Production control method and system of high-frequency data cable

Abstract

The invention belongs to the technical field of production control, and provides a production control method and system for a high-frequency data cable. The method comprises the following steps: detecting a group of thickness distribution data of a foaming belt at a belt placing end of the foaming belt by adopting a laser thickness gauge, and calculating to obtain a thickness distribution deviation value; when the thickness distribution deviation value exceeds a deviation threshold value, depth prediction is carried out on the group of thickness distribution data and the foaming belt type by using a conjoint analysis model based on lightweight temporary case data, and joint compensation parameters are obtained and comprise associated line pressing pressure and line pressing temperature; and a line pressing device integrated with a heating element is controlled to mechanically press the foaming belt, so that the density uniformity of the foaming layer is improved to be higher than a specified value. Through the temperature and pressure synergistic effect, the foaming belt is slightly softened and uniformly fills the gaps of the core wires, the density uniformity of the foaming layer can be further improved, and therefore the characteristic impedance consistency and the anti-interference performance of the cable are remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of production control, and in particular, relates to a production control method and system for high-frequency data cables. Background Art

[0002] In the production of high-frequency data cables, achieving uniform coverage of the core wire foam layer is a key technical challenge in improving signal transmission stability. Prior art CN110570996A provides a process for laminating assembled cables using a dual crimping wheel structure. However, relying solely on mechanical pressure cannot completely eliminate the gap between the foam tape and the core wire, nor can it completely eliminate wrapping stress. This is especially true when the foam tape thickness is uneven or the core wire tension fluctuates, which can easily lead to regional density differences in the foam layer after lamination.

[0003] It can be seen that the above-mentioned pressing process only relies on mechanical pressure to compact the foam tape, which cannot effectively eliminate the internal stress generated by factors such as tension fluctuations and core wire displacement during the wrapping process, resulting in looseness or excessive compression in local areas of the foam tape, and ultimately causing insufficient density uniformity of the foam layer (usually 90%-95%), affecting the characteristic impedance consistency and anti-interference performance of the cable.

[0004] Therefore, how to improve the above process to further enhance the uniformity of the foaming layer on the surface of the two core wires, for example, to more than 98%, is a key technical issue that currently needs further improvement. Summary of the Invention

[0005] To solve the above technical problems, the present invention provides a production control method, system, electronic equipment, storage medium and computer program product for a high-frequency data cable.

[0006] The present invention discloses a production control method for high-frequency data cables, the method comprising the following steps:

[0007] At the unwinding end of the foaming tape, a laser thickness gauge is used to detect a set of thickness distribution data of the foaming tape, and the thickness distribution deviation value is calculated;

[0008] When the thickness distribution deviation value exceeds the deviation threshold, a joint analysis model is used to perform an in-depth prediction of the thickness distribution data and foam tape type based on lightweight temporary case data, and derive joint compensation parameters, including associated crimping pressure and crimping temperature; wherein the temporary case data is derived from actual mechanical crimping data of the same type of high-frequency data cable within a specified recent period;

[0009] The pressing device integrated with the heating element is controlled to mechanically press the foaming tape so as to increase the density uniformity of the foaming layer to above a specified value.

[0010] The present invention also discloses a production control system for high-frequency data cables, characterized in that the system includes a processing module and a storage module, and the processing module executes a computer program in the storage module to implement the following steps:

[0011] At the unwinding end of the foaming tape, a laser thickness gauge is used to detect a set of thickness distribution data of the foaming tape, and the thickness distribution deviation value is calculated;

[0012] When the thickness distribution deviation value exceeds the deviation threshold, a joint analysis model is used to perform an in-depth prediction of the thickness distribution data and foam tape type based on lightweight temporary case data, and derive joint compensation parameters, including associated crimping pressure and crimping temperature; wherein the temporary case data is derived from actual mechanical crimping data of the same type of high-frequency data cable within a specified recent period;

[0013] The pressing device integrated with the heating element is controlled to mechanically press the foaming tape so as to increase the density uniformity of the foaming layer to above a specified value.

[0014] The present invention also discloses an electronic device, comprising: a memory for storing computer-readable instructions; a processor for reading and executing the computer-readable instructions, wherein the computer-readable instructions are executed by the processor to execute any of the aforementioned methods.

[0015] The present invention also discloses a storage medium that non-temporarily stores computer-readable instructions, and when the computer-readable instructions are executed by a computer, any of the aforementioned methods is implemented.

[0016] The present invention also discloses a computer program product, which is used to implement any of the above methods.

[0017] This invention uses a joint analysis model, guided by recent data from similar temporary cases, to predict depth based on thickness distribution data and foam tape type to derive the optimal crimping pressure and temperature. This method then drives a crimping device with an integrated heating element, using the combined effects of temperature and pressure to gently soften the foam tape and evenly fill the gaps between the core wires. This solution can further improve the density uniformity of the foam layer, significantly enhancing the characteristic impedance consistency and anti-interference performance of the cable. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 This is a flow chart of a production control method for a high-frequency data cable disclosed in an embodiment of the present invention;

[0020] Figure 2 It is a structural diagram of the joint analysis model disclosed in the embodiment of the present invention;

[0021] Figure 3 It is a structural diagram of a production control system for high-frequency data cables disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0022] The following specific embodiments illustrate the implementation of this application. Those familiar with the art can easily understand the other advantages and functions of this application from the content disclosed in this specification. Obviously, the described embodiments are part of the embodiments of this application, but not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0023] In addition, the technical features involved in the different embodiments of the present application described below can be combined with each other as long as they do not conflict with each other.

[0024] In high-frequency data cable production, uneven foam tape thickness is a key factor contributing to insufficient density uniformity within the foam layer. Existing crimping processes rely solely on fixed mechanical pressure (such as the dual-crimping wheel structure described in patent CN110570996A), which cannot dynamically adjust the lamination parameters to account for variations in foam tape thickness. This can easily lead to: Insufficient compaction in thick areas, resulting in gaps between cells and uneven dielectric constants; and excessive compression in thin areas, leading to material degradation and increased signal loss.

[0025] In response to the above-mentioned shortcomings of the existing wire crimping process, the present invention designs and adopts a closed-loop control scheme of "real-time detection-intelligent modeling-temperature and pressure coordinated compensation", which can further improve the uniformity of the foaming layer of high-frequency data cables to ensure its characteristic impedance consistency and anti-interference performance.

[0026] like Figure 1 As shown, an embodiment of the present invention discloses a production control method for high-frequency data cables, the method comprising the following steps:

[0027] Step 100: At the unwinding end of the foam tape, a laser thickness gauge is used to detect a set of thickness distribution data of the foam tape, and a thickness distribution deviation value is calculated.

[0028] A laser thickness gauge is installed at the unwinding end of the foam tape. This instrument collects a set of thickness data (e.g., 10-20 measurement points) along the width of the tape in real time and calculates the thickness distribution deviation (e.g., standard deviation σ). This allows for early identification of defects in the tape. For example, if a thickness deviation exceeding ±3% in a specific area is detected, the subsequent compensation process is triggered.

[0029] Step 200: When the thickness distribution deviation value exceeds the deviation threshold, a joint analysis model is used to perform an in-depth prediction of the set of thickness distribution data and the foam tape type based on lightweight temporary case data to obtain joint compensation parameters, including associated pressing pressure and pressing temperature; wherein the temporary case data is based on actual mechanical pressing data of the same type of high-frequency data cable within a specified recent period.

[0030] Step 300 : Controlling a pressing device integrated with a heating element to mechanically press the foaming tape so as to increase the density uniformity of the foaming layer to above a specified value.

[0031] Compared to existing crimping processes that rely solely on fixed mechanical pressure, this invention integrates a heating element into the crimping device (e.g., a patented, improved heated crimping wheel) and uses a joint analysis model to predict joint compensation parameters, namely, the associated crimping pressure and crimping temperature. This allows the crimping device to press the foam tape at an optimal set of crimping pressures and temperatures. Through the synergistic effect of temperature and pressure, the density standard deviation of the foam layer is significantly reduced compared to traditional processes, further improving uniformity to above a specified value (e.g., 98%), approaching the level of an ideal uniform medium. Furthermore, by eliminating wrapping stress, the bond strength between the foam layer and the core wire is significantly enhanced, making it less susceptible to cracking or delamination over long-term use.

[0032] The pressing temperature should be between 60-80°C. The heating will slightly soften the foam tape, reducing the elastic modulus by 30%-50% (for example, from 15MPa to 7.5-10.5MPa). The material enters a viscoelastic state, making it easier for pressure to drive the material to evenly fill the gaps between the core wires. The pressing pressure should be between tens and hundreds of Newtons (N) to compact the foam tape without damaging the core wire and the foam tape.

[0033] The present invention also pre-trains the constructed joint analysis model to learn the cross-type mapping relationship between thickness deviation, foam tape type, and lamination parameters. This allows the joint compensation parameters to be derived based on the aforementioned set of thickness distribution data and the depth prediction of the foam tape type. The foam tape types include polytetrafluoroethylene (PTFE), expanded polyethylene (PE) / polypropylene (PP), cross-linked polyethylene (XLPE), and polystyrene (PS).

[0034] Generally speaking, the historical pressing data of a single type of product is obviously insufficient for the training of the joint analysis model, and the joint analysis model is only used to predict the joint compensation parameters of a single type of product, which requires the construction of multiple joint analysis models, which is not economical. Therefore, the present invention collects historical pressing data of multiple types of products in advance, and then uses these historical pressing data to construct training data and pre-train the joint analysis model. At the same time, the present invention also collects the actual pressing data of the same type of products in the recent period (for example, data from the past 24 hours), including thickness deviation values, pressing line pressure, pressing line temperature, and uniformity results, and constructs a lightweight temporary case database based on this. The small amount of temporary case data (compared to the historical pressing data during pre-training) in the temporary case database can be used to guide the deep prediction of the joint analysis model this time, so that the obtained joint compensation parameters are more accurate.

[0035] This invention uses a joint analysis model, guided by recent data from similar temporary cases, to predict depth based on thickness distribution data and foam tape type to derive the optimal crimping pressure and temperature. This method then drives a crimping device with an integrated heating element, using the combined effects of temperature and pressure to gently soften the foam tape and evenly fill the gaps between the core wires. This solution can further improve the density uniformity of the foam layer, significantly enhancing the characteristic impedance consistency and anti-interference performance of the cable.

[0036] Alternatively, as Figure 2 As shown, the laser thickness gauge is used to detect a set of thickness distribution data of the foaming tape, and the thickness distribution deviation value is calculated, including:

[0037] Determine the type of foaming tape, retrieve all historical thickness distribution data corresponding to the foaming tape type, calculate the divergence of each of the historical thickness distribution data, and determine the number of detection points based on the divergence;

[0038] A laser thickness gauge is used to detect a set of thickness distribution data of the foaming tape corresponding to the number of detection points, and the thickness distribution deviation value is calculated.

[0039] There are many types of foam tape used in high-frequency data cables, and different foam tapes exhibit varying thickness deviation patterns. To address this, the present invention first retrieves historical thickness distribution data corresponding to the type of foam tape being used and calculates the divergence of this historical thickness distribution data, such as the Kullback-Leibler divergence or Jensen-Shannon divergence. A higher divergence indicates a more complex thickness distribution for that type of foam tape (i.e., greater randomness in thickness fluctuations); a lower divergence indicates a less complex thickness distribution for that type of foam tape (i.e., less randomness in thickness fluctuations).

[0040] Therefore, when the divergence is higher, the present invention controls the laser thickness gauge to measure more different points on the foam tape to obtain more diverse thickness distribution data. This allows the thickness distribution deviation value obtained through analysis to be closer to the actual foam tape condition. Conversely, the laser thickness gauge is controlled to measure fewer different points on the foam tape, significantly reducing the difficulty of sampling thickness distribution data and the time required to calculate the thickness distribution deviation value.

[0041] It is understandable that a movable track can be configured for the laser thickness gauge so that it can change the detection position. Furthermore, more laser thickness gauges can be configured on the movable track, which can improve the detection speed of more historical thickness distribution data.

[0042] Optionally, the joint analysis model includes a spatiotemporal graph neural network module, a meta-learning module, and a fusion prediction module; the joint analysis model is used to perform in-depth prediction of the set of thickness distribution data and foaming tape type based on lightweight temporary case data to obtain joint compensation parameters, including:

[0043] The spatiotemporal graph neural network module constructs the set of thickness distribution data into spatiotemporal graph structure data according to the spatial position of the detection points and the detection time sequence, performs graph convolution and time convolution operations on the spatiotemporal graph structure data, and extracts the spatial distribution characteristics and temporal evolution characteristics of the thickness deviation;

[0044] The meta-learning module encodes the foam tape type as material property features including the material molecular structure, thermal expansion coefficient, and elastic modulus. Based on lightweight temporary case data, the model-agnostic meta-learning (MAML) algorithm is used to quickly fine-tune the model parameters and establish a mapping relationship between "thickness feature-material property-pressing parameter" across material types.

[0045] The fusion prediction module performs tensor splicing on the extracted spatial distribution features, temporal evolution features and material property features, generates prior information on the pressing parameters based on the mapping relationship, and outputs the joint compensation parameters including the pressing line pressure and the pressing line temperature through the fully connected layer.

[0046] Traditional deep learning requires a large amount of labeled data, but trial production data for new materials in high-frequency data cable production is generally scarce, making it difficult for joint analysis models to obtain sufficient actual pressing data for a single type of foam tape. To address this, the present invention utilizes a joint analysis model that combines a spatiotemporal graph neural network with meta-learning. The model is first trained using historical data from multiple materials (e.g., PTFE, PE, XLPE, PS) for universal mapping, learning common cross-material patterns (e.g., "the greater the thickness deviation, the greater the proportion of pressure required to increase the pressing force"). For the specific foam tape type being used, only a lightweight set of temporary case data (e.g., 10 or 20 groups) is required. Through rapid gradient updates, the general model parameters can be adapted to the specific parameters of the foam tape type being used, thus overcoming data scarcity limitations.

[0047] The specific processing process is as follows:

[0048] 1. Spatiotemporal Graph Neural Network Module:

[0049] The set of thickness distribution data includes thickness values of detection points along the width direction of the foaming tape (eg, 10-20 points) and corresponding detection timestamps (eg, time series data collected at a production line speed of 20 m / min).

[0050] This thickness distribution data is constructed into a spatiotemporal graph structured by the spatial location and time sequence of the test points. Each test point corresponds to a graph node, with a feature vector consisting of [thickness value, spatial coordinate (x), timestamp (t)], where the spatial coordinate (x) represents the position of the test point along the width of the foam strip (e.g., the left edge is 0, the right edge is 1), and the timestamp (t) is converted to the normalized time relative to the start of the production line.

[0051] Edge connection rules: (1) Spatial edge: adjacent detection points (such as Point and (2) Time edge: a directed edge is established between the nodes at the same spatial position (for example, the same point at time t and time t-1), and the weight is the inverse of the time interval (reflecting the strength of temporal dependence).

[0052] Next, a double convolution operation based on a graph convolution layer and a one-dimensional convolution layer is used for feature extraction:

[0053] Perform operations through graph convolutional layers (such as GCN):

[0054]

[0055] in, For the In the graph convolution layer, nodes The eigenvector of For nodes The number of neighbor nodes, For nodes The number of neighbor nodes; It is The weight matrix of the graph convolution layer; Indicates in In the graph convolution layer, nodes The eigenvector of It is The bias vector of the graph convolution layer.

[0056] Arrange the graph nodes in chronological order and extract temporal features through a one-dimensional convolutional layer (TCN):

[0057]

[0058] in, Represents the feature vector obtained after calculation at time step t; It is the weight matrix of the convolution kernel, which determines the weighted summation of the input features in the one-dimensional convolution operation. The convolution kernel extracts the characteristic patterns in the input data by sliding and calculating on the input feature sequence. The weight value is continuously learned and adjusted during the model training process. Indicates that from the time step arrive The feature vector sequence of is the bias vector of the convolutional layer.

[0059] The extracted time series features include the periodicity of thickness fluctuations (e.g., a sudden increase in thickness due to equipment vibration at the 10th minute of each hour) or the trend (e.g., a continuous increase in thickness due to temperature drift).

[0060] Generate spatiotemporal feature tensors ,in is the number of detection points, is the number of time steps, is the number of feature channels (e.g., 64 dimensions), including spatial distribution features (e.g., “abnormal thickness area on the left edge”) and time series evolution features (e.g., “the thickness fluctuation period is 15 minutes”).

[0061] 2. Meta-learning module:

[0062] The material properties corresponding to the input foam tape type (such as PTFE, PE, XLPE, PS) include: molecular structure characteristics: represented by one-hot encoding (for example, PTFE is [1,0,0,0], PE is [0,1,0,0]); continuous physical properties: thermal expansion coefficient (10⁻ 5 / ℃)、elastic modulus (MPa), glass transition temperature (℃).

[0063] Map the above material properties into high-dimensional vectors through a fully connected network (e.g. 128 dimensions) so that the model can distinguish the different effects of "PTFE's low dielectric constant and PE's high elastic modulus" on the pressing parameters. The mapping process is as follows:

[0064]

[0065] in, is a weight matrix that performs a linear transformation on the input data in a fully connected network.

[0066] Next, fast fine-tuning is performed based on the Model-Agnostic Meta-Learning (MAML) algorithm:

[0067] The joint analysis model has learned the common rules across materials in the pre-training stage (for example, "for every 1% increase in thickness deviation, the pressing pressure needs to increase by 2-5N"), and the parameters Contains general mapping capabilities.

[0068] When entering temporary case data (e.g., 10 sets of PE material thickness deviations and lamination parameters), perform the following steps:

[0069] (1) Inner loop update: calculate the loss using temporary data through 1-2 gradient descents , update the parameters to ;

[0070] (2) Outer loop optimization: ensuring updated parameters Maintains generalization capabilities to other material tasks (achieved through multi-task loss weighting).

[0071] Establish a specialized mapping function for the current material type , the spatiotemporal characteristics and material properties are associated with the prior range of pressing parameters (e.g., the pressing line temperature of PE material should be between 60-80 °C).

[0072] 3. Fusion prediction module:

[0073] For the input features, the spatiotemporal features output by the spatiotemporal graph neural network , material property features output by the meta-learning module , perform tensor splicing and merging , extracting nonlinear relationships through a two-layer fully connected network:

[0074]

[0075]

[0076] in, The weight matrix of is integrated into the cross-material pressing rule learned in the pre-training stage (e.g. “for every 5°C increase in temperature, the pressure can be reduced by 10N”).

[0077] It should be noted that the above-mentioned spatiotemporal characteristics , which is obtained by global average pooling the spatial distribution features extracted by spatial graph convolution operation and the temporal evolution features extracted by temporal convolution operation (such as the periodic fluctuation law of thickness over time, the time position of abnormal mutation points, etc.) output by the spatiotemporal graph neural network.

[0078] Finally, the fully connected layer directly predicts the line pressure and wire pressing temperature , the output range is constrained by material physics:

[0079] ,

[0080] For example, PE material (to avoid failure to compact), (Avoid material degradation).

[0081] Through the comparative learning mechanism, the lamination parameters of similar materials are ensured to be closer in the embedding space, thus improving the consistency of the same type of parameters. For example:

[0082]

[0083] in, The loss value is used to measure the degree of difference between the feature vector output by the model and the expected similarity. It guides the model optimization during the model training process, making the feature representations of the same type of materials closer and the feature representations of different types of materials farther apart. Represents the distance metric function, which is used to measure the distance between two feature vectors; 、 They represent the eigenvectors corresponding to two different polyethylene (PE) materials. These eigenvectors are obtained after model processing and contain a comprehensive representation of material-related properties, thickness distribution, and other information. The eigenvector corresponding to polytetrafluoroethylene (PTFE) material is also a comprehensive representation after model processing, which is used to compare the distance with the eigenvector of PE material to reflect the differences between different material categories; is an interval parameter, which is an artificially set positive constant used to control the minimum distance that should be maintained between the characteristic vectors of different types of materials.

[0084] The above-mentioned layered processing mode of the present invention not only ensures the ability to capture data details, but also realizes cross-material knowledge reuse through meta-learning, and ultimately generates high-precision pressing parameters under small sample conditions, which can meet the real-time and reliability requirements of high-frequency cable production.

[0085] Optionally, before quickly fine-tuning the model parameters using a model-independent meta-learning algorithm based on lightweight temporary case data to establish a mapping relationship between "thickness characteristics-material properties-pressing parameters" across material types, the following steps are also included:

[0086] Based on the data labels, the ratio of positive data to negative data in the lightweight temporary case data is calculated. The inner loop learning rate and update steps of the model-independent meta-learning algorithm are adjusted according to this ratio. Specifically:

[0087] When the ratio is greater than a first threshold, the inner loop learning rate corresponding to the inverse data is increased, and the number of steps for inner loop updates based on the inverse data is decreased; when the ratio is less than a second threshold, the inner loop learning rate corresponding to the inverse data is decreased, and the number of steps for inner loop updates based on the inverse data is increased; the first threshold is greater than the second threshold.

[0088] When fine-tuning the model parameters of a joint analysis model based on lightweight temporary case data using a model-independent meta-learning algorithm, the ratio of positive data to negative data in the lightweight temporary case data may vary significantly. If the ratio of positive and negative data is unbalanced, the model learning effect will be affected. For example, when there is too little negative data, its gradient contribution in model training is easily overwhelmed by the positive data, making it difficult for the joint analysis model to effectively learn the ability to handle abnormal situations (situations represented by negative data); when there is too much negative data, it may cause the model to over-focus on abnormal situations and ignore the general rules contained in the positive data. To address this situation, the present invention adjusts the inner loop learning rate and update steps of the model-independent meta-learning algorithm according to the ratio of positive and negative data to optimize the model learning process.

[0089] First, based on data labels, we distinguish positive and negative data from the lightweight temporary case data and then calculate the ratio between the two. Data labels are used to identify whether the data is normal (positive data) or abnormal (negative data).

[0090] Then, when the ratio of positive to negative data is greater than the first threshold, it means that the proportion of negative data is too small. At this time, the inner loop learning rate corresponding to the negative data is increased to make the joint analysis model pay more attention to the feature information of the negative data. At the same time, the number of steps of the inner loop update based on the negative data is reduced to prevent the joint analysis model from overfitting due to excessive updates due to too little negative data.

[0091] When the ratio of positive to negative data falls below the second threshold, it indicates that the negative data accounts for a large proportion. In this case, the inner loop learning rate corresponding to the negative data is lowered to prevent the joint analysis model from being overly influenced by the negative data. At the same time, the number of steps in the inner loop update based on the negative data is increased to give the joint analysis model more opportunities to learn patterns in the negative data.

[0092] The quantitative adjustment method of the inner loop learning rate and update steps is illustrated as follows:

[0093] 1. Inner loop learning rate:

[0094] set up It is the ratio of positive data to negative data, that is, the ratio of the number of positive data divided by the number of negative data. is the initial inner loop learning rate.

[0095] when ( When the higher ratio threshold is set, that is, the first threshold, for example, 10), it indicates that the proportion of negative data is small, and the learning rate of negative data is high. , .in is the adjustment coefficient, for example , used to control the learning rate adjustment amplitude.

[0096] when ( When the lower ratio threshold is set, that is, the second threshold, for example, 2), it indicates that the proportion of negative data is large, and the learning rate of negative data is high. , the learning rate for positive data .

[0097] when When , the learning rate of negative data and positive data can be calculated by linear interpolation. That is:

[0098] ; .

[0099] 2. Number of steps in the inner loop update

[0100] set up Update the number of steps for the initial inner loop according to the ratio of positive data to negative data Make the following adjustments:

[0101] when When the number of update steps based on the inverse data is ; Update steps based on positive data .in is the adjustment factor (e.g. 0.2).

[0102] when When the number of update steps based on the inverse data is ; The number of update steps based on positive data. The number of update steps based on positive data .

[0103] when When , the number of update steps is also determined by linear interpolation, that is:

[0104] ; .

[0105] By properly adjusting the learning rate and the number of update steps, the joint analysis model can balance the learning of normal and abnormal situations under different positive and negative data ratios, avoiding over-learning or under-learning of a certain type of situation due to data imbalance, thereby improving the generalization ability of the joint analysis model in various scenarios such as different material types, and more accurately establishing the "thickness characteristics-material properties-pressing parameters" mapping relationship.

[0106] Optionally, before using the joint analysis model to perform in-depth prediction on the set of thickness distribution data and foaming tape type based on the lightweight temporary case data, the method further includes:

[0107] Calculate the real-time divergence of the set of thickness distribution data, and adjust the temporal convolution kernel size of the spatiotemporal graph neural network module and the inner loop learning rate of the meta-learning module based on the real-time divergence.

[0108] In order to simultaneously ensure the accuracy and speed of the joint analysis model's depth prediction of the joint compensation parameters, the present invention further designs an approach to adjust the temporal convolution kernel size of the spatiotemporal graph neural network module and the inner loop learning rate of the meta-learning module based on the real-time divergence of the input thickness distribution data. The details are as follows:

[0109] In scenarios with high real-time divergence (e.g., equipment failure), sudden thickness changes appear as high-frequency signals. Small convolution kernels (e.g., k=3) focus on the mutation point through a small receptive field, avoiding feature smoothing caused by large kernels and significantly reducing the detection latency of sudden thickness anomalies. In scenarios with low real-time divergence (e.g., stable production), thickness changes slowly. Large convolution kernels (e.g., k=7) can integrate longer time series information, improving the ability to model the long-term dependencies between material properties and pressing parameters, thereby helping to reduce parameter prediction errors.

[0110] At the same time, in high-divergence scenarios, the data distribution changes dramatically, requiring rapid adaptation to new features. Increasing the learning rate (for example, from 1e-3 to 5e-3) allows meta-learning to complete parameter updates in 2-3 steps, compared to 5-8 steps with a traditional fixed learning rate. In low-divergence scenarios, the data distribution is stable, and excessively high inner-loop learning rates can easily lead to overfitting. Lowering the inner-loop learning rate (for example, to 1e-4) preserves the prior knowledge advantage of meta-learning and significantly reduces the number of samples required for parameter convergence in pilot production of new materials.

[0111] (1) Setting a high divergence threshold , low divergence threshold .in, 、 are the mean and standard deviation of the historical divergence, , .

[0112] The dynamic adjustment formula of the temporal convolution kernel size of the spatiotemporal graph neural network module is as follows:

[0113]

[0114] in, (capturing high-frequency fluctuations), (extracting long-term trends), , .

[0115] In addition, the temporal convolution layer can use a 1D convolution kernel, and when the kernel size is dynamically adjusted, the output dimension is kept consistent by zero padding.

[0116] (2) The learning rate mapping function is:

[0117]

[0118] in: (base learning rate), (increase learning rate when divergence is high), (Reduce the learning rate when the divergence is low).

[0119] Table 1 below shows the comparative data of production tests of high-frequency data cables using PE materials:

[0120] Anomaly detection delay Parameter prediction error Number of samples required for convergence Fixed parameters 150ms 4.2% 20 groups Only adjust the convolution kernel size 90ms 3.8% 18 groups Only adjust the learning rate 120ms 3.5% 15 groups Joint dynamic adjustment 80ms 3.1% 10 groups

[0121] It can be seen that the solution adopted by the present invention to jointly and dynamically adjust the temporal convolution kernel size of the spatiotemporal graph neural network module and the inner loop learning rate of the meta-learning module can significantly reduce the anomaly detection delay and parameter prediction error, and only requires a smaller number of samples to achieve convergence, thereby ensuring the accuracy of the predicted joint compensation parameters while improving the prediction speed.

[0122] It should be noted that the inner loop learning rate derived from optimizing the ratio of positive data to negative data in lightweight temporary case data is for the negative data, while the inner loop learning rate derived from optimizing the real-time divergence of the measured thickness distribution data is for the entire meta-learning module, that is, for both positive and negative data. In actual operation, the average of the two inner loop learning rates for the negative data can be used as the actual inner loop learning rate.

[0123] like Figure 3 As shown, an embodiment of the present invention further discloses a production control system 10 for high-frequency data cables, characterized in that the system includes a processing module 100 and a storage module 200, and the processing module 100 executes a computer program in the storage module 200 to implement the following steps:

[0124] At the unwinding end of the foaming tape, a laser thickness gauge is used to detect a set of thickness distribution data of the foaming tape, and the thickness distribution deviation value is calculated;

[0125] When the thickness distribution deviation value exceeds the deviation threshold, a joint analysis model is used to perform an in-depth prediction of the thickness distribution data and foam tape type based on lightweight temporary case data, and derive joint compensation parameters, including associated crimping pressure and crimping temperature; wherein the temporary case data is derived from actual mechanical crimping data of the same type of high-frequency data cable within a specified recent period;

[0126] The pressing device integrated with the heating element is controlled to mechanically press the foaming tape so as to increase the density uniformity of the foaming layer to above a specified value.

[0127] An embodiment of the present invention further discloses an electronic device, comprising: a memory for storing computer-readable instructions; a processor for reading and executing the computer-readable instructions, wherein the computer-readable instructions are executed by the processor to execute the method described in any one of the aforementioned embodiments.

[0128] An embodiment of the present invention further discloses a storage medium that non-temporarily stores computer-readable instructions, and when the computer-readable instructions are executed by a computer, the method described in any one of the aforementioned embodiments is implemented.

[0129] An embodiment of the present invention further discloses a computer program product, which is used to implement the method according to any one of the aforementioned embodiments.

[0130] Although the embodiments or examples of the present invention have been described with reference to the accompanying drawings, it should be understood that the above-mentioned methods, systems, electronic devices, storage media and computer program products are merely exemplary embodiments or examples, and the scope of the present invention is not limited by these embodiments or examples, but is only limited by the claims after authorization and their equivalents. Various elements in the embodiments or examples can be omitted or replaced by their equivalents. In addition, the steps can be performed in an order different from that described in the present invention. Further, the various elements in the embodiments or examples can be combined in various ways. It is important that as technology evolves, many of the elements described here can be replaced by equivalent elements that appear after the present invention.

Claims

1. A production control method for high-frequency data cables, characterized in that: The method comprises the following steps: At the unwinding end of the foaming tape, a laser thickness gauge is used to detect a set of thickness distribution data of the foaming tape, and the thickness distribution deviation value is calculated; When the thickness distribution deviation value exceeds the deviation threshold, a joint analysis model is used to perform an in-depth prediction of the thickness distribution data and foam tape type based on lightweight temporary case data, and derive joint compensation parameters, including associated crimping pressure and crimping temperature; wherein the temporary case data is derived from actual mechanical crimping data of the same type of high-frequency data cable within a specified recent period; The pressing device integrated with the heating element is controlled to mechanically press the foaming tape so as to increase the density uniformity of the foaming layer to above a specified value.

2. The production control method of a high-frequency data cable according to claim 1, characterized in that: A set of thickness distribution data of the foam tape is tested using a laser thickness gauge, and the thickness distribution deviation value is calculated, including: Determine the type of foaming tape, retrieve all historical thickness distribution data corresponding to the foaming tape type, calculate the divergence of each of the historical thickness distribution data, and determine the number of detection points based on the divergence; A laser thickness gauge is used to detect a set of thickness distribution data of the foaming tape corresponding to the number of detection points, and the thickness distribution deviation value is calculated.

3. The production control method of a high-frequency data cable according to claim 1, characterized in that: The joint analysis model includes a spatiotemporal graph neural network module, a meta-learning module, and a fusion prediction module. The joint analysis model is used to perform in-depth prediction of the thickness distribution data and foaming belt type based on lightweight temporary case data to obtain joint compensation parameters, including: The spatiotemporal graph neural network module constructs the set of thickness distribution data into spatiotemporal graph structure data according to the spatial position of the detection points and the detection time sequence, performs graph convolution and time convolution operations on the spatiotemporal graph structure data, and extracts the spatial distribution characteristics and temporal evolution characteristics of the thickness deviation; The meta-learning module encodes the foam tape type as material property features including the material molecular structure, thermal expansion coefficient, and elastic modulus. Based on lightweight temporary case data, a model-independent meta-learning algorithm is used to quickly fine-tune the model parameters and establish a mapping relationship between "thickness feature-material property-pressing parameter" across material types. The fusion prediction module performs tensor splicing on the extracted spatial distribution features, temporal evolution features and material property features, generates prior information on the pressing parameters based on the mapping relationship, and outputs the joint compensation parameters including the pressing line pressure and the pressing line temperature through the fully connected layer.

4. The production control method of a high-frequency data cable according to claim 3, characterized in that: Before quickly fine-tuning model parameters using a model-independent meta-learning algorithm based on lightweight temporary case data and establishing a mapping relationship between "thickness characteristics-material properties-pressing parameters" across material types, the following also applies: Based on the data labels, the ratio of positive data to negative data in the lightweight temporary case data is calculated. The inner loop learning rate and update steps of the model-independent meta-learning algorithm are adjusted according to this ratio. Specifically: When the ratio is greater than a first threshold, the inner loop learning rate corresponding to the inverse data is increased, and the number of steps for inner loop updates based on the inverse data is decreased; when the ratio is less than a second threshold, the inner loop learning rate corresponding to the inverse data is decreased, and the number of steps for inner loop updates based on the inverse data is increased; the first threshold is greater than the second threshold.

5. The production control method of a high-frequency data cable according to claim 3, characterized in that: Before using the joint analysis model to make in-depth predictions on this set of thickness distribution data and foaming tape types based on lightweight temporary case data, it also includes: Calculate the real-time divergence of the set of thickness distribution data, and adjust the temporal convolution kernel size of the spatiotemporal graph neural network module and the inner loop learning rate of the meta-learning module based on the real-time divergence.

6. A production control system for high-frequency data cables, characterized by: The system includes a processing module and a storage module. The processing module executes a computer program in the storage module to implement the following steps: At the unwinding end of the foaming tape, a laser thickness gauge is used to detect a set of thickness distribution data of the foaming tape, and the thickness distribution deviation value is calculated; When the thickness distribution deviation value exceeds the deviation threshold, a joint analysis model is used to perform an in-depth prediction of the thickness distribution data and foam tape type based on lightweight temporary case data, and derive joint compensation parameters, including associated crimping pressure and crimping temperature; wherein the temporary case data is derived from actual mechanical crimping data of the same type of high-frequency data cable within a specified recent period; The pressing device integrated with the heating element is controlled to mechanically press the foaming tape so as to increase the density uniformity of the foaming layer to above a specified value.

7. A production control system for high-frequency data cables according to claim 6, characterized in that: A set of thickness distribution data of the foam tape is tested using a laser thickness gauge, and the thickness distribution deviation value is calculated, including: Determine the type of foaming tape, retrieve all historical thickness distribution data corresponding to the foaming tape type, calculate the divergence of each of the historical thickness distribution data, and determine the number of detection points based on the divergence; A laser thickness gauge is used to detect a set of thickness distribution data of the foaming tape corresponding to the number of detection points, and the thickness distribution deviation value is calculated.

8. An electronic device comprising: a memory for storing computer-readable instructions; A processor is configured to read and execute the computer-readable instructions, wherein the computer-readable instructions are executed by the processor to perform the method according to any one of claims 1 to 5.

9. A storage medium that non-transitorily stores computer-readable instructions, characterized in that: When the computer-readable instructions are executed by a computer, the method according to any one of claims 1 to 5 is implemented.

10. A computer program product, characterized in that: The computer program product is used to implement the method according to any one of claims 1 to 5.

Images