Natural rubber raw material acceptance data intelligent grading method and system

Abstract

This invention discloses an intelligent grading method and system for natural rubber raw material acceptance data, specifically relating to the field of rubber raw material acceptance, to solve the problem of batch status misjudgment caused by missing acceptance data. The method involves collecting multi-source heterogeneous detection parameters, forming a physicorheological parameter set and a chemical component parameter set according to physical and chemical properties, and generating a standard dual-parameter sequence. This standard dual-parameter sequence is then mapped to a dual-parameter fusion state matrix according to orthogonal arrangement principles to identify data gaps. The remaining valid data undergoes intrinsic confidence verification, and compensatory state vectors or null placeholders are generated based on the verification results, forming a holographic fusion topology matrix. Subsequently, a physicochemical interaction topology graph is constructed based on node connectivity constraints, and nonlinear spatial deformation mapping is performed to extract a high-dimensional manifold topological feature set. Finally, the high-dimensional manifold topological feature set is projected onto the macroscopic mechanical steady-state boundary, and adaptive grading is performed based on the spatial deviation distribution pattern.

Description

Technical Field

[0001] This invention relates to the field of rubber raw material acceptance, and more specifically, to an intelligent grading method and system for natural rubber raw material acceptance data. Background Technology

[0002] In the manufacturing of high-performance tires and industrial polymer materials, natural rubber is not a homogeneous raw material that can be consistently used simply by meeting one of its test indicators. Natural rubber originates from bio-based materials, and its Mooney viscosity, plasticity retention, volatile matter, and ash content are affected by factors such as origin, storage time, drying degree, coagulation process, and batch mixing. Raw material acceptance typically involves using equipment such as Mooney testers and component analyzers to output physical rheological and chemical composition parameters, which quality inspectors then use to determine whether to accept the material for storage. This approach can complete basic acceptance when indicators are complete, equipment is stable, and batch fluctuations are minimal. However, in highly sensitive formulations and continuous mass production scenarios, the rheological state and compositional state of the raw material are not independent. Slight deviations in a single parameter may be offset by the structural state of other parameters, or deviations in multiple parameters may be amplified into abnormal torque during the mixing process. Existing acceptance records, which primarily rely on independent indicator comparisons or simple statistical processing, are insufficient to express the interaction between physical rheological and chemical composition parameters.

[0003] Existing invention patent application CN115563803A discloses a method for evaluating the service life of natural rubber. This method selects different natural rubbers as preset rubber compounds, compares torque values ​​to select the maximum torque, calculates a correction factor, and uses the correction data to perform linear fitting on time and torque to obtain an equivalent aging rate equation. The service life of the rubber compound under evaluation is then evaluated based on this equation. This disclosed method demonstrates that a technical approach exists in the field of natural rubber performance evaluation that utilizes test parameters such as torque to establish evaluation relationships. However, its focus is on evaluating the service life of the rubber compound, rather than the structured fusion of multi-source heterogeneous test data during raw material acceptance at the factory, the judgment of the reliability of data gaps, and batch warehousing grading.

[0004] In actual acceptance testing, testing equipment may experience localized data gaps due to fieldbus interference, workstation vibration, uneven sampling, or repeated testing failures. Furthermore, high-frequency pulse noise may superimpose on the test values. If these gaps are directly filled with averages, nearest neighbor values, or manual experience, "not measured by the equipment" will be mistakenly treated as "material has been determined," leading to subsequent evaluations based on unverified residual data. For materials like natural rubber, where the physical network state and chemical composition state are coupled, this problem further manifests as: batches that could meet the target formulation requirements through comprehensive physicochemical properties are mistakenly rejected, while batches with superficially acceptable individual indicators but deviating overall characteristics are mistakenly released. Therefore, an adaptive grading method for natural rubber raw material acceptance data is needed. This method should be able to organize multi-source acceptance data within a unified metric space, first verify the reliability of residual valid data when data gaps occur, and then establish a correspondence between physicochemical interaction characteristics and macroscopic mechanical acceptance boundaries.

[0005] To address the aforementioned problems, a technical solution is provided. Summary of the Invention

[0006] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide an intelligent grading method and system for natural rubber raw material acceptance data. This method involves collecting multi-source heterogeneous detection parameters, forming a set of physical rheological parameters and a set of chemical component parameters based on physical and chemical properties, and generating a standard dual-parameter sequence. The standard dual-parameter sequence is then mapped to a dual-parameter fusion state matrix according to an orthogonal arrangement principle. Data gaps are identified, and intrinsic confidence is verified on the remaining valid data. Based on the verification results, a compensating state vector or null placeholder is generated, forming a holographic fusion topology matrix. Subsequently, a physicochemical interaction topology graph is constructed based on node connectivity constraints, and nonlinear spatial deformation mapping is performed to extract a high-dimensional manifold topological feature set. Finally, the high-dimensional manifold topological feature set is projected onto the macroscopic mechanical steady-state boundary, and adaptive grading is performed based on the spatial deviation distribution pattern to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A method for intelligent grading of natural rubber raw material acceptance data, comprising the following steps: Step S1: In response to the material acceptance command, collect multi-source heterogeneous detection parameters, form a set of physical rheological parameters and a set of chemical composition parameters according to physical and chemical properties, and perform dimensional normalization and heterogeneous noise smoothing filtering to output a standard dual-parameter sequence. Step S2: Receive the standard dual-parameter sequence and map it into a dual-parameter fusion state matrix according to the orthogonal arrangement principle. When a data gap is identified, perform a credit check on the remaining valid data, generate a compensating state vector or a null space placeholder based on the check result, and output the holographic fusion topology matrix. Step S3: For the holographic fusion topology matrix, construct a physical-chemical interaction topology graph based on the node association connectivity constraints, and perform nonlinear spatial deformation mapping to extract the high-dimensional manifold topology feature set; Step S4: Project the high-dimensional manifold topological feature set onto the macroscopic mechanical steady-state boundary, compare the spatial deviation distribution of the projection points, perform adaptive classification and division, and generate closed-loop control messages.

[0008] Furthermore, in step S1, the multi-source heterogeneous detection parameters are written into the current batch according to the binding relationship between the detection equipment channel and the acceptance items. Detection values ​​belonging to shear stress or rheological response are assigned to the physical rheological parameter set, and detection values ​​belonging to component ratio or chemical state characterization are assigned to the chemical component parameter set. Data gap markers are generated for positions where no valid values ​​are returned. For the valid original detection values ​​in the physical rheological parameter set and the chemical component parameter set, the minimum usable boundary and the maximum usable boundary are determined based on the intersection of the acceptance valid range corresponding to the target production formula and the detection equipment range. The valid original detection values ​​are converted into dimensionless detection values. Data gap markers do not participate in the dimension normalization process.

[0009] Furthermore, in step S1, a sliding window is established within the same detection parameter, and the sliding window only includes valid dimensionless detection values; when the valid dimensionless detection values ​​within the window reach the valid number requirement of the sliding window, the value in the middle after sorting is taken to generate a smooth value; when the valid number requirement of the sliding window is not reached, a data gap marker is retained at the corresponding position, and the physical rheological parameter sequence and the chemical component parameter sequence are organized into a standard dual-parameter sequence.

[0010] Furthermore, in step S2, the matrix nodes of the dual-parameter fusion state matrix are jointly determined by the chemical composition smoothing value of the corresponding row and the physical rheological smoothing value of the corresponding column, and stored as a physicochemical interaction node vector; when any source position involved in the formation of the matrix node has a data gap marker, the matrix node does not generate a valid physicochemical interaction node vector, but is recorded as a gap.

[0011] Furthermore, in step S2, the confidence level review includes: reading continuous and effective physical rheological smoothing values ​​along the time axis and accumulating the fluctuation contribution values ​​of adjacent sampling segments to form a viscoelastic fluctuation evolution quantity; organizing the effective chemical component smoothing values ​​into chemical component vectors, constructing an autocorrelation matrix and extracting the maximum eigenvalue to form the component distribution principal element degree; inputting the viscoelastic fluctuation evolution quantity and the component distribution principal element degree into the isolated forest model in a fixed order, and outputting the abnormal isolated state quantity.

[0012] Furthermore, in step S2, the safety envelope boundary is generated by the historical abnormal isolated state quantity corresponding to the gold production batch dataset; when the abnormal isolated state quantity falls into the safety envelope boundary, a compensatory state vector is generated and written into the vacancy; when the abnormal isolated state quantity falls out of the safety envelope boundary, the vacancy is frozen with a zero space placeholder and a frozen vacancy mark is generated.

[0013] Furthermore, in step S3, the matrix position in the holographic fusion topology matrix is ​​defined as a physical-chemical interaction node. The physical-chemical interaction node includes an effective physical-chemical interaction node and a frozen physical-chemical interaction node. The effective physical-chemical interaction node participates in the calculation of node association connectivity, node association strength and graph Laplacian matrix, while the frozen physical-chemical interaction node only retains the original matrix coordinates and the frozen vacancy mark.

[0014] Furthermore, in step S3, each effective physicochemical interaction node is treated as a graph node, and the node pairs that are adjacent in the row direction or column direction and whose two ends are both effective physicochemical interaction nodes are treated as graph edges to form a physicochemical interaction topology graph; the node association strength is calculated based on the number of adjacent effective nodes and the dimensionless distance between nodes, written into the association strength matrix, and Laplace eigenmap is performed accordingly to form a high-dimensional manifold topology feature set.

[0015] Furthermore, in step S4, the macroscopic mechanical steady-state boundary is jointly formed by the upper and lower boundaries of the Mooney viscosity dimension, the rotor torque peak dimension, and the tensile modulus dimension; the high-dimensional manifold topological feature set is converted into a projection feature vector and input into the manifold-mechanical projection model to generate the macroscopic mechanical spatial projection position; the spatial deviation and the number of deviation dimensions outside the boundary are calculated based on the macroscopic mechanical spatial projection position, and a closed-loop control message is generated by combining the number of frozen vacancy markers.

[0016] A smart grading system for natural rubber raw material acceptance data includes: The parameter pre-tuning module is used to collect multi-source heterogeneous detection parameters in response to material acceptance instructions, form a set of physical rheological parameters and a set of chemical composition parameters according to physical and chemical properties, and perform dimensional normalization and heterogeneous noise smoothing filtering to output a standard dual-parameter sequence. The missing data verification module is used to receive standard dual-parameter sequences and map them into dual-parameter fusion state matrices according to the orthogonal arrangement principle. When a data missing is identified, the module performs a credit verification on the remaining valid data, generates a compensating state vector or a null space placeholder based on the verification result, and outputs a holographic fusion topology matrix. The topology deformation module is used to construct a physical-chemical interaction topology graph based on the node association connectivity constraints for the holographic fusion topology matrix, and to perform nonlinear spatial deformation mapping to extract a high-dimensional manifold topology feature set; The projection classification module projects the high-dimensional manifold topological feature set onto the macroscopic mechanical steady-state boundary, compares the spatial deviation distribution of the projection landing point, performs adaptive classification, and generates closed-loop control messages.

[0017] The technical effects and advantages of the intelligent grading method and system for natural rubber raw material acceptance data of the present invention are as follows: This invention integrates the physical rheological parameters and chemical composition parameters in the acceptance of natural rubber raw materials into a unified processing chain. First, it performs dimensional normalization and heterogeneous noise smoothing filtering on multi-source heterogeneous detection parameters, and then uses a standard dual-parameter sequence for matrix processing. This processing reduces the interference of inconsistent dimensions between different testing equipment, different sampling sequences, and local impulse noise on acceptance judgment, enabling discrete detection values ​​such as Mooney viscosity, ash content, volatile matter, and initial plasticity to participate in batch status expression within the same structure.

[0018] When data gaps are identified in the dual-parameter fusion state matrix, this invention does not directly fill in the missing positions with numerical values. Instead, it first performs a confidence level audit on the remaining valid data, and then generates a compensating state vector or a zero-space placeholder based on the audit result. This process can distinguish between physically maintainable gaps and gaps that should be frozen, reducing erroneous compensations caused by equipment packet loss, abnormal sampling, or noise interference, and ensuring that the handling of acceptance data gaps is consistent with the physicochemical state of natural rubber itself.

[0019] This invention establishes a correspondence between the physicochemical interaction state of raw material batches and manufacturing acceptance criteria such as Mooney viscosity, rotor torque peak, and tensile modulus by transforming holographic fusion topological matrices, high-dimensional manifold topological feature sets, and macroscopic mechanical steady-state boundaries. The resulting closed-loop control message can simultaneously carry spatial deviation distribution patterns, adaptive grading results, and frozen vacancy markers, enabling raw material storage, quality inspection verification, and internal mixing feeding to be handled based on the same batch state, reducing the risk of misjudgment during acceptance due to individual indicators meeting the requirements but overall state deviation. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating an intelligent grading method for natural rubber raw material acceptance data according to the present invention.

[0021] Figure 2 This is a schematic diagram of the structure of an intelligent grading system for natural rubber raw material acceptance data according to the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Please see Figure 1 This invention provides an intelligent grading method for natural rubber raw material acceptance data, comprising: This invention focuses on processing multi-source heterogeneous detection parameters generated during the acceptance process of natural rubber raw materials. The acceptance data processing server first reads the detection values ​​of the current batch according to the material acceptance instructions, and forms a physicorheological parameter set and a chemical component parameter set according to physical and chemical properties, respectively. Then, dimensional normalization and heterogeneous noise smoothing filtering are performed on the two parameter sets, ensuring that the data output from different testing devices enters a unified standard dual-parameter sequence. The acceptance data processing server then converts the standard dual-parameter sequence into a dual-parameter fusion state matrix according to the orthogonal arrangement principle, and identifies data gaps in the matrix. For batches with data gaps, the acceptance data processing server performs a confidence level review on the remaining valid data, and writes compensating state vectors or null placeholders into the gap positions based on the review results, thereby forming a holographic fusion topology matrix. The holographic fusion topology matrix is ​​converted into a physicochemical interaction topology graph based on node connectivity constraints, and then a high-dimensional manifold topology feature set is generated through nonlinear spatial deformation mapping. The acceptance data processing server projects the high-dimensional manifold topological feature set onto the macroscopic mechanical steady-state boundary, compares the spatial deviation distribution of the projection points in the Mooney viscosity dimension, the rotor torque peak dimension, and the tensile modulus dimension, and generates adaptive classification results and closed-loop control messages accordingly.

[0024] After natural rubber raw materials enter the acceptance stage, testing equipment such as Mooney testers and component analyzers will output physical rheological parameters and chemical composition parameters, respectively. However, the output dimensions, sampling order, and noise patterns of different testing equipment are not consistent, and some testing locations may not return valid values. Therefore, step S1 needs to respond to the material acceptance command first, collect, classify, normalize the dimensions, and smooth and filter heterogeneous noise from the multi-source heterogeneous testing parameters, so that subsequent processing can receive the physical rheological parameter set and the chemical composition parameter set on a unified structural metric basis.

[0025] Step S1: In response to the material acceptance command, collect multi-source heterogeneous detection parameters, form a set of physical rheological parameters and a set of chemical composition parameters according to physical and chemical properties, and perform dimensional normalization and heterogeneous noise smoothing filtering to output a standard dual-parameter sequence. The specific implementation method is as follows: In response to the material acceptance command for the current batch of natural rubber raw materials, the acceptance data processing server reads the raw test values ​​returned by the Mooney tester, component analyzer, and testing equipment channels bound to the target production formula. The material acceptance command is a trigger signal generated when the current batch enters the raw material acceptance stage, and it is bound to the testing cycle, testing equipment channels, and the acceptance item scope corresponding to the target production formula for the current batch.

[0026] The acceptance data processing server, based on the binding relationship between the testing equipment channels and the acceptance items, writes the original test values ​​into the multi-source heterogeneous testing parameters of the current batch, and classifies them according to physical and chemical properties. Test values ​​belonging to shear stress or rheological response are assigned to the physical rheological parameter set, while test values ​​belonging to component ratio or chemical state characterization are assigned to the chemical component parameter set. The test values ​​in the physical rheological parameter set are sorted by sampling time, and the test values ​​in the chemical component parameter set are sorted by inspection sequence number. If a certain testing equipment channel does not return a valid value within the testing cycle, a data gap marker is generated at the corresponding sampling time or inspection sequence number. The data gap marker is transmitted synchronously with the test parameter name, sampling time, or inspection sequence number, and is converted into a vacancy position in step S2.

[0027] After parameter aggregation is completed, the acceptance data processing server performs extreme value normalization on the physical rheological parameter set and the chemical composition parameter set, so that the different dimensional detection values ​​output by different testing equipment enter the same dimensionless algebraic metric space. Before performing extreme value normalization, the acceptance data processing server first determines whether there is a valid intersection between the valid acceptance range corresponding to the target production formula and the range of the testing equipment. When there is no valid intersection, or the minimum usable boundary is equal to the maximum usable boundary, the acceptance data processing server does not perform dimensional normalization on the detection parameter, but writes a data gap marker at the corresponding position of the detection parameter, and retains the name of the detection parameter, sampling time or inspection sequence number. For any detection parameter In the Valid raw detection value at each sampling time or test sequence number Dimensionless detection value after normalization We obtain the following relationship:

[0028] in, To detect parameters In the The valid original detection values ​​at each sampling time or test sequence number are derived from the physical rheological parameter set or the chemical component parameter set; To detect parameters The minimum usable boundary is determined by the intersection of the acceptance effective range corresponding to the target production formula and the range of the testing equipment; To detect parameters The maximum usable boundary is determined by the intersection of the acceptance effective range corresponding to the target production formula and the range of the testing equipment; To detect parameters The dimensionless detection value generated at the corresponding location. The location of the data gap marker is not included in the extreme value normalization calculation and continues to retain its original location.

[0029] After normalization, the acceptance data processing server establishes a sliding window within the same detection parameter to perform heterogeneous noise smoothing filtering on the dimensionless detection values. For physical rheological parameters with continuous sampling characteristics, the sliding window length is the shortest time window covering three adjacent valid sampling points; for chemical composition parameters formed by repeated detections, the sliding window length is the sequence number of three consecutive tests. The center position of the window is the smoothing position. When the smoothing position is located at the beginning or end of the detection parameter sequence, causing the sliding window to be unable to fully expand to both sides, the acceptance data processing server expands in the available direction with the smoothing position as the center, and only uses the actual dimensionless detection values ​​that do not have data gap markers for sorting; if the actual number of valid dimensionless detection values ​​included does not meet the valid number requirement of the sliding window, a data gap marker is retained at the smoothing position. Only valid dimensionless detection values ​​are included in the window, and data gap markers do not participate in the sorting. When there are at least 3 valid dimensionless detection values ​​in the window, the acceptance data processing server sorts the valid dimensionless detection values ​​in the window, takes the median value, and replaces the dimensionless detection value at the center of the window with the median value; when there are less than 3 valid dimensionless detection values ​​in the window, the acceptance data processing server does not generate a smoothed value at that position, but writes a data gap marker at that position.

[0030] After smoothing and filtering, the acceptance data processing server organizes the smoothed physical rheological values ​​into a physical rheological parameter sequence according to the detection parameter name and sampling time, and organizes the smoothed chemical component values ​​into a chemical component parameter sequence according to the detection parameter name and inspection number. The physical rheological parameter sequence and the chemical component parameter sequence together constitute a standard dual-parameter sequence. The standard dual-parameter sequence retains the column order of the physical rheological parameter sequence, the row order of the chemical component parameter sequence, and the original position of the data gap markers, serving as the direct input for step S2 to construct the dual-parameter fusion state matrix.

[0031] In one implementation, after the current batch of natural rubber raw materials has completed barcode verification, the acceptance data processing server reads the Mooney viscosity sampling values ​​continuously output by the Mooney tester, and reads the ash, volatile matter, and initial plasticity test values ​​output by the component analyzer. The Mooney viscosity sampling values ​​are entered into the physical rheological parameter set, and the ash, volatile matter, and initial plasticity values ​​are entered into the chemical composition parameter set. If the Mooney tester loses a packet at a certain sampling time, that sampling time is written into a data gap marker; if a valid result is not returned in a repeated ash content test, that test number is written into a data gap marker. The above data gap markers proceed to step S2 along with the standard dual-parameter sequence.

[0032] After step S1 is completed, the multi-source heterogeneous detection parameters of the current batch are organized into a set of physical rheological parameters and a set of chemical composition parameters, and then subjected to dimensional normalization and heterogeneous noise smoothing to form a standard dual-parameter sequence. This standard dual-parameter sequence retains the original positions of the physical rheological parameter sequence, the chemical composition parameter sequence, and the data gap markers, providing direct input for step S2 to construct the dual-parameter fusion state matrix according to the orthogonal arrangement principle.

[0033] After outputting the standard dual-parameter sequence in step S1, the physical rheological parameters and chemical composition parameters are already in a unified dimensionless algebraic metric space, but they still exist in sequence form and cannot yet express the matrix interaction relationship between the two types of parameters. Step S2 needs to receive the standard dual-parameter sequence, map it into a dual-parameter fusion state matrix, and when a data gap is identified, perform a intrinsic confidence check on the remaining valid data, and then decide whether to use a compensating state vector to maintain the gap or to freeze it with a zero-space placeholder based on the check result.

[0034] Step S2: Receive the standard dual-parameter sequence and map it into a dual-parameter fusion state matrix according to the orthogonal arrangement principle. When a data gap is detected, perform a confidence check on the remaining valid data, generate a compensating state vector or a null space placeholder based on the check result, and output the holographic fusion topology matrix. The specific implementation method is as follows: The acceptance data processing server receives the standard dual-parameter sequence output in step S1 and reads the physical rheological parameter sequence, chemical composition parameter sequence, and data gap markers. The physical rheological parameter sequence is expanded along the column direction, and the chemical composition parameter sequence is expanded along the row direction. Following an orthogonal arrangement principle, a dual-parameter fusion state matrix is ​​established within the same dimensionless algebraic metric space. .

[0035] Dual-parameter fusion state matrix The first in The first row corresponds to the chemical component parameter sequence. The smoothing value of the first chemical component, the... The column corresponds to the first in the sequence of physical rheological parameters. A physical rheological smoothing value. Matrix node. Physicochemical interaction node vector storage:

[0036] in, For the first Line 1 The physicochemical interaction node vectors corresponding to the column matrix nodes; The first in the chemical component parameter sequence Smoothing values ​​for each chemical component; The first in the physical rheological parameter sequence A physical rheological smoothing value. Physicochemical interaction node vector. The first component represents the chemical composition state of the corresponding row, and the second component represents the physicorheological state of the corresponding column. Within the same batch, the physicochemical interaction node vectors, compensatory state vectors, and node vectors used for calculating dimensionless distances between nodes all use the same component order, and the component positions are not changed in subsequent processing. Subsequent calculations of node distances, node association strengths, high-dimensional manifold coordinates, and macroscopic mechanical space projection positions are all based on the physicochemical interaction node vectors. As basic input. When or When any source location has a data gap marker, the matrix node Instead of generating valid physical-chemical interaction node vectors, they are recorded as empty spaces.

[0037] Dual-parameter fusion state matrix After construction is complete, the acceptance data processing server scans all matrix nodes. If no missing nodes are identified, the dual-parameter fused state matrix is ​​then processed. The structurally complete holographic fusion topology matrix is ​​output to step S3. If at least one missing position is identified, anomaly interception processing based on intrinsic creditworthiness is initiated.

[0038] Anomaly interception processing first calculates the viscoelastic wave evolution based on the set of physical rheological parameters formed in step S1. The acceptance data processing server reads continuous and valid physical rheological smoothing values ​​along the time axis, forms differential rates of change based on adjacent sampling points, and then performs square integration on the differential rates of change to obtain the viscoelastic wave evolution. :

[0039] in, For viscoelastic wave evolution; For the first physical rheological parameter set One effective physical rheological smoothing value; To and The adjacent previous effective physirographic smoothing value; and For the corresponding sampling time; This represents the number of valid physical rheological smoothing values ​​used in the calculation. When adjacent valid sampling points cross data gap markers, this adjacency relationship is not included. The calculation and acceptance data processing server only calculates the differential rate of change between consecutive valid sampling points.

[0040] Simultaneously, the acceptance data processing server calculates the principal component degree of the component distribution based on the chemical component parameter set formed in step S1. The acceptance data processing server organizes the smoothed values ​​of valid chemical components without data gap markers into chemical component vectors according to the test sequence number in the chemical component parameter sequence. The original test number is saved simultaneously. Chemical component positions marked with missing data are not included in the autocorrelation matrix calculation; they are only treated as missing positions in the dual-parameter fusion state matrix for anomaly interception. Autocorrelation matrix Principal element degree of component distribution We obtain the following relationship:

[0041] in, It is a column vector consisting of smoothed values ​​of effective chemical components; Chemical component vector Transpose of; This is an autocorrelation matrix used to express the proportionate clustering relationships of chemical component vectors; Autocorrelation matrix The largest eigenvalue; Principal element degree of component distribution. The acceptance data processing server will handle the viscoelastic fluctuation evolution. Principal element degree of component distribution The input vectors are organized into an isolated forest in a fixed order. .

[0042] The acceptance data processing server calls a pre-built isolated forest model to process the isolated forest input vectors. In one embodiment, the pre-built isolated forest model employs a randomized isolated tree ensemble structure. The isolated forest model comprises 100 randomized isolated trees, each with 256 historical batch samples as training subsamples in a single run. When the number of historical batches in the gold production batch dataset is less than 256, all historical batch samples are used as training subsamples. Each historical batch sample consists of viscoelastic fluctuation evolution and component distribution principal element degrees in a fixed order, and extreme value normalization is performed before inputting them into the model. The maximum tree depth of each randomized isolated tree is set to 8 layers. Splitting stops when a sample is isolated, only one sample remains in a node, or the tree depth reaches 8 layers. During model training, the acceptance data processing server independently extracts training subsamples from each randomized isolated tree, randomly selects either the viscoelastic fluctuation evolution dimension or the component distribution principal element degree dimension at each tree node, and randomly generates a split point between the minimum and maximum values ​​of the current node sample in the selected dimension, dividing the sample into left and right child nodes. The average isolation path length formed by the random splitting is used as the basis for calculating the abnormal isolated state quantity during the training process. The model's pre-training parameters include 100 random isolation trees, 256 training subsamples per run, a maximum tree depth of 8, and a fixed random seed of 2026. These pre-training parameters are determined using the gold production batch dataset during historical batch replay testing and are updated with the gold production batch dataset after any changes to the target production formula. When the current batch enters the review process, the acceptance data processing server sequentially inputs the isolated forest input vector corresponding to the current batch into 100 random isolation trees, records the path length traversed when isolated in each random isolation tree, and calculates the average of all path lengths as the abnormal isolated state quantity for the current batch. This abnormal isolated state quantity is then compared with the safety envelope boundary.

[0043] Each historical batch in the gold production batch dataset generates its historical viscoelastic fluctuation evolution in the same manner as the current batch. Principal element degree of historical component distribution And form the input vector of the historical isolated forest. Before model training, the acceptance data processing server performs extreme value normalization on the historical viscoelastic fluctuation evolution sequence and the historical component distribution principal element degree sequence, respectively, so that the two input dimensions received by the isolated forest model are in the same dimensionless algebraic metric space.

[0044] Each random isolated tree in the isolated forest model is based on the historical isolated forest input vector. For training samples, either the viscoelastic fluctuation evolution dimension or the principal component degree dimension of the component distribution is randomly selected, and split points are generated between the maximum and minimum values ​​of the samples in the selected dimension until the samples are isolated or the preset maximum tree depth is reached. The current batch's abnormal isolated state quantity. The input vector for the current isolated forest The average path length required to be isolated across all random isolation trees:

[0045] in, This refers to an abnormal, isolated state quantity; This represents the number of random isolation trees in the isolated forest model. Input vector for isolated forest In the The path length traversed when isolated in a random isolation tree.

[0046] The safety envelope boundary is defined by the lower safety envelope limit. and upper limit of safety envelope Composition. The acceptance data processing server assembles the historical anomaly isolated state quantities corresponding to the gold production batch dataset into a sequence. The kernel density of the sequence is estimated using a Gaussian kernel function to obtain a one-dimensional probability density curve of the isolated historical anomalous states. The lower 5 percentiles of this probability density curve are then extracted as the lower bound of the safety envelope based on the cumulative distribution function of the probability density curve. The upper 95 percentile was extracted as the upper limit of the safe envelope. When the target production formula is updated, only gold production batches under the same target production formula will be regenerated. and .

[0047] The acceptance data processing server will handle abnormal orphaned states. With lower bound of safety envelope Upper limit of security envelope Compare. When satisfied... When the safety pass determination result is reached, dynamic maintenance of vacancy is performed; when or When the time comes, output the security interception judgment result and freeze the empty space with a zero space placeholder.

[0048] When the output safety pass judgment result is obtained, the acceptance data processing server generates a compensatory state vector based on the physical correlation state of macromolecular entanglement resistance. For vacancy... The acceptance data processing server first searches for the location at the sampling time in the physical rheological parameter sequence. The most effective physirheological smoothing value on both sides and the closest distance and The corresponding sampling times are respectively and When a valid physical rheological smoothing value exists on only one side at the missing sampling time, the acceptance data processing server reads the two closest consecutive valid physical rheological smoothing values ​​on that side and forms an extrapolation direction along the time axis based on these two consecutive valid physical rheological smoothing values. When there are fewer than two consecutive valid physical rheological smoothing values ​​on the same side, the acceptance data processing server does not generate a physical continuity component and freezes the corresponding missing position into a zero-space placeholder. Then, based on the viscoelastic fluctuation evolution... Generation of entanglement resistance attenuation factor :

[0049] in, It is an entanglement resistance attenuation factor used to compress the extension of vacancies along the time axis according to the degree of macromolecular network fluctuation in the current batch; This represents the evolution of viscoelastic waves.

[0050] Physical maintenance components Generate according to the following relationship:

[0051] in, The component that sustains the vacancy in the physical rheological direction; The nearest valid physiorheological smoothing value to the left of the missing sampling time; The nearest valid physiorheological smoothing value to the right of the missing sampling time; The sampling time corresponding to the vacancy; and They are respectively and The corresponding sampling time.

[0052] For the chemical composition direction, the acceptance data processing server reads the autocorrelation matrix. The largest eigenvalue in the middle corresponding unit eigenvector The unit eigenvector is in the th... The components at each position of a chemical component are denoted as When the first Each chemical component represents a missing location and is the current unit feature vector. When this location is not included, the acceptance data processing server reads the unit eigenvector corresponding to the largest eigenvalue of the autocorrelation matrix of the chemical composition of the gold production batch of the same target production formula, and assigns this unit eigenvector to the first... The components at each position of a chemical component are used as Chemical maintenance components Generate according to the following relationship:

[0053] in, This refers to the component that maintains the vacancy in the direction of the chemical composition; To determine the principal element degree of component distribution The principal component magnitude obtained by square root processing is used to bring the chemical continuity component and the smoothed chemical component value to the same numerical scale. Used to indicate the first The proportion of each chemical component in the principal component direction.

[0054] Compensation state vector It consists of chemical and physical components arranged in the same order of the physicochemical interaction node vectors:

[0055] When the vacancy is caused solely by the absence of physical rheological parameters and the chemical component parameters are valid, the smoothed value of the original valid chemical component is used as the replacement. When the vacancy is caused solely by the absence of chemical component parameters and the physical rheological parameters are valid, the original valid physical rheological smoothing value shall be used as the replacement. Compensation state vector After writing the vacancy, it is used as the effective physical and chemical interaction node vector in the subsequent calculation of node distance and node association strength.

[0056] When outputting the security interception judgment result, the acceptance data processing server freezes vacant positions using zero-space placeholders. Zero-space placeholders are vacant nodes that do not carry physicochemical compensation values ​​and are only used to maintain the integrity of the matrix dimensions. After writing the zero-space placeholders, the row and column coordinates of the vacant positions remain unchanged, and a frozen vacant position marker is generated simultaneously. The dual-parameter fusion state matrix, after writing the compensation state vector or the zero-space placeholders, forms a holographic fusion topology matrix.

[0057] After step S2 is completed, the standard dual-parameter sequence is converted into a holographic fusion topology matrix. For matrix nodes without data gaps, the holographic fusion topology matrix retains valid physicochemical interaction nodes; for matrix nodes with data gaps, the holographic fusion topology matrix writes a compensating state vector or a null space placeholder according to the intrinsic confidence level, and simultaneously retains the frozen vacancy marker. This holographic fusion topology matrix provides the input basis for constructing the physicochemical interaction topology graph and performing nonlinear spatial deformation mapping in step S3.

[0058] After forming the holographic fusion topology matrix in step S2, the effective physicochemical interaction nodes, compensatory state vectors, null space placeholders, and frozen vacancy markers of the current batch have been incorporated into the same matrix structure. However, this matrix still mainly expresses the row and column coordinate relationships and has not yet extracted the connectivity relationships and spatial deformation features between nodes. Step S3 requires constructing a physicochemical interaction topology graph based on the node association connectivity constraints and extracting a high-dimensional manifold topology feature set through nonlinear spatial deformation mapping.

[0059] Step S3: For the holographic fusion topology matrix, construct a physicochemical interaction topology graph based on node connectivity constraints, and perform nonlinear spatial deformation mapping to extract a high-dimensional manifold topological feature set. The specific implementation method is as follows: The acceptance data processing server receives the holographic fusion topology matrix output in step S2. The matrix positions in the holographic fusion topology matrix are defined as physicochemical interaction nodes, which include valid physicochemical interaction nodes and frozen physicochemical interaction nodes. Valid physicochemical interaction nodes are those carrying valid detection values ​​or compensatory state vectors, participating in the calculation of node connectivity, node association strength, and graph Laplacian matrix; frozen physicochemical interaction nodes are those carrying null space placeholders, retaining only the original matrix coordinates and frozen vacancy markers, and do not participate in graph edge construction and eigenvalue decomposition.

[0060] The acceptance data processing server traverses all physicochemical interaction nodes in matrix coordinate order, establishing a set of adjacent nodes for each node. The adjacent node set consists of horizontally adjacent physicochemical interaction nodes within the same row and vertically adjacent physicochemical interaction nodes within the same column. Adjacent directions that do not exist at the matrix boundary are not included in the adjacent node set. Node connectivity. Calculated according to the following relationship:

[0061] in, For the first Line 1 The node connectivity degree corresponding to the physical-chemical interaction node; For the first Line 1 The set of neighboring nodes of the physical interaction node; It is any adjacent node in the set of adjacent nodes; It is a valid marker for adjacent nodes. It is set to 1 when the adjacent node carries a valid detection value or a compensation state vector, and to 0 when the adjacent node carries a zero space placeholder.

[0062] After the node connectivity calculation is completed, the acceptance data processing server treats each valid physicochemical interaction node as a graph node, and node pairs that satisfy row-wise or column-wise adjacency and whose two ends are both valid physicochemical interaction nodes as graph edges, forming a physicochemical interaction topology graph. For connecting physicochemical interaction nodes... Interaction nodes between physicochemical processes The graph edges, the strength of the association between the computing nodes of the acceptance data processing server. :

[0063] in, For physical and chemical interaction nodes Interaction nodes with physicochemical processes The strength of the node association between them; and These are the physical and chemical interaction nodes. Interaction nodes between physicochemical processes The corresponding node connectivity; The dimensionless distance between the content carried by two physical and chemical interaction nodes.

[0064] When the physical and chemical interaction node Bearing the physical and chemical interaction node vector Physicochemical interaction nodes Bearing the physical and chemical interaction node vector hour, The calculation is based on Euclidean distance, which involves first calculating the difference between two chemical components and the difference between two physical rheological components, then squaring each difference separately, adding them together, and finally taking the square root of the sum.

[0065] in, and These are the chemical component components of the two nodes, respectively; and These are the physical rheological components of the two nodes, respectively. The chemical component component originates from the chemical component state in the physicochemical interaction node vector, and the physical rheological component originates from the physical rheological state in the physicochemical interaction node vector. When any node carries a compensating state vector, this compensating state vector is used as the corresponding physicochemical interaction node vector in the same distance calculation; when any node carries a null space placeholder, that node does not generate... The corresponding graph edges are not written into the correlation strength matrix.

[0066] The acceptance data processing server writes the association strength of all nodes into the association strength matrix in the order of physicochemical interaction nodes. And based on the correlation strength matrix Construct the degree matrix Degree matrix The diagonal elements represent the cumulative value of the node association strength between the corresponding effective physicochemical interaction node and all adjacent effective physicochemical interaction nodes. (Graph Laplacian matrix) Generate according to the following relationship:

[0067] in, The graph Laplacian matrix is ​​used to represent the local curved structure of a physicochemical interaction topology graph under node connectivity constraints. It is a degree matrix; This is the correlation strength matrix.

[0068] Acceptance data processing server for graph Laplace matrix Eigenvalue decomposition is performed, selecting eigenvectors corresponding to eigenvalues ​​other than zero in ascending order of eigenvalues. During eigenvector selection, the acceptance data processing server treats eigenvalues ​​whose absolute value is not greater than the eigenvalue discrimination tolerance as zero eigenvalues. The eigenvalue discrimination tolerance is determined by the graph Laplacian matrix. The maximum value of the diagonal elements is obtained by multiplying by the preset floating-point calculation precision; when the graph Laplacian matrix... When the maximum value of a diagonal element is zero, Laplacian eigenmapping is not performed, and valid physicochemical interaction nodes are written to the zero coordinate placeholder in the original matrix coordinate order. The number of selected eigenvectors is denoted as... , Let be the manifold coordinate dimension of the high-dimensional manifold topological feature set. For implementations where a macroscopically mechanical steady-state boundary has been established for the target production formula, We set the value to 3 to establish a projected correspondence between the high-dimensional manifold coordinates and the Mooney viscosity dimension, the rotor torque peak dimension, and the tensile modulus dimension. If the number of non-zero eigenvalues ​​in the graph Laplacian matrix is ​​less than 3, we use the existing non-zero eigenvectors to form the low-dimensional coordinates and write 0 coordinates as placeholders in the missing dimensions; if the number of non-zero eigenvalues ​​is more than 3, we select the eigenvectors corresponding to the three smallest non-zero eigenvalues.

[0069] The coordinates of the effective physical-chemical interaction nodes in the selected feature vector are written into the high-dimensional manifold coordinates. The high-dimensional manifold topological feature set consists of the high-dimensional manifold coordinates of the effective physical-chemical interaction nodes, the node association connectivity of the effective physical-chemical interaction nodes, the original matrix coordinates of the frozen physical-chemical interaction nodes, and the frozen vacancy markers. The acceptance data processing server organizes the high-dimensional manifold topological feature set in the order of the original matrix coordinates and outputs it to step S4.

[0070] In one implementation, if there are continuous valid nodes in the Mooney viscosity sampling direction of a certain batch of natural rubber, and the chemical component nodes corresponding to ash and volatile matter form a valid connection with the sampling direction, then these nodes form continuous graph edges in the physicochemical interaction topology graph; if a certain vacancy is maintained by a compensated state vector in step S2, then the node participates in the node association strength calculation as a valid physicochemical interaction node; if a certain vacancy is frozen by a null space placeholder, then the node only retains the original matrix coordinates and the frozen vacancy mark, and does not participate in the eigenvalue decomposition of the graph Laplacian matrix.

[0071] After step S3 is completed, the holographic fusion topology matrix is ​​converted into a high-dimensional manifold topology feature set. This high-dimensional manifold topology feature set includes the high-dimensional manifold coordinates of the effective physicochemical interaction nodes, the node connectivity, the original matrix coordinates of the frozen physicochemical interaction nodes, and the frozen vacancy markers. It can express the connectivity state and spatial distribution relationship between the current batch of physicochemical interaction nodes, and serves as the input object for performing macroscopic mechanical steady-state boundary projection in step S4.

[0072] After outputting the high-dimensional manifold topological feature set in step S3, the physicochemical interaction state of the current batch has formed a projectable high-dimensional feature expression, but a classification relationship has not yet been established with the Mooney viscosity dimension, rotor torque peak dimension, and tensile modulus dimension corresponding to the target production formula. Step S4 requires projecting the high-dimensional manifold topological feature set onto the macroscopic mechanical steady-state boundary, comparing the spatial deviation distribution of the projection points, and forming an adaptive classification result accordingly.

[0073] Step S4: Project the high-dimensional manifold topological feature set onto the macroscopic mechanical steady-state boundary, compare the spatial deviation distribution of the projected points, perform adaptive classification and partitioning, and generate closed-loop control messages. The specific implementation is as follows: The acceptance data processing server receives the high-dimensional manifold topology feature set output in step S3, and reads the high-dimensional manifold coordinates, node connectivity, original matrix coordinates of frozen physicochemical interaction nodes, and frozen vacancy markers. The high-dimensional manifold coordinates represent the spatial position of effective physicochemical interaction nodes after nonlinear spatial deformation mapping; the node connectivity represents the local transmission capability of the node in the holographic fusion topology matrix; and the frozen vacancy markers represent non-expandable nodes frozen by null space placeholders.

[0074] The acceptance data processing server synchronously reads the macroscopic mechanical steady-state boundary corresponding to the target production formula. The macroscopic mechanical steady-state boundary is jointly constituted by the Mooney viscosity tolerance band, the extreme range of the peak tolerance of the downstream internal mixer rotor torque, and the upper and lower limits of the tensile modulus of the final vulcanized product, serving as the first, second, and third dimensions of the three-dimensional tolerance space, respectively. The macroscopic mechanical steady-state boundary is defined by the lower boundary vector. and upper boundary vector Enclosure formation:

[0075] in, is the lower boundary vector of the macroscopic mechanical steady-state boundary; is the upper boundary vector of the macroscopic mechanical steady-state boundary; and These are the lower and upper limits of the Mooney viscosity tolerance zone corresponding to the target production formulation, respectively. and These are the lower and upper limits of the extreme range of the peak torque tolerance of the downstream internal mixer rotor, respectively. and These represent the lower and upper limits of the tensile modulus of the final vulcanized product, respectively. All these boundaries are based on the acceptance standard version corresponding to the current target production formulation.

[0076] The acceptance data processing server invokes a pre-defined manifold-mechanical projection model to convert the high-dimensional manifold topological feature set into a macroscopic mechanical spatial projection position. In one embodiment, the pre-defined manifold-mechanical projection model is implemented using a kernel ridge regression model. The training dataset for the manifold-mechanical projection model consists of gold production batch data under the same target production formula, with a training sample size of no less than 300 batches; when the number of historical gold production batches is less than 300, all historical batches under the current target production formula that have been verified to have no process anomalies are used. Each training sample includes a projected feature vector and back-end measured mechanical results. The projected feature vector is formed by concatenating the high-dimensional manifold coordinates, node connectivity, and number of frozen vacancy markers in the original matrix coordinate order from the high-dimensional manifold topological feature set. The back-end measured mechanical results include measured Mooney viscosity, measured peak torque of the downstream internal mixer rotor, and measured tensile modulus of the final vulcanized product. During model training, the acceptance data processing server first calculates the Euclidean distance between the projected feature vectors of each historical batch and takes the median of all pairwise distances as the kernel width. Then, a Gaussian kernel function is used to generate the kernel matrix between historical batches, and a ridge regression regularization term is superimposed on the kernel matrix. The training objective of the model is to minimize the squared error between the three-dimensional mechanical prediction results output by the model and the back-end measured mechanical results of historical batches, and to constrain the model coefficient amplitude through the ridge regression regularization term. The pre-training parameters of the manifold-mechanical projection model include 300 training samples, a kernel width of the median of pairwise distances, a regularization coefficient of 0.01, a maximum number of solution iterations of 1000, and a convergence error limit of 0.000001. The aforementioned pre-training parameters were determined by the playback test of the gold production batch dataset and were retrained when the target production formula or macroscopic mechanical steady-state boundary changed. When the current batch enters the projection processing, the acceptance data processing server generates the projection feature vector of the current batch in the same matrix coordinate order, calculates the kernel similarity between the projection feature vector and the projection feature vector of each historical batch, and then multiplies and sums the kernel similarity with the corresponding model coefficients to obtain the macroscopic mechanical spatial projection position of the current batch on the Mooney viscosity dimension, the rotor torque peak dimension, and the tensile modulus dimension.

[0077] The acceptance data processing server first converts the high-dimensional manifold topological feature set of the current batch into projected feature vectors. Projected feature vectors The high-dimensional manifold coordinates of all effective physicochemical interaction nodes, the connectivity of corresponding nodes, and the number of frozen vacancy markers are concatenated in the original matrix coordinate order. For frozen physicochemical interaction nodes, 0 coordinates are written at the corresponding positions and frozen vacancy markers are added to ensure that the projected feature vectors of different batches under the same target production formula have the same length.

[0078] Training phase, historical batches The projected eigenvector is denoted as The measured mechanical results at the back end are recorded as ,in This is the measured Mooney viscosity result corresponding to this historical batch. This is the measured peak torque result of the downstream internal mixer rotor. This represents the measured tensile modulus of the final vulcanized product. The kernel ridge regression model uses a Gaussian kernel function.

[0079] in, Projected feature vectors for the current batch Projected feature vectors of historical batches Kernel similarity between them; The kernel width is determined by the median of the pairwise distances between the projected feature vectors in the gold production batch dataset; Let be the Euclidean distance between the two projected eigenvectors.

[0080] Model coefficient matrix The following ridge regression relation is obtained by solving:

[0081] in, This is the kernel matrix between historical batches, and the matrix elements are composed of... constitute; , which is the ridge regression regularization coefficient, used to suppress the amplification of projection results by historical sample noise; It is the identity matrix; This is the model coefficient matrix; This is a matrix of measured mechanical results from the back end of historical batches.

[0082] When the current batch enters the projection processing, the acceptance data processing server generates the projection feature vector of the current batch according to the same matrix coordinate order as in the training phase, and calculates the kernel similarity between this projection feature vector and the projection feature vector of each historical batch. Subsequently, the acceptance data processing server multiplies each kernel similarity with the model coefficients of the corresponding historical batch and sums them to obtain the macroscopic mechanical spatial projection position of the current batch on the Mooney viscosity dimension, the rotor torque peak dimension, and the tensile modulus dimension. Macroscopic mechanical spatial projection position of the current batch. Generate according to the following relationship:

[0083] in, The number of historical batches used in training the manifold-mechanical projection model; Model coefficient matrix China and historical batches The corresponding coefficient vector; The projected coordinates of the Mooney viscosity dimension; The projected coordinates of the corresponding rotor torque peak dimension; The projected coordinates of the corresponding tensile modulus dimension. This macroscopic mechanical spatial projection position is used to determine the spatial deviation distribution pattern.

[0084] The acceptance data processing server will project the macroscopic mechanical space position. The data is compared dimension-by-dimensionally with the macroscopic mechanical steady-state boundary. For any mechanical dimension, when the corresponding projected coordinates are between the lower and upper boundaries, the dimension is recorded as the point within the boundary; when the corresponding projected coordinates are less than the lower boundary, the dimension is recorded as the lower deviation point; and when the corresponding projected coordinates are greater than the upper boundary, the dimension is recorded as the upper deviation point. The acceptance data processing server organizes the comparison results of the Mooney viscosity dimension, the rotor torque peak dimension, and the tensile modulus dimension into a spatial deviation distribution pattern in sequence.

[0085] Spatial deviation Calculated according to the following relationship:

[0086] in, This refers to the spatial deviation. The offset distance from the boundary of the Mooney viscosity dimension, when lie in and When the interval is 0, the value is 0. Time to take ,when Time to take ; represents the deviation distance outside the boundary of the rotor torque peak dimension, according to and , The comparison relationship is determined; The offset distance from the boundary of the tensile modulus surface is given by the following formula: and , The comparison relationship is determined. When calculating the spatial deviation, the acceptance data processing server first determines the boundary deviation distances of the Mooney viscosity dimension, the rotor torque peak dimension, and the tensile modulus dimension. When the projected coordinates of any dimension are between the corresponding lower and upper boundaries, the boundary deviation distance of that dimension is zero. When the projected coordinates of any dimension are lower than the corresponding lower boundary, the boundary deviation distance is obtained by subtracting the projected coordinates from the corresponding lower boundary. When the projected coordinates of any dimension are higher than the corresponding upper boundary, the boundary deviation distance is obtained by subtracting the corresponding upper boundary from the projected coordinates. The acceptance data processing server squares the three boundary deviation distances respectively, adds the squared results, and then performs square root processing to obtain the spatial deviation. Simultaneously, it counts the number of dimensions with boundary deviations and the number of frozen vacancy markers.

[0087] The acceptance data processing server calls the grade classification boundary corresponding to the target production formula. The grade classification boundary is determined by the distribution of projected distances of historical gold production batches within the macroscopic mechanical steady-state boundary. The historical projected distances are generated using the same calculation caliber as the current spatial deviation, that is, the deviation distances of historical batches outside the boundaries in the Mooney viscosity dimension, rotor torque peak dimension, and tensile modulus dimension are calculated separately, and the square root of the sum of squares is applied to the deviation distances outside the three boundaries. Specifically, the 75th percentile of the historical projected distances in ascending order is taken as the controlled warehousing boundary. The 90th percentile was taken as the boundary for review and diversion. The 75th percentile corresponds to the outer edge of the main projection distribution in historical gold production batches, used to distinguish whether the projection point of the current batch is still within the normal process fluctuation range; the 90th percentile corresponds to the outer edge of the edge projection distribution in historical gold production batches, used to distinguish whether the projection point of the current batch has entered the process edge range that needs to be reviewed. The acceptance data processing server is updated only under the same target production formula, the same macroscopic mechanical steady-state boundary, and the same back-end acceptance standard. and .

[0088] When spatial deviation When the number of frozen vacancy markers is 0, the acceptance data processing server will classify the current batch as a direct entry level; when However, when the number of frozen vacancy markers is greater than 0, it is classified as a controlled inbound level; when When the number of deviations from the boundary surface is 1, it is classified as a formula-restricted level; when When the number of deviations from the boundary surface reaches 2, it is classified as a review and diversion level; when If all three mechanical dimensions deviate from the boundary, the system is classified as a withdrawal interception level. The above classification result is the adaptive classification result.

[0089] The acceptance data processing server generates a closed-loop control message based on the adaptive grading results. This message includes the macroscopic mechanical spatial projection position, spatial deviation distribution, spatial deviation amount, adaptive grading results, original matrix coordinates of the frozen vacancy markers, and the material feeding status for the current batch. The direct warehousing level adds the current batch to the available batch queue; the controlled warehousing level adds the original matrix coordinates of the frozen vacancy markers to the available batch queue; the formula-restricted level only opens the batch to feeding tasks matching the current target production formula; the review and diversion level generates a re-inspection task for the deviation dimension and suspends the batch from entering the internal mixing feeding queue; and the return-to-warehouse interception level disables the batch's material feeding access. After the closed-loop control message is generated, the acceptance grading process for the current natural rubber raw material batch is complete.

[0090] After step S4 is completed, the acceptance data processing server obtains the macroscopic mechanical spatial projection position, spatial deviation distribution, spatial deviation amount, number of deviation surfaces outside the boundary, and number of frozen vacancy markers for the current batch within the macroscopic mechanical steady-state boundary. Based on this, it generates adaptive grading results and closed-loop control messages. The closed-loop control messages include the current batch's material feeding status, the original matrix coordinates of frozen vacancy locations, and corresponding handling information, used to complete the adaptive grading processing of natural rubber raw material acceptance data.

[0091] Please see Figure 2 This invention provides an intelligent grading system for natural rubber raw material acceptance data, comprising: The parameter pre-tuning module is used to collect multi-source heterogeneous detection parameters in response to material acceptance instructions, form a set of physical rheological parameters and a set of chemical composition parameters according to physical and chemical properties, and perform dimensional normalization and heterogeneous noise smoothing filtering to output a standard dual-parameter sequence. The missing data verification module is used to receive standard dual-parameter sequences and map them into dual-parameter fusion state matrices according to the orthogonal arrangement principle. When a data missing is identified, the module performs a credit verification on the remaining valid data, generates a compensating state vector or a null space placeholder based on the verification result, and outputs a holographic fusion topology matrix. The topology deformation module is used to construct a physical-chemical interaction topology graph based on the node association connectivity constraints for the holographic fusion topology matrix, and to perform nonlinear spatial deformation mapping to extract a high-dimensional manifold topology feature set; The projection classification module projects the high-dimensional manifold topological feature set onto the macroscopic mechanical steady-state boundary, compares the spatial deviation distribution of the projection landing point, performs adaptive classification, and generates closed-loop control messages.

[0092] The above embodiments are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for intelligent grading of natural rubber raw material acceptance data, characterized in that, Including the following steps: Step S1: In response to the material acceptance command, collect multi-source heterogeneous detection parameters, form a set of physical rheological parameters and a set of chemical composition parameters according to physical and chemical properties, and perform dimensional normalization and heterogeneous noise smoothing filtering to output a standard dual-parameter sequence. Step S2: Receive the standard dual-parameter sequence and map it into a dual-parameter fusion state matrix according to the orthogonal arrangement principle. When a data gap is identified, perform a credit check on the remaining valid data, generate a compensating state vector or a null space placeholder based on the check result, and output the holographic fusion topology matrix. Step S3: For the holographic fusion topology matrix, construct a physical-chemical interaction topology graph based on the node association connectivity constraints, and perform nonlinear spatial deformation mapping to extract the high-dimensional manifold topology feature set; Step S4: Project the high-dimensional manifold topological feature set onto the macroscopic mechanical steady-state boundary, compare the spatial deviation distribution of the projection points, perform adaptive classification and division, and generate closed-loop control messages.

2. The intelligent grading method for natural rubber raw material acceptance data according to claim 1, characterized in that, In step S1, the multi-source heterogeneous detection parameters are written into the current batch according to the binding relationship between the detection equipment channel and the acceptance item. The detection values ​​belonging to shear force or rheological response are assigned to the physical rheological parameter set, and the detection values ​​belonging to component ratio or chemical state characterization are assigned to the chemical component parameter set. Data gap markers are generated at the positions where no valid values ​​are returned. For the valid raw test values ​​in the physical rheological parameter set and the chemical component parameter set, the minimum and maximum usable boundaries are determined based on the intersection of the acceptance valid range corresponding to the target production formula and the range of the testing equipment. The valid raw test values ​​are then converted into dimensionless test values, and data gap markers are not included in the dimension normalization process.

3. The intelligent grading method for natural rubber raw material acceptance data according to claim 2, characterized in that, In step S1, a sliding window is established within the same detection parameter. The sliding window only includes valid dimensionless detection values. When the number of valid dimensionless detection values ​​in the window reaches the valid number requirement of the sliding window, the value in the middle after sorting is taken to generate a smooth value. When the effective number of sliding windows is not met, data gap markers are retained at the corresponding positions, and the physical rheological parameter sequence and chemical component parameter sequence are organized into a standard dual-parameter sequence.

4. The intelligent grading method for natural rubber raw material acceptance data according to claim 3, characterized in that, In step S2, the matrix nodes of the dual-parameter fusion state matrix are determined by the smoothed values ​​of the chemical composition in the corresponding row and the smoothed values ​​of the physical rheology in the corresponding column, and stored as a physicochemical interaction node vector. When any source position involved in the formation of a matrix node has a data gap marker, the matrix node does not generate a valid physical-chemical interaction node vector, but is recorded as a gap.

5. The intelligent grading method for natural rubber raw material acceptance data according to claim 4, characterized in that, In step S2, the confidence level review includes: reading continuous and effective physical rheological smoothing values ​​along the time axis and accumulating the fluctuation contribution values ​​of adjacent sampling segments to form viscoelastic fluctuation evolution; organizing the effective chemical component smoothing values ​​into chemical component vectors, constructing an autocorrelation matrix and extracting the maximum eigenvalue to form the component distribution principal element; inputting the viscoelastic fluctuation evolution and the component distribution principal element into the isolated forest model in a fixed order, and outputting the abnormal isolated state quantity.

6. The intelligent grading method for natural rubber raw material acceptance data according to claim 5, characterized in that, In step S2, the safety envelope boundary is generated from the historical abnormal isolated state quantities corresponding to the gold production batch dataset; when the abnormal isolated state quantity falls into the safety envelope boundary, a compensatory state vector is generated and written into the vacancy; when the abnormal isolated state quantity falls out of the safety envelope boundary, the vacancy is frozen with a zero space placeholder and a frozen vacancy marker is generated.

7. The intelligent grading method for natural rubber raw material acceptance data according to claim 6, characterized in that, In step S3, the matrix position in the holographic fusion topology matrix is ​​defined as a physical-chemical interaction node. The physical-chemical interaction node includes an effective physical-chemical interaction node and a frozen physical-chemical interaction node. The effective physical-chemical interaction node participates in the calculation of node association connectivity, node association strength and graph Laplacian matrix, while the frozen physical-chemical interaction node only retains the original matrix coordinates and the frozen vacancy mark.

8. The intelligent grading method for natural rubber raw material acceptance data according to claim 7, characterized in that, In step S3, each effective physicochemical interaction node is treated as a graph node, and the node pairs that are adjacent in the row direction or column direction and whose two ends are both effective physicochemical interaction nodes are treated as graph edges to form a physicochemical interaction topology graph. The node association strength is calculated based on the number of adjacent effective nodes and the dimensionless distance between nodes, written into the association strength matrix, and Laplace eigenmap is performed accordingly to form a high-dimensional manifold topology feature set.

9. The intelligent grading method for natural rubber raw material acceptance data according to claim 8, characterized in that, In step S4, the macroscopic mechanical steady-state boundary is jointly formed by the upper and lower boundaries of the Mooney viscosity dimension, the rotor torque peak dimension, and the tensile modulus dimension. The high-dimensional manifold topological feature set is converted into a projection feature vector and input into the manifold-mechanical projection model to generate the macroscopic mechanical spatial projection position. Based on the macroscopic mechanical spatial projection position, the spatial deviation and the number of deviation dimensions outside the boundary are calculated, and a closed-loop control message is generated by combining the number of frozen vacancy markers.

10. A natural rubber raw material acceptance data intelligent grading system, used to implement the natural rubber raw material acceptance data intelligent grading method according to any one of claims 1-9, characterized in that, include: The parameter pre-tuning module is used to collect multi-source heterogeneous detection parameters in response to material acceptance instructions, form a set of physical rheological parameters and a set of chemical composition parameters according to physical and chemical properties, and perform dimensional normalization and heterogeneous noise smoothing filtering to output a standard dual-parameter sequence. The missing data verification module is used to receive standard dual-parameter sequences and map them into dual-parameter fusion state matrices according to the orthogonal arrangement principle. When a data missing is identified, the module performs a credit verification on the remaining valid data, generates a compensating state vector or a null space placeholder based on the verification result, and outputs a holographic fusion topology matrix. The topology deformation module is used to construct a physical-chemical interaction topology graph based on the node association connectivity constraints for the holographic fusion topology matrix, and to perform nonlinear spatial deformation mapping to extract a high-dimensional manifold topology feature set; The projection classification module projects the high-dimensional manifold topological feature set onto the macroscopic mechanical steady-state boundary, compares the spatial deviation distribution of the projection landing point, performs adaptive classification, and generates closed-loop control messages.

Images