influencing factors in the EVA hot melt adhesive production process. These data provide a rich and reliable data basis for subsequent data processing and model training. Through in-depth analysis and processing of these data, more accurate prediction models can be trained, indirectly improving the production efficiency and product quality of EVA hot melt adhesive.

[0085] Reference Figure 1S20 in the method is applied to an EVA hot melt adhesive production monitoring method, including outlier and missing value processing, timestamp alignment, normalization processing and data integration, which is used to pre-process the EVA hot melt adhesive production data and obtain the input data of the production stability monitoring model;

[0086] In this embodiment, the production data processing unit performs comprehensive processing on the production data to ensure the consistency and availability of the production data. The normalization process includes normalizing the environmental condition data, the production process parameter data and the EVA hot melt adhesive quality data. By normalizing the environmental condition data such as ambient temperature, relative humidity, vibration intensity and light intensity, as well as the production process parameter data such as melting temperature, extrusion pressure, extrusion speed and cooling temperature, the consistency and comparability of the data are ensured. This process makes the data more accurate and effective in subsequent model training and analysis by comprehensively considering the weight and range of each parameter, and ultimately improves the performance of the prediction model and the stability of the production process.

[0087] Furthermore, the normalization process includes normalization of environmental condition data, normalization of production process parameter data and normalization of EVA hot melt adhesive quality data;

[0088] The normalized calculation formula for the environmental condition data is:

[0089]

[0090] Where E' represents the comprehensive normalized environmental condition data; T, H, V, and L represent the ambient temperature, relative humidity, vibration intensity, and light intensity, respectively; T min &T max , H min &H max 、V min &V max , L min &L max Respectively express the minimum and maximum values ​​of the parameters; ω T ,ω H ,ω V ,ω L are the weight coefficients of ambient temperature, relative air humidity, vibration intensity and light intensity respectively; Δ is a comprehensive function; and the correlation coefficient is set according to professionals in this field.

[0091] The calculation formula for data normalization processing of the production process parameter data is:

[0092]

[0093] Among them, F' represents the comprehensive normalized production process parameter data; M, P, S, and C represent the melting temperature, extrusion pressure, extrusion speed, and cooling temperature, respectively; M min &M max , P min &P max , S min &S max , C min &C max Respectively express the minimum and maximum values ​​of the parameters; μ T , μ H , μ V , μ L are weight coefficients of melting temperature, extrusion pressure, extrusion speed and cooling temperature respectively; Δ is a comprehensive function; and the correlation coefficient is set according to professionals in this field.

[0094] Reference Figure 1 S30 in the embodiment is applied to an EVA hot melt adhesive production monitoring method, which is used to obtain the prediction results of EVA hot melt adhesive quality data through the feature extraction layer, representation learning layer, feature fusion layer and output layer in the production stability monitoring model, and issue a monitoring reminder according to the prediction results; the training process of the production stability prediction model is as follows Figure 3 shown.

[0095] The training process of the production stability monitoring model includes:

[0096] Step 1: Collect historical production data, including historical environmental condition data, historical production process parameter data and historical EVA hot melt adhesive quality data;

[0097] Step 2: preprocessing the historical production data, and then dividing the training data according to the ratio to obtain training data and test data; the preprocessing operation includes normalization, timestamp alignment and data integration;

[0098] Step 3: Extracting environmental time series features and production parameter time series features; the extraction of environmental time series features and production parameter time series features includes:

[0099] Applying a sliding window to the historical environmental condition data and the historical EVA hot melt adhesive quality data in the training data to generate an environmental condition window sequence;

[0100] Input each environmental condition window sequence into the LSTM network, calculate the hidden state and output of the LSTM, and obtain the environmental time series characteristics;

[0101] Applying a sliding window to the historical production parameter data and the historical EVA hot melt adhesive quality data in the training data to generate a production parameter window sequence;

[0102] Input each production parameter window sequence into the LSTM network, calculate the hidden state and output of the LSTM, and obtain the sequence characteristics between production parameters;

[0103] For input data Apply sliding window to generate window sequence

[0104]

[0105] Each window Input to the LSTM network and calculate the hidden state and output of LSTM:

[0106]

[0107] Among them, the LSTM function is further expanded as:

[0108]

[0109] in, represents the input matrix, represents the number of time steps, and N represents the number of features; Indicates the size of the sliding window, and s is the step size of the sliding window; represents the i-th window sequence; represents the hidden state of the i-th window at time step t, W h and U h is the weight matrix of LSTM, b h is the bias term of the LSTM function, f is the activation function; H lstm is the time feature sequence of the final output.

[0110] By applying a sliding window to the training data to generate a window sequence, and inputting each window into the LSTM network to calculate the hidden state and output, we obtain time series features, which can effectively capture the temporal relationship in the data, provide support for the subsequent relationship learning among environmental condition data, production process parameter data and EVA hot melt adhesive quality data, and improve the effect of relationship learning.

[0111] Step 4: Acquisition of environmental condition correlation features and production parameter correlation features; inputting the historical environmental condition data and the historical EVA hot melt adhesive quality data into the convolution layer in combination with the environmental time series features to obtain an environmental correlation feature vector; inputting the historical production process parameter data and the historical EVA hot melt adhesive quality data into the convolution layer in combination with the production parameter time series features to obtain a production parameter correlation feature vector;

[0112] The process of obtaining the environment-related feature vector and the production parameter-related feature vector is as follows:

[0113] Environment-related feature vector acquisition:

[0114] Input the historical environmental condition data, historical EVA hot melt adhesive quality data and environmental time series features in the training data into a one-dimensional convolution layer with a convolution kernel size of 3×3 and a number of 12 to obtain the first layer of environmental data association features;

[0115] The first layer of environmental data features is input into a one-dimensional convolution layer with a convolution kernel size of 3×3 and 24 convolution kernels, and an attention mechanism is added to the convolution layer to obtain the second layer of environmental data association features;

[0116] The second layer of environmental data features is input into a one-dimensional convolution layer with a convolution kernel size of 3×3 and 48 convolution kernels, and a nonlinear transformation is performed through a ReLU activation function to obtain the third layer of environmental data association features;

[0117] The first-layer environment data association features, the second-layer environment data association features, and the third-layer environment data association features are integrated to obtain an environment association feature vector;

[0118] The calculation formula of the environment-related feature vector is:

[0119]

[0120] represents the environment-related feature vector output after fusion, F1, F2 and F3 represent the first-layer environment data-related features, the second-layer environment data-related features and the third-layer environment data-related features respectively; W1, W2 and W3 represent the weight matrices of the three convolutional layers respectively, b1, b2 and b3 are the bias items of the three convolutional layers, and σ is the activation function; and ω3 are fusion weight coefficients, and satisfy

[0121] In this embodiment, based on the time series characteristics of the environmental condition data in the training data, through multi-layer convolution and attention mechanism weighted calculation, the correlation between the environmental condition data and the EVA hot melt adhesive quality data is captured, and the impact of changes in environmental conditions on changes in the quality of EVA hot melt adhesive can be predicted.

[0122] Production parameter associated feature vector acquisition:

[0123] Input the production process parameter data, EVA hot melt adhesive quality data and production parameter time series features in the training data into a one-dimensional convolution layer with a convolution kernel size of 3×3 and a number of 12 to obtain the first layer of production parameter association features;

[0124] The first layer of production parameter association features is input into a one-dimensional convolution layer with a convolution kernel size of 3×3 and 24 convolution kernels, and an attention mechanism is added to the convolution layer to obtain the second layer of production parameter association features;

[0125] The second layer of production parameter association feature features are input into a one-dimensional convolution layer with a convolution kernel size of 3×3 and 48 convolution kernels, and nonlinearly transformed through the ReLU activation function to obtain the third layer of production parameter association features;

[0126] The first-layer production parameter association features, the second-layer production parameter association features, and the third-layer production parameter association features are merged to obtain a production parameter association feature vector;

[0127] The calculation formula of the production parameter associated feature vector is:

[0128]

[0129] in, Represents the associated feature vector of production parameters output after fusion, F1 prod 、F2 prod and F3 prod Respectively represent the first-layer production parameter association features, the second-layer production parameter association features, and the third-layer production parameter association features; W1 prod 、W2 prod and W3 prod Represents the weight matrices of the three convolutional layers, b1 prod 、b2 prod and b3 prod are the bias terms of the three convolutional layers, σ is the activation function; λ1, λ2 and λ3 are the fusion weight coefficients, and they satisfy λ1+λ2+λ3=1;

[0130] In this embodiment, based on the time series characteristics of the production parameter data in the training data, through multi-layer convolution and attention mechanism weighted calculation, the correlation between the production parameter data and the EVA hot melt adhesive quality data is captured, and the impact of production parameter changes on EVA hot melt adhesive quality changes can be predicted.

[0131] Step 5: Input the environment-related feature vector and the production parameter-related feature vector into the feature fusion layer for fusion, and use the test data for evaluation and optimization;

[0132] Step 6: After evaluation and optimization, the best hyperparameter combination is used as the model hyperparameter of the production stability monitoring model;

[0133]

[0134] in, is the output of the feature fusion layer; W out is the weight matrix of the output layer, represents the environment-related feature vector, represents the production parameter associated feature vector, b out is the bias term of the output layer, φ is the activation function of the output layer; α and β are the weight coefficients of feature fusion, and α+β=1;

[0135] Define hyperparameters and their value ranges; evaluate the model on test data;

[0136] Combining the evaluation index with the hyperparameters to perform multiple fittings until convergence, to obtain a production stability monitoring model with the best hyperparameter combination;

[0137] Define the hyperparameters and their value ranges:

[0138]

[0139] Model Evaluation:

[0140]

[0141] Hyperparameter iterative optimization:

[0142]

[0143] Among them, Θ is defined as the set of hyperparameters, θ i represents the i-th hyperparameter, Θ i Defined as the value range of the i-th hyperparameter; represents the output of the initial training model, D test represents the standardized preparation data test set, M is defined as the evaluation indicator set, m j represents the jth evaluation indicator; Indicates that in the test set D test Calculate the jth evaluation index m j The value of θ * is the hyperparameter combination that finally minimizes the sum of the evaluation indicators on the test set.

[0144] In this embodiment, hyperparameters and their value ranges are defined, and a model evaluation is performed on test data. The evaluation indicators and hyperparameters are combined for multiple fittings until convergence, and the optimal hyperparameter combination is obtained as the model hyperparameters of the production stability monitoring model. This iterative optimization process improves the prediction accuracy of the quality of EVA hot melt adhesive.

[0145] According to the fitting results of the production stability monitoring model, the comprehensive stability range of the EVA hot melt adhesive quality data is limited; the EVA hot melt adhesive quality data includes bonding strength, flexibility and appearance quality scores;

[0146] Through the production stability monitoring model, the quality data of EVA hot melt adhesive in the current real-time production data is predicted and monitored, and based on the prediction results, it is determined whether the current real-time production conditions are stable;

[0147] The prediction monitoring process includes:

[0148]

[0149] in, It represents the prediction result of real-time collected data on the production stability prediction model. represents the production stability prediction model after iterative optimization; g represents the preprocessing of the real-time environmental condition data and the production process parameter data; E real-time ,F real-time The current environmental condition data and production environment data collected in real time. A function representing the limited integrated stability range, based on the model fitting results; S feedback Defined as the feedback signal, S feedback ∈{0,1}.

[0150] Through the fitting results of the production stability monitoring model, the comprehensive stability range of the EVA hot melt adhesive quality data is limited, and the current real-time production data is predicted to obtain the stability evaluation of the EVA hot melt adhesive quality data. It can accurately determine whether the current production conditions are stable and issue monitoring reminders, thereby improving the stability of the production process and the consistency of product quality.

[0151] In order to verify the effectiveness of the stability monitoring model proposed in this embodiment, a comparative experiment was carried out in this embodiment;

[0152] The experimental steps include: re-collecting the real-time environmental condition data and production process parameter data of the EVA hot melt adhesive production site, and simultaneously collecting the corresponding EVA production quality data, including:

[0153] Bonding strength: tensile testing machine, used to test the bonding strength of EVA hot melt adhesive;

[0154] Flexibility: Bending tester, used to test the flexibility of EVA hot melt adhesive;

[0155] Appearance quality scoring: Computer vision software captures the appearance of EVA hot melt adhesive, analyzes the image and scores it;

[0156] The comparison objects are: (1) the quality data of EVA hot melt adhesive produced without stability monitoring model monitoring, and (2) the quality data of EVA hot melt adhesive produced after the EVA hot melt adhesive was monitored and adjusted by the stability monitoring model;

[0157] The comparative experimental results are shown in Table 3.

[0158] Table 3 Comparative experimental results

[0159]

[0160] Referring to Table 3, the means and standard deviations of the three quality data in the two groups of data are calculated, as shown in Table 4.

[0161] Table 4 Comparison of experimental results mean and variance

[0162]

[0163] Referring to Table 4, the quality data of EVA hot melt adhesive produced after the EVA hot melt adhesive was monitored and adjusted by the stability monitoring model showed higher stability and consistency in the three aspects of bonding strength, flexibility and appearance quality score. This shows that the quality of EVA hot melt adhesive produced after the EVA hot melt adhesive was monitored and adjusted by the stability monitoring model is more stable, the product consistency is better, and the overall quality is better than the quality of EVA hot melt adhesive produced without the stability monitoring model monitoring.

[0164] The production stability monitoring model uses multi-layer learning to achieve real-time monitoring and prediction of the quality stability in the production process of EVA hot melt adhesive. By comprehensively considering the interrelationships and influences of various features, the final output module generates the prediction results of EVA quality data. In this way, the system can monitor and predict the quality problems that may occur in the production process of EVA hot melt adhesive in real time. Through real-time evaluation and prediction of changes in production conditions, the model can promptly discover and predict potential problems, ensure the stability of the production process, and reduce variability and uncertainty in production. In this way, not only the stability and efficiency of production are improved, but also the consistency of the quality of EVA hot melt adhesive products is ensured, thereby improving the market competitiveness and user satisfaction of the products.

[0165] Embodiment 2

[0166] This example collects and analyzes data for the EVA hot melt adhesive preparation and processing workshop of Factory B;

[0167] The collection method of the real-time environmental condition data and the production process parameter data includes:

[0168] Ambient temperature: Temperature sensors are deployed at key locations in the production workshop to reflect the temperature of the entire workshop;

[0169] Relative humidity of air: Humidity sensors are deployed near temperature sensors to ensure coordinated collection of workshop ambient temperature and humidity;

[0170] Equipment vibration intensity: deployed on extruders and cooling equipment to ensure that the vibration of the equipment during operation can be monitored;

[0171] Workshop light intensity: Light sensors are deployed near the main light sources in the production workshop to ensure that they can accurately reflect the lighting conditions in the workshop;

[0172] Melt temperature: High temperature sensor, deployed in the heating zone of the extruder, ensures that the temperature in the melt state can be monitored;

[0173] Extrusion pressure: The pressure sensor is deployed at the discharge port of the extruder to ensure that the pressure during extrusion can be monitored;

[0174] Extrusion speed: The speed sensor is deployed at the discharge port of the extruder to ensure that the extrusion speed can be monitored;

[0175] Cooling temperature: Temperature sensors are deployed in the cooling water tank to ensure that the temperature of the cooling medium can be monitored;

[0176] Reference Figure 2 , Figure 2 It is applied to an EVA hot melt adhesive production monitoring system, which is used to collect real-time environmental condition data and production process parameter data of the EVA hot melt adhesive production site; in this embodiment, the real-time environmental condition data and production process parameter data of the EVA hot melt adhesive preparation production site of Factory B are collected every 5 minutes, and an example of environmental condition data collection is shown in Table 5:

[0177] Table 5 Environmental Condition Data Collection Example

[0178]

[0179] An example of production process parameter data collection is shown in Table 6:

[0180] Table 6 Example of production process parameter data collection

[0181]

[0182] Reference Figure 1 S20 in the embodiment is applied to an EVA hot melt adhesive production monitoring system, including outlier and missing value processing, timestamp alignment, normalization processing and data integration, and is used to pre-process the EVA hot melt adhesive production data and obtain the input data of the production stability monitoring model;

[0183] Furthermore, the normalization process includes normalization of environmental condition data, normalization of production process parameter data and normalization of EVA hot melt adhesive quality data;

[0184] The normalized calculation formula for the environmental condition data is:

[0185]

[0186] Where E' represents the comprehensive normalized environmental condition data; T, H, V, and L represent the ambient temperature, relative humidity, vibration intensity, and light intensity, respectively; T min &T max , H min &H max 、V min &V max , L min &L max Respectively express the minimum and maximum values ​​of the parameters; ω T ,ω H ,ω V ,ω L are the weight coefficients of ambient temperature, relative air humidity, vibration intensity and light intensity respectively; Δ is a comprehensive function; and the correlation coefficient is set according to professionals in this field.

[0187] The calculation formula for data normalization processing of the production process parameter data is:

[0188]

[0189] Among them, F' represents the comprehensive normalized production process parameter data; M, P, S, and C represent the melting temperature, extrusion pressure, extrusion speed, and cooling temperature, respectively; M min &M max , P min &P max , S min &S max , C min &C max Respectively express the minimum and maximum values ​​of the parameters; μ T , μ H , μ V , μ L are weight coefficients of melting temperature, extrusion pressure, extrusion speed and cooling temperature respectively; Δ is a comprehensive function; and the correlation coefficient is set according to professionals in this field.

[0190] Reference Figure 1S30 in the embodiment is applied to an EVA hot melt adhesive production monitoring method, which is used to obtain the prediction results of EVA hot melt adhesive quality data through the feature extraction layer, representation learning layer, feature fusion layer and output layer in the production stability monitoring model, and issue a monitoring reminder according to the prediction results; the training process of the production stability prediction model is as follows Figure 3 shown.

[0191] The training process of the production stability monitoring model includes:

[0192] Step 1: Collect historical production data, including historical environmental condition data, historical production process parameter data and historical EVA hot melt adhesive quality data;

[0193] Step 2: preprocessing the historical production data, and then dividing the training data according to the ratio to obtain training data and test data; the preprocessing operation includes normalization, timestamp alignment and data integration;

[0194] Step 3: Extracting environmental time series features and production parameter time series features; the extraction of environmental time series features and production parameter time series features includes:

[0195] Applying a sliding window to the historical environmental condition data and the historical EVA hot melt adhesive quality data in the training data to generate an environmental condition window sequence;

[0196] Input each environmental condition window sequence into the LSTM network, calculate the hidden state and output of the LSTM, and obtain the environmental time series characteristics;

[0197] Applying a sliding window to the historical production parameter data and the historical EVA hot melt adhesive quality data in the training data to generate a production parameter window sequence;

[0198] Input each production parameter window sequence into the LSTM network, calculate the hidden state and output of the LSTM, and obtain the sequence characteristics between production parameters;

[0199] For input data Apply sliding window to generate window sequence

[0200]

[0201] Each window Input to the LSTM network and calculate the hidden state and output of LSTM:

[0202]

[0203] Among them, the LSTM function is further expanded as:

[0204]

[0205] in, represents the input matrix, represents the number of time steps, and N represents the number of features; Indicates the size of the sliding window, and s is the step size of the sliding window; represents the i-th window sequence; represents the hidden state of the i-th window at time step t, W h and U h is the weight matrix of LSTM, b h is the bias term of the LSTM function, f is the activation function; H lstm is the time feature sequence of the final output.

[0206] Step 4: Acquisition of environmental condition correlation features and production parameter correlation features; inputting the historical environmental condition data and the historical EVA hot melt adhesive quality data into the convolution layer in combination with the environmental time series features to obtain an environmental correlation feature vector; inputting the historical production process parameter data and the historical EVA hot melt adhesive quality data into the convolution layer in combination with the production parameter time series features to obtain a production parameter correlation feature vector;

[0207] The process of obtaining the environment-related feature vector and the production parameter-related feature vector is as follows:

[0208] Environment-related feature vector acquisition:

[0209] Input the historical environmental condition data, historical EVA hot melt adhesive quality data and environmental time series features in the training data into a one-dimensional convolution layer with a convolution kernel size of 3×3 and a number of 12 to obtain the first layer of environmental data association features;

[0210] The first layer of environmental data features is input into a one-dimensional convolution layer with a convolution kernel size of 3×3 and 24 convolution kernels, and an attention mechanism is added to the convolution layer to obtain the second layer of environmental data association features;

[0211] The second layer of environmental data features is input into a one-dimensional convolution layer with a convolution kernel size of 3×3 and 48 convolution kernels, and a nonlinear transformation is performed through a ReLU activation function to obtain the third layer of environmental data association features;

[0212] The first-layer environment data association features, the second-layer environment data association features, and the third-layer environment data association features are integrated to obtain an environment association feature vector;

[0213] The calculation formula of the environment-related feature vector is:

[0214]

[0215] in, represents the environment-related feature vector output after fusion, F1, F2 and F3 represent the first-layer environment data-related features, the second-layer environment data-related features and the third-layer environment data-related features respectively; W1, W2 and W3 represent the weight matrices of the three convolutional layers respectively, b1, b2 and b3 are the bias items of the three convolutional layers, and σ is the activation function; and ω3 are fusion weight coefficients, and satisfy

[0216] Input the production process parameter data, EVA hot melt adhesive quality data and production parameter time series features in the training data into a one-dimensional convolution layer with a convolution kernel size of 3×3 and a number of 12 to obtain the first layer of production parameter association features;

[0217] The first layer of production parameter association features is input into a one-dimensional convolution layer with a convolution kernel size of 3×3 and 24 convolution kernels, and an attention mechanism is added to the convolution layer to obtain the second layer of production parameter association features;

[0218] The second layer of production parameter association features is input into a one-dimensional convolution layer with a convolution kernel size of 3×3 and 48 convolution kernels, and a nonlinear transformation is performed through a ReLU activation function to obtain the third layer of production parameter association features;

[0219] The first-layer production parameter association features, the second-layer production parameter association features, and the third-layer production parameter association features are merged to obtain a production parameter association feature vector;

[0220] The calculation formula of the production parameter associated feature vector is:

[0221]

[0222] in, Represents the associated feature vector of production parameters output after fusion, F1 prod 、F2 prod and F3 prod Respectively represent the first-layer production parameter association features, the second-layer production parameter association features, and the third-layer production parameter association features; W1 prod 、W2 prod and W3 prod Represents the weight matrices of the three convolutional layers, b1 prod 、b2 prod and b3 prod are the bias terms of the three convolutional layers, σ is the activation function; λ1, λ2 and λ3 are the fusion weight coefficients, and they satisfy λ1+λ2+λ3=1;

[0223] Step 5: Input the environment-related feature vector and the production parameter-related feature vector into the feature fusion layer for fusion, and use the test data for evaluation and optimization;

[0224] Step 6: The best hyperparameter combination after evaluation iteration until the result converges is used as the model hyperparameter of the production stability monitoring model; the process formula is:

[0225]

[0226] in, is the output of the feature fusion layer; W out is the weight matrix of the output layer, represents the environment-related feature vector, represents the production parameter associated feature vector, b out is the bias term of the output layer, φ is the activation function of the output layer; α and β are the weight coefficients of feature fusion, and α+β=1;

[0227] Define hyperparameters and their value ranges; evaluate the model on test data;

[0228] Combining the evaluation index with the hyperparameters to perform multiple fittings until convergence, to obtain a production stability monitoring model with the best hyperparameter combination;

[0229] Define the hyperparameters and their value ranges:

[0230]

[0231] Model Evaluation:

[0232]

[0233] Hyperparameter iterative optimization:

[0234]

[0235] Among them, Θ is defined as the set of hyperparameters, θ i represents the i-th hyperparameter, Θ i Defined as the value range of the i-th hyperparameter; represents the output of the initial training model, D test represents the standardized preparation data test set, M is defined as the evaluation indicator set, m j represents the jth evaluation indicator; Indicates that in the test set D test Calculate the jth evaluation index m j The value of

[0236] According to the fitting results of the production stability monitoring model, the comprehensive stability range of the EVA hot melt adhesive quality data is limited; the EVA hot melt adhesive quality data includes bonding strength, flexibility and appearance quality scores;

[0237] Through the production stability monitoring model, the quality data of EVA hot melt adhesive in the current real-time production data is predicted and monitored, and based on the prediction results, it is determined whether the current real-time production conditions are stable;

[0238] The prediction monitoring process includes:

[0239]

[0240] in, It represents the prediction result of real-time collected data on the production stability prediction model. represents the production stability prediction model after iterative optimization; g represents the preprocessing of the real-time environmental condition data and the production process parameter data; E real-time ,F real-time The current environmental condition data and production environment data collected in real time. A function representing the limited integrated stability range, based on the model fitting results; S feedback Defined as the feedback signal, S feedback ∈{0,1}.

[0241] In this embodiment, the input data of the production stability monitoring model is obtained by comprehensively collecting the real-time production data of EVA hot melt adhesive. After multiple data preprocessing such as timestamp alignment, normalization processing and data integration, the consistency and availability of the data are ensured. By improving the quality of input data and reducing the interference of noise and outliers, the quality prediction of EVA hot melt adhesive is made more efficient and accurate, thereby further improving the accuracy and reliability of production monitoring and helping to achieve precise control and optimization of the production process.

[0242] In this embodiment, by using the historical production data of EVA hot melt adhesive, a production stability monitoring model of EVA hot melt adhesive is constructed. The model obtains the time series features of environmental condition data and production process parameter data through feature extraction, and then uses representation learning to generate production parameter-related feature vectors and environment-related feature vectors, and then fuses these features to finally output the prediction results of EVA quality data; among them, the iterative optimization process further improves the prediction effect of the model.

[0243] In this embodiment, by collecting and analyzing the real-time production data of EVA hot melt adhesive in real time and building a production stability monitoring model, the impact of changes in production conditions on product quality is predicted and evaluated. By timely monitoring and feedback of EVA hot melt adhesive production conditions, production personnel are helped to discover and predict potential problems, ensure the stability of the production process, reduce variability and uncertainty in production, thereby ensuring the consistency of product quality, and then improving production quality and efficiency, and enhancing the market competitiveness of enterprises.

[0244] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for monitoring the production of EVA hot melt adhesive, characterized in that: include: Collect real-time environmental condition data and production process parameter data at the EVA hot melt adhesive production site; The environmental condition data include ambient temperature, relative air humidity, equipment vibration intensity and workshop light intensity; the production process parameter data include melting temperature, extrusion pressure, extrusion speed and cooling temperature; Pre-processing environmental condition data and production process parameter data; Real-time monitoring of production stability: input the environmental condition data and production process parameter data into the production stability monitoring model, and obtain the EVA hot melt adhesive quality data prediction results through the feature extraction layer, representation learning layer, feature fusion layer and output layer in sequence. The prediction results are used to determine whether the current production conditions are stable and issue stability monitoring reminders; the EVA hot melt adhesive quality data includes bonding strength, flexibility and appearance quality scores; The processing formula for real-time monitoring of production stability is: in, It represents the prediction result of real-time collected data on the production stability prediction model. represents a production stability monitoring model; g represents preprocessing of the real-time environmental condition data and production process parameter data; E real-time ,F real-time The current environmental condition data and production environment data collected in real time. A function representing the limited integrated stability range, based on the model fitting results; S feedback Defined as the feedback signal, S feedback ∈{0,1}.

2. The EVA hot melt adhesive production monitoring method according to claim 1, characterized in that: The pre-processing comprises: Processing missing values, outliers and noise data of the EVA hot melt adhesive production data, and converting all numerical data into unified floating point data to obtain standardized data; The standardized data are scaled to the same range through normalization processing to obtain data of the same dimension; For the standardized data, aligning timestamps; The normalized and timestamp-aligned data are integrated to obtain the EVA hot melt adhesive production data.

3. The EVA hot melt adhesive production monitoring method according to claim 1, characterized in that: The training process of the production stability monitoring model includes: Step 1: Collect historical production data, including historical environmental condition data, historical production process parameter data and historical EVA hot melt adhesive quality data; Step 2: preprocessing the historical production data, and then dividing it according to the ratio to obtain training data and test data; the preprocessing includes normalization, timestamp alignment and data integration; Step 3: Input the training data into the feature extraction layer to extract environmental time series features and production parameter time series features; Step 4: Input the environmental condition data and EVA hot melt adhesive quality data in the training data into the representation learning layer in combination with the environmental time series features to obtain the environmental associated feature vector; input the production process parameter data and historical EVA hot melt adhesive quality data in the training data into the representation learning layer in combination with the production parameter time series features to obtain the production parameter associated feature vector; Step 5: Input the environment-related feature vector and the production parameter-related feature vector into the feature fusion layer for fusion, and use the test data for evaluation and optimization; Step 6: The best hyperparameter combination after evaluation iteration until the results converge is used as the model hyperparameters of the production stability monitoring model.

4. The EVA hot melt adhesive production monitoring method according to claim 3, characterized in that: The extraction of the environmental time series features and the production parameter time series features includes: Apply sliding windows to the historical environmental condition data and historical EVA hot melt adhesive quality data in the training data to generate an environmental condition window sequence; Input each environmental condition window sequence into the LSTM network, calculate the hidden state and output of the LSTM, and obtain the environmental time series characteristics; Applying a sliding window to the historical production parameter data and the historical EVA hot melt adhesive quality data in the training data to generate a production parameter window sequence; Input each production parameter window sequence into the LSTM network, calculate the hidden state and output of the LSTM, and obtain the production parameter time series characteristics.

5. The EVA hot melt adhesive production monitoring method according to claim 3, characterized in that: The process of obtaining the environment-related feature vector is as follows: The historical environmental condition data, historical EVA hot melt adhesive quality data and environmental time series features in the training data are input into a one-dimensional convolution layer with a convolution kernel size of 3×3 and a number of 12 to obtain the first layer of environmental data association features; The first layer of environmental data features is input into a one-dimensional convolution layer with a convolution kernel size of 3×3 and 24 convolution kernels, and an attention mechanism is added to the convolution layer to obtain the second layer of environmental data association features; The second layer of environmental data features is input into a one-dimensional convolution layer with a convolution kernel size of 3×3 and 48 convolution kernels, and a nonlinear transformation is performed through a ReLU activation function to obtain the third layer of environmental data association features; The first-layer environment data association features, the second-layer environment data association features, and the third-layer environment data association features are integrated to obtain an environment association feature vector; The calculation formula of the environment-related feature vector is: represents the environment-related feature vector output after fusion, F1, F2 and F3 represent the first-layer environment data-related features, the second-layer environment data-related features and the third-layer environment data-related features respectively; W1, W2 and W3 represent the weight matrices of the three convolutional layers, b1, b2 and b3 are the bias terms of the three convolutional layers, and σ is the activation function; and ω3 are fusion weight coefficients, and satisfy 6. The EVA hot melt adhesive production monitoring method according to claim 3, characterized in that: The process of obtaining the production parameter associated feature vector is as follows: Input the production process parameter data, EVA hot melt adhesive quality data and production parameter time series features in the training data into a one-dimensional convolution layer with a convolution kernel size of 3×3 and a number of 12 to obtain the first layer of production parameter association features; The first layer of production parameter association features is input into a one-dimensional convolution layer with a convolution kernel size of 3×3 and 24 convolution kernels, and an attention mechanism is added to the convolution layer to obtain the second layer of production parameter association features; The second layer of production parameter association features is input into a one-dimensional convolution layer with a convolution kernel size of 3×3 and 48 convolution kernels, and a nonlinear transformation is performed through a ReLU activation function to obtain the third layer of production parameter association features; The first-layer production parameter association features, the second-layer production parameter association features, and the third-layer production parameter association features are merged to obtain a production parameter association feature vector; The calculation formula of the production parameter associated feature vector is: in, Represents the associated feature vector of production parameters output after fusion, F1 prod 、F2 prod and F3 prod Respectively represent the first-layer production parameter association features, the second-layer production parameter association features, and the third-layer production parameter association features; W1 prod 、W2 prod and W3 prod Represents the weight matrices of the three convolutional layers, b1 prod 、b2 prod and b3 prod are the bias terms of the three convolutional layers, σ is the activation function; λ1, λ2 and λ3 are the fusion weight coefficients, and satisfy λ1+λ2+λ3=1.

7. The EVA hot melt adhesive production monitoring method according to claim 3, characterized in that: The evaluation and optimization process includes: in, is the output of the feature fusion layer; W out is the weight matrix of the output layer, represents the environment-related feature vector, represents the production parameter associated feature vector, b out is the bias term of the output layer, φ is the activation function of the output layer; α and β are the weight coefficients of feature fusion, and α+β=1; Define hyperparameters and their value ranges; evaluate the model on test data; Combine the evaluation indicators with the hyperparameters and perform multiple fittings until convergence to obtain the production stability monitoring model with the best hyperparameter combination; Define the hyperparameters and their value ranges: Model Evaluation: Hyperparameter iterative optimization: Among them, Θ is defined as the set of hyperparameters, θ i represents the i-th hyperparameter, Θ i Defined as the value range of the i-th hyperparameter; represents the output of the feature fusion layer, D test represents the standardized preparation data test set, M is defined as the evaluation indicator set, m j represents the jth evaluation indicator; Indicates that in the test set D test Calculate the jth evaluation index m j The value of θ * is the hyperparameter combination that finally minimizes the sum of the evaluation indicators on the test set.

8. An EVA hot melt adhesive production monitoring system, characterized in that: include: Data acquisition unit: used to collect real-time environmental condition data and production process parameter data at the EVA hot melt adhesive production site; The environmental condition data include ambient temperature, relative air humidity, equipment vibration intensity and workshop light intensity; the production process parameter data include melting temperature, extrusion pressure, extrusion speed and cooling temperature; Data processing unit: used to pre-process the EVA hot melt adhesive production data and obtain the input data of the production stability monitoring model; The data preprocessing includes timestamp alignment, normalization processing and data integration; The timestamp alignment is used to synchronize the times of different data sources; The normalization process scales data of different dimensions to the same range; Production stability monitoring unit: used to output the corresponding EVA hot melt adhesive quality data prediction results; input the environmental condition data and production process parameter data into the production stability monitoring model, and obtain the EVA hot melt adhesive quality data prediction results through the feature extraction layer, representation learning layer, feature fusion layer and output layer in sequence. The prediction results are used to determine whether the current production conditions are stable and issue stability monitoring reminders.

9. The EVA hot melt adhesive production monitoring system according to claim 8, characterized in that: The process of production stability monitoring is: Input real-time production data, obtain the EVA hot melt adhesive quality prediction results through the production stability monitoring model, and then judge whether the current production conditions are stable based on the prediction results, and then issue monitoring reminders; The processing formula for real-time monitoring of production stability is: in, It represents the prediction result of real-time collected data on the production stability prediction model. represents a production stability monitoring model with an optimal hyperparameter combination; g represents preprocessing of the real-time environmental condition data and production process parameter data; E real-time ,F real-time The current environmental condition data and production environment data collected in real time. A function representing the limited integrated stability range, based on the model fitting results; S feedback Defined as the feedback signal, S feedback ∈{0,1}.

10. The EVA hot melt adhesive production monitoring system according to claim 8, characterized in that: The training process of the production stability monitoring model includes: Step 1: Collect historical production data, including historical environmental condition data, historical production process parameter data and historical EVA hot melt adhesive quality data; Step 2: preprocessing the historical production data, and then dividing it according to the ratio to obtain training data and test data; the preprocessing operation includes normalization, timestamp alignment and data integration; Step 3: Input the training data into the feature extraction layer to extract environmental time series features and production parameter time series features; Step 4: Input the environmental condition data and EVA hot melt adhesive quality data in the training data into the representation learning layer in combination with the environmental time series features to obtain the environmental associated feature vector; input the production process parameter data and historical EVA hot melt adhesive quality data in the training data into the representation learning layer in combination with the production parameter time series features to obtain the production parameter associated feature vector; Step 5: Input the environment-related feature vector and the production parameter-related feature vector into the feature fusion layer for fusion, and use the test data for evaluation and optimization; Step 6: The best hyperparameter combination after evaluation iteration until the results converge is used as the model hyperparameters of the production stability monitoring model.

Images