A method for controlling the vulcanization process of a silicone rubber product

Abstract

This invention relates to a control method for the vulcanization process of silicone products in the field of next-generation information technology, comprising: if the parameter adjustment requirement exceeds a preset threshold, optimizing the vulcanization process control model through gradient descent to determine the optimal temperature-time combination and obtain dynamic process instructions; updating the equipment actuator with the dynamic process instructions, obtaining curing uniformity data from real-time feedback of the vulcanization process and determining performance non-uniformity deviations to obtain correction coefficients; predicting the impact of raw material batch fluctuations based on the correction coefficients using a random forest model, determining the stability of the curing state to obtain an optimized parameter set; driving a vulcanization process simulation module to generate a virtual test scenario through the optimized parameter set, obtaining simulation performance indicators and determining actual production adaptability to obtain a final control scheme; and obtaining abnormal signals from real-time changes in production conditions according to the prevention and adjustment rules, determining the stability of the curing process, and obtaining a batch production optimization path.

Description

Technical Field

[0001] This invention relates to the field of next-generation information technology, and in particular to a method for controlling the vulcanization process of silicone products. Background Technology

[0002] Silicone products are crucial components in consumer electronics, automotive electronics, and communication devices. These products, such as waterproof seals and cushioning pads, are widely used in smartphones, wearable devices, and other fields, directly impacting product reliability and lifespan. Especially in precision electronic devices, these silicone components require extremely high dimensional accuracy and performance stability, while simultaneously meeting the demands of large-scale production, making them an indispensable key element in ensuring the quality of end products.

[0003] However, current mainstream production methods in the industry cannot fully meet this demand. Traditional processes often separate multiple production steps, relying on independent equipment, with frequent material transfers between stages, resulting in high time costs. A deeper problem is that this approach lacks the ability to respond promptly to key changes in the production process, especially when facing complex product structures and variable production conditions. It struggles to ensure consistent performance across all components, thus affecting the overall reliability of downstream products. Focusing on specific technical challenges, the most critical challenge in silicone product manufacturing lies in the precise control of the vulcanization process. Vulcanization is a crucial step determining the product's hardness, elasticity, and durability, but the structural characteristics of different products, such as thin-walled sections or thick connecting sections, result in significant differences in their response to temperature and time. This variation makes it difficult to adapt to fixed-parameter control methods, leading to some areas not fully cured while others are over-cured, causing uneven performance. Furthermore, this unevenness can be further amplified by different batches of raw materials or fluctuations in equipment condition, ultimately affecting the product's waterproofing effect and lifespan.

[0004] Therefore, how to dynamically adjust process parameters during vulcanization based on the diversity of product structures and changes in real-time production conditions to ensure that each component reaches its optimal curing state in different parts has become a key issue in improving the quality and production efficiency of precision silicone products. Solving this problem not only concerns the performance of individual products but also directly relates to how to reduce scrap rates and improve consistency in large-scale production, meeting the stringent reliability requirements of the precision electronics industry. Summary of the Invention

[0005] This invention provides a method for controlling the vulcanization process of silicone products, mainly including: Temperature and pressure distribution data during the vulcanization process are collected using a sensor array. Humidity field data, obtained by fusing the temperature and pressure data, is then used to form a synchronous dataset. Real-time change features of product structural parts are extracted from this synchronous dataset. Abnormal fluctuation values ​​within these real-time change features are identified, and data weights are adjusted to obtain an adjusted feature set. A curing state index is determined based on this adjusted feature set. A convolutional neural network is used to extract structural difference features of the product based on the curing state index. The reaction differences between thin-walled sections and thick connecting sections are determined, and parameter adjustment requirements are obtained. If the parameter adjustment requirements exceed a preset threshold, the vulcanization process control model is optimized using gradient descent to determine the optimal temperature-time combination and obtain dynamic process instructions. These dynamic process instructions are then applied. The instruction updates the equipment actuator, obtains curing uniformity data from real-time feedback of the vulcanization process, determines performance non-uniformity deviation, and obtains correction coefficients. Based on the correction coefficients, a random forest model is fused to predict the impact of raw material batch fluctuations, and the stability of the curing state is judged to obtain an optimized parameter set. The optimized parameter set drives the vulcanization process simulation module to generate a virtual test scenario, obtains simulation performance indicators, determines actual production adaptability, and obtains the final control scheme. If the final control scheme meets the preset consistency threshold, a support vector machine classifier is trained using historical data of the vulcanization process to judge the curing anomaly pattern under product structure diversity and obtain prevention and adjustment rules. Based on the prevention and adjustment rules, abnormal signals are obtained from real-time changes in production conditions to determine the stability of the curing process and obtain a batch production optimization path.

[0006] Furthermore, the step of extracting real-time change features of product structural parts from the synchronous dataset, determining abnormal fluctuation values ​​in the real-time change features and adjusting data weights to obtain an adjusted feature set, and determining the curing state index based on the adjusted feature set includes: performing spatiotemporal matching of temperature field data and pressure distribution data for corresponding product structural parts in the synchronous dataset to form humidity field data; extracting temperature change rate, pressure gradient change, and humidity inference values ​​for multiple product structural parts from the synchronous dataset as the real-time change features; determining whether there are abnormal fluctuation values ​​in the real-time change features that exceed a preset fluctuation range; if so, assigning reduced weights to the data at the corresponding time point and spatial location in the synchronous dataset according to the magnitude of the abnormal fluctuation values ​​to obtain the adjusted feature set; and obtaining the curing state index characterizing the current curing degree through weighted fusion calculation based on the adjusted feature set.

[0007] Furthermore, the step of extracting product structural difference features using a convolutional neural network based on the curing state index, determining the reaction difference between thin-walled parts and thick connecting sections, and obtaining parameter adjustment requirements includes: reconstructing the curing state index into a multi-channel feature map according to the spatial distribution of product structural parts; inputting the multi-channel feature map into the convolutional neural network, which is then processed through multiple convolutional layers, activation layers, and pooling layers to obtain a deep feature representation; mapping the reaction difference vectors of different structural regions of the product through fully connected layers or global pooling; and calculating the parameter adjustment requirements based on the curing rate difference and temperature gradient difference between thin-walled parts and thick connecting sections in the reaction difference vector.

[0008] Furthermore, if the parameter adjustment requirement exceeds a preset threshold, the vulcanization process control model is optimized through gradient descent to determine the optimal temperature-time combination and obtain dynamic process instructions. This includes: inputting the parameter adjustment requirement as a component of the loss function into the vulcanization process control model; the vulcanization process control model uses the current temperature curve and pressure curve as state inputs and the next setpoint temperature and pressure values ​​as action outputs; iteratively updating the model parameters through the gradient descent algorithm until the loss converges; outputting multiple temperature-time combinations based on the converged model, and selecting the optimal combination that satisfies the curing uniformity constraint and energy consumption constraint as the dynamic process instructions.

[0009] Furthermore, the step of updating the equipment actuator using the dynamic process command, obtaining curing uniformity data from real-time feedback of the vulcanization process, determining performance non-uniformity deviation, and obtaining a correction coefficient includes: adjusting the execution parameters of the heating device and the pressurizing device in real time according to the dynamic process command; collecting feedback data from temperature and pressure sensors at multiple locations, calculating the statistical variance of the curing degree of each part as the curing uniformity data; fusing the curing uniformity data with the expanded pressure field distribution to obtain the performance non-uniformity deviation characterizing the overall non-uniformity; extracting response hysteresis difference characteristics of different parts for the performance non-uniformity deviation, and determining the correction coefficient based on the intensity of the difference characteristics.

[0010] Furthermore, the step of predicting the impact of raw material batch fluctuations using a random forest model based on the correction coefficients to determine the stability of the solidification state and obtain an optimized parameter set includes: collecting physical property fluctuation data of the current raw material batch and weighting and concatenating it with the correction coefficients to form a prediction input set; inputting the prediction input set into a pre-trained random forest model, which uses multiple decision trees to predict in parallel and integrate the impact values ​​of batch fluctuations on the solidification endpoint; comparing the impact values ​​with historical standard solidification indicators and calculating the deviation direction and magnitude as the basis for judging the stability of the solidification state; and adjusting the temperature benchmark value, pressure benchmark value, and holding time according to the judgment result through parameter mapping relationships to generate the optimized parameter set.

[0011] Furthermore, the step of generating a virtual test scenario by driving the vulcanization process simulation module through the optimized parameter set, obtaining simulation performance indicators and determining the adaptability to actual production, and obtaining the final control scheme includes: extracting temperature threshold sequence, pressure range sequence, and time node sequence from the optimized parameter set as driving signals; inputting the driving signals into a pre-established vulcanization process simulation module, which performs coupled calculations of material flow and heat conduction based on the finite element method to generate the temperature field evolution of the entire virtual vulcanization process; extracting simulation performance indicators such as maximum temperature difference, uniformity of curing degree distribution, and residual stress from the temperature field evolution; comparing the simulation performance indicators with preset production standard thresholds, and fusing historical optimal control schemes according to the deviation magnitude to obtain the final control scheme.

[0012] Furthermore, if the final control scheme meets a preset consistency threshold, a support vector machine classifier is trained using historical data from the vulcanization process to determine curing anomaly patterns under product structure diversity and obtain preventive adjustment rules. This includes: extracting temperature fields, pressure fields, and curing result labels from multiple batches of products with different structures from historical data of the vulcanization process to form a training sample set; inputting the training sample set into the support vector machine classifier for training to obtain a classification model that can distinguish multiple curing anomaly patterns; predicting possible curing anomaly patterns using the classification model for newly input product structure parameters and process parameters and outputting the set of anomaly categories with the highest probability; and generating preventive adjustment rules for material composition and structural characteristics based on the set of anomaly categories and historical deviation data to obtain the preventive adjustment rules.

[0013] Furthermore, the step of obtaining abnormal signals from real-time changes in production conditions based on the preventive adjustment rules, determining the stability of the curing process, and obtaining a batch production optimization path includes: real-time monitoring of changes in production environment temperature, humidity, and raw material moisture content, and extracting a set of abnormal signals that exceed the normal range; calculating a process stability index for the abnormal signal set, including a weighted sum of fluctuation amplitude and duration; matching the stability index with threshold conditions in the preventive adjustment rules to determine the degree of stability deviation of the current curing process; selecting a corresponding subset of preventive adjustment rules based on the degree of stability deviation, and generating a temperature compensation sequence, a pressure compensation sequence, and a time node adjustment sequence for batch production to obtain the batch production optimization path.

[0014] Furthermore, the step of acquiring temperature field data and pressure distribution data during the vulcanization process through a sensor array, and obtaining humidity field data based on the fusion of the temperature field data and the pressure distribution data to form a synchronous dataset includes: synchronously acquiring multi-channel temperature and pressure values ​​through temperature and pressure sensors arranged at multiple locations on the mold; for each acquisition moment, aligning the temperature and pressure values ​​according to spatial location in a grid; using a preset humidity inference model to jointly calculate the temperature and pressure values ​​at each grid location to obtain the humidity field estimate for the corresponding location; and aligning the temperature field data, pressure distribution data, and humidity field data at all times according to the time series to form the synchronous dataset containing multimodal information.

[0015] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses an intelligent dynamic control method for improving the quality and production efficiency of silicone products. It uses a sensor array to collect real-time temperature and pressure distribution data, obtaining curing state indicators for various structural parts of the product. A convolutional neural network is used to extract the reaction differences between thin-walled sections and thick connecting segments, accurately determining parameter adjustment needs. When the deviation exceeds a threshold, a gradient descent optimization control model is used to obtain the optimal dynamic temperature-time combination command. Furthermore, real-time feedback of curing uniformity data is fused with a random forest model to predict the impact of raw material batch fluctuations, generating an optimized parameter set to drive a simulation module for virtual testing, verifying its adaptability to actual production, and ultimately forming a highly consistent control scheme. Simultaneously, a support vector machine classifier is trained based on historical data to identify abnormal curing patterns under diverse product structures, constructing preventative adjustment rules, and achieving closed-loop intelligent control from abnormal signals to optimized paths for batch production. This invention significantly improves the uniformity, stability, and consistency of the vulcanization process of complex composite materials, effectively reducing performance inconsistencies and scrap rates, and improving production efficiency and product quality reliability. Attached Figure Description

[0016] Figure 1This is a flowchart of a method for controlling the vulcanization process of silicone products according to the present invention.

[0017] Figure 2 This is a schematic diagram of a method for controlling the vulcanization process of silicone products according to the present invention.

[0018] Figure 3 This is a schematic diagram of the framework of a method for controlling the vulcanization process of silicone products according to the present invention.

[0019] Figure 4 This is another schematic diagram of the framework of a method for controlling the vulcanization process of silicone products according to the present invention. Detailed Implementation

[0020] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0021] like Figures 1-4 This embodiment of a method for controlling the vulcanization process of silicone products may specifically include: Step S101: Temperature field data and pressure distribution data are collected from the vulcanization process using a sensor array to obtain real-time change characteristics of the corresponding product structural parts and obtain curing state indicators.

[0022] Temperature and pressure distribution data are collected from the vulcanization process using a sensor array. These data are then fused using a thermodynamic model such as the Clausius-Clapeyron equation to calculate humidity data, forming a synchronous dataset. Real-time change features of the product's structural components are extracted from this synchronous dataset, specifically displacement and deformation changes based on a time series. Abnormal fluctuation values, or the standard deviation of the features, are identified within these real-time change features. If the abnormal fluctuation value exceeds a preset threshold of 5, the data weights within the synchronous dataset are adjusted according to the proportion of abnormal values, i.e., weight W_i = 1 / (1 + abnormal proportion), resulting in an adjusted feature set. A curing state index, i.e., the weighted average of the features, is calculated for the adjusted feature set to determine whether the curing state index meets the standards for real-time change features of the product's structural components.

[0023] In one implementation, acquiring temperature field data and pressure distribution data from the vulcanization process using a sensor array first requires the placement of suitable sensor equipment. Specifically, the sensor array may include multiple infrared temperature sensors and pressure sensors distributed on the inner surface of the vulcanization mold to cover the main structural parts of the product.

[0024] For example, during the vulcanization process of rubber tires, sensor arrays are installed in the tire crown, sidewall, and bead areas of the mold to ensure comprehensive capture of temperature and pressure changes. The data acquisition is achieved through real-time sampling, collecting data once per second to obtain continuous temperature field distribution maps and pressure distribution matrices. This arrangement accurately reflects the heat conduction and pressure application during vulcanization, providing fundamental data for subsequent analysis. Furthermore, based on the acquired temperature field and pressure distribution data, the real-time change characteristics of corresponding structural parts of the product are obtained. This step is crucial, requiring data processing to extract features.

[0025] Specifically, image processing techniques, such as the Canny edge detection algorithm, are first applied to the temperature field data to identify contour changes in the product's structural parts. This algorithm extracts edges by smoothing the image using Gaussian filtering, calculating gradient strength and direction, non-maximum suppression, and double threshold detection. For example, during vulcanization, the gradient change in the temperature field can indicate the internal heat flow of the material, while the pressure distribution reflects the material deformation. Real-time feature vectors are obtained by calculating the spatiotemporal change rates of the temperature and pressure gradients. These feature vectors include indicators such as the deformation rate of the part D=ΔL / L0 (D is the deformation rate, ΔL is the length change, and L0 is the initial length) and the temperature uniformity deviation U=σ / T_avg (U is the deviation, σ is the standard deviation, and T_avg is the average temperature).

[0026] In one possible implementation, for complex products such as rubber seals, feature extraction can be combined with finite element simulation to mesh the data, thereby quantifying the variation characteristics of each structural component. This method ensures the accuracy and real-time nature of the features, avoiding the errors of traditional manual monitoring.

[0027] It should be noted that acquiring real-time change features involves a data fusion process. Specifically, temperature field data and pressure distribution data are combined using a weighted fusion algorithm to form a comprehensive change matrix, i.e., C = w_T × T + w_P × P, where C is the comprehensive change matrix, T is the temperature field data, P is the pressure distribution data, w_T and w_P are the weights of temperature and pressure, respectively, and w_T + w_P = 1. For example, the fusion weights are set according to the elastic modulus E of the product material; for high-elasticity rubber, E is greater than 5 MPa, so w_P = 0.7. Through this fusion, the gradient of each element in the comprehensive change matrix is ​​calculated as the extracted feature, which can reflect the dynamic progress of the vulcanization reaction, such as local differences in the degree of material crosslinking. The difficulty in this step lies in handling noisy data, which mainly consists of sensor measurement errors and environmental interference. Through the prediction and update steps of Kalman filtering, state estimation is achieved to reduce the impact of noise and ensure data accuracy.

[0028] Preferably, a Kalman filter is used to smooth the raw data to ensure the reliability of the features. In vulcanization operations, this feature extraction helps to detect uneven heating problems in a timely manner, thereby allowing for adjustments to process parameters.

[0029] For example, after acquiring real-time change characteristics, a solidification state index is obtained. This index is obtained by quantitatively evaluating the characteristics. Specifically, the solidification state index can be defined as a multi-dimensional vector, including solidification completion degree, uniformity, and defect risk value.

[0030] For example, the curing completion rate is determined by calculating the Euclidean distance between the current temperature (T) and pressure (P) features and a preset threshold model. If the distance is less than the threshold (e.g., temperature greater than 150 degrees Celsius and pressure greater than 2 MPa), it is considered complete. The threshold model is trained using linear regression based on historical vulcanization data, with real-time temperature and pressure features as input and the probability of curing completion as output. In one embodiment, for the vulcanization of automotive rubber hoses, the index calculation uses a support vector machine classifier within this threshold model framework. The temperature and pressure features are input into the model, and the probability distribution vector of the curing state is output. This approach enables independent evaluation of different parts of the product, ensuring the accuracy of the index. In another embodiment, the aforementioned sensor array can be extended to integrate multimodal data such as temperature and pressure.

[0031] Specifically, in addition to temperature and pressure, vibration data can be collected to enhance the comprehensiveness of feature extraction. For example, micro-vibration signals during vulcanization can be captured by vibration sensors, and their amplitude can be combined with the rate of change of temperature and pressure to calculate composite change characteristics. This integration improves the robustness of curing state indicators reflecting the degree of rubber vulcanization, enabling it to cope with deviations caused by environmental disturbances such as mold aging in actual rubber product production. Furthermore, real-time change characteristics of product structural parts are extracted through time series analysis, which is essentially the result of time series analysis of the data.

[0032] Specifically, for the temperature field, the time series data of the temperature field is used as input, and the frequency spectrum components are extracted using Fourier transform as output, where the sampling frequency is f_s (f_s represents the sampling rate), which represents the dynamics of heat wave propagation; for the pressure distribution, the pressure spatiotemporal data matrix is ​​used as input, principal component analysis is used for dimensionality reduction, and the first three principal components are extracted as output. The main deformation modes are highlighted by visualizing these principal components.

[0033] For example, in a vulcanizing mold, the pressure variation characteristics of the tire crown region (i.e., the part of the tire in contact with the road surface) can be obtained by calculating the average pressure deviation and peak offset. These deviations reflect the real-time state of material flow. The average pressure deviation is defined as the relative deviation with respect to the initial pressure, calculated as ΔP_avg=(P_t-P_0) / P_0, where ΔP_avg is the average pressure deviation, P_t is the real-time pressure, and P_0 is the initial pressure. The peak offset is ΔP_peak=(P_peak-P_0_peak), where ΔP_peak is the peak offset, P_peak is the peak pressure, and P_0_peak is the initial peak value. This principle ensures a direct correspondence between the characteristics and the curing process, avoiding overgeneralized descriptions.

[0034] Preferably, the process of obtaining the curing state index includes threshold judgment and feedback mechanism. Specifically, multiple threshold ranges are preset, such as when the curing completion rate is greater than 80%, it is considered qualified. By comparing the characteristics such as temperature and viscosity extracted from the sensor with the threshold in real time, the curing state index value is generated and fed back to the process control system to automatically adjust the heating parameters.

[0035] In one embodiment, for the vulcanization of industrial rubber gaskets, the indicators also include a defect prediction sub-item, such as calculating the risk of bubble formation based on pressure unevenness. Pressure unevenness is obtained by calculating the standard deviation P_sd, using the formula P_sd=sqrt(sum((P_i-P_mean)^2) / n), where P_i is the pressure value at each measurement point, P_mean is the average pressure, and n is the number of measurement points. The risk of bubble formation R=P_sd / 10; if R>1, the risk is high. This detailed calculation process improves the practicality of monitoring. In simulation experiments, it can reduce the scrap rate from 15% to 5%, increasing production efficiency by 10%.

[0036] The feature acquisition parameters can be adjusted according to the complexity of the product under different vulcanization scenarios. For simple rubber blocks, feature extraction only requires basic statistical calculations on the temperature sensor data, i.e., calculating the mean and variance of the temperature at each measuring point; for complex tires, convolution operations are required to extract local features from the temperature field image. This parameter adjustment method makes the technical solution highly versatile without changing the core application in the vulcanization field. The above feature acquisition and analysis process can be integrated into the control system.

[0037] Specifically, the sensor array is connected to the central processing unit. After transmitting floating-point temperature and pressure data in real time, it executes principal component analysis feature extraction algorithms and neural network index calculation algorithms to output the vulcanization index S, where S = a × T + b × P, and a and b are weighting coefficients. This systematic implementation ensures seamless integration from data acquisition to output. In rubber product manufacturing, based on a comparison of S with a threshold of 3.5, fuzzy control methods can be used to automatically adjust vulcanization time and pressure; if S is greater than 3.5, the pressure is reduced.

[0038] For example, the results show that the curing state index, obtained through the above method, can accurately guide process optimization. In practical applications, this index detects early over-curing risk by calculating S=(T_max-T_current) / T_max (where T_max is the maximum torque and T_current is the current torque). When S>0.9, a risk is indicated, thus preventing product performance degradation. This objective effect stems from detailed feature analysis, supporting the comprehensive effectiveness of the method.

[0039] Step S102: Based on the obtained curing state indicators, a convolutional neural network is used to extract the structural difference features of the product, determine the reaction difference between the thin-walled part and the thick connecting section, and obtain the parameter adjustment requirements.

[0040] Temperature and pressure data are collected from the vulcanization process using a sensor array. Humidity data is obtained by fusing the temperature and pressure data using a saturated vapor pressure model, with the fusion formula H = f(T, P), where H represents the humidity field, T represents the temperature field, and P represents the pressure distribution. Synchronous datasets are obtained through fusion calculations using the ideal gas law and phase equilibrium. Real-time variation characteristics of product structural parts are extracted from these synchronous datasets using time series analysis. These characteristics include the deformation rate and stress change rate of the parts. Abnormal fluctuation values ​​are identified within these real-time variation characteristics; these values ​​represent deviations exceeding the mean. If the abnormal fluctuation value exceeds a preset threshold of three standard deviations, the data weights within the synchronous dataset are adjusted according to the degree of abnormality, using the adjustment formula Wi = W_0. (1-k Let D_i represent the adjusted weights, W_i represent the initial weights, k represent the adjustment coefficient, and D_i represent the abnormal deviation. Obtain the adjusted feature set. Use a convolutional neural network to extract product structural difference features from the adjusted feature set, determining the response differences between thin-walled sections and thick connecting segments. Based on these response differences, obtain the parameter adjustment requirements.

[0041] In one implementation, a convolutional neural network is used to extract product structural difference features based on the acquired curing state indicators. This process first requires the curing state indicators as input data, which include quantified values ​​of temperature field and pressure distribution.

[0042] Specifically, solidification status indicators can be converted into image formats, such as temperature distribution heatmaps or pressure matrix diagrams, which are easier for convolutional neural networks to process. Through this conversion, the network can capture the spatial feature differences of different parts of the product.

[0043] For example, in the vulcanization of rubber tires, curing state indicators reflect the thermal pressure changes of the tire crown and sidewalls. Inputting these indicators into a network extracts the edge features of the structural contours. This method ensures targeted feature extraction and is applicable to products with various mold shapes in rubber product manufacturing. Furthermore, the extraction process using a convolutional neural network involves multiple convolutional operations.

[0044] Specifically, the network structure includes an input layer, multiple convolutional layers, pooling layers, and fully connected layers. First, the input curing state index data passes through convolutional layers, where filters scan the image to calculate local region feature responses, such as edge detection of temperature gradients. These responses are then processed by activation functions to generate feature maps. Subsequently, pooling layers reduce dimensionality, retaining key information and avoiding overfitting. In the vulcanization scenario of rubber seals, this network can handle pressure distribution data, extracting the difference between deformation features of thin-walled sections and rigidity features of thick connecting sections. The entire extraction process emphasizes spatial invariance, ensuring high sensitivity of features to specific parts of the product. This detailed network operation supports subsequent difference determination steps.

[0045] It should be noted that the principle of convolutional neural networks here is based on capturing local patterns through convolution operations.

[0046] Specifically, convolutional layers use a kernel sliding window approach to perform weighted summation on the input data, extracting low-level features such as texture or gradients. As the layers deepen, the network learns high-level abstract features, such as the overall difference patterns of product structural parts. During the training phase, historical vulcanization data is used as samples, and the network adjusts the weights through backpropagation to optimize its ability to distinguish between thin-walled and thick parts.

[0047] For example, in the vulcanization of rubber pipe fittings, the training data includes labeled samples of temperature unevenness in thin-walled areas. After learning, the network can automatically identify similar patterns. This principle clearly explains how the network can extract hidden differences from curing indicators, avoiding the limitations of traditional manual analysis. In practical applications, this extraction method improves the accuracy of features and provides a reliable basis for parameter adjustment.

[0048] Preferably, the extracted product structure difference features are used to determine the response difference between the thin-walled section and the thick connecting section. Specifically, these features include the thermal gradient vector of the thin-walled section and the pressure gradient vector of the thick section, and the numerical representations of the two are output through a regression model; the response imbalance is calculated by comparing the vector differences between the two, where the response imbalance is defined as a quantitative index of the response difference between the thin-walled section and the thick connecting section, and the calculation formula is D=‖V1-V2‖2, where D represents the response imbalance, V1 represents the thermal gradient vector of the thin-walled section, V2 represents the pressure gradient vector of the thick section, and ‖·‖2 represents the Euclidean norm (2-norm) of the vector.

[0049] In one possible implementation, for the vulcanization of industrial rubber gaskets, a feature vector (composed of curing parameters such as temperature, pressure, and time for thin-walled and thick sections) is input into a difference calculation module (input is the feature vector of the thin-walled and thick sections, output is the difference). The difference is measured using Euclidean distance, calculated as d=√(∑(x_i-y_i)^2), where x_i and y_i are the corresponding elements of the two vectors. If the deviation exceeds a preset threshold of 10%, it is identified as a high-difference region. This determination process considers the dynamic nature of the vulcanization reaction; for example, thin-walled sections are prone to overheating leading to uneven cross-linking, while thick sections may experience under-vulcanization due to insufficient pressure. This analysis clearly reveals the inconsistencies in the internal reaction of the product and maps high-difference regions to parameter adjustments, supporting decisions for process optimization.

[0050] For example, after identifying the reaction differences, parameter adjustment requirements are obtained. These requirements are generated by mapping the differences to specific process parameters. Specifically, based on the difference values, adjustment suggestions are generated, such as increasing the cooling time of thin-walled sections or increasing the pressure value of thick sections. In rubber product manufacturing, this requirement can be output as a control signal and directly fed back to the vulcanizing equipment.

[0051] For example, in the vulcanization of complex rubber blocks, if there are significant differences in reaction between thin-walled and thick-walled sections, adjustments may be needed, including extending the heating cycle to even out the temperature distribution. This method ensures the parameters are tailored to the specific needs of the product, reducing defects and improving product consistency in business operations.

[0052] Understandably, determining the difference in response between thin-walled sections and thick connecting sections involves data fusion. Specifically, features extracted by the CNN are fused with curing state indicators, and a weighted average method is used to calculate a comprehensive difference index. The curing state indicators are material hardening degree data obtained through real-time sensor monitoring; the CNN input is a multi-scale infrared image, and the output is a high-dimensional feature vector. The weighted average calculation formula is D=w1. F+w2 S, where D is the comprehensive difference index, F is the CNN feature, S is the solidification state index, and w1 and w2 are weights of 0.6 and 0.4 respectively, determined based on experience.

[0053] In one embodiment, for rubber tire production, the fused difference index reflects the difference in vulcanization rates between the thinner bead wall and the thicker crown, thereby generating adjustment requirements such as modifying the mold pressure distribution. This fusion enhances the comprehensiveness of the analysis and avoids the bias of single features. In another embodiment, convolutional neural networks can be extended to multi-scale feature extraction.

[0054] Specifically, the network design incorporates convolutional kernels of varying sizes to simultaneously capture both fine-grained and coarse-grained differences, such as the microscopic deformation of thin-walled sections and the macroscopic stress of thick sections. This extension, applied in the vulcanization of rubber seals, enables more precise determination of reaction differences and outputs refined parameter adjustment requirements. Furthermore, the generation of these parameter adjustment requirements includes a threshold mapping mechanism.

[0055] Specifically, a preset difference threshold range is defined. Based on the determined reaction difference falling into this range, an adjustment range is mapped to the desired adjustment using a proportional function. The reaction difference refers to the deviation between the actual vulcanization parameters and the target value. For example, in the rubber hose scenario, a high difference (deviation greater than 10%) corresponds to a significant 20% increase in pressure, while a low difference (deviation less than 5%) results in a slight temperature adjustment of one degree Celsius. This mechanism ensures the gradual nature of the adjustment. In one embodiment, this threshold range mapping adjustment process is integrated into the vulcanization control system.

[0056] Specifically, after the curing status indicators are input into the CNN module, adjustment requirements are automatically output, achieving closed-loop optimization. In rubber gasket production, this integration reduces manual intervention, resulting in lower scrap rates and shorter production cycles.

[0057] For example, in the actual vulcanization of rubber products, the above method allows for timely parameter adjustments to address differences in product structure, ensuring uniform curing. This objective application demonstrates the versatility of the technical solution within the same field, supporting the scope of protection of the claims.

[0058] Step S103: If the parameter adjustment requirement exceeds the preset threshold, the vulcanization process control model is optimized by gradient descent to determine the optimal temperature-time combination and obtain dynamic process instructions.

[0059] Auxiliary data is collected from the vulcanization process using vibration signal sensors. This data is then fused with temperature field data from temperature sensors and pressure distribution data from pressure sensors to obtain extended process parameter field data, forming a complete synchronous dataset. From this complete synchronous dataset, time-domain statistical methods are used to extract real-time variation features of different product sections, calculating the mean and standard deviation to determine potential abnormal fluctuation values. If the potential abnormal fluctuation value exceeds 0.05, the weight of the vibration-related data is adjusted to 0.8 of the original weight, resulting in an optimized feature set. A convolutional neural network is then used to extract structural difference features between thin-walled and thick-walled sections of the product from the optimized feature set. The convolutional neural network uses the optimized feature set as input, processes the data through convolutional and pooling layers, and outputs the structural difference features to determine the corresponding reaction differences. Based on these reaction differences, parameter adjustment requirements are obtained. Based on these parameter adjustment requirements, a gradient descent optimization model of the vulcanization process is used to optimize the model. This model takes the parameter adjustment requirements as input and outputs the optimal temperature-time combination, thus obtaining dynamic process instructions.

[0060] In one implementation, if the parameter adjustment requirement exceeds a preset threshold, optimization of the vulcanization process control model is initiated. This threshold is typically set as a percentage threshold for reaction difference deviation, such as 10%. When the adjustment requirement exceeds this value, it indicates that the current process parameters cannot meet the requirements for uniform curing.

[0061] Specifically, the vulcanization process control model is a mathematical framework used to simulate the reaction dynamics of rubber products under heating and pressure, including temperature distribution and time series prediction functions. By monitoring and comparing adjustment requirements with thresholds, the system automatically triggers an optimization mechanism. In rubber tire vulcanization, this judgment ensures timely response to structural differences, avoiding product defects. Furthermore, the vulcanization process control model is optimized using gradient descent. This optimization method is based on iteratively adjusting model parameters to minimize a loss function that quantifies the error between the predicted cured state and actual indicators.

[0062] It's important to note that gradient descent works by calculating the partial derivatives of the loss function with respect to the parameters, then updating the parameter values ​​along the negative gradient direction to gradually approach the optimal solution. In the rubber seal vulcanization scenario, the initial model parameters include the temperature curve and time intervals. The optimization process starts with the current adjustment requirements and iteratively updates these parameters.

[0063] For example, the loss function can be defined as the sum of squares of temperature deviations. Through multiple iterations (e.g., 50 cycles), the model parameters gradually converge, ensuring that the optimized model more accurately reflects the thermal conductivity characteristics of thin-walled sections. This method supports refined process control in business operations, reducing the need for manual intervention. The entire optimization process emphasizes step size control of parameter updates, typically using a learning rate of 0.01 to avoid oscillations or excessively slow convergence. In actual rubber hose production, the optimized model can handle pressure distribution differences under complex mold shapes, providing more reliable curing predictions. This detailed explanation of optimization highlights its practicality in vulcanization control and supports subsequent combined judgment steps.

[0064] Preferably, the optimal temperature-time combination is determined. This determination is based on an optimized model, evaluating the curing effects of various temperature and time pairs. Specifically, by simulating reaction curves under different combinations, the pair that minimizes the overall curing deviation is selected.

[0065] In one possible implementation, for rubber gasket vulcanization, the model inputs candidate combinations such as 150 degrees Celsius for 20 minutes, calculates the difference in curing completion between thin-walled and thick sections, and selects the one with the smallest deviation as the optimal one. This judgment process considers the interaction between heat conduction rate and pressure accumulation to ensure the applicability of the combination.

[0066] For example, dynamic process instructions are obtained. These instructions are generated from an optimal combination of process parameters obtained through an optimization algorithm, which is directly applied to the vulcanizing equipment. For instance, in rubber block vulcanization, the instructions include real-time adjustments to heating power and duration, such as extending the cooling phase in thin-walled regions. This generation method achieves closed-loop feedback, specifically by monitoring the vulcanization process through equipment sensor data and adjusting the generation of the next optimal combination accordingly, thus improving production consistency. It is understood that the entire optimization process can be integrated into the vulcanization system.

[0067] Specifically, when the parameter adjustment requirement exceeds a threshold, the system sequentially executes gradient descent, combined judgment, and instruction output. In rubber product manufacturing, this integration is applicable to various product forms, ensuring the dynamic adaptability of the process. In another implementation, gradient descent can be combined with an adaptive learning rate.

[0068] Specifically, the learning rate is dynamically adjusted based on gradient changes during iterations, such as gradually decreasing from an initial high value to accelerate convergence. In rubber tire production, this extension optimizes the model's response to sudden variations, generating more precise process instructions. Furthermore, the output of dynamic process instructions includes multi-parameter fusion.

[0069] For example, combining temperature and time with pressure values ​​can form a comprehensive command sequence, supporting automated control in complex vulcanization scenarios.

[0070] Step S104: The device actuator is updated with dynamic process instructions to obtain curing uniformity data from the real-time feedback of the vulcanization process, and the performance non-uniformity deviation is determined to obtain a correction coefficient.

[0071] Dynamic process commands are used to update the actuators, acquiring curing uniformity data from real-time feedback during the vulcanization process. This curing uniformity data is then fused with extended pressure field information (based on extended distribution data from pressure sensors) to determine performance non-uniformity deviations. For these deviations, regional response differences (differences in response to vulcanization pressure at different locations) are extracted to determine the deviation correction requirement (the amount of deviation to be corrected). If the deviation correction requirement exceeds a preset threshold of 5%, the optimal pressure-time combination is determined by optimizing the vulcanization control model (a mathematical model with pressure P and time T as the optimal input and output combination). Based on this optimal pressure-time combination, correction coefficients are generated to update the actuator feedback loop (adjusting actuator parameters using these correction coefficients to achieve closed-loop control).

[0072] In one implementation, dynamic process instructions are used to update the equipment actuators. This update process generates instructions based on the aforementioned optimized model and applies them directly to the control unit of the vulcanizing equipment.

[0073] Specifically, dynamic process instructions include real-time modification signals for temperature adjustments and time sequences, transmitted via an interface to actuators such as heating elements or pressure pumps. These instructions ensure that the equipment responds to changes during the vulcanization process, enabling continuous operation. In rubber tire production, this update mechanism supports synchronous control of different areas within the mold, avoiding manual intervention. Furthermore, curing uniformity data is obtained from the real-time feedback of the vulcanization process. This acquisition step involves deploying a sensor network, such as temperature and pressure sensors, distributed at key locations within the vulcanization mold. Real-time feedback refers to the continuously collected data stream during equipment operation, including curing status indicators for both thin-walled and thick sections.

[0074] It's important to note that the curing uniformity data is a quantified value of the temperature distribution and curing completion rate monitored by these sensors, such as calculating the percentage difference in curing in different areas. This data acquisition forms a feedback loop, supporting subsequent analysis. In the rubber seal manufacturing scenario, the sensors collect data once per second to ensure timely feedback.

[0075] Specifically, the performance non-uniformity deviation is determined. This determination process is based on the acquired curing uniformity data, calculating the difference between different regions. For example, comparing the curing completion of a thin-walled region with that of a thicker region, the deviation is defined as the absolute value or percentage of the difference between the two. Performance non-uniformity deviation reflects product quality issues caused by uneven heat conduction during the vulcanization process.

[0076] In one possible implementation, deviation calculation uses simple statistical methods, such as mean deviation or standard deviation, with input data derived from real-time feedback.

[0077] For example, in the vulcanization of rubber gaskets, if the curing completion rate of the thin-walled area is 85% and that of the thick area is 95%, the deviation is 10%, and this value is used to quantify the degree of non-uniformity.

[0078] Understandably, this determination process emphasizes the real-time nature of the data to ensure that deviations reflect the current process status.

[0079] Preferably, a correction coefficient is obtained. This process involves analyzing performance non-uniformity deviations to generate numerical factors for adjusting process parameters.

[0080] Specifically, the correction factor is a multiplier or addition / subtraction term calculated based on the deviation value. For example, if the deviation exceeds the threshold, the factor is 1.2, which is used to amplify the heating power.

[0081] It should be noted that the principle of the correction coefficient lies in compensating for unevenness, gradually reducing deviations through iterative application of feedback data. In rubber hose production, the coefficient calculation considers mold shape factors, such as the heat distribution characteristics of cylindrical molds, and is directly integrated into dynamic instructions after generation. This method achieves adaptive adjustment of the process. In another implementation, the entire process is integrated into the vulcanization system, automating the process from feedback acquisition to coefficient acquisition.

[0082] For example, in the vulcanization of rubber blocks, the system first updates the actuator, then acquires data to determine the deviation, and finally outputs a correction coefficient, forming a closed loop. This integration supports the processing of various product forms and improves operational efficiency. Furthermore, the application of the correction coefficient is extended to multi-parameter scenarios.

[0083] Specifically, the coefficient is combined with the pressure value to generate a comprehensive adjustment signal. In rubber product manufacturing, this extension addresses the unevenness issues under complex vulcanization conditions, ensuring uniform curing of the final product.

[0084] Step S105: Based on the correction coefficient, the random forest model is fused to predict the impact of raw material batch fluctuations, and the stability of the solidified state is determined to obtain the optimized parameter set.

[0085] Fluctuation data is acquired from raw material batches through real-time monitoring. A weighted average is used to fuse correction coefficients to obtain a prediction input set, which consists of fluctuation amplitude, frequency, and correction coefficients. A random forest model (a pre-trained model) is applied to this prediction input set to construct 100 decision trees. The batch fluctuation impact is determined by selecting ensemble prediction values ​​through random feature selection. The stability of the solidified state is assessed by comparing the batch fluctuation impact with a standard indicator (a preset fluctuation threshold of 0.05) to obtain deviation features (fluctuation deviation vector). Based on these deviation features, an optimized parameter set is generated through a linear function mapping. This mapping takes the deviation vector as input and outputs the corresponding adjustment parameters.

[0086] In one implementation, the random forest model is fused based on correction coefficients. This fusion process uses the previously obtained correction coefficients as input features and incorporates them into the construction of the decision tree of the random forest model.

[0087] Specifically, the random forest model consists of multiple decision trees, each constructed using randomly sampled training data and features. Correction coefficients are used as weighting factors to adjust the splitting criteria of the tree nodes. This fusion ensures the model considers deviation compensation during the vulcanization process, making it suitable for batch data processing in rubber product manufacturing.

[0088] For example, in the scenario of rubber tire vulcanization, the model input includes temperature and pressure parameters. By fusing correction coefficients, the model's responsiveness to variations is improved, thereby determining the stability of the cured state. Furthermore, based on the fused random forest model, the impact of batch fluctuations in raw materials such as rubber compounds on the stability of the cured state is predicted.

[0089] Specifically, raw material batch fluctuations refer to variations in composition between different suppliers or production batches, such as changes in vulcanizing agent content or filler ratios. Random forest models handle these variations through ensemble learning. They first extract features such as density and viscosity values, and then calculate the average output of the leaf nodes in each decision tree to generate a probability distribution of the impact of fluctuations.

[0090] It should be noted that this prediction considers training the model with historical data, such as using records from the past 100 batches as a training set, to ensure that the model captures fluctuation patterns. In the production of rubber seals, if the vulcanizing agent content fluctuates by 5% in a batch, the model can predict an increase in curing time of 10 minutes, forming a quantitative assessment.

[0091] Understandably, this involves assessing the stability of the cured state. This assessment process utilizes predicted fluctuation impact values ​​to calculate stability indicators for the curing process.

[0092] Specifically, stability is achieved by comparing predicted values ​​with a standard threshold. For example, the predicted impact refers to the proportion of deviation between the predicted value and the standard threshold, and stability is defined as the proportion of predicted impacts below 0.1. If the predicted impact exceeds the threshold, it is judged as unstable.

[0093] For example, in the vulcanization of rubber gaskets, after the neural network model (inputting temperature and time, outputting predicted fluctuation values) outputs the impact of fluctuations, the system checks the variance of the curing completion degree (the percentage of rubber crosslinking degree measured by infrared spectroscopy). If the variance is less than 0.05, it is considered stable. This judgment forms a feedback mechanism; that is, if the variance is greater than 0.05, the parameters are iteratively adjusted to support subsequent adjustments. Preferably, an optimized parameter set is obtained. This obtaining step, based on the judgment result, uses a genetic algorithm to generate adjusted process parameters, such as temperature and time values.

[0094] Specifically, the parameter set is modified from the initial parameters through an iterative optimization algorithm. For example, the temperature parameter under unstable conditions is increased with a correction factor. In the context of rubber hose manufacturing, the optimized parameter set includes heating power and pressure sequences to ensure adaptability to batch fluctuations.

[0095] In one possible implementation, the entire fusion-optimized process is integrated into the vulcanization control system. Specifically, starting with the correction coefficient input, the model predicts fluctuations, then assesses stability, and finally outputs a parameter set. This integration handles variations in rubber block vulcanization, ensuring process continuity. In another implementation, model training considers multiple batches of data, for example, using cross-validation to verify prediction accuracy and avoid overfitting. Furthermore, the prediction of raw material batch fluctuations is extended to multi-feature scenarios.

[0096] Specifically, by incorporating humidity or environmental factors as additional input, random forests prioritize key variations by ranking features by importance. In the vulcanization of rubber products, this extension addresses stability assessments under complex conditions.

[0097] For example, in the production of rubber tires, after determining the stability of the cured state, the optimized parameter set is applied to the mold control to achieve adaptive adjustment of the parameters.

[0098] It should be noted that the generation of the optimized parameter set emphasizes real-time performance and is transmitted to the actuator through the system interface.

[0099] In one embodiment, the fusion model refers to a random forest ensemble learning model that takes an optimized parameter set as input, with 50 decision trees to ensure prediction robustness. This model outputs an assessment of the negative impact of production parameter fluctuations on the performance of rubber seals, specifically referring to performance deviations such as pressure resistance and sealing caused by fluctuations in parameters such as temperature, pressure, and vulcanization time. In the context of rubber seals, this setup supports accurate assessment of the impact of fluctuations.

[0100] Step S106: Drive the vulcanization process simulation module to generate a virtual test scenario by optimizing the parameter set, obtain simulation performance indicators, determine the adaptability to actual production, and obtain the final control scheme.

[0101] The driving signal, including temperature threshold and pressure range, is obtained from the optimized parameter set and input into the vulcanization process simulation module. This module, pre-built based on the finite element method, generates a virtual test scenario after the driving signal is input. Temperature distribution is simulated within the virtual test scenario, using a heat conduction equation to track material flow. Material flow refers to the flow of rubber material at high temperatures and its filling of the mold, which is closely related to temperature distribution. The variables in the heat conduction equation are time (t) and position (x). After solving, a performance index set is obtained through integration. Production adaptability is determined from the performance index set by comparing it with a standard threshold to determine the deviation value. A preset control scheme is then fused based on the deviation value. The fusion method is as follows: if the deviation value is greater than 5%, the scheme is replaced; otherwise, parameters are adjusted using a weighted average. The preset control scheme is obtained from historical data to obtain the final control scheme.

[0102] In one implementation, the vulcanization process simulation module is driven by an optimized set of parameters. This driving process takes the previously obtained optimized parameters as input and activates the simulation module's operating mechanism.

[0103] Specifically, the vulcanization process simulation module is a physical model-based software component that simulates the chemical reaction process of rubber materials under heating and pressure. The optimized parameter set includes adjusted temperature profiles and pressure values, which are passed into the module via an interface to guide the initial condition settings for the simulation.

[0104] For example, in the vulcanization of rubber tires, a parameter set driving module (whose inputs are process parameters such as temperature and pressure, and whose outputs are simulated temperature fields and crosslinking degrees, simulating heat conduction and chemical reaction processes using the finite element method) simulates heat conduction and crosslinking reactions within the vulcanization chamber, ensuring that the virtual environment reflects actual fluctuations (such as random changes in temperature, pressure, and reaction rates). Further, virtual test scenarios are generated. This generation step, based on the driven parameter set, randomly generates multiple sets of parameter combinations using Monte Carlo sampling methods to construct different virtual environments (such as variable temperature scenarios and variable pressure scenarios) to test the vulcanization process.

[0105] Specifically, the virtual testing scenario simulates the impact of different batches of raw materials through parameter variations, such as changes in the density of rubber compounds or adjustments in the proportion of vulcanizing agents. The simulation module uses the finite element analysis method to divide the grid and calculate the state evolution of each grid point.

[0106] It should be noted that this generation takes into account the injection of random factors, such as introducing a 5% batch deviation, to cover potential production variations. In the rubber seal vulcanization scenario, the virtual scenario can simulate 10 consecutive cycles of operation to form a test sequence.

[0107] Preferably, simulated performance indicators are obtained. This acquisition process extracts key data from the virtual test scenario, such as curing completion, hardness distribution, and time consumption.

[0108] Specifically, performance indicators are calculated through the output function of the simulation module, which is a virtual environment simulation unit based on molecular dynamics. Its output function takes simulation data (such as molecular positions and bonding states) as input, calculates indicator values ​​through statistical analysis, and outputs them. For example, the curing completion rate is defined as the ratio when the crosslinking density (the number of crosslinks per unit volume) reaches a preset proportion of 0.95. At the end of each virtual scene, this module summarizes the data and generates an indicator report.

[0109] In one possible implementation, for rubber gasket production, indicators include elastic modulus and residual stress, which are derived from simulated stress-strain curves to ensure quantitative evaluation.

[0110] Understandably, the step involves determining the suitability for actual production. This determination process utilizes the acquired performance indicators, comparing them with actual production standards.

[0111] Specifically, adaptability is achieved through threshold checks; for example, if the proportion of simulated hardness indicators within the target range exceeds 90%, it is considered adaptable.

[0112] It should be noted that this determination considers multiple indicators comprehensively, such as combining time and quality indicators to calculate the fitness score. The specific calculation can use a weighted average method: S = w1 × T + w2 × Q, where S is the fitness score, T is the time indicator, Q is the quality indicator, and w1 and w2 are their respective weights (e.g., 0.5 each). If the virtual indicator (i.e., the indicator obtained through simulation prediction) shows excessive deviation, it is marked as maladaptive, supporting subsequent adjustments.

[0113] Specifically, the final control scheme includes a refined version of the preset parameter set in the initial control model. This refinement adjusts the parameter values ​​based on the adaptive determination results, such as adding an auxiliary heating sequence to compensate for the unadaptive part of the system's insufficient response. This compensation process involves adding heating pulses in the region where the response curve is below the threshold to improve efficiency.

[0114] For example, in a rubber block vulcanization system, the scheme is transmitted to the controller via a feedback loop for automatic adjustment. This scheme ensures production continuity and reduces scrap rates in sealing applications. Furthermore, the driving mechanism of the vulcanization process simulation module is extended to multiple application scenarios.

[0115] Specifically, by incorporating environmental variables such as humidity as additional parameters, the simulation module adapts to complex conditions through iterative calculations. In rubber tire production, this extended support scheme for real-time optimization.

[0116] For example, the entire process from inputting drive parameters to obtaining the solution is integrated into the control system.

[0117] Specifically, starting with parameter input, multiple scenarios are generated through simulation. Then, key indicators (such as strength S and durability D, where S represents tensile strength and D represents fatigue life) are obtained based on the scenario data. The adaptability of the solution is determined accordingly, and finally, an optimized solution is output. This integrated approach handles parameter variations in the rubber product manufacturing process, ensuring process stability.

[0118] Step S107: If the final control scheme meets the preset consistency threshold, the historical data of the vulcanization process is used to train a support vector machine classifier to determine the solidification abnormality pattern under the product structure diversity and obtain the prevention and adjustment rules.

[0119] If the final control scheme meets the preset consistency threshold, sample data containing product structure diversity is extracted from historical data of the vulcanization process. This sample data is then classified and labeled, including normal and abnormal modes during the curing process, forming an initial sample set. A support vector machine classifier is trained on this initial sample set to obtain classification rules that can distinguish between normal and abnormal modes. These rules are used to assess curing parameters under product structure diversity to determine the possible set of abnormal modes. Feature parameters corresponding to each abnormal mode are extracted from the abnormal mode set. These parameters are then fused with performance index deviation values ​​obtained from historical data to generate deviation distribution characteristics for different product structures. Based on these deviation distribution characteristics, the initial prevention rules obtained from historical data are adjusted. Corresponding prevention and adjustment rules are generated for different product structure curing abnormal modes to ensure that the rules can adapt to the curing process requirements under product structure diversity.

[0120] The generation steps are as follows: 1. In one embodiment, when extracting sample data containing product structural diversity from historical vulcanization process data, the historical data is first filtered to select product records involving thin-walled sections and thick connecting sections. For example, a batch of tire vulcanization data is selected, where thin-walled sections correspond to the tire sidewall area and thick connecting sections correspond to the tire crown area. This data includes temperature and pressure change records. This extraction method can cover the differences in curing behavior under different structures, which is beneficial to the comprehensiveness of subsequent classification. When classifying and labeling these sample data, normal patterns are labeled as records with uniform temperature distribution and stable pressure without fluctuations, while abnormal patterns are labeled as records with uneven curing caused by localized excessively high temperatures. The initial sample set formed in this way can provide reliable basic data to support the accuracy of subsequent training.

[0121] 2. Specifically, when training an initial sample set using a Support Vector Machine (SVM) classifier, a supervised learning method, it maximizes the margin between different categories by finding a hyperplane. For example, during training, normal pattern samples are treated as one class and abnormal patterns as another. The classifier calculates the distance between feature vectors to construct a decision boundary, and the resulting classification rules can effectively distinguish these patterns. When judging the solidification parameters under product structure diversity using these classification rules, such as inputting the temperature curve and pressure distribution parameters of a new product, the classifier will output whether it belongs to an abnormal pattern, thereby determining the abnormal pattern set. The beneficial effect of this method is that it improves the accuracy of anomaly detection and reduces solidification defects in production.

[0122] 3. In one embodiment, when extracting the characteristic parameters corresponding to each abnormal mode from the abnormal mode set, such as extracting the temperature peak and pressure fluctuation amplitude as characteristic parameters, these parameters reflect specific problems in the curing process. When combining the performance index deviation values ​​obtained from historical data, the deviation values, such as hardness deviation or strength deviation, are weighted and fused with the characteristic parameters to generate deviation distribution characteristics for different product structures. For example, the deviation distribution characteristics of thin-walled parts show higher temperature sensitivity. The beneficial effect of this fusion calculation is to provide a more comprehensive abnormal assessment and help optimize production parameters.

[0123] 4. Specifically, when adjusting the initial prevention rules obtained from historical data based on the characteristics of the deviation distribution, for example, if the initial rule is to keep the temperature within a certain range, the adjusted rule lowers the upper limit of the temperature for thin-walled parts to avoid overheating. In this way, the generated prevention adjustment rules can target the curing anomaly patterns of different product structures, ensuring that the rules are highly adaptable, which is beneficial to preventing curing anomalies and improving overall product quality and production efficiency.

[0124] Step S108: Obtain abnormal signals from real-time changes in production conditions according to the prevention and adjustment rules, determine the stability of the solidification process, and obtain the optimized path for mass production.

[0125] Anomaly signals are acquired by monitoring real-time changes in production conditions, including temperature, pressure, and vulcanization time. A stability index is used to calculate the stability of the vulcanization process, yielding a stability deviation value. This stability deviation value is then combined with the real-time production condition changes to calculate a deviation assessment result. Based on this deviation assessment result, preventative adjustment rules are applied: if the deviation assessment result is greater than 0.15, the vulcanization temperature is preferentially increased by 3–5°C or the vulcanization time is extended by 1–3 minutes to generate an optimized batch production path.

[0126] In one implementation, abnormal signals are acquired from real-time changes in production conditions according to preventative adjustment rules. This is first achieved by collecting data in real-time through a sensor network deployed on the vulcanizing equipment. This data includes temperature fluctuation sequences, pressure change curves, and real-time torque readings. Specifically, the system collects data at preset intervals, such as five seconds, forming a continuous time-series data stream.

[0127] It should be noted that abnormal signals are defined as deviations from the standard process curve, such as a sudden slowdown in the rate of temperature rise or an abnormal peak in torque growth. By matching these deviations with a preventative adjustment rule base, the system identifies whether they correspond to known abnormal patterns, such as the potential risk of undersulfurization due to temperature lag in the central region. Furthermore, after acquiring abnormal signals, determining the stability of the curing process involves a multi-dimensional assessment.

[0128] For example, in the vulcanization process of rubber seals, stability indices are calculated, including temperature uniformity coefficient and torque fluctuation variance. The temperature uniformity coefficient is obtained by comparing the temperature differences in different parts of the mold; if the difference exceeds a threshold, it is considered unstable. The torque fluctuation variance is based on statistical data from the isothermal stage and reflects the stability of the crosslinking reaction.

[0129] Understandably, this assessment takes into account product structure factors. For example, thin-walled products focus more on rapid heating stability, while thick-walled products focus on uniformity in the later stages of isothermal control, thus ensuring the accuracy of the assessment.

[0130] Preferably, based on the stability assessment results, a rule matching algorithm is used to generate an optimized path for batch production. Specifically, the preset threshold is 0.85. If the stability is lower than this threshold, the system extracts the corresponding adjustment parameters from the pre-built expert experience rule base according to the type of stability deviation (such as time or pressure related), and considers the coupling effect between parameters, sorting them by priority to form an ordered optimization sequence. For example, the vulcanization time is first extended to 1.2 times the standard value, and then the pressure is finely adjusted to 1.05 times the original value.

[0131] In one possible implementation, for scenarios involving continuous production of rubber hoses, the optimization path includes a batch adjustment strategy: first, implement immediate corrections for the current batch, and then update the process parameters for subsequent batches to minimize the propagation of anomalies.

[0132] In one embodiment, for mass production of rubber gaskets, anomaly signal acquisition focuses on real-time monitoring during the pressure holding phase. If an abnormal pressure drop is detected, the system immediately triggers a rule query to assess whether it affects curing uniformity. Further stability evaluation incorporates historical benchmark comparisons, such as similarity calculations with torque curves of normal batches, to ensure accurate quantification. In practical applications, this method enables adaptive process adjustment through continuous monitoring.

[0133] It should be noted that the generation of optimized paths emphasizes operability, such as outputting a detailed list of steps, including parameter modification values ​​and expected recovery time. For production lines with different batch sizes, the paths can be flexibly expanded; for example, manual intervention can be prioritized for small batches, while automation can be implemented for large batches, thus adapting to production needs.

[0134] For example, in the vulcanization of rubber tire components, after acquiring abnormal signals from real-time changes in conditions, the stability determination process integrates multi-sensor data fusion. The fusion algorithm simply weights and averages temperature and pressure readings to calculate a comprehensive stability score. If the score is low, an optimization path is suggested for phased intervention: first optimizing the heating parameters, then adjusting the cooling rate. This layered approach helps maintain overall production efficiency. Furthermore, the robustness of the entire process is enhanced through a closed-loop feedback mechanism.

[0135] Specifically, after the optimized path is implemented, the system continues to monitor subsequent condition changes. If the number of abnormal signals decreases, the path is confirmed to be effective; otherwise, iterative rules are applied. This mechanism provides a basis for continuous optimization in the diversified production of rubber products, ensuring the long-term stability of the curing process.

[0136] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method of controlling the vulcanization process of a silicone rubber article, characterized by, include: Temperature field data and pressure distribution data during the vulcanization process are collected by a sensor array, and humidity field data obtained by fusing the temperature field data and the pressure distribution data are used to form a synchronous dataset. Real-time change features of product structure parts are extracted from the synchronous dataset, abnormal fluctuation values ​​in the real-time change features are determined and data weights are adjusted to obtain an adjusted feature set, and solidification status indicators are determined based on the adjusted feature set. Based on the curing state index, a convolutional neural network is used to extract the structural difference features of the product, determine the reaction difference between thin-walled parts and thick connecting sections, and obtain parameter adjustment requirements. If the parameter adjustment requirement exceeds the preset threshold, the vulcanization process control model is optimized by gradient descent to determine the optimal temperature-time combination and obtain dynamic process instructions. The dynamic process instructions are used to update the equipment actuators, obtain curing uniformity data from real-time feedback of the vulcanization process, determine performance non-uniformity deviations, and obtain correction coefficients. Based on the correction coefficients, the random forest model is fused to predict the impact of raw material batch fluctuations, and the stability of the solidified state is determined to obtain the optimized parameter set. The optimized parameter set drives the vulcanization process simulation module to generate a virtual test scenario, obtain simulation performance indicators and determine the adaptability to actual production, and obtain the final control scheme. If the final control scheme meets the preset consistency threshold, a support vector machine classifier is trained using historical data of the vulcanization process to determine the solidification anomaly pattern under product structure diversity and obtain the prevention and adjustment rules. Based on the aforementioned prevention and adjustment rules, abnormal signals are obtained from real-time changes in production conditions, and the stability of the curing process is determined to obtain an optimized path for batch production.

2. A method of controlling the vulcanization process of a silicone rubber article according to claim 1, wherein The process of extracting real-time change features of product structural parts from the synchronous dataset, determining abnormal fluctuation values ​​in the real-time change features and adjusting data weights to obtain an adjusted feature set, and determining solidification status indicators based on the adjusted feature set includes: The humidity field data is generated by performing spatiotemporal matching of temperature field data and pressure distribution data for corresponding product structural parts within the synchronous dataset. The temperature change rate, pressure gradient change, and humidity inference values ​​of multiple product structural parts are extracted from the synchronous dataset as the real-time change features. Determine whether there are any abnormal fluctuation values ​​in the real-time change features that exceed the preset fluctuation range. If so, assign reduced weights to the data at the corresponding time point and spatial location in the synchronous dataset according to the magnitude of the abnormal fluctuation value to obtain the adjusted feature set. The solidification state index, which characterizes the current degree of solidification, is obtained by weighted fusion calculation based on the adjusted feature set.

3. The method for controlling the vulcanization process of silicone products as described in claim 1, characterized in that, The step of extracting product structural difference features using a convolutional neural network based on the curing state index, determining the reaction differences between thin-walled sections and thick connecting sections, and obtaining parameter adjustment requirements includes: The curing state indicators are reconstructed into a multi-channel feature map according to the spatial distribution of the product structure. The multi-channel feature map is input into the convolutional neural network, which is then processed sequentially through multiple convolutional layers, activation layers, and pooling layers to obtain a deep feature representation. The response difference vectors of different structural regions of the product are obtained by mapping through a fully connected layer or global pooling. The parameter adjustment requirements are calculated based on the difference in curing rate and temperature gradient between the thin-walled section and the thick connecting section in the reaction difference vector.

4. The method for controlling the vulcanization process of silicone products as described in claim 1, characterized in that, If the parameter adjustment requirement exceeds a preset threshold, the vulcanization process control model is optimized through gradient descent to determine the optimal temperature-time combination and obtain dynamic process instructions, including: The parameter adjustment requirements are input into the vulcanization process control model as part of the loss function; The vulcanization process control model takes the current temperature curve and pressure curve as state inputs and the temperature setpoint and pressure setpoint of the next moment as action outputs. The model parameters are iteratively updated using the gradient descent algorithm until the loss converges. Based on the converged model, multiple temperature-time combinations are output, and the optimal combination that satisfies the curing uniformity constraint and energy consumption constraint is selected as the dynamic process instruction.

5. The method for controlling the vulcanization process of silicone products as described in claim 1, characterized in that, The process of updating the equipment actuator using the dynamic process instructions, obtaining curing uniformity data from real-time feedback of the vulcanization process and determining performance non-uniformity deviations, and obtaining correction coefficients includes: The execution parameters of the heating and pressurizing devices are adjusted in real time according to the dynamic process instructions. Feedback data is collected from temperature and pressure sensors at multiple locations, and the statistical variance of the curing degree at each location is calculated as the curing uniformity data. The performance non-uniformity deviation, which characterizes the overall non-uniformity, is obtained by fusing the curing uniformity data with the expanded pressure field distribution. To address the performance unevenness deviation, the response hysteresis difference characteristics of different parts are extracted, and the correction coefficient is determined based on the intensity of the difference characteristics.

6. The method of claim 1, wherein the step of determining the amount of time comprises the step of: The optimized parameter set obtained by fusing the random forest model with the correction coefficient to predict the impact of raw material batch fluctuations and judging the stability of the solidified state includes: ​ Collect the physical property fluctuation data of the current batch of raw materials, and weight and concatenate it with the correction coefficient to form a prediction input set; The predicted input set is input into a pre-trained random forest model, which uses multiple decision trees to predict in parallel and integrate the results to obtain the impact value of batch fluctuation on the fixed endpoint. The influence value is compared with the historical standard curing index, and the deviation direction and magnitude are calculated as the basis for judging the stability of the curing state. Based on the judgment result, the temperature reference value, pressure reference value, and holding time are adjusted through parameter mapping relationship to generate the optimized parameter set.

7. The method for controlling the vulcanization process of silicone products as described in claim 1, characterized in that, The process of generating a virtual test scenario by driving the vulcanization process simulation module through the optimized parameter set, obtaining simulation performance indicators and determining actual production adaptability, and obtaining the final control scheme includes: Temperature threshold sequence, pressure range sequence, and time node sequence are extracted from the optimized parameter set as driving signals; The driving signal is input into a pre-established vulcanization process simulation module. The simulation module performs coupled calculations of material flow and heat conduction based on the finite element method to generate a virtual temperature field evolution of the entire vulcanization process. The simulation performance indicators such as maximum temperature difference, uniformity of curing degree distribution, and residual stress are extracted from the temperature field evolution. The simulated performance indicators are compared with preset production standard thresholds, and the historical best control schemes are integrated based on the deviation magnitude to obtain the final control scheme.

8. The method for controlling the vulcanization process of silicone products as described in claim 1, characterized in that, If the final control scheme meets the preset consistency threshold, a support vector machine classifier is trained using historical data from the vulcanization process to determine abnormal curing patterns under product structure diversity and obtain preventive adjustment rules, including: A training sample set was formed by extracting temperature field, pressure field, and curing result labels of multiple batches of products with different structures from historical data of the vulcanization process. The training sample set is input into a support vector machine classifier for training to obtain a classification model that can distinguish multiple fixed abnormal patterns; For newly input product structure parameters and process parameters, the classification model predicts possible solidification anomaly patterns and outputs the set of anomaly categories with the highest probability. Based on the set of anomaly categories and historical deviation data, preventive adjustment rules for material composition and structural characteristics are generated, thus obtaining the preventive adjustment rules.

9. The method for controlling the vulcanization process of silicone products as described in claim 1, characterized in that, The step of obtaining abnormal signals from real-time changes in production conditions based on the prevention and adjustment rules, determining the stability of the solidification process, and obtaining a batch production optimization path includes: Real-time monitoring of changes in production environment temperature, humidity, and raw material moisture content; extraction of abnormal signal sets that exceed normal ranges. The stability index for the abnormal signal set is calculated, including a weighted sum of fluctuation amplitude and duration; The stability index is matched with the threshold conditions in the prevention and adjustment rules to determine the degree of stability deviation of the current curing process. Based on the degree of stability deviation, a corresponding subset of prevention and adjustment rules is selected to generate a temperature compensation sequence, a pressure compensation sequence, and a time node adjustment sequence for batch production, thereby obtaining the batch production optimization path.

Images