Automatic control method and system for production line of special equipment for rubber processing

Abstract

The invention provides an automatic control method and system for a production line of special equipment for rubber processing, relates to the technical field of rubber processing, and aims at obtaining process parameter information such as energy absorption performance and shear response performance of materials in an initial processing link and analyzing and determining initial internal state indexes of the materials. Recording the initial internal state index and the energy input information as a processing resume, transmitting the processing resume to a subsequent processing link, and in the subsequent processing link, according to the received processing resume and real-time process information, deducing the current internal state of the material, and adjusting processing parameters to eliminate or weaken the internal deviation of the material. And the processing resume is updated in time. The problems that in a traditional method, manual experience is excessively relied on, and the cooperation efficiency between equipment is not high are effectively solved, the quality stability, the size stability and the fatigue resistance of rubber products are remarkably improved, and intelligent and automatic control over a production line is achieved.

Description

Technical Field

[0001] This application relates to the field of rubber processing technology, and more specifically, to an automated control method and system for a production line of special equipment for rubber processing. Background Technology

[0002] In the rubber product manufacturing industry, traditional production equipment often faces problems such as disjointed production processes, excessive reliance on manual experience for parameter adjustments, and low efficiency in collaboration between equipment. Key parameters such as temperature and pressure at each stage of the production line need to be coordinated in real time and flexibly. However, existing technologies lack intelligent decision-making capabilities when handling the linkage of multiple machines, often leading to unstable final product quality. Specifically, when introducing new raw materials, even if their surface properties appear similar to existing materials, subtle differences at the molecular level can trigger unexpected changes during processing. Current control systems, which typically only focus on whether the output of each independent stage meets basic standards, struggle to detect these subtle internal changes. This allows latent problems within the material, such as inhomogeneous molecular structures or the quiet accumulation of internal stress, to accumulate throughout the production line. Each processing step may attempt to correct obvious problems, but these corrections often exacerbate unseen internal issues, ultimately significantly reducing the overall performance of the product, such as its stability and fatigue resistance. Even more troubling is that existing systems do not issue any alarms because all easily measurable parameters show normal readings, giving the illusion that production is proceeding smoothly.

[0003] For example, in a rubber product production line, when the upstream mixing process makes local compensatory adjustments due to minor differences in raw materials, it leads to changes in the material's intrinsic physical properties that are not directly monitored (such as molecular chain structure and internal stress distribution). These implicit deviations accumulate in subsequent processes such as calendering and vulcanization, ultimately affecting the long-term stability of the product's overall performance (such as dimensional stability and fatigue resistance). Current technology lacks a correlational control method that spans multiple processing units. This method can infer the continuous evolution of the material's intrinsic state in real time and feed this information forward to downstream processes, thereby guiding downstream systems to make predictive, non-compensatory parameter optimization adjustments. This would fundamentally eliminate or mitigate the implicit deviations introduced upstream, ensuring the stability of the final product's overall performance. Summary of the Invention

[0004] This application provides an automated control method and system for a production line of special equipment for rubber processing, which aims to solve the technical problems in traditional rubber product manufacturing, such as discontinuous production processes, excessive reliance on manual experience for parameter adjustment, low efficiency of cooperation between equipment, and difficulty in detecting subtle differences at the molecular level of materials, which lead to the accumulation of internal problems and ultimately affect the overall performance stability of the product.

[0005] On the one hand, this application provides an automated control method for a rubber processing equipment production line, including: Obtain process parameter information of the material in the initial processing stage, including the material's energy absorption performance and shear response performance; analyze the process parameter information to determine the material's initial intrinsic state index. The initial intrinsic state indicators and energy input information in the initial processing stage are recorded as the processing history of the material, and the processing history is transferred to the subsequent processing stage. In subsequent processing stages, based on the received processing history and real-time process information in the subsequent processing stages, the current internal state of the material is inferred. Then, based on the current internal state, the processing parameters of the subsequent processing stages are adjusted to eliminate or reduce the inherent deviation of the material, and the current internal state is updated to the processing history.

[0006] On the other hand, this application provides an automated control system for a rubber processing production line, the system comprising: The information acquisition module is used to acquire process parameter information of the material in the initial processing stage. The process parameter information includes the material's energy absorption performance and shear response performance. The process parameter information is analyzed to determine the initial intrinsic state index of the material. The history recording and transmission module is used to record the initial internal state indicators and energy input information in the initial processing stage as the processing history of the material, and to transmit the processing history to the subsequent processing stage. The state inference and parameter adjustment module is used to infer the current internal state of the material in subsequent processing stages based on the received processing history and real-time process information in subsequent processing stages, and then adjust the processing parameters of the subsequent processing stages based on the current internal state to eliminate or reduce the inherent deviation of the material, and update the current internal state to the processing history.

[0007] This application relates to an automated control method and system for a rubber processing production line. By acquiring process parameter information such as energy absorption and shear response of materials in the initial processing stage and analyzing and determining the initial intrinsic state indicators of the materials, it solves the problem in existing technologies where subtle differences at the molecular level of materials are difficult to detect, leading to the accumulation of intrinsic problems. This method records the initial intrinsic state indicators and energy input information as a processing history and transmits it to subsequent processing stages, thus overcoming the problems of discontinuous and isolated information in traditional production processes. In subsequent processing stages, the current intrinsic state of the materials can be inferred based on the received processing history and real-time process information, and processing parameters can be adjusted accordingly to eliminate or reduce the intrinsic deviations of the materials, while updating the processing history in a timely manner. This proactive, non-compensatory parameter optimization and adjustment mechanism fundamentally solves the problem in existing technologies where the accumulation of implicit deviations from upstream processes ultimately affects the overall performance stability of the product. By inferring the continuous evolution of the material's internal state in real time and controlling the cross-process correlation, this application effectively avoids the problems of over-reliance on human experience and low efficiency of collaboration between equipment in traditional methods. It significantly improves the quality stability, dimensional stability and fatigue resistance of rubber products, realizes intelligent and automated control of the production line, and has significant and excellent technical effects. Attached Figure Description

[0008] To illustrate this application more clearly, the accompanying drawings used in the embodiments will be briefly described below. Obviously, those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0009] Figure 1 The diagram above illustrates a flow chart of an automated control method for a rubber processing equipment production line. Figure 2 The diagram above illustrates the structure of an automated control system for a rubber processing production line.

[0010] Figure reference numerals: 100, Automation control system for rubber processing special equipment production line; 10, Information acquisition module; 20, History recording and transmission module; 30, Status inference and parameter adjustment module. Detailed Implementation

[0011] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0012] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0013] In the rubber product manufacturing industry, traditional production equipment often faces problems such as disjointed production processes, excessive reliance on manual experience for parameter adjustments, and low efficiency in collaboration between equipment. Key parameters such as temperature and pressure at each stage of the production line need to be coordinated in real time and flexibly. However, existing technologies lack intelligent decision-making capabilities when handling the linkage of multiple machines, often leading to unstable final product quality. Specifically, when introducing new raw materials, even if their surface properties appear similar to existing materials, subtle differences at the molecular level can trigger unexpected changes during processing. Current control systems, which typically only focus on whether the output of each independent stage meets basic standards, struggle to detect these subtle internal changes. This allows latent problems within the material, such as inhomogeneous molecular structures or the quiet accumulation of internal stress, to accumulate throughout the production line. Each processing step may attempt to correct obvious problems, but these corrections often exacerbate unseen internal issues, ultimately significantly reducing the overall performance of the product, such as its stability and fatigue resistance. Even more troubling is that existing systems do not issue any alarms because all easily measurable parameters show normal readings, giving the illusion that production is proceeding smoothly.

[0014] like Figure 1 The diagram illustrates an exemplary flow chart of an automated control method for a rubber processing production line. This application proposes an automated control method for a rubber processing production line, comprising: S10, Obtain process parameter information of the material in the initial processing stage, the process parameter information includes the energy absorption performance and shear response performance of the material, analyze the process parameter information, and determine the initial intrinsic state index of the material; Among them, the energy absorption performance of a material refers to the characteristic of the material absorbing external energy (such as mechanical energy and thermal energy) during processing. This usually reflects the macroscopic and microscopic physical properties of the material, such as viscoelasticity and intermolecular forces. For example, in the internal mixing process, the degree to which rubber materials absorb mechanical shear energy can reflect their degree of plasticization and the breaking and recombination of molecular chains.

[0015] Shear response refers to the deformation and flow characteristics of a material when subjected to shear force, which is closely related to the material's molecular weight distribution, crosslinking density, and filler dispersion state. For example, during extrusion or calendering, the shear thinning or shear thickening behavior of a material can reflect its processing fluidity and internal structural uniformity.

[0016] Initial intrinsic state indices are quantitative characterizations of the internal microstructure and physicochemical properties of materials after the initial processing stage. They comprehensively reflect the processing history and potential inherent deviations of the material in the initial stage. This index is an important basis for adjusting parameters in subsequent processing stages.

[0017] S20, the initial internal state indicators and the energy input information in the initial processing stage are recorded as the processing history of the material, and the processing history is transferred to the subsequent processing stage; Among them, processing history refers to the data set that accompanies the material throughout the entire production line, recording its key processing history and internal state information. It contains all the important information about the material from the initial processing to the current stage, and is the core of achieving cross-stage collaborative control.

[0018] S30, in the subsequent processing stage, based on the received processing history and the real-time process information in the subsequent processing stage, the current internal state of the material is inferred, and then the processing parameters of the subsequent processing stage are adjusted according to the current internal state to eliminate or reduce the internal deviation of the material, and the current internal state is updated to the processing history.

[0019] Intrinsic deviation refers to the difference between the microstructure or physicochemical properties of a material and its ideal state during processing. This deviation may originate from batch differences in raw materials, fluctuations in initial processing parameters, or the influence of environmental factors, and may accumulate in subsequent processing stages, ultimately affecting product performance.

[0020] This application provides an automated control method for a production line of special equipment for rubber processing. Its core lies in the accurate perception, recording and cross-process transmission of the internal state of the material, and the dynamic adjustment of processing parameters based on this.

[0021] Specifically, this method first involves acquiring process parameter information of the material during the initial processing stage. This process parameter information includes the material's energy absorption and shear response characteristics. The process parameter information is then analyzed to determine the material's initial intrinsic state indicators. In the initial processing stage, such as the internal mixing process, process parameter information can be acquired through various methods. For example, sensors installed on the internal mixer can be used to collect data such as torque, power, temperature, and pressure in real time. These data can directly or indirectly reflect the material's energy absorption and shear response characteristics. For instance, torque and power variation curves can reflect energy dissipation and viscosity changes under shear, while temperature and pressure changes may be related to plasticization and molecular chain motion. Furthermore, non-contact methods such as online spectral analysis and acoustic detection can be used to acquire microstructural information of the material, further enriching the process parameter information. After acquiring this process parameter information, in-depth analysis is required to determine the material's initial intrinsic state indicators. For example, statistical methods, such as principal component analysis (PCA) or partial least squares (PLS), can be used to extract key features characterizing the material's intrinsic state from the multidimensional process parameters. Machine learning models, such as support vector machines (SVM) or neural networks, can also be used to establish a mapping relationship between process parameters and the intrinsic state of materials, thereby predicting or calculating initial intrinsic state indices. For example, by analyzing the energy absorption rate and shear stress relaxation curve of materials during the mixing process, the degree of breakage of rubber molecular chains and the formation of cross-linked networks can be inferred and quantified as initial intrinsic state indices.

[0022] Furthermore, the initial intrinsic state indicators and energy input information from the initial processing stage are recorded as the material's processing history, and this processing history is transmitted to subsequent processing stages. Once the initial intrinsic state indicators of the material are determined, they need to be recorded together with the energy input information from the initial processing stage (e.g., total energy input of the internal mixer, shearing work, etc.) as the material's "processing history." This recording can take various forms, such as being stored in the database of the production line's central control system or attached to the material batch in the form of an electronic tag. The processing history can be recorded as a simple text file, a structured database entry, or even a distributed ledger based on blockchain technology to ensure data integrity and traceability. After recording, this processing history needs to be effectively transmitted to subsequent processing stages. For example, data can be transmitted in real time to subsequent processes such as calendering, extrusion, and vulcanization via industrial Ethernet, wireless communication modules, etc. Alternatively, the processing history can be transmitted in tandem via physical media, such as material containers with RFID tags. This transmission mechanism ensures that subsequent stages can obtain the material's "historical information," providing a basis for subsequent intelligent decision-making.

[0023] Furthermore, in subsequent processing stages, based on the received processing history and real-time process information from those stages, the current intrinsic state of the material is inferred. Then, based on this current intrinsic state, the processing parameters of the subsequent stages are adjusted to eliminate or reduce inherent deviations in the material, and the current intrinsic state is updated in the processing history. In subsequent processing stages, such as the calendering process, processing history from upstream stages is received. Simultaneously, this stage also collects its own process information in real time, such as the calender's roll temperature, roll gap, linear speed, material thickness, and surface roughness. Combining this information, the current intrinsic state of the material needs to be inferred. For example, inference algorithms based on physical models or data-driven models can be used to combine the initial intrinsic state indicators from the processing history with the real-time process information from the current stage to predict the material's microscopic characteristics at that stage, such as molecular orientation, crystallinity, and internal stress distribution. For instance, by analyzing the thickness uniformity and surface defects of the material during calendering, combined with the initial intrinsic state recorded in the processing history, the molecular chain orientation and internal stress accumulation of the material during the calendering stage can be inferred. Once the current intrinsic state of the material is determined, processing parameters for subsequent processing stages can be dynamically adjusted based on this state. For example, if an inherent deviation of uneven molecular orientation is detected in the material, the roller speed difference or roller temperature of the calender can be adjusted to promote the rearrangement of molecular chains, thereby eliminating or reducing this deviation. Parameter adjustment strategies can be based on a pre-defined rule base or optimized using adaptive control algorithms or reinforcement learning models. After each adjustment, the current intrinsic state of the material also needs to be updated in the processing history, forming a closed-loop feedback mechanism to ensure that the processing history always reflects the latest state of the material, providing accurate information for downstream processes.

[0024] The advantage of this application lies in its ability to fundamentally solve the problem of product quality instability caused by the accumulation of inherent material deviations. Traditional control systems often only focus on whether the output of each independent link meets basic standards, making it difficult to detect and correct subtle differences and the accumulation of inherent stress at the molecular level of materials. For example, in a traditional rubber production line, when the upstream mixing process makes local compensatory adjustments due to minor differences in raw materials, it may lead to changes in the inherent physical properties of the material that are not directly monitored. This implicit state deviation accumulates step by step in subsequent processes such as calendering and vulcanization, ultimately affecting the long-term stability of the overall product performance. This application infers the continuous evolution of the material's inherent state in real time and feeds this information forward to downstream processes, thereby guiding the downstream system to make predictive, non-compensatory parameter optimization adjustments, fundamentally eliminating or reducing the implicit deviations introduced upstream. This interconnected control method across multiple processing units ensures the stability of the overall performance of the final product and significantly improves the automation level of the production line and product consistency.

[0025] In some embodiments, the step of acquiring process parameter information of the material in the initial processing stage, wherein the process parameter information includes the material's energy absorption performance and shear response performance, and analyzing the process parameter information to determine the initial intrinsic state index of the material includes: Acquire internal mixing process data, environmental data, and raw material pre-scanning feature data, and perform time synchronization and noise reduction processing on the internal mixing process data, environmental data, and raw material pre-scanning feature data; Wavelet transform is performed on the torque and power signals in the processed internal mixing process data to obtain energy in a specific frequency band. Fourier transform is performed on the temperature rise rate signal in the processed internal mixing process data to obtain the main frequency components of the temperature rise rate signal in different time periods. Mel frequency cepstral coefficients are extracted from the acoustic signal in the processed raw material pre-scan feature data to obtain acoustic feature vectors characterizing the raw material properties. A comprehensive feature vector is constructed, which includes the specific frequency band energy, the main frequency components, the acoustic feature vector, and the environmental humidity and temperature in the environmental data. Correlation analysis is performed on the comprehensive feature vector to identify feature combinations that are weakly correlated with environmental changes. Sensitivity analysis was performed on the feature patterns in the identified feature combinations to calculate the correlation coefficients between the feature patterns and environmental humidity and temperature. Features with correlation coefficients lower than a preset threshold are selected, and the selected features are weighted and summed to calculate the initial intrinsic state index.

[0026] Specifically, in the initial processing stage, it is first necessary to acquire multi-source data, including mixing process data, environmental data, and raw material pre-scanning characteristic data. Mixing process data can be understood as various physical quantities collected in real time during the operation of the mixing mill, such as torque, power, temperature, and pressure. Environmental data refers to the environmental conditions surrounding the mixing mill, such as ambient humidity and temperature. Raw material pre-scanning characteristic data refers to data obtained by pre-scanning the raw material using non-contact or contact sensors before it enters the mixing mill, such as acoustic signals and spectral signals. To ensure the accuracy of subsequent analysis, these data from different sources need to be time-synchronized to ensure that all data points are aligned on the time axis. Simultaneously, to eliminate the impact of sensor noise, environmental interference, and other factors on data quality, noise reduction processing is also required, such as using filtering algorithms and wavelet denoising methods.

[0027] The process involves several key steps. First, wavelet transforms are applied to the torque and power signals in the processed internal mixing data. This allows for simultaneous analysis of local signal characteristics in both the time and frequency domains, revealing the energy distribution across different frequency bands. These specific frequency bands effectively characterize the energy absorbed and shear response of the material during the internal mixing process. Second, Fourier transforms are performed on the temperature rise rate signal in the processed internal mixing data to convert the time-domain signal into a frequency-domain signal. This identifies the primary frequency components of the temperature rise rate signal across different time periods, reflecting the dynamic characteristics of changes in the material's internal structure. Third, Mel-frequency cepstral coefficients (MFCCs) are extracted from the acoustic signals in the processed raw material pre-scan feature data to obtain acoustic feature vectors characterizing the raw material's properties. MFCCs, a widely used feature extraction method in speech recognition, effectively capture the spectral envelope information of acoustic signals, thereby reflecting the physical structure and compositional characteristics of the raw material.

[0028] In practical applications, the extracted specific frequency band energy, main frequency components, acoustic feature vectors, and environmental humidity and temperature data are combined to construct a comprehensive feature vector. This comprehensive feature vector includes the dynamic response of the material during processing, the inherent properties of the raw materials, and environmental condition information. To identify feature combinations with weak correlation to environmental changes, correlation analysis is required on this comprehensive feature vector. Correlation analysis aims to discover the interrelationships between different features in the dataset, thereby distinguishing the inherent characteristics of materials that are less affected by environmental factors.

[0029] Furthermore, sensitivity analysis is performed on the feature patterns within the identified feature combinations to quantify the correlation between these feature patterns and ambient humidity and temperature. By calculating the correlation coefficients between feature patterns and ambient humidity and temperature, the strength of the influence of environmental factors on each feature pattern can be accurately assessed.

[0030] Ultimately, features with correlation coefficients below a preset threshold are selected. These features are considered to have a weaker correlation with environmental changes and are more likely to reflect the stable internal state of the material. The selected features are then weighted and summed to comprehensively consider the contribution of different features to the internal state of the material, thereby calculating an index that can accurately characterize the initial internal state of the material.

[0031] Through the above technical solution, this application enables the accurate and robust determination of the initial intrinsic state indicators of materials. Compared with traditional methods that rely solely on a single process parameter or fail to adequately consider environmental factors, this application significantly improves the ability to capture the energy absorption performance, shear response performance, and raw material characteristics of materials by integrating mixing process data, environmental data, and raw material pre-scanning characteristic data, and employing advanced signal processing and feature extraction techniques. In particular, by performing correlation analysis and sensitivity analysis on the comprehensive feature vector, interference caused by environmental changes is effectively identified and eliminated, making the determined initial intrinsic state indicators more accurately reflect the inherent microstructure and performance potential of the material. This provides a more reliable basis for precise control in subsequent processing stages, ultimately contributing to improved quality consistency and production efficiency of rubber products.

[0032] For example, suppose that on a rubber tire production line, an initial condition assessment is required for a new batch of rubber compounds.

[0033] First, during the operation of the internal mixer, real-time data on the mixing process, such as torque, power, and chamber temperature, are collected. Simultaneously, environmental sensors acquire ambient humidity and temperature within the workshop. Before the rubber compound is fed into the internal mixer, an ultrasonic sensor pre-scans the raw rubber block to obtain its acoustic signals. All collected data is sent to the data processing unit for time synchronization, ensuring all data points are aligned on the same timeline. Subsequently, noise reduction processing is performed on the data, for example, using Kalman filtering to remove sensor noise.

[0034] Next, wavelet transforms were performed on the torque and power signals in the processed mixing process data. For example, the Daubechies wavelet basis function was used to extract the energy in specific frequency bands such as 0-5Hz and 5-20Hz. These energies characterize the macroscopic flowability and microscopic shear response of the compound, respectively. Simultaneously, Fourier transforms were performed on the cavity temperature rise rate signal to identify its main frequency components in time periods such as 0.1-0.5Hz and 0.5-1Hz. These components reflect the dynamic characteristics of the crosslinking reaction and thermodynamic changes within the compound. For the acoustic signal of the raw material pre-scan, Mel-frequency cepstral coefficients were extracted to obtain a 13-dimensional acoustic feature vector, which can characterize the physical properties of the raw material block, such as hardness and viscosity.

[0035] Subsequently, the extracted specific frequency band energy, the main frequency components of the temperature rise rate, the acoustic feature vector, and the ambient humidity and temperature are combined into a high-dimensional comprehensive feature vector. Correlation analysis is then performed on this comprehensive feature vector, such as using partial least squares (PLS) or canonical correlation analysis (CCA), to identify feature combinations with weak correlations to changes in ambient humidity and temperature. For example, it was found that certain acoustic features related to the internal structure of the rubber compound and low-frequency torque energy had low correlations with environmental factors.

[0036] Furthermore, sensitivity analysis was performed on the feature patterns within these identified feature combinations. For example, the Pearson correlation coefficient between each feature pattern and ambient humidity and temperature was calculated. Assuming a preset threshold of 0.3, all features with correlation coefficients below 0.3 were selected. These selected features were considered to be more stable characteristics reflecting the inherent intrinsic state of the compound.

[0037] Finally, these selected features are weighted and summed, for example, by quantifying the importance of different features based on weights obtained from expert experience or machine learning model training. This weighted summation yields a single numerical value, which is the initial intrinsic state index of the batch of rubber compound. This index will be used as part of the processing history and passed on to subsequent processing stages such as calendering and vulcanization, guiding the adjustment of subsequent processing parameters.

[0038] In some embodiments, the step of performing correlation analysis on the comprehensive feature vector to identify feature combinations that are weakly correlated with environmental changes includes: Perform feature pattern separation operation on the comprehensive feature vector to weaken the interference signal caused by environmental factors or their interaction with materials; Principal component analysis is performed on the comprehensive feature vector after feature pattern separation to identify the orthogonal directions with the largest variance in the data and obtain multiple principal components. Calculate the correlation coefficient between each principal component and the environmental data, and select the principal components whose correlation coefficient with the environmental data is lower than a preset threshold as feature combinations representing the inherent microstructural differences of the materials.

[0039] Specifically, the feature pattern separation operation aims to distinguish the signal patterns in the comprehensive feature vector caused by environmental factors or their interaction with the material from the inherent feature patterns of the material itself, using a specific algorithm or model. Its purpose is to remove or reduce environmental noise and interference to the greatest extent possible before subsequent analysis, allowing the subsequent principal component analysis to focus more on the essential properties of the material. For example, this can be achieved using independent component analysis (ICA), blind source separation (BSS), or other signal decoupling techniques.

[0040] Principal Component Analysis (PCA) performed on the composite eigenvectors after feature pattern separation can be understood as a dimensionality reduction technique used to transform high-dimensional data into a low-dimensional space while preserving the main information. Specifically, PCA projects the original data onto a new coordinate system through orthogonal transformation, ensuring that the new coordinate axes (i.e., the principal components) are the directions of maximum data variance, and that the principal components are mutually orthogonal. This allows for the identification of the orthogonal directions of maximum variance in the data, yielding multiple principal components that effectively capture the main variability in the data.

[0041] In practical applications, the correlation coefficient between each principal component and the environmental data is calculated to quantify the degree of association between each principal component and environmental factors. Correlation coefficients such as Pearson correlation coefficient and Spearman correlation coefficient can be used. By selecting principal components with correlation coefficients below a preset threshold, it can be ensured that the selected principal components have a low correlation with environmental changes, thus more accurately representing the inherent microstructural differences of the materials. The preset threshold can be set according to the actual application scenario and the requirements for environmental sensitivity; for example, it can be set to 0.1, 0.2, etc.

[0042] Through the above technical solution, this application effectively solves the problem that traditional correlation analysis struggles to accurately identify inherent microstructural differences in materials under complex environmental interference. The feature pattern separation operation significantly improves data purity, providing more reliable input for principal component analysis. Therefore, the identified principal components can more accurately characterize the intrinsic properties of the material, rather than being products of the environment or its interactions. Finally, through correlation coefficient screening, the high independence of the selected feature combinations from environmental changes is ensured, resulting in higher accuracy and stability of the determined initial intrinsic state indicators. This provides a more reliable basis for parameter adjustments in subsequent processing stages, thereby improving the overall accuracy and adaptability of the automated control of rubber processing equipment production lines.

[0043] For example, suppose that during the internal mixing process, significant fluctuations in ambient temperature and humidity have a complex impact on the torque and power signals of the internal mixer, while batch-to-batch variations in the raw materials are also reflected in the acoustic characteristics. First, after preprocessing the collected internal mixing process data, environmental data, and raw material pre-scan feature data, a comprehensive feature vector incorporating this information is constructed. Next, to mitigate environmental interference, a pre-trained multilayer perceptron (MLP) can be used as a feature pattern separator. This MLP is trained to identify and separate specific interference patterns generated by the interaction between ambient temperature, ambient humidity, and materials. For instance, when the ambient temperature rises, the internal mixing torque may exhibit a specific drift pattern; the MLP can learn and "subtract" this pattern from the original torque signal, thereby obtaining a purer, inherent torque response of the material.

[0044] After feature pattern separation, principal component analysis (PCA) is performed on the processed composite eigenvectors. For example, PCA might identify the first three principal components, explaining 80%, 10%, and 5% of the total variance, respectively. The first principal component might primarily reflect the basic viscosity characteristics of the rubber formulation, the second might be related to the uniformity of filler dispersion, and the third might be related to some microstructural defect. Subsequently, Pearson correlation coefficients between these three principal components and ambient temperature and humidity are calculated. If the correlation coefficients between the first principal component and ambient temperature are found to be 0.05 and 0.03, respectively; the second principal component 0.08 and 0.06; and the third principal component 0.35 and 0.28, a preset threshold of 0.1 is set. Then, the first and second principal components will be selected because their correlation coefficients with environmental data are both below 0.1, indicating that they primarily represent inherent microstructural differences in the material, while the third principal component may still contain significant environmental influences and is therefore excluded. Ultimately, the two selected principal components will serve as a combination of features representing the inherent microstructural differences of the materials, and will be used for subsequent sensitivity analysis and calculation of initial intrinsic state indices.

[0045] In some embodiments, the step of performing feature pattern separation on the integrated feature vector to weaken interference signals caused by environmental factors or their interaction with materials includes: A multi-layer sensing network is established, using the refining process data, the environmental data, and the raw material pre-scan feature data as inputs; The multilayer sensing network is trained to identify interference signal patterns generated by the interaction between environmental factors and materials; The inherent characteristics of the material are reconstructed from the comprehensive feature vector using the multilayer sensing network.

[0046] Specifically, the multilayer sensing network is a feedforward artificial neural network, which typically includes at least one input layer, one output layer, and one or more hidden layers. This network can achieve pattern recognition and feature extraction by learning the complex nonlinear relationship between input and output. The mixing process data may include, but is not limited to, parameters such as torque, power, temperature, and pressure monitored in real time during the mixing process. The environmental data may include real-time environmental monitoring values ​​such as humidity and temperature in the processing workshop. The raw material pre-scan feature data may include feature vectors characterizing the physicochemical properties of the raw materials obtained through acoustic signal analysis, spectral analysis, or other non-destructive testing techniques. When establishing the multilayer sensing network, its specific network structure, such as the number of hidden layers, the number of neurons in each layer, and the activation function used, can be optimized according to the complexity of the actual data and the required model accuracy.

[0047] This process involves iteratively optimizing the network using a large amount of historical processing data to train a multilayer sensing network. This historical data typically includes refining process data, environmental data, and raw material pre-scanning feature data collected under different environmental conditions. The training aims to enable the network to accurately learn and distinguish signal patterns caused by environmental factors and their interactions with materials—i.e., interference signal patterns. This training process is usually implemented using backpropagation and gradient descent optimizers to minimize the error between the network's predicted interference signals and the actual interference signals, thereby enabling the network to identify and separate interference.

[0048] In practical applications, after the network training is complete and reaches the expected performance, a new comprehensive feature vector is input into the trained network. Based on the interference signal patterns it has learned, the network "strips" or "filters out" the interference components caused by environmental factors and their interactions from the input comprehensive feature vector, thereby obtaining a purer and more accurate reflection of the inherent characteristics of the material's microstructure and properties. This process can be understood as the network performing an intelligent denoising and feature extraction on the original comprehensive feature vector, aiming to highlight the intrinsic properties of the material itself.

[0049] Through the above technical solution, this application can significantly improve the accuracy and robustness of feature pattern separation. Specifically, by utilizing multilayer sensing networks to process complex nonlinear interactions, the interference of environmental factors on the assessment of the material's intrinsic state can be more thoroughly reduced, resulting in purer and more accurate determination of the material's inherent characteristics. This not only improves the reliability of the initial intrinsic state indicators but also provides a more solid data foundation for inferring the material's current intrinsic state and precisely adjusting processing parameters in subsequent processing stages. This helps to more effectively eliminate or reduce the material's inherent deviations, optimizing the quality and production efficiency of rubber products.

[0050] For example, suppose that during the rubber mixing process, it is necessary to accurately assess the intrinsic state of the rubber material. First, a large amount of historical mixing batch data is collected, including mixing process data such as torque, power, and temperature curves of the mixer under different ambient temperatures and humidity levels, as well as pre-scan feature data of the raw materials obtained through near-infrared spectroscopy (NIR) or ultrasonic detection. This data, along with corresponding environmental data (ambient temperature, ambient humidity), is used to train a multilayer perceptron. This network is designed with three hidden layers, each containing 128, 64, and 32 neurons respectively, and employs the ReLU activation function. The network's input layer receives the mixing process data, environmental data, and pre-scan feature data of the raw materials after time synchronization and noise cancellation processing. The network's output layer is then trained to predict or reconstruct the inherent characteristics of the material unaffected by environmental disturbances. During training, the network weights are adjusted by comparing the differences between the features reconstructed by the network and known true inherent characteristics of the material (e.g., obtained through standard laboratory tests). Once the network training is complete and reaches the expected accuracy, in a new batch of intensive mixing, real-time data on the mixing process, environmental data, and pre-scanned feature data of the raw materials are acquired and input into the trained multilayer perceptron. Based on the complex nonlinear patterns it has learned, the network automatically identifies and filters out interference from environmental temperature and humidity on the material's energy absorption and shear response, thereby outputting a purer and more accurate feature vector characterizing the inherent microstructure and properties of the rubber material in the current batch. This feature vector is then used to construct a comprehensive feature vector and for subsequent correlation analysis and intrinsic state index calculations.

[0051] In some embodiments, the step of performing principal component analysis on the comprehensive feature vector after feature pattern separation to identify the orthogonal direction with the largest variance in the data and obtain multiple principal components includes: Perform a nonlinear transformation on the composite feature vector to convert it into a high-dimensional feature space; Principal component analysis is performed in the high-dimensional feature space to identify the orthogonal directions with the largest variance in the data, thereby obtaining multiple principal components.

[0052] Specifically, nonlinear transformation is achieved by mapping the original composite feature vector from a low-dimensional space to a high-dimensional feature space using nonlinear functions. This transformation allows nonlinear relationships that were difficult to separate or identify in the low-dimensional space to be converted into linearly separable or easier-to-analyze relationships in the high-dimensional feature space. For example, this nonlinear transformation can be performed using kernel functions (such as Gaussian kernels, polynomial kernels, etc.) or through feature extraction using multi-layer neural networks. The high-dimensional feature space refers to the more abstract space in which the data resides after the nonlinear transformation. In this space, the nonlinear relationships between data points may be "straightened out" or "expanded," allowing linear methods to process them more effectively.

[0053] Furthermore, after the data is mapped to a high-dimensional feature space, standard principal component analysis (PCA) algorithms are applied to these high-dimensional features. This allows for the more effective capture of the main variance directions in the data within the high-dimensional feature space. These directions may exhibit linear relationships in the high-dimensional space, but complex nonlinear relationships in the original low-dimensional space. By performing PCA in the high-dimensional space, features that better represent the inherent microstructural differences of the material can be extracted, thereby improving the accuracy of the initial intrinsic state indicators.

[0054] Through the above technical solution, this application can more effectively handle the complex nonlinear data relationships existing in the processing of rubber materials. Compared with performing linear principal component analysis directly in the original space, introducing nonlinear transformation can better linearize the inherent microstructural differences of the material in a high-dimensional space, thereby enabling principal component analysis to capture deeper and more essential characteristic patterns. As a result, the determined initial intrinsic state indices will be more accurate and robust, more precisely reflecting the true state of the material, providing a more reliable basis for parameter adjustments in subsequent processing stages, and further improving the accuracy and effectiveness of automated control.

[0055] For example, kernel principal component analysis (PCA) can be used before performing principal component analysis on the composite eigenvector. Specifically, a suitable kernel function, such as the Gaussian kernel (RBF kernel), is first selected. This kernel function can nonlinearly map the original composite eigenvector to an infinite-dimensional feature space. Then, in this high-dimensional feature space, the covariance matrix of the data is calculated, and eigenvalue decomposition is performed to obtain the principal components. By adjusting the parameters of the Gaussian kernel function (such as the width parameter sigma), it can be adapted to the nonlinear characteristics of different batches of materials. This method can effectively reveal the nonlinear patterns hidden in the material's energy absorption and shear response performance, thereby extracting more representative principal components in the high-dimensional feature space for subsequent calculations of initial intrinsic state indices.

[0056] In some embodiments, the step of performing a nonlinear transformation on the composite feature vector to convert the composite feature vector to a high-dimensional feature space includes: Based on the energy absorption performance, shear response performance, and raw material pre-scanning characteristic data in the comprehensive feature vector, the heterogeneity and nonlinear coupling strength of the current batch of materials are evaluated to obtain the evaluation results; Based on the evaluation results and raw material pre-scanning characteristic data, one or more nonlinear transformation functions that match the characteristics of the current batch of materials are selected from a preset set of nonlinear transformation functions. For the selected nonlinear transformation function, the parameters of the nonlinear transformation function are adjusted based on the energy absorption performance, shear response performance, and raw material pre-scanning characteristic data to obtain the target nonlinear transformation function; By using the target nonlinear transformation function, a nonlinear transformation is performed on the comprehensive feature vector, transforming the comprehensive feature vector into a high-dimensional feature space.

[0057] Specifically, by analyzing the energy absorption performance, shear response performance, and pre-scanning characteristic data of the raw materials during processing, the heterogeneity (degree of heterogeneity) of the component distribution within the current batch of materials and the complexity of the interactions between components (nonlinear coupling strength) are quantified, thereby assessing the degree of heterogeneity and the strength of nonlinear coupling in the current batch of materials. For example, these data can be processed using statistical methods (such as analysis of variance and entropy calculation) or machine learning models (such as cluster analysis and regression models) to obtain quantitative evaluation results. These evaluation results aim to provide a basis for subsequent selection and adjustment of the nonlinear transformation function.

[0058] The "preset set of nonlinear transformation functions" can include various types of nonlinear transformation functions, such as polynomial kernel functions, radial basis function (RBF) kernel functions, and sigmoid kernel functions. These functions each have different nonlinear mapping capabilities and applicable ranges. Based on the aforementioned evaluation results, for example, if the evaluation results show that the material has high heterogeneity and strong nonlinear coupling, a kernel function capable of handling complex nonlinear relationships may be selected; if the heterogeneity is low, a relatively simple function may be selected. The selection process can be based on predefined rules, lookup tables, or implemented through machine learning classifiers.

[0059] In practical applications, the selected nonlinear transformation function is finely configured to more accurately adapt to the specific characteristics of the current batch of materials, thereby adjusting the parameters of the nonlinear transformation function. For example, for a polynomial kernel function, its order can be adjusted; for a radial basis function, its width parameter (gamma value) can be adjusted. The parameter adjustment is based on the material's energy absorption performance, shear response performance, and pre-scanned characteristic data of the raw materials. Optimal parameter combinations can be found through optimization algorithms (such as grid search, genetic algorithms, and Bayesian optimization) to maximize the discriminative power of the transformed features or minimize the reconstruction error.

[0060] Therefore, the nonlinear transformation function, after parameter adjustment, is applied to the original comprehensive feature vector, mapping it to a higher-dimensional feature space. In this higher-dimensional space, material properties that were difficult to distinguish in the lower-dimensional space may become easier to separate and identify, thus providing higher-quality input data for subsequent principal component analysis.

[0061] Through the above technical solution, this application can dynamically adjust the nonlinear transformation strategy according to the actual characteristics of different batches of rubber materials, avoiding the problem of inaccurate feature representation caused by using a general transformation function. This allows the intrinsic state of the material to be more accurately characterized in the high-dimensional feature space, significantly improving the effectiveness of subsequent principal component analysis in identifying inherent microstructural differences in the material. Therefore, this application can more accurately determine the initial intrinsic state indicators of the material, providing a more reliable basis for parameter adjustment in subsequent processing stages, thereby effectively eliminating or reducing the intrinsic deviation of the material and improving the automation control accuracy and product quality stability of the rubber processing equipment production line.

[0062] For example, suppose that during the rubber mixing process, the energy absorption performance, shear response performance, and pre-scanning characteristic data of the current batch of material are first acquired. By analyzing this data, for example using a pre-trained neural network model, it is assessed that the current batch of material has a high degree of heterogeneity and exhibits significant nonlinear coupling strength. Based on this assessment result, from a pre-defined set of nonlinear transformation functions, such as a set including polynomial kernel functions, radial basis function (RBF) kernel functions, and sigmoid kernel functions, the RBF kernel function is selected as the initial matching function because it performs well in handling high-dimensional nonlinear data.

[0063] Furthermore, for the selected RBF kernel function, the width parameter (gamma value) of the RBF kernel function is adjusted using an optimization algorithm (e.g., cross-validation combined with grid search) based on the energy absorption performance, shear response performance, and pre-scanned characteristic data of the current batch of materials. For example, if the energy absorption performance of the material is found to fluctuate significantly, a smaller gamma value may be needed to capture more refined local features; conversely, if the fluctuation is gradual, a larger gamma value may be used to focus on the global structure. After optimization, a target RBF kernel function customized for the characteristics of the current batch of materials is obtained. Finally, this target RBF kernel function is used to perform a nonlinear transformation on the comprehensive feature vector, mapping it to a high-dimensional feature space. In this high-dimensional space, the differences in the microstructure of the material are effectively amplified and separated, providing a clearer and more discriminative feature representation for subsequent principal component analysis, thereby ensuring the accuracy of the initial intrinsic state indicators.

[0064] In some embodiments, the step of performing a nonlinear transformation on the composite feature vector to convert the composite feature vector to a high-dimensional feature space includes: The nonlinear coupling strength of the current batch of materials is evaluated based on the energy absorption performance and shear response performance in the comprehensive feature vector. Based on the nonlinear coupling strength, one or more kernel functions are selected from a preset set of kernel functions; For a selected kernel function, the parameters of the kernel function are adjusted based on the energy absorption performance and shear response performance. The parameters include the width parameter of the kernel function or the order of the polynomial kernel function to obtain the target kernel function. Using the target kernel function, a nonlinear mapping is performed on the comprehensive feature vector to transform the comprehensive feature vector into a high-dimensional feature space; In the high-dimensional feature space, redundant feature dimensions are identified and removed based on the energy absorption performance and shear response performance, thereby optimizing the dimensionality of the high-dimensional feature space.

[0065] Specifically, assessing the nonlinear coupling strength of the current batch of materials refers to quantifying the complexity of the interactions between its internal components and with the external environment by analyzing the material's energy absorption and shear response during processing. For example, the coupling strength can be assessed by calculating the mutual information and correlation coefficients between these performance data, or by employing a nonlinear regression model, providing a basis for selecting an appropriate nonlinear transformation method. The kernel function set can be understood as a predefined set of mathematical functions used to map data in a low-dimensional space to a high-dimensional feature space, thus making originally linearly inseparable data linearly separable in the high-dimensional space. Common kernel functions include radial basis function (RBF) kernels, polynomial kernels, and sigmoid kernels. Based on the assessed nonlinear coupling strength, one or more kernel functions that best capture the nonlinear characteristics of the current batch of materials can be intelligently selected from this set. For example, for materials with high nonlinear coupling strength, an RBF kernel or a higher-order polynomial kernel might be chosen.

[0066] In practical applications, adjusting the parameters of the kernel function is a crucial step in ensuring optimal nonlinear transformation results. For example, for the RBF kernel, its width parameter determines the range of influence of data points in the high-dimensional space; for the polynomial kernel, its order determines the complexity of the polynomial mapping. These parameters can be iteratively optimized based on energy absorption and shear response performance, for example, through cross-validation or grid search, to maximize the discriminative power of the material's intrinsic state, thereby obtaining the target kernel function best suited to the current material characteristics. Furthermore, performing a nonlinear mapping on the comprehensive feature vector using the target kernel function transforms the original comprehensive feature vector into a higher-dimensional feature space through the selected and adjusted kernel function. This mapping can reveal nonlinear structures and patterns that are difficult to discover in the original low-dimensional space, providing richer and more discriminative feature representations for subsequent principal component analysis. In addition, after the high-dimensional mapping is completed, by analyzing the correlation or contribution of the high-dimensional features to the material's energy absorption and shear response performance, redundant dimensions that contribute little to the characterization of the material's intrinsic state or are highly correlated with other features are identified, thus enabling the identification and removal of redundant feature dimensions and optimizing the dimensionality of the high-dimensional feature space. For example, feature selection algorithms (such as L1 regularization and recursive feature elimination) or information gain-based methods can be used to identify and remove these redundant dimensions. The aim is to reduce the complexity of high-dimensional spaces, reduce computational burden, and improve the efficiency and accuracy of subsequent principal component analysis, thereby avoiding the "curse of dimensionality" problem.

[0067] Through the above technical solution, this application can adaptively select and optimize the kernel function and its parameters of the nonlinear transformation according to the actual nonlinear coupling strength of the current batch of materials. This ensures that when mapping the comprehensive feature vector to a high-dimensional feature space, the complex nonlinear characteristics and intrinsic microstructural differences of the materials can be captured more accurately and effectively. Compared with the general nonlinear transformation that may be used in basic technical solutions, this technical solution significantly improves the ability of the high-dimensional feature space to represent the intrinsic state of the materials, avoiding information distortion or redundancy caused by mismatched transformation functions. In addition, by identifying and removing redundant feature dimensions in the high-dimensional feature space, the data dimensionality is effectively reduced. This not only reduces the computational burden but also improves the efficiency and accuracy of subsequent principal component analysis, making the final determined intrinsic state indicators of the materials more accurate and reliable. This provides a more solid data foundation for the adjustment of subsequent processing parameters.

[0068] For example, suppose that during the rubber mixing process, it is necessary to evaluate the intrinsic state of rubber materials with different formulations. First, the energy absorption and shear response of the materials during the mixing process are continuously monitored. For example, for rubber formulations containing a high proportion of fillers, their energy absorption and shear response may exhibit stronger nonlinear coupling characteristics. Based on this real-time data, the nonlinear coupling strength of the current batch of materials is evaluated. Based on the evaluation results, if the material exhibits a high nonlinear coupling strength, a radial basis function (RBF) kernel may be preferentially selected from a preset set of kernel functions. Subsequently, based on the current energy absorption and shear response performance, the width parameter (gamma value) of the RBF kernel is finely adjusted through iterative optimization algorithms (such as grid search combined with cross-validation) to ensure that the kernel function can optimally map the comprehensive feature vector to a high-dimensional feature space, thereby maximizing the distinguishability of the intrinsic states between different batches of materials. After completing the nonlinear mapping, the correlation between each feature dimension and the material's energy absorption and shear response performance is further analyzed in the generated high-dimensional feature space. For example, by calculating the mutual information or correlation coefficient between each high-dimensional feature dimension and these representations, dimensions with low contribution or high redundancy with other features are identified. These redundant dimensions are then removed, thereby optimizing the dimensionality of the high-dimensional feature space and ensuring that the final feature set used for principal component analysis is concise and efficient, more accurately reflecting the inherent microstructural differences of the material. In this way, even with complex and varied rubber formulations, this technical solution can provide accurate assessment of the intrinsic state.

[0069] In some embodiments, the step of identifying and removing redundant feature dimensions and optimizing the dimensions of the high-dimensional feature space based on the energy absorption performance and shear response performance includes: Continuously monitor the real-time fluctuation patterns of the material's energy absorption and shear response performance; Based on the real-time fluctuation pattern, feature dimension weight adjustment coefficients are generated and input into the weight allocation mechanism to dynamically quantify the contribution of each high-dimensional feature dimension to the energy absorption performance and shear response performance. Identify and remove feature dimensions whose contribution is below a preset threshold under a dynamically adjusted weight allocation mechanism.

[0070] Specifically, by using sensor arrays or high-frequency data acquisition systems, real-time data on the energy absorption and shear response of materials during processing are acquired and continuously analyzed to capture their dynamic characteristics over time, such as instantaneous peak values, trend changes, and periodic oscillations. This enables continuous monitoring of the real-time fluctuation patterns of the material's energy absorption and shear response. Energy absorption can be understood as the rate or total amount of energy absorbed by the material during processing, while shear response can be understood as the deformation or flow characteristics of the material when subjected to shear force.

[0071] Specifically, based on the real-time fluctuation pattern, feature dimension weight adjustment coefficients are generated and input into the weight allocation mechanism to dynamically quantify the contribution of each high-dimensional feature dimension to the energy absorption and shear response performance. This means that the importance of different high-dimensional feature dimensions is intelligently adjusted according to the dynamic characteristics of the material's real-time changes. For example, when the material's energy absorption performance fluctuates drastically, feature dimensions related to energy absorption may be assigned higher weights to ensure that this key information is fully considered during the redundant feature removal process. The weight allocation mechanism can be an adaptive algorithm based on a machine learning model (such as a neural network or support vector machine), which can dynamically calculate and update the weight of each feature dimension based on real-time data input. The degree of contribution refers to the importance or influence of a certain high-dimensional feature dimension in accurately describing the material's energy absorption and shear response performance.

[0072] In practical applications, at each time step or within a preset evaluation period, the actual contribution of each high-dimensional feature dimension to the material's energy absorption and shear response performance is evaluated based on dynamically updated weights. If the contribution of a certain feature dimension consistently falls below a preset threshold, that feature dimension is considered redundant or unimportant and can be removed from the high-dimensional feature space, thereby optimizing and reducing the dimensionality of the feature space. The preset threshold can be determined based on experience, experiments, or through cross-validation to balance the dimensionality of the feature space and model performance.

[0073] Through the above technical solution, this application overcomes the limitations of traditional static feature dimension optimization methods in handling dynamically changing rubber processing processes. By introducing continuous real-time monitoring of the fluctuation patterns of material energy absorption and shear response, and dynamically adjusting the weights of feature dimensions based on this, the feature space optimization process can adaptively respond to changes in material state and processing environment. This significantly improves the accuracy and real-time performance of redundant feature identification, ensuring that the feature dimensions retained in the high-dimensional feature space are always the most relevant and contributing parts to the material's intrinsic state. Compared to the above methods, this application has higher adaptability and robustness, and can more effectively eliminate or reduce the inherent deviations of materials, thereby improving the accuracy and stability of automated control in rubber processing equipment production lines.

[0074] For example, it is assumed that during the rubber mixing process, the energy absorption and shear response of the material are affected by a combination of factors such as the speed of the mixer, the feeding sequence, and the ambient temperature, exhibiting nonlinear real-time fluctuations.

[0075] First, high-frequency sensors will be deployed to continuously collect data such as torque, power, material temperature, and shear force from the internal mixer. This data will then be preprocessed to extract real-time fluctuation patterns in the material's energy absorption and shear response. For example, by performing sliding window analysis on the torque and power signals, statistical characteristics such as mean, variance, and kurtosis will be calculated to characterize the intensity and frequency of fluctuations.

[0076] Secondly, based on these real-time fluctuation patterns, a pre-trained deep learning model (e.g., a recurrent neural network or a Transformer model) is used to generate feature dimension weight adjustment coefficients. During training, this model learns how to assign different weights to each feature dimension in the high-dimensional feature space according to different fluctuation patterns. For example, when an abnormal peak is detected in the material's energy absorption performance, feature dimensions related to material viscosity, cross-linking degree, etc., may be assigned higher weights, because these peaks may indicate significant changes in the material's internal structure.

[0077] These weighting adjustment coefficients are then input into a weighting allocation mechanism that dynamically calculates the contribution of each high-dimensional feature dimension to the current material energy absorption and shear response performance. For example, weighted correlation analysis or information gain-based evaluation methods can be used.

[0078] Finally, a preset threshold, such as 0.05, is set. Under the dynamically adjusted weight allocation mechanism, if the contribution of a certain high-dimensional feature dimension to the material's energy absorption and shear response performance is below 0.05 for multiple consecutive time steps, then that feature dimension is identified as a redundant feature and removed. In this way, the high-dimensional feature space can be dynamically optimized, always maintaining its conciseness and effectiveness, thereby providing more accurate and efficient input for subsequent inference of the material's intrinsic state and adjustment of processing parameters.

[0079] In some embodiments, the step of generating feature dimension weight adjustment coefficients based on the real-time fluctuation pattern and inputting them into the weight allocation mechanism to dynamically quantify the contribution of each high-dimensional feature dimension to the energy absorption performance and shear response performance includes: Continuously monitor the real-time fluctuation patterns of the material's energy absorption and shear response performance; Abnormal signal detection is performed on the real-time fluctuation pattern to identify and isolate abnormal spikes or drifts caused by occasional external interference, thereby obtaining the target real-time fluctuation pattern. Based on the target's real-time fluctuation pattern, feature dimension weight adjustment coefficients are generated and input into the weight allocation mechanism to dynamically quantify the contribution of each high-dimensional feature dimension to the energy absorption performance and shear response performance.

[0080] Specifically, statistical methods, machine learning algorithms, or rule-based methods are employed to identify points or regions in a data sequence that deviate from normal behavioral patterns, thereby achieving anomaly signal detection. For example, a threshold-based method can be used, setting a dynamic or static threshold; when data points in a real-time fluctuation pattern exceed this threshold, they are judged as anomalies. Alternatively, time series analysis methods such as moving averages and exponential smoothing can be used to detect anomalies by comparing the deviation between current data points and historical data points. Furthermore, unsupervised learning algorithms such as Isolation Forest and Local Outlier Factor (LOF) can be used to automatically identify anomalous patterns in the data. The aim is to accurately identify instantaneous or persistent disturbances caused by factors other than inherent material properties.

[0081] Upon detecting abnormal signals, appropriate processing measures are taken to remove or correct them from the original data. For example, for abnormal spikes, methods such as median filtering, mean filtering, or wavelet denoising can be used for smoothing, or the spike data points can be directly replaced with the average or interpolation of adjacent normal data points. For abnormal drift, techniques such as trend decomposition, high-pass filtering, or baseline correction can be used to eliminate their influence, ensuring that the data basis for subsequent analysis is stable and accurate. The goal is to ensure that the obtained real-time fluctuation patterns accurately reflect the intrinsic changes of the material, rather than external interference.

[0082] In practical applications, the target real-time fluctuation mode refers to the purer and more stable real-time fluctuation data of material energy absorption and shear response performance after abnormal signal detection and isolation, removing occasional external interference (such as spikes and drifts). This mode more accurately reflects the true dynamic behavior and internal state changes of materials during processing.

[0083] Through the above technical solution, this application can effectively eliminate or reduce the impact of occasional external interference on the real-time fluctuation pattern data of materials, thereby obtaining more realistic and stable dynamic behavior data of materials. Therefore, when generating feature dimension weight adjustment coefficients, calculations can be performed based on more reliable data, significantly improving the accuracy and robustness of the weight adjustment coefficients. This further ensures the accuracy of subsequent redundant feature dimension identification and removal, avoiding erroneous judgments caused by data noise, thus optimizing the dimensionality of the high-dimensional feature space. This allows the selected feature dimensions to more accurately and effectively characterize the inherent microstructural differences of materials, thereby improving the reliability and accuracy of the entire automated control method in inferring the internal state of materials and adjusting processing parameters.

[0084] This application also proposes an automated control system for a rubber processing equipment production line, such as... Figure 2 As shown, an automated control system 100 for a rubber processing production line includes: The information acquisition module 10 is used to acquire process parameter information of the material in the initial processing stage. The process parameter information includes the energy absorption performance and shear response performance of the material. The process parameter information is analyzed to determine the initial intrinsic state index of the material. The history recording and transmission module 20 is used to record the initial internal state indicators and the energy input information in the initial processing stage as the processing history of the material, and to transmit the processing history to the subsequent processing stage. The state inference and parameter adjustment module 30 is used to infer the current internal state of the material in subsequent processing stages based on the received processing history and real-time process information in subsequent processing stages, and then adjust the processing parameters of the subsequent processing stages based on the current internal state to eliminate or reduce the internal deviation of the material, and update the current internal state to the processing history.

[0085] The system in this application achieves precise perception of the material's internal state through an information acquisition module, seamless information transmission through a history recording and transmission module, and predictive, non-compensatory parameter optimization and adjustment through a state inference and parameter adjustment module. This eliminates or reduces hidden deviations introduced upstream at the system level. This integrated, interconnected control system spanning multiple processing units ensures the stability of the overall performance of the final product and significantly improves the automation level of the production line and product consistency.

[0086] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. An automated control method for a rubber processing production line, characterized in that, include: Obtain process parameter information of the material in the initial processing stage, including the material's energy absorption performance and shear response performance; analyze the process parameter information to determine the material's initial intrinsic state index. The initial intrinsic state indicators and energy input information in the initial processing stage are recorded as the processing history of the material, and the processing history is transferred to the subsequent processing stage. In subsequent processing stages, based on the received processing history and real-time process information in the subsequent processing stages, the current internal state of the material is inferred. Then, based on the current internal state, the processing parameters of the subsequent processing stages are adjusted to eliminate or reduce the inherent deviation of the material, and the current internal state is updated to the processing history.

2. The automated control method for a rubber processing production line according to claim 1, characterized in that, The step of acquiring process parameter information of the material in the initial processing stage, including the material's energy absorption performance and shear response performance, and analyzing the process parameter information to determine the initial intrinsic state index of the material includes: Acquire internal mixing process data, environmental data, and raw material pre-scanning feature data, and perform time synchronization and noise reduction processing on the internal mixing process data, environmental data, and raw material pre-scanning feature data; Wavelet transform is performed on the torque and power signals in the processed internal mixing process data to obtain energy in a specific frequency band. Fourier transform is performed on the temperature rise rate signal in the processed internal mixing process data to obtain the main frequency components of the temperature rise rate signal in different time periods. Mel frequency cepstral coefficients are extracted from the acoustic signal in the processed raw material pre-scan feature data to obtain acoustic feature vectors characterizing the raw material properties. A comprehensive feature vector is constructed, which includes the specific frequency band energy, the main frequency components, the acoustic feature vector, and the environmental humidity and temperature in the environmental data. Correlation analysis is performed on the comprehensive feature vector to identify feature combinations that are weakly correlated with environmental changes. Sensitivity analysis was performed on the feature patterns in the identified feature combinations to calculate the correlation coefficients between the feature patterns and environmental humidity and temperature. Features with correlation coefficients lower than a preset threshold are selected, and the selected features are weighted and summed to calculate the initial intrinsic state index.

3. The automated control method for a rubber processing production line according to claim 2, characterized in that, The step of performing correlation analysis on the comprehensive feature vector to identify feature combinations that are weakly correlated with environmental changes includes: Perform feature pattern separation operation on the comprehensive feature vector to weaken the interference signal caused by environmental factors or their interaction with materials; Principal component analysis is performed on the comprehensive feature vector after feature pattern separation to identify the orthogonal directions with the largest variance in the data and obtain multiple principal components. Calculate the correlation coefficient between each principal component and the environmental data, and select the principal components whose correlation coefficient with the environmental data is lower than a preset threshold as feature combinations representing the inherent microstructural differences of the materials.

4. The automated control method for a rubber processing production line according to claim 3, characterized in that, The step of performing feature pattern separation on the comprehensive feature vector to weaken interference signals caused by environmental factors or their interaction with materials includes: A multi-layer sensing network is established, using the refining process data, the environmental data, and the raw material pre-scan feature data as inputs; The multilayer sensing network is trained to identify interference signal patterns generated by the interaction between environmental factors and materials; The inherent characteristics of the material are reconstructed from the comprehensive feature vector using the multilayer sensing network.

5. The automated control method for a rubber processing production line according to claim 3, characterized in that, The step of performing principal component analysis on the comprehensive feature vector after feature pattern separation to identify the orthogonal direction with the largest variance in the data and obtain multiple principal components includes: Perform a nonlinear transformation on the composite feature vector to convert it into a high-dimensional feature space; Principal component analysis is performed in the high-dimensional feature space to identify the orthogonal directions with the largest variance in the data, thereby obtaining multiple principal components.

6. The automated control method for a rubber processing production line according to claim 5, characterized in that, The step of performing a nonlinear transformation on the composite feature vector to convert the composite feature vector to a high-dimensional feature space includes: Based on the energy absorption performance, shear response performance, and raw material pre-scanning characteristic data in the comprehensive feature vector, the heterogeneity and nonlinear coupling strength of the current batch of materials are evaluated to obtain the evaluation results; Based on the evaluation results and raw material pre-scanning characteristic data, one or more nonlinear transformation functions that match the characteristics of the current batch of materials are selected from a preset set of nonlinear transformation functions. For the selected nonlinear transformation function, the parameters of the nonlinear transformation function are adjusted based on the energy absorption performance, shear response performance, and raw material pre-scanning characteristic data to obtain the target nonlinear transformation function; By using the target nonlinear transformation function, a nonlinear transformation is performed on the comprehensive feature vector, transforming the comprehensive feature vector into a high-dimensional feature space.

7. The automated control method for a rubber processing production line according to claim 5, characterized in that, The step of performing a nonlinear transformation on the composite feature vector to convert the composite feature vector to a high-dimensional feature space includes: The nonlinear coupling strength of the current batch of materials is evaluated based on the energy absorption performance and shear response performance in the comprehensive feature vector. Based on the nonlinear coupling strength, one or more kernel functions are selected from a preset set of kernel functions; For a selected kernel function, the parameters of the kernel function are adjusted based on the energy absorption performance and shear response performance. The parameters include the width parameter of the kernel function or the order of the polynomial kernel function to obtain the target kernel function. Using the target kernel function, a nonlinear mapping is performed on the comprehensive feature vector to transform the comprehensive feature vector into a high-dimensional feature space; In the high-dimensional feature space, redundant feature dimensions are identified and removed based on the energy absorption performance and shear response performance, thereby optimizing the dimensionality of the high-dimensional feature space.

8. The automated control method for a rubber processing production line according to claim 7, characterized in that, The step of identifying and removing redundant feature dimensions and optimizing the dimensions of the high-dimensional feature space based on the energy absorption performance and shear response performance includes: Continuously monitor the real-time fluctuation patterns of the material's energy absorption and shear response performance; Based on the real-time fluctuation pattern, feature dimension weight adjustment coefficients are generated and input into the weight allocation mechanism to dynamically quantify the contribution of each high-dimensional feature dimension to the energy absorption performance and shear response performance. Identify and remove feature dimensions whose contribution is below a preset threshold under a dynamically adjusted weight allocation mechanism.

9. The automated control method for a rubber processing production line according to claim 8, characterized in that, The step of generating feature dimension weight adjustment coefficients based on the real-time fluctuation pattern and inputting them into the weight allocation mechanism to dynamically quantify the contribution of each high-dimensional feature dimension to the energy absorption performance and shear response performance includes: Continuously monitor the real-time fluctuation patterns of the material's energy absorption and shear response performance; Abnormal signal detection is performed on the real-time fluctuation pattern to identify and isolate abnormal spikes or drifts caused by occasional external interference, thereby obtaining the target real-time fluctuation pattern. Based on the target's real-time fluctuation pattern, feature dimension weight adjustment coefficients are generated and input into the weight allocation mechanism to dynamically quantify the contribution of each high-dimensional feature dimension to the energy absorption performance and shear response performance.

10. An automated control system for a rubber processing production line, characterized in that, The system includes: The information acquisition module is used to acquire process parameter information of the material in the initial processing stage. The process parameter information includes the material's energy absorption performance and shear response performance. The process parameter information is analyzed to determine the initial intrinsic state index of the material. The history recording and transmission module is used to record the initial internal state indicators and energy input information in the initial processing stage as the processing history of the material, and to transmit the processing history to the subsequent processing stage. The state inference and parameter adjustment module is used to infer the current internal state of the material in subsequent processing stages based on the received processing history and real-time process information in subsequent processing stages, and then adjust the processing parameters of the subsequent processing stages based on the current internal state to eliminate or reduce the inherent deviation of the material, and update the current internal state to the processing history.

Images