A dynamic optimization control method for process parameters in a high-toughness product production process

Abstract

The present application relates to the technical field of parameter regulation, in particular to a high-toughness product production process parameter dynamic optimization control method, the present application extracts the derivative of the time sequence signal of each subzone pressure of the product, determines the starting time of the subzone based on the continuity of the sign of the derivative from negative to positive, and constructs a time sequence difference matrix; according to the difference matrix, the heating power and the heating duration of each subzone are regulated in groups and layers to form a thermal performance control vector; collect the full-field deformation data in the process of graded pressure, and construct a deformation reference based on the history of qualified products, identify the nonlinear amplification characteristics through the deviation and its increment slope, realize defect positioning and form a defect feature vector; through the spatial homodirection of the thermal performance control vector and the defect feature vector across batches, the inflection point identification sensitivity parameter is adaptively calibrated. The present application can improve the process control precision and is suitable for high consistency manufacturing scenarios.

Description

Technical Field

[0001] This invention relates to the field of parameter control technology, specifically to a method for dynamic optimization and control of process parameters in the production process of high-toughness products. Background Technology

[0002] In continuous batch production, the toughness of high-toughness, compression-resistant products depends primarily on the diffusion depth at the interface of the heat-sealing process. Existing production control systems often employ static parameter table settings or single-process, single-loop PID feedback control.

[0003] Existing control methods rely on the assumptions of a stable process environment and direct online measurement of the controlled variable. They lack real-time characterization of variables and cannot adaptively perceive the dynamic shift of the optimal window. Therefore, when faced with multiple time-varying disturbances, existing systems cannot automatically optimize parameters and struggle to implement dynamic collaborative compensation and adaptive control of multiple variables such as heat sealing temperature, pressure, and speed. This results in the system's inability to self-adjust to maintain the set optimal performance, ultimately leading to poor product toughness consistency and large batch-to-batch fluctuations in compressive strength.

[0004] To address this, a dynamic optimization control method for process parameters in the production of high-toughness products is proposed. Summary of the Invention

[0005] The purpose of this invention is to provide a dynamic optimization control method for process parameters in the production of high-toughness products. By identifying the derivative inflection point, a difference matrix is ​​constructed to achieve vectorization of heating regulation. Defect location is achieved by combining graded pressure deformation analysis. Cross-batch isotropic statistics are used to achieve adaptive parameter calibration, forming a closed-loop optimization control.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for dynamic optimization and control of process parameters in the production of high-toughness products includes: Pressure timing signals are passively collected in each zone of the product, and the first derivative of the product is extracted in real time. The start time of each zone is determined by the moment when the derivative sign changes from negative to positive, and a zone start time difference matrix is ​​constructed. Based on the difference matrix, the heating power is adjusted for the earliest positive zone, and the heating duration is adjusted for the lagging zone according to the difference ratio. The power switching time and duration of each zone are recorded as the thermal performance control vector. The product is graded and controlled, and deformation data is collected in the entire field at each pressure level. The deformation data is compared with the statistical deformation benchmark table at each level. The defect location is obtained by the characteristic of nonlinear amplification of layered defect deformation with pressure. The detection resource allocation is optimized based on the partition with the largest difference in the thermal performance control vector, and the defect feature vector is output. The spatial co-directionality of thermal performance control vectors and defect feature vectors in each batch is statistically analyzed. When the co-directionality exceeds the judgment threshold, the sensitivity parameter for identifying the inflection point of the corresponding calibrated zone is adjusted.

[0007] Preferably, the steps of extracting the first derivative of the product and determining the start time of each partition specifically include: for the pressure time series signal passively collected from each partition, sliding forward step by step with a preset time window, calculating the rate of change of pressure value with respect to time within each time window, and forming the derivative time series of the partition; The continuity of the derivative time series is confirmed. Only when the derivative sign changes from negative to positive and remains positive for a preset number of consecutive time steps, the first sign flip time is determined as the start time of the corresponding partition. A difference matrix is ​​formed by using the time difference between the start times of any two partitions as matrix elements and arranging them with the partition number as the row and column index. The number of time steps used for inflection point continuity confirmation is archived as a sensitivity parameter for inflection point identification.

[0008] Preferably, the steps of adjusting heating power and heating duration based on the difference matrix specifically include: extracting the absolute value of the time difference between each partition and the earliest starting partition from the difference matrix; dividing the absolute value of the time difference into control levels according to preset segment intervals, and grouping partitions whose absolute values ​​of time difference fall within the same segment interval into the same control group; for the earliest starting partition, switching the heating power from the current value to the target value with a predetermined reference slope, while keeping the heating duration unchanged from the reference duration; for partitions within the same control group, sharing the heating power slope, and extending the heating duration proportionally above the reference duration according to the ratio of the absolute value of the time difference of this partition to the maximum absolute value of the time difference within the group; after the heating control is completed, each partition exits the heating stage and enters a unified graded pressurization stage, ensuring that the heating control and graded pressurization do not overlap in timing; The power switching time, the power slope value after switching, and the heating duration after regulation for each zone are recorded as ternary components in the thermal performance regulation vector. The set of ternary components constitutes the complete thermal performance regulation vector.

[0009] Preferably, the steps for obtaining defect location specifically include: full-field deformation data is collected by a distributed sensor array at the corresponding location of each partition, each sensor node outputs the local deformation amount of the corresponding partition under the current pressure level, and the overall set of data from each node constitutes the full-field deformation data for each pressure level. For historical qualified batches of products, record the total deformation of each zone at the end of each pressure level under the same pressure classification sequence. Statistically analyze the historical deformation of each zone under the same pressure level. The center value of the distribution and the diffusion range value together constitute the deformation benchmark entry of the zone under the pressure level. All deformation benchmark entries are summarized into a zone-pressure level deformation benchmark table. For the current batch, the deformation at the end of the current pressure level of each partition is compared with the corresponding entry in the partition-pressure level deformation reference table to calculate the deviation. The incremental slope of the deviation between adjacent pressure levels is used as a nonlinear amplification feature. When the incremental slope of a certain partition shows a continuous increasing trend as the pressure level advances and exceeds the preset multiple, it is marked as a suspected defect partition. The partition numbers of each suspected defect partition, the deviation of each pressure level, and the incremental slope value are summarized to form a defect feature vector.

[0010] Preferably, the step of outputting a complete defect feature vector specifically includes: extracting the heating duration component value of each zone from the thermal performance control vector; calculating the absolute value of the difference between the duration component value of each zone and the mean of the duration component values ​​of all zones to obtain the thermal deviation of each zone; sorting each zone according to the thermal deviation value from largest to smallest to form a thermal deviation sorting list; cross-comparing the thermal deviation sorting list with the list of suspected defect zones in the initial content of the defect feature vector: assigning denser subsequent pressure level detection interval values ​​to zones with higher thermal deviation values ​​that are also listed as suspected defect zones, and using the upper limit value for deviation judgment; relaxing the detection interval values ​​for zones with lower thermal deviation values ​​that are not listed as suspected defect zones, and using the upper limit value for deviation judgment; appending the final detection pressure level interval value and the upper limit value for deviation judgment to the defect feature vector, which together with the zone number, the deviation value of each pressure level, and the incremental slope value constitutes a complete defect feature vector.

[0011] Preferably, the step of statistically analyzing the spatial homogeneity of thermal performance control vectors and defect feature vectors across batches and calibrating sensitivity parameters specifically includes: for continuously accumulated batches, extracting the heating duration component value from the thermal performance control vector of each batch and the incremental slope value of the corresponding batch in the defect feature vector; under each batch dimension, pairing and arranging the heating duration component value and incremental slope value corresponding to each batch, and statistically analyzing the proportion of batches whose changes are in the same direction to the total number of batches participating in the statistics, as the spatial homogeneity index of the batch; when the spatial homogeneity index of a certain batch exceeds the judgment threshold, the time step used for confirming the continuity of the inflection point of the batch is reduced by a preset number of steps based on the current value; after each calibration, the adjusted continuous time step value, the corresponding batch number, and the spatial homogeneity index value triggered at the time of calibration are recorded together.

[0012] Preferably, the judgment threshold is adaptively updated as batches accumulate. Specifically, this includes: using a fixed number of consecutive batches as a statistical window, calculating the difference between the maximum and minimum values ​​of the spatial homogeneity index of each partition within the current window, which is used as the homogeneity distribution range of the current window; taking the median of the homogeneity distribution range of all historically completed statistical windows, which is used as the benchmark value of the global judgment threshold; for partitions that have triggered sensitivity parameter calibration at least once in historical statistical windows, the individual judgment threshold is lowered based on the global benchmark value according to the ratio of the number of historical triggers to the length of the statistical window, so that partitions that have frequently triggered calibration in the past can trigger a new round of calibration at a lower level of spatial homogeneity index. The updated global baseline value and individual judgment thresholds for each partition, along with the current spatial homogeneity index values ​​for each partition, will be uniformly incorporated into the next round of cross-batch homogeneity statistics. Together with the thermal performance control vector and the complete defect feature vector, they will participate in subsequent iterations to form a closed-loop optimization of cross-batch process parameters.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention extracts the first derivative of the pressure timing signal of each zone in real time and uses the stable reversal of the derivative sign from negative to positive as the criterion for the zone's start time, constructing a difference matrix for the zone's start time. This replaces the traditional method that relies on absolute thresholds or empirical time nodes. This method utilizes derivative continuity constraints to improve the noise resistance and robustness of inflection point identification, avoiding false triggering of start-up determination due to signal fluctuations. Simultaneously, by characterizing the relative timing relationship of each zone's response through the difference matrix, process control is upgraded from "single-point triggering" to "global timing coordination," providing a precise basis for subsequent zone-specific control, thereby significantly improving the precision and consistency of process control.

[0014] 2. This invention employs a grouped and stratified coordinated control strategy for thermal power and heating duration in each zone based on a difference matrix. It maps time differences to control levels, achieving a structured control mechanism of "same slope within the same group, different duration between different groups." This mechanism can compensate for lagging zones while ensuring a consistent overall heating trend, effectively reducing the non-uniformity of thermal history in different regions. Simultaneously, it structures the control process into a thermal performance control vector, providing a quantifiable and traceable data representation for the process. This transforms thermal parameters from discrete control variables into data objects with a vectorized analysis foundation, significantly improving the computability and reusability of process optimization.

[0015] 3. This invention constructs a defect identification mechanism based on full-field deformation data under graded pressure and introduces the "deviation increment slope" as a nonlinear amplification feature, achieving early identification and precise location of latent defects. Furthermore, by combining cross-batch spatial homogeneity analysis of thermal performance control vectors and defect feature vectors, a statistical correlation between process parameters and defect evolution is established, and the inflection point identification sensitivity parameter is dynamically calibrated accordingly, forming a closed-loop adaptive optimization mechanism. This method breaks through the traditional static parameter setting mode, enabling the system to continuously self-correct with the accumulation of batch data, improving defect prediction accuracy and long-term system stability, and significantly enhancing the intelligence level of the production process. Attached Figure Description

[0016] Figure 1 A flowchart illustrating a method for dynamic optimization and control of process parameters in the production process of high-toughness products provided by this invention; Figure 2 A schematic diagram of the logic flow for optimizing detection resources and outputting defect feature vectors provided by this invention; Figure 3 A schematic diagram of the cross-batch adaptive update logic flow for the determination threshold provided by the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention.

[0018] Example 1:

[0019] This invention relates to the dynamic optimization and control of process parameters in the production of high-toughness products. The high-toughness products are fiber-reinforced rubber composite materials, and their production process includes a heating curing stage and a pressure curing stage. The two stages are carried out sequentially and do not overlap.

[0020] The partitioning method in this invention is as follows: Based on the geometry of the mold and the heat transfer path, the mold cavity surface is divided into several partitions according to spatial location. Heat conduction exists between adjacent partitions through the mold's metal substrate, but each partition is equipped with an independent heating control loop. Typically, these are divided into three categories: central partition, transition partition, and edge partition. The edge partition, due to its larger contact area with the outer edge of the mold for heat dissipation, has a higher heating rate than the central partition, which is the fundamental reason for the timing differences in the process response of each partition. The partition number is indexed by the spatial coordinates of the partition on the mold cavity surface, maintaining a unique correspondence throughout the entire text.

[0021] Please see Figure 1This invention provides a method for dynamic optimization and control of process parameters in the production of high-toughness products. The technical solution is as follows: Pressure timing signals are passively collected in each zone of the product. The first derivative of the product is extracted in real time. The starting time of each zone is determined by the moment when the derivative sign changes from negative to positive, and a difference matrix of the starting time of each zone is constructed. Based on the difference matrix, the heating power of the zone that enters positive value earliest is adjusted, and the heating duration of the lagging zone is adjusted according to the difference ratio. The power switching time and duration of each zone are recorded as a thermal performance control vector. The product is controlled in stages, and deformation data is collected across all pressure levels. The deformation data is compared level by level with a statistical deformation benchmark table. The defect location is obtained by using the nonlinear amplification of layered defect deformation with pressure. The detection resource allocation is optimized based on the zone with the largest difference in the thermal performance control vector, and a defect feature vector is output. The spatial co-directionality of the thermal performance control vector and the defect feature vector of each zone is statistically analyzed across batches. When the co-directionality exceeds the judgment threshold, the sensitivity parameter for identifying the inflection point of the corresponding zone is calibrated.

[0022] Furthermore, the steps for extracting the first derivative of the product in real time and determining the start time of each partition specifically include: The pressure timing signals passively collected from each partition are gradually slid backward in a preset time window. The rate of change of pressure value with respect to time is calculated within each time window to form the derivative timing of that partition. The pressure timing signals are led out through sensing channels corresponding to the location of each partition. The sensing nodes are arranged on the outside of the mold, and the pressure of the medium inside the mold cavity is used to indirectly characterize the pressure on each partition. The continuity of the derivative timing is confirmed. Only when the derivative sign changes from negative to positive and remains positive for a subsequent preset number of time steps, is the first sign flip time determined as the start time of the corresponding partition. A difference matrix is ​​constructed by using the time difference between the start times of any two partitions as matrix elements and arranging them with the partition number as the row and column index. The sign of the matrix elements reflects the order of the process response in the two partitions, and the absolute value of the matrix elements reflects the degree of response lag. The number of time steps used for inflection point continuity confirmation is archived as a sensitivity parameter for inflection point identification and used for subsequent cross-batch calibration.

[0023] Specifically, during the mold design phase, sensing channels are pre-reserved at corresponding positions on the mold wall according to the spatial location of each zone. The channel axis is perpendicular to the mold cavity surface, and the inner diameter meets the installation requirements of the sensing element. The sensing element is arranged at the outer end of the sensing channel, and the pressure-sensing surface of the sensing element maintains pressure transmission coupling with the working medium in the mold cavity through a thin-walled structure. To prevent venting and pressure relief from interfering with the actual pressure fluctuations at the material's gel point, when the mold has an open venting channel, the venting valve must be closed within a preset time after the heating stage begins to ensure that the mold cavity is a closed system when the material reaches near the gel point.

[0024] The pressure-sensing surface of the sensing element is filled with a high-temperature resistant sealing material between itself and the inner wall of the sensing channel to prevent high-temperature materials from entering the channel. Each zone corresponds to at least one sensing node, and each sensing node is connected to the data acquisition module on the outside of the mold through a lead wire channel pre-embedded inside the template. The data acquisition module synchronously records the pressure readings of the sensing nodes in each zone at a frequency of ten samples per second, generating a set of pressure timing signals indexed by the zone number.

[0025] The physical evolution of the pressure time-series signals for each zone is as follows: In the initial stage of heating and curing of the high-toughness composite material, as the temperature rises, the material viscosity decreases, and the volume undergoes a brief expansion or localized flow, causing a slight decrease in the pressure of the medium within the mold cavity. This stage corresponds to the decreasing segment of the pressure time-series signal for each zone, and its first-order time derivative is negative. When the degree of cross-linking reaction inside the material reaches the gel point, the material transforms from a viscous flow state to an elastic solid state, the modulus rises sharply, and the volume shrinkage trend turns into a pressure rebound. This stage corresponds to the starting point of the rebound of the pressure time-series signal for each zone, and its first-order time derivative changes from negative to positive. The moment when the derivative sign changes from negative to positive is the characteristic moment of the start of material curing and cross-linking in each zone. This invention uses this moment as the starting moment for the corresponding zone. Since the heating rate of the edge zone is higher than that of the center zone, the moment when its derivative changes from negative to positive is earlier than that of the center zone. The difference in the starting moment of each zone quantifies the difference in the thermal response time-series of each zone under the current batch.

[0026] The data acquisition module calculates the first derivative estimate in real time for the pressure time series signals collected from each partition in the following way: a time window of preset length is gradually slid backward on the time series signal, moving backward by one sampling step each time. Within the current time window, the difference between the pressure value at the end sampling point and the pressure value at the beginning sampling point is divided by the window time span to obtain the derivative estimate at the center time of the window, thereby forming the derivative time series of each partition.

[0027] The method for determining the preset window length: The selection of the time window length is based on the criterion of effectively suppressing random noise in the sensing signal, while ensuring that the fluctuations of the real physical signal are not excessively smoothed. Before implementation, statistical analysis is performed on the historical pressure time-series signals of each zone of the same type of product, and the amplitude standard deviation of the high-frequency noise in the signal is extracted. The initial window length is determined with the constraint that the statistical average of the noise fluctuation within the time span corresponding to the window length does not exceed 20% of the absolute value of the derivative at the real physical inflection point. For the typical high-toughness product of this invention, the initial window length corresponds to five sampling steps, that is, the pressure difference between the current and the previous four sampling points is used as the derivative estimate. In the subsequent batch accumulation process, if systematic misidentification occurs, the window length can be fine-tuned based on the batch historical data, with an adjustment range of three to eight sampling steps. When this method is extended to other material systems (such as thermoplastic elastomers), since different materials have different viscosity change rates, the above-mentioned initial window length needs to be recalibrated according to the characteristic noise frequency band of the new product.

[0028] Determining the start time solely based on a single derivative sign flip may lead to misjudgment due to local signal noise. Therefore, the derivative timing is validated as follows: when the derivative estimate of a certain partition first becomes positive, it is further checked whether the derivative estimates for several consecutive sampling steps remain positive. If the condition of continuous positive values ​​is met, the moment when the first derivative estimate is positive is determined as the start time of that partition. If the derivative becomes negative again within a subsequent validation window, the validation is canceled, and the system waits for the next derivative sign flip.

[0029] The initial value for the number of time steps required for continuous confirmation is determined as follows: Based on the statistical characteristics of the derivative time series near the true inflection point of each partition in historical batches, the minimum number of continuous confirmation time steps that ensures a false positive rate of no more than 5% is selected as the initial value. For the typical high-toughness product of this invention, the initial number of continuous confirmation time steps is set to five sampling steps. That is, after the first positive derivative appears, the derivative of the next five consecutive sampling steps is positive before the inflection point is confirmed. This number of continuous confirmation time steps is the sensitivity parameter for inflection point identification described in this invention, and it is archived using the partition number as an index. In the initial stage of production of new products where no historical batch data is available, the initial value of the time steps is determined based on the ratio of the nominal gel time of the material at the target curing temperature to the system sampling period. The sampling steps corresponding to 1 / 20 to 1 / 10 of the gel time are taken as the initial time steps to solve the cold start problem.

[0030] After the start time of each partition is determined, a partition start time difference matrix is ​​constructed using the time difference between the start times of any two partitions as matrix elements, arranged by partition number as row and column index. The element in the i-th row and j-th column of the matrix represents the difference between the start time of partition j and the start time of partition i. A positive value indicates that the process response of partition j lags behind that of partition i, while a negative value indicates that it leads. The absolute value reflects the degree of lag. The diagonal elements of the matrix are zero, and the matrix satisfies antisymmetry.

[0031] Furthermore, the steps for adjusting the heating power and heating duration based on the difference matrix specifically include: Extract the absolute value of the time difference between each partition and the earliest starting partition from the difference matrix; divide the absolute value of the time difference into several control levels according to the preset segment intervals, and group the partitions whose absolute value of the time difference falls into the same segment interval into the same control group. For the earliest starting zone, the heating power is switched from the current value to the target value with a predetermined reference slope, and the heating duration remains unchanged with the reference duration. For the other zones in the same control group, the heating power slope is shared with the group, the heating duration is above the reference duration, and is extended proportionally according to the ratio of the absolute value of the time difference of this zone to the absolute value of the maximum time difference in the group. After the heating control is completed, each zone exits the heating stage and enters the unified stage of graded pressurization, ensuring that the heating control and graded pressurization do not overlap in time. The power switching time, the power slope value after switching, and the heating duration after regulation for each zone are recorded as ternary components in the thermal performance regulation vector for that zone. The set of ternary components constitutes the complete thermal performance regulation vector.

[0032] Specifically, from the difference matrix, using the earliest starting partition (i.e., the partition with the smallest starting time, usually a marginal partition) as a reference, the absolute value of the time difference of each partition relative to the reference partition is extracted. The absolute values ​​of the time differences of each partition are then classified according to preset segment intervals, and partitions whose absolute values ​​of time differences fall within the same segment interval are grouped into the same control group.

[0033] The method for determining the preset segmentation intervals is as follows: Based on the typical distribution range of the difference matrix at the initial time of each partition in historical batches of similar products, the absolute value distribution range of each non-zero element in the difference matrix is ​​evenly divided into several segments. This ensures that the partitions within each segment can be considered equivalent in terms of heating control response, and that the control differences between adjacent segments are sufficient to significantly affect the final curing uniformity of the product. The initial segmentation intervals are divided using the quartiles of the difference matrix elements from historical batches as boundaries, and these boundary values ​​can be updated subsequently based on batch data.

[0034] For the earliest starting zone, based on the arrival of the confirmation signal at its starting time in the difference matrix, the heating power of the zone is switched down from the current operating power level to the target heat preservation power level with a predetermined reference slope, and the heating duration remains unchanged with the reference duration.

[0035] The predetermined baseline slope is determined using thermal simulation. Given known parameters such as the target process temperature, material thermal diffusivity, and mold heat capacity, the power switching slope is calculated to ensure that the temperature of the earliest starting zone remains within the acceptable tolerance range of the target curing temperature (within 3°C above and below the target curing temperature) during the subsequent curing time, without causing excessive temperature rise due to continuous power maintenance. This slope is used to determine the baseline slope. The specific value corresponding to the target insulation power level is also determined through thermal simulation to balance the heat generated at this power with the heat dissipation from the mold to the environment, ensuring that the temperature of the earliest starting zone does not continue to rise within the baseline duration.

[0036] For the remaining zones within the same control group, the heating power switching slope is the same as that of the group, and the heating duration is longer than the baseline duration. The heating is extended proportionally according to the ratio of the absolute value of the time difference in this zone to the absolute value of the maximum time difference within the control group. The physical meaning of this extension method is: the larger the absolute value of the time difference in a zone, the greater the lag in its curing process, requiring a longer heating time to ensure its curing degree catches up with the earliest starting zone. The extension amount is proportional to the lag. Since the isothermal curing kinetics of polymer materials typically exhibit autocatalytic nonlinear characteristics, the above proportional linear extension rule applies when the absolute value of the time difference between each zone does not exceed 30% of the baseline duration. When the absolute value of the time difference exceeds 30%, a nonlinear compensation rule is adopted: using the isothermal curing kinetic curve of the product material as a mapping function, the additional heating time required to ensure that the difference in curing degree between the lagging zone and the earliest starting zone does not exceed 5% when exiting the heating stage is calculated, replacing the simple proportional extension.

[0037] After all zones have completed differentiated heating according to the power switching times and durations specified by the thermal performance control vector, each zone uniformly exits the heating phase before entering the next stage of graded pressurization. The heating control phase is completed before the start of the graded pressurization phase, and the two phases do not overlap in time, thus avoiding interference from changes in heating power on the pressure baseline during the graded pressurization process. A short stabilization waiting interval is set between the end of the heating phase and the start of the graded pressurization phase to ensure that the temperature field of the mold and the product reaches a relatively steady state before pressurization begins.

[0038] The following three pieces of information for each partition are recorded as ternary components in the thermal performance control vector: power switching time (relative time value with the curing start time as zero), the slope of the heating power after switching (i.e., the rate of power reduction per unit time), and the duration of heating after control. The set of ternary components for all partitions, arranged by partition number as index, constitutes the complete thermal performance control vector.

[0039] Furthermore, the steps of comparing deformation data step by step based on the statistical deformation benchmark table and obtaining defect location specifically include: The full-field deformation data is collected by a distributed sensor array located at the corresponding positions of each partition of the mold. Each sensor node outputs the local deformation of the corresponding partition under the current pressure level. The overall collection of data from each node constitutes the full-field deformation data for each pressure level. For historical qualified batches of products, record the total deformation of each zone at the end of each pressure level under the same pressure classification sequence. Statistically analyze the historical deformation of each zone under the same pressure level. The center value of the distribution and the diffusion range value together constitute the deformation benchmark entry of the zone under the pressure level. All deformation benchmark entries are summarized into a zone-pressure level deformation benchmark table. For the current batch, the deformation at the end of the current pressure level of each partition is compared with the corresponding entry in the partition-pressure level deformation reference table to calculate the deviation. The incremental slope of the deviation between adjacent pressure levels is used as a nonlinear amplification characteristic. When the incremental slope of a certain partition shows a continuous increasing trend as the pressure level advances and exceeds the preset multiple, the partition is marked as a suspected defect partition. The initial content of the defect feature vector is formed by summarizing the partition number of each suspected defect partition, the deviation of each pressure level, and the incremental slope value.

[0040] Specifically, after the product is demolded, it enters the inflation and testing station. During the graded pressurization process of each air chamber of the product, after the pressure at each pressure level stabilizes, the normal displacement of each zone is collected by a distributed sensor array arranged at each zone detection position. Each sensor node outputs the local deformation of its corresponding zone under the current pressure level. The overall set of data from all sensor nodes constitutes the full-field deformation data under that pressure level. The graded pressurization pressure levels are inflated sequentially at five levels: 20%, 40%, 60%, 80%, and 100% of the rated working pressure. The stabilization time for each pressure level is determined by the fluctuation amplitude of the sensor node readings being less than 1% of the range. Once this condition is met, the full-field deformation data collection for the current pressure level is triggered.

[0041] The historical deformation benchmark table for qualified products by region is established based on the data of historical qualified batches of products that have passed full-item performance testing. For each product configuration, no fewer than thirty products are randomly selected from qualified batches that have passed all acceptance inspections. Under the same pressure grading sequence as the current test, the total deformation of each region at the end of each pressure level is recorded sequentially, thereby accumulating a set of historical deformation data for each region at each pressure level.

[0042] For each partition at each pressure level, historical deformation data sets are statistically analyzed. The distribution center value and diffusion range value of the set are extracted. The center value minus twice the diffusion range value is used as the lower limit of the deformation benchmark entry for that partition at that pressure level, and the center value plus twice the diffusion range value is used as the upper limit, forming the deformation benchmark interval for that partition at that pressure level. The deformation benchmark intervals for all partitions and all pressure levels are summarized to form a partition-pressure level deformation benchmark table, which is stored in the benchmark database using both the partition number and the pressure level number as dual indexes.

[0043] The center value is taken as the arithmetic mean of the historical deformation data set for the corresponding zone and pressure level, and the diffusion range value is taken as the standard deviation of the set. The physical meaning of using the mean plus or minus two standard deviations as the baseline interval is that if the deformation of the zone in the historical data approximately follows a normal distribution, then about 95% of the deformation of qualified products will fall within this interval. The boundary of the baseline interval should not frequently trigger false alarms due to random disturbances during processing, while maintaining sufficient sensitivity to abnormal deformations exceeding two standard deviations. When the number of historical samples is less than thirty, the upper limit of the baseline interval should be temporarily widened, using the mean plus three standard deviations instead, and the standard setting should be restored after the sample size reaches thirty.

[0044] For the current batch of products, after the pressure holding at each pressure level is stable, the current total deformation of each zone is compared with the deformation reference interval corresponding to that zone and pressure level in the zone-pressure level deformation reference table. The deviation of each zone under that pressure level is calculated. The deviation is defined as the difference between the current deformation and the center value of the deformation reference interval. A positive value indicates that the current deformation exceeds the center value, and a negative value indicates that it is lower than the center value.

[0045] The difference in deviation between two adjacent pressure levels for each zone is divided by the pressure difference between the two adjacent pressure levels to obtain the slope of the deviation increment for that zone at the current pressure level. This slope quantifies the rate at which the deviation increases with increasing pressure.

[0046] When the incremental slope of a certain partition shows a continuous increasing trend as the pressure level advances, that is, when the incremental slope in two or more consecutive adjacent pressure level intervals is greater than the value of the previous interval, and the incremental slope of the highest pressure level interval exceeds the preset multiple, the partition is marked as a suspected defect partition.

[0047] The preset multiplier is based on the statistical values ​​of the incremental slope of each partition in the historical qualified product database across each pressure level range. The mean of the incremental slope of historical qualified products plus three times the standard deviation is used as the multiplier determination benchmark. When the incremental slope of a partition in the current batch exceeds this benchmark, it is judged as a suspected defect. For marginal partitions in historical qualified products that do have slight deformation differences, their benchmarks are calculated separately and are not used in conjunction with those of the central partitions to reduce misjudgments caused by spatial location differences.

[0048] The physical mechanism of nonlinear amplification of deformation in delamination defect areas with pressure explains that when delamination defects exist within a product, the reinforcing fiber fabric and rubber layer in that area lose their adhesive constraint, forming a localized cavitation within the delamination region. During inflation and pressurization, this cavitation region is subjected to internal gas pressure, and its deformation behavior is similar to that of a localized thin film bending under pressure. In the small deformation stage, the deformation amount has an approximately linear relationship with pressure, but as the pressure increases, the degree of film bending intensifies. After entering the large deformation stage, the rate of increase in deformation with increasing pressure is significantly higher than the linear expectation, resulting in a characteristic of the incremental slope continuously increasing with the pressure level. In contrast, the deformation in defect-free areas is constrained by the overall reinforcing fabric, and the rate of increase in deformation with increasing pressure tends to stabilize or decrease, with the incremental slope not showing a continuously increasing trend. The difference in the evolution trend of the incremental slope between the two types of regions is the physical basis for identifying delamination defects.

[0049] The initial content of the defect feature vector is formed by summing up the partition number of each suspected defect partition, the deviation value of each pressure level, and the incremental slope value of each adjacent pressure level interval.

[0050] Furthermore, the steps of optimizing the allocation of detection resources and outputting a complete defect feature vector based on the thermal performance control vector are described in reference to... Figure 2 Specifically, it includes: Extract the heating duration component value of each zone from the thermal performance control vector, calculate the absolute value of the difference between the duration component value of each zone and the mean of the duration component values ​​of all zones, and obtain the thermal deviation of each zone; sort each zone in descending order of thermal deviation to form a thermal deviation sorting list. Cross-compare the thermal deviation sorting list with the suspected defect partition list in the initial content of the defect feature vector: for partitions that are ranked high in thermal deviation and are also listed in the suspected defect partition, assign denser subsequent pressure level detection interval values ​​and adopt stricter deviation judgment upper limit values; for partitions that are ranked low in thermal deviation and are not listed in the suspected defect partition, relax the detection interval values ​​and adopt deviation judgment upper limit values. The final detection pressure level interval value and deviation judgment upper limit value adopted by each zone are added to the defect feature vector, and together with the zone number, the deviation amount of each pressure level and the incremental slope value, they form a complete defect feature vector.

[0051] Specifically, from the thermal performance control vector executed during the completed heating control phase, the heating duration component value of each zone is extracted. The absolute value of the difference between the duration component value of each zone and the arithmetic mean of the duration component values ​​of all zones is calculated as the thermal deviation of that zone. The larger the thermal deviation, the higher the degree of differentiated heating control received by that zone in this batch, that is, the more significant the deviation of that zone from the average heating level. The zones are sorted from largest to smallest thermal deviation to form a thermal deviation ranking list, with the zones at the top of the list being those with higher degrees of thermal history anomalies.

[0052] The physical meaning of thermal deviation: The difference in heating duration originates from the lag at the start time of each partition in the difference matrix. A partition with a large thermal deviation means that it has undergone a significantly different thermal history from other partitions in this batch. Whether the heating time is significantly longer (corresponding to a larger lag in curing in the central area) or significantly shorter (corresponding to a larger advance in curing in the edge area), it increases the prior probability of finding potential defects in this partition in subsequent inspections.

[0053] The thermal deviation sorting list is cross-compared with the suspected defect partition list in the initial content of the defect feature vector to determine whether each partition simultaneously meets the two conditions of being ranked high in thermal deviation and being marked as a suspected defect.

[0054] For zones with high thermal deviation rankings and also listed in suspected defect zones, more frequent testing intervals will be allocated in subsequent pressure level tests for that batch of products: the testing pressure level interval will be no more than 10% of the rated pressure between two consecutive adjacent pressure levels. Simultaneously, a stricter upper limit for deviation judgment will be applied to this zone, with 90% of the upper limit of the historical benchmark range as the effective judgment upper limit. This ensures that a defect record is triggered when the deviation exceeds 90% of the upper limit. For zones with low thermal deviation rankings and not listed in suspected defect zones, the testing pressure level interval will be relaxed to 20% of the rated pressure between two consecutive adjacent pressure levels, and the upper limit for deviation judgment will be relaxed to 110% of the upper limit of the historical benchmark range.

[0055] The parameters for the detection interval and the upper limit of the deviation judgment are derived as follows: The specific interval ratios mentioned above are determined based on the following considerations: Smaller pressure level intervals can capture subtle changes in the incremental slope as the pressure level advances, which is beneficial for identification when the defect deformation just begins to show nonlinear amplification, making it suitable for high-risk areas; larger intervals reduce the number of samplings, making it suitable for low-risk areas to improve detection efficiency. The extent to which the upper limit of the deviation judgment is tightened or relaxed is determined based on historical false alarm rate statistics, ensuring that the false alarm rate in high-risk areas is less than 5% and the false alarm rate in low-risk areas is less than 10%.

[0056] The final pressure level interval value and deviation judgment upper limit value adopted for each zone are added to the defect feature vector, which, together with the zone number, the deviation amount of each pressure level, and the incremental slope value of each adjacent pressure level interval, constitutes a complete defect feature vector. The complete defect feature vector is written into the batch database and stored in association with the thermal performance control vector and difference matrix of the same batch, forming a complete process-quality data record for that batch.

[0057] Furthermore, the steps of statistically analyzing the spatial co-orientation of the thermal performance control vector and defect feature vector across batches and calibrating the sensitivity parameters specifically include: For multiple batches accumulated continuously, the heating duration component value in the thermal performance control vector of each partition of each batch and the incremental slope value of the corresponding partition in the defect feature vector are extracted. Under each partition dimension, the heating duration component value and incremental slope value of each batch are paired and arranged. The proportion of batches in which the two change in the same direction, that is, when the heating duration component value increases, the incremental slope value increases synchronously, or when the heating duration component value decreases, the incremental slope value decreases synchronously, is used as the spatial homogeneity index of the partition. When the spatial homogeneity index of a certain partition exceeds the judgment threshold, the number of time steps used to confirm the continuity of the inflection point of that partition will be reduced by a preset number of steps based on the current value, so that the determination of the start time of that partition responds more promptly to the local minimum value of the pressure signal, thereby reducing the system deviation introduced by the inflection point recognition lag in the thermal performance control vector. After each calibration, the adjusted continuous time step values, along with the corresponding partition number and the spatial homogeneity index value at the time of calibration, are recorded together and included in the accumulation as reference information for the next round of homogeneity statistics.

[0058] To eliminate the interference of raw material fluctuations and ambient temperature on causal inference, before performing unidirectional pairing statistics, the batch number of raw materials and the real-time workshop average temperature of each batch are extracted from the batch database. If there is a change of raw material batch or a sudden seasonal change in temperature, the system first uses the preset material vulcanization coefficient and temperature compensation coefficient to normalize and correct the heating duration component of the current batch, so as to remove the false correlation of the environment / material dimension and ensure that the unidirectional index truly reflects the deviation of process control.

[0059] Specifically, in the batch management database, a thermal history-quality correlation data table is maintained using the partition number as an index. After each production batch is completed, the heating duration component value of each partition in the thermal performance control vector of that batch, as well as the incremental slope value of each partition in the complete defect feature vector, are written into the historical record column of the corresponding partition, forming batch-by-batch data accumulation.

[0060] For multiple batches accumulated consecutively, under each partition dimension, the heating duration component value and the increment slope value of each batch in that partition are paired and arranged to form a batch-by-batch data pair sequence. The proportion of batches exhibiting the same trend of increasing heating duration component value and decreasing increment slope value, relative to the total number of batches included in the statistics, is used as the spatial homogeneity indicator for the current statistical window of that partition.

[0061] The physical meaning of the spatial homogeneity index: An increase in the heating duration component indicates that the partition in this batch has undergone a relatively longer heating time. If the incremental slope value of the corresponding partition in the same batch also increases synchronously (i.e., the degree of defect deformation amplification intensifies), it reflects a systematic co-variation relationship between the longer heating time and the higher defect density. This suggests that the current inflection point identification is too late, causing the partition to continue to be subjected to unnecessary overheating, thereby leading to the deterioration of the interface adhesion performance. The higher the homogeneity index, the higher the frequency of the above co-variation relationship in historical batches, and the more evidence there is to adjust the timing of inflection point identification.

[0062] The initial batch size is determined as follows: The initial batch size is set to meet the minimum requirement of statistical stability for the same-direction index. Based on the production batch plan, a minimum initial statistical window of ten consecutive batches is selected. After accumulating a sufficient amount of paired data within this window, the same-direction index calculation is initiated. If the production batch size is large and the process fluctuations between batches are small, the statistical window can be appropriately extended to improve the stability of the judgment.

[0063] When the spatial homogeneity index of a certain partition exceeds the judgment threshold, the number of time steps used to confirm the continuity of the inflection point of that partition will be adjusted downward by a preset number of steps based on the current value, so that the inflection point identification of that partition is more timely and the systematic deviation of the heating duration being too long due to the lag in inflection point identification is reduced.

[0064] The initial threshold is set to aim for a false trigger rate below 20%. In the absence of historical statistical window data, the initial threshold is fixed at 60, meaning calibration is triggered only when more than 60% of the statistical batches show the same direction of change. This initial value is determined based on the following considerations: In the theoretical case where process parameters between batches are completely random and have no systematic correlation, the probability of any two variables changing in the same direction is approximately 50%. Setting a threshold of 60% effectively distinguishes between random and systematic same-direction changes, keeping the false trigger rate within an acceptable range. This initial value is continuously adjusted in subsequent batches using an adaptive update mechanism.

[0065] The preset number of steps is determined as follows: The number of steps to be adjusted in each calibration is based on the average impact of a single step adjustment on the change of the homogeneity index in subsequent batches in the historical calibration records. Initially, it is set to adjust by one time step each time. If the homogeneity index does not decrease within the next 3 statistical windows, the adjustment will be increased to two time steps on the next trigger, and so on. However, the number of consecutive confirmation time steps after the adjustment is not less than 3 time steps to maintain basic noise resistance.

[0066] After each calibration is performed, the adjusted continuous confirmation time step value, the corresponding partition number, and the spatial homogeneity index value triggered at the time of this calibration are recorded in the calibration log.

[0067] Furthermore, the judgment threshold is adaptively updated as batches accumulate, referring to... Figure 3 Specifically, it includes: Using a fixed number of consecutive batches as a statistical window, calculate the difference between the maximum and minimum values ​​of the spatial homogeneity index for each partition within the current window, and use this difference as the homogeneity distribution range of the current window; take the median of the homogeneity distribution range of all historically completed statistical windows, and use this median as the benchmark value for the global judgment threshold. For partitions that have triggered sensitivity parameter calibration at least once in the historical statistical window, their individual judgment threshold is lowered based on the global benchmark value according to the ratio of the number of historical triggers to the length of the statistical window. This allows partitions that have frequently triggered calibration in the past to trigger a new round of calibration with a lower spatial homogeneity index level, so as to respond more quickly to the continuous process fluctuations in the partition. The updated global baseline value and individual judgment thresholds for each partition, along with the current spatial homogeneity index values ​​for each partition, will be uniformly incorporated into the next round of cross-batch homogeneity statistics. Together with the thermal performance control vector and the complete defect feature vector, they will participate in subsequent iterations to form a closed-loop optimization of cross-batch process parameters.

[0068] Specifically, a fixed number of consecutive batches are used as a statistical window. After each statistical window ends, the difference between the maximum and minimum values ​​of the spatial homogeneity index of each partition within the current window is calculated. This difference is used as the homogeneity distribution range of the current statistical window, reflecting the spatial dispersion of the degree of process fluctuation in different partitions within the current window.

[0069] The statistical window length is determined by using the production scheduling cycle as the basic unit, initially set at ten consecutive batches. The window length should not be too short to avoid occasional fluctuations in a single batch dominating statistical conclusions; nor should it be too long to ensure timely response to phased changes in process conditions (such as changing raw material batches or adjusting process specifications). During stable production, the window length can be extended to twenty batches; after a change in raw material batches or a significant adjustment in process parameters, the window length is temporarily shortened to five batches to accelerate statistical response.

[0070] For all windows that have been statistically analyzed in history, the range of values ​​corresponding to the same direction of distribution for each window is summarized, and the median is taken as the benchmark value for the global judgment threshold. The median is robust to occasional extreme windows in historical data and will not have an excessive impact on the global benchmark value due to the excessively large distribution range of a single abnormal window.

[0071] For a partition that has triggered sensitivity parameter calibration at least once in the historical statistical window, its individual judgment threshold is adjusted downward based on the global benchmark value as follows: Calculate the ratio of the cumulative number of calibrations triggered in all historical statistical windows to the total number of historical statistical windows that have been completed. Multiply this ratio by a preset adjustment coefficient and subtract it from the global benchmark value to obtain the adjusted value of the individual judgment threshold for the partition. The adjusted value is not less than 60% of the global benchmark value to prevent oversensitivity from causing frequent false triggers.

[0072] The method for determining the preset adjustment coefficient is as follows: The adjustment coefficient is determined with reference to the typical trigger frequency of frequently triggered partitions in history, so that the individual judgment threshold of the partition with the highest historical trigger frequency is about 70% of the global benchmark value. That is, when the sameness index of the partition exceeds 70%, a new round of calibration is triggered, which is about 15% lower than the initial threshold, in order to reflect the persistence of its process fluctuation.

[0073] The updated global baseline value and individual judgment thresholds for each partition, along with the current spatial homogeneity index values ​​for each partition, will be uniformly incorporated into the next round of cross-batch homogeneity statistics. These will then be used together with the thermal performance control vector and complete defect feature vector generated in the next batch of production for subsequent analysis.

[0074] If a systematic jump in the mean defect density occurs in a future batch, manifested as the mean of the same-direction distribution range of three consecutive statistical windows being more than twice the previous historical mean, it is determined that the process conditions or raw material characteristics have fundamentally changed. The system automatically clears the historical median calculation basis of the current global benchmark value, reinitializes it with the distribution range value of the most recent statistical window as the new global benchmark value, and reverts the individual judgment thresholds of each partition to the global benchmark value, starting a new round of adaptive convergence process.

[0075] Through the execution of the above-mentioned entire process, this invention constitutes a complete closed-loop optimization control system with "zoned pressure inflection point sensing, differentiated heating control, graded pressurization deformation detection, and cross-batch sensitivity adaptive calibration" as the main data chain. This enables dynamic adaptive optimization of process parameters during the production of high-toughness products. The complete data link for each zone in each batch is as follows: pressure timing signal, difference matrix, thermal performance control vector, full-field deformation data, complete defect feature vector, spatial homogeneity index, sensitivity parameter calibration value, and the difference matrix input for the next batch, forming a batch-by-batch self-optimizing closed loop for data flow.

[0076] Example 2:

[0077] Based on the data accumulation of the process in Example 1, this embodiment adds two parallel cross-batch feature tracking paths, which are respectively applied to the batch mean stability of the difference matrix and the rise waveform characteristics of the pressure time series signal.

[0078] The established control link in Example 1 is as follows: each zone passively acquires pressure timing signals, extracts the start time of each zone using a sliding window derivative calculation and continuous confirmation mechanism, and constructs a difference matrix; the difference matrix drives zone-specific heating control to generate a thermal performance control vector; after product demolding, the defect feature vector is obtained by comparing the deformation of the entire field under graded pressure and the deformation benchmark table of historical qualified products in the zone-pressure level; the sensitivity parameters for identifying inflection points are calibrated across batches using spatial unidirectional statistical calibration, and the real-time effectiveness of the judgment threshold is maintained by an adaptive update mechanism.

[0079] The steps for cross-batch drift detection of the difference matrix of consecutive batches specifically include: Using a fixed number of consecutive batches as a detection statistical window, for each partition pair, the mean value of the element in the current detection statistical window is calculated for the difference matrix element. The difference between the current window mean and the overall mean of all completed detection statistical windows is used as the cross-batch drift of that partition pair. When the absolute value of the cross-batch drift of any partition pair exceeds the drift warning threshold, it is determined that a systematic shift in the thermal response time sequence relationship between each partition has occurred in the current batch. The current batch is marked as the baseline drift warning batch, and the statistical update cycle of the partition-pressure level deformation benchmark table is shortened in the subsequent batches. The deformation benchmark entries of the corresponding partition are re-statistically re-statistically re-statisticalized based on the most recently accumulated qualified product data. The current drift value, the partition pair number that triggered the warning, and the trigger time are recorded together in the batch database and stored in association with the difference matrix of the same batch.

[0080] Specifically, in the existing batch database, the difference matrix is ​​stored by batch number. Using five consecutive batches as a detection and statistical window, the arithmetic mean of the five element values ​​within the current window is calculated for each partition pair's difference matrix elements. The arithmetic mean of the corresponding element values ​​from all completed historical detection windows is used as the overall reference mean. When the absolute value of the difference between the current window mean and the overall reference mean for a partition pair exceeds the drift warning threshold, a baseline drift warning is triggered.

[0081] The drift warning threshold is determined as follows: The initial drift warning threshold is twice the batch-to-batch standard deviation of the difference matrix elements for each zone during a period of stable historical production conditions. This standard deviation can be statistically obtained from the historical difference matrix records of the first thirty batches after production starts. If the initial batch size is less than thirty batches, a temporary warning threshold is set by multiplying the existing batch standard deviation by a coefficient of 1.5. Once thirty batches are reached, the threshold is switched to the official standard. The basis for using twice the standard deviation as the threshold is that, assuming an approximately normal distribution among the batches of the difference matrix elements, the mean of approximately 95% of normal batch windows does not exceed this threshold; exceeding it indicates a higher probability of systematic drift.

[0082] After the drift warning is triggered, the historical qualified products generated in the subsequent three detection and statistics windows will be given priority in the statistics when they are written into the partition-pressure level deformation benchmark table. The deformation benchmark entries of the corresponding partition will be recalculated in three batches as one update cycle (replacing the ten-batch update cycle under normal conditions). When the change of the center value of the benchmark entry in two consecutive update cycles does not exceed one-tenth of the diffusion range value, it is determined that the benchmark table has converged to the new material property level, exits the accelerated update mode, and resumes the normal update cycle.

[0083] During the accelerated update phase, the minimum valid sample size for each update cycle is no less than ten items. If the number of qualified items is less than ten within three batches, the collection is extended until ten items are met before the baseline entry update is performed to ensure the robustness of the diffusion range value estimate.

[0084] The steps for extracting the slope feature value of the pressure recovery segment from the derivative time series of pressure time series signals of each partition and performing batch-to-batch consistency analysis specifically include: After the start time of each partition is confirmed, the first consecutive positive value in the derivative time series is confirmed as the starting point, and the first time when the change of the derivative time series within a continuous period of no less than the preset stable confirmation time step is not more than the preset stable judgment range is taken as the ending point. The average value of the derivative time series between the start and end times is extracted as the slope characteristic value of the partition in the pressure recovery segment of the current batch. For the slope characteristic values ​​of the pressure recovery segment of each zone accumulated in multiple batches, the standard deviation within the current sliding statistical window is calculated according to the zone number, and the standard deviation is used as the batch consistency index of the slope characteristic value of the zone. When the batch consistency index of a certain partition exceeds the batch consistency warning threshold, the current batch is marked as an abnormal batch in terms of slope characteristics. The corresponding partition number and the slope characteristic value of the current batch are written into the batch database and stored in association with the difference matrix.

[0085] Specifically, based on the execution of the inflection point continuous confirmation logic, for each partition that has completed continuous confirmation, the slope feature value of the recovery segment is further extracted from its derivative time series. The starting point of the extraction interval is the first continuous confirmation time (i.e., the inflection point confirmation time), and the endpoint is determined as follows: sliding backward from the starting point, the absolute value of the change in the derivative time series at each time step is recorded. When the change in each step does not exceed the preset stability judgment range for at least 5 consecutive time steps, the first time of the continuous stable interval is set as the endpoint.

[0086] The preset stability judgment amplitude is obtained by taking 1.5 times the statistical standard deviation of the gradual change of the derivative time series stability stage (after entering the positive sulfurization plateau period) of each partition in the historical batch as the initial setting value, so that the amplitude can distinguish between normal derivative oscillation noise and true derivative stabilization signal.

[0087] The arithmetic mean of all derivative estimates between the start and end times is used as the slope characteristic value of the pressure recovery segment of the current batch in this partition, stored in the batch database, and associated with the difference matrix.

[0088] For continuously accumulated batches, a sliding statistical window of ten batches is used, and the standard deviation of the slope characteristic value within the window is calculated according to the partition number of each batch as the consistency index between batches.

[0089] The batch consistency warning threshold is obtained as follows: the initial warning threshold is twice the standard deviation of the slope characteristic value of each zone in the stable process stage after production starts (20 consecutive batches of raw material batches without changes and with normal equipment operation); in the initial stage, the existing batch standard deviation is multiplied by 2 as the temporary threshold, and the formal value is switched after 20 batches.

[0090] Before the cross-batch spatial homogeneity statistics are executed, a step is also included to screen the batches participating in the statistics based on the batch consistency index, specifically including: Extract the slope characteristic values ​​of each batch and each partition within the current statistical window to obtain the batch consistency index. Identify batches with consistency indices exceeding the batch consistency warning threshold as batches with abnormal material properties and remove them from the spatial isotropic statistical sample of the current statistical window. Use the heating duration component value and incremental slope value of the remaining batches to complete the calculation of the spatial isotropic index. Mark the number of valid batches actually participating in the statistics and the number of abnormal batches that were removed in the calculation results. When the number of valid batches is lower than the preset minimum number of valid batches, the spatial homogeneity statistics and sensitivity parameter calibration for that period are suspended and will be postponed until the number of valid batches in the next statistical window meets the requirements. At the same time, the suspension record is written to the batch database. If several consecutive statistical windows are suspended due to insufficient valid batches, a raw material consistency warning is issued to the operation management terminal.

[0091] Specifically, before each trigger of spatial homogeneity statistics, the slope characteristics of each batch and each partition within the current statistical window are extracted as batch consistency indicators. These indicators are compared with the batch consistency warning threshold, and batches exceeding the threshold are marked as batches with abnormal material properties and removed from the statistical sample.

[0092] The minimum number of valid batches is preset to ensure that the statistical error of the spatial homogeneity index does not exceed 15%. Statistical inference suggests that under the binomial distribution assumption, this constraint can be met when the sample size is no less than seven batches. Therefore, the minimum number of valid batches is initially set to 7. If the number of valid batches in a statistical window is less than seven, the current statistical analysis is paused. If three consecutive statistical windows are paused, an alert is sent to the management, requiring quality engineers to verify the consistency of raw material batches to prevent abnormal conditions from being continuously masked.

[0093] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and 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 dynamic optimization and control of process parameters in the production of high-toughness products, characterized in that, include: Pressure timing signals are passively collected in each zone of the product, and the first derivative of the product is extracted in real time. The start time of each zone is determined by the moment when the derivative sign changes from negative to positive, and a zone start time difference matrix is ​​constructed. Based on the difference matrix, the heating power is adjusted for the earliest positive zone, and the heating duration is adjusted for the lagging zone according to the difference ratio. The power switching time and duration of each zone are recorded as the thermal performance control vector. The product is graded and controlled, and deformation data is collected in the entire field at each pressure level. The deformation data is compared with the statistical deformation benchmark table at each level. The defect location is obtained by the characteristic of nonlinear amplification of layered defect deformation with pressure. The detection resource allocation is optimized based on the partition with the largest difference in the thermal performance control vector, and the defect feature vector is output. The spatial co-directionality of thermal performance control vectors and defect feature vectors in each batch is statistically analyzed. When the co-directionality exceeds the judgment threshold, the sensitivity parameter for identifying the inflection point of the corresponding calibrated zone is adjusted.

2. The method for dynamic optimization and control of process parameters in the production process of high-toughness products according to claim 1, characterized in that: The steps for extracting the first derivative of the product and determining the start time of each partition include: passively collecting the pressure time-series signal of each partition, gradually sliding backward with a preset time window, calculating the rate of change of the pressure value with respect to time within each time window, and forming the derivative time series of the partition; The continuity of the derivative time series is confirmed. Only when the derivative sign changes from negative to positive and remains positive for a preset number of consecutive time steps, the first sign flip time is determined as the start time of the corresponding partition. A difference matrix is ​​formed by using the time difference between the start times of any two partitions as matrix elements and arranging them with the partition number as the row and column index. The number of time steps used for inflection point continuity confirmation is archived as a sensitivity parameter for inflection point identification.

3. The method for dynamic optimization and control of process parameters in the production process of high-toughness products according to claim 2, characterized in that: The steps for adjusting heating power and heating duration based on the difference matrix specifically include: extracting the absolute value of the time difference between each partition and the earliest starting partition from the difference matrix; dividing the absolute value of the time difference into control levels according to preset segment intervals, and grouping partitions whose absolute values ​​of time difference fall within the same segment interval into the same control group; for the earliest starting partition, switching the heating power from the current value to the target value with a predetermined reference slope, while keeping the heating duration unchanged from the reference duration; for partitions within the same control group, the heating power slope is shared, and the heating duration is above the reference duration, extended proportionally according to the ratio of the absolute value of the time difference of this partition to the maximum absolute value of the time difference within the group; after the heating control is completed, each partition exits the heating stage and enters a unified graded pressurization stage, ensuring that the heating control and graded pressurization do not overlap in timing; The power switching time, the power slope value after switching, and the heating duration after regulation for each zone are recorded as ternary components in the thermal performance regulation vector. The set of ternary components constitutes the complete thermal performance regulation vector.

4. The method for dynamic optimization and control of process parameters in the production process of high-toughness products according to claim 1, characterized in that: The steps for obtaining defect location specifically include: full-field deformation data is collected by a distributed sensor array at the corresponding location of each partition, each sensor node outputs the local deformation amount of the corresponding partition under the current pressure level, and the overall set of data from each node constitutes the full-field deformation data for each pressure level. For historical qualified batches of products, record the total deformation of each zone at the end of each pressure level under the same pressure classification sequence. Statistically analyze the historical deformation of each zone under the same pressure level. The center value of the distribution and the diffusion range value together constitute the deformation benchmark entry of the zone under the pressure level. All deformation benchmark entries are summarized into a zone-pressure level deformation benchmark table. For the current batch, the deformation at the end of the current pressure level of each partition is compared with the corresponding entry in the partition-pressure level deformation reference table to calculate the deviation. The incremental slope of the deviation between adjacent pressure levels is used as a nonlinear amplification feature. When the incremental slope of a certain partition shows a continuous increasing trend as the pressure level advances and exceeds the preset multiple, it is marked as a suspected defect partition. The partition numbers of each suspected defect partition, the deviation of each pressure level, and the incremental slope value are summarized to form a defect feature vector.

5. The method for dynamic optimization and control of process parameters in the production process of high-toughness products according to claim 4, characterized in that: The steps for outputting a complete defect feature vector specifically include: extracting the heating duration component value of each zone from the thermal performance control vector; calculating the absolute value of the difference between the duration component value of each zone and the mean of the duration component values ​​of all zones to obtain the thermal deviation amount of each zone; sorting each zone in descending order of thermal deviation amount to form a thermal deviation sorting list; cross-comparing the thermal deviation sorting list with the list of suspected defect zones in the initial content of the defect feature vector: assigning denser subsequent pressure level detection interval values ​​to zones with higher thermal deviation amounts that are also listed as suspected defect zones, and using the upper limit value for deviation judgment; relaxing the detection interval values ​​for zones with lower thermal deviation amounts that are not listed as suspected defect zones, and using a more lenient upper limit value for deviation judgment; appending the final detection pressure level interval value and the upper limit value for deviation judgment used for each zone to the defect feature vector, which, together with the zone number, the deviation amount of each pressure level, and the incremental slope value, constitutes a complete defect feature vector.

6. The method for dynamic optimization and control of process parameters in the production process of high-toughness products according to claim 2, characterized in that: The steps for cross-batch statistical analysis of the spatial homogeneity between the thermal performance control vector and the defect feature vector of each partition, and for calibrating the sensitivity parameters, specifically include: for continuously accumulated batches, extracting the heating duration component value from the thermal performance control vector of each partition in each batch, and the incremental slope value of the corresponding partition in the defect feature vector; under each partition dimension, pairing and arranging the heating duration component value and incremental slope value corresponding to each batch, and statistically analyzing the proportion of batches whose values ​​change in the same direction to the total number of batches participating in the statistics, as the spatial homogeneity index of the partition; when the spatial homogeneity index of a partition exceeds the judgment threshold, the time step used for confirming the continuity of the partition inflection point is reduced by a preset number based on the current value; after each calibration, the adjusted continuous time step value, the corresponding partition number, and the spatial homogeneity index value triggered at the time of calibration are recorded together.

7. The method for dynamic optimization and control of process parameters in the production process of high-toughness products according to claim 6, characterized in that: The judgment threshold is adaptively updated as batches accumulate. Specifically, it includes: using a fixed number of consecutive batches as a statistical window, calculating the difference between the maximum and minimum values ​​of the spatial homogeneity index of each partition within the current window, which is used as the homogeneity distribution range of the current window; taking the median of the homogeneity distribution range of all historically completed statistical windows, which is used as the benchmark value of the global judgment threshold; for partitions that have triggered sensitivity parameter calibration at least once in historical statistical windows, the individual judgment threshold is lowered based on the global benchmark value according to the ratio of the number of historical triggers to the length of the statistical window, so that partitions that have frequently triggered calibration in the past can trigger a new round of calibration with a lower level of spatial homogeneity index. The updated global baseline value and individual judgment thresholds for each partition, along with the current spatial homogeneity index values ​​for each partition, will be uniformly incorporated into the next round of cross-batch homogeneity statistics. Together with the thermal performance control vector and the complete defect feature vector, they will participate in subsequent iterations to form a closed-loop optimization of cross-batch process parameters.

Images