A continuous production method and system for rubber-based brake linings

Abstract

The application discloses a continuous production method and system of rubber-based brake lining, collects process parameters and quality detection records of each process, generates deviation response characteristics through adjacent process qualified rate synchronous fluctuation attribution analysis, generates candidate abnormal co-occurrence groups through abnormal co-occurrence correlation analysis, generates a hypothetical deviation amount through undisturbed reference comparison, and statistically calculates cross-process recurrence rates, combines the hypothetical deviation amount, corrects recurrence rate sudden drop nodes to generate a corrected abnormal group, merges and screens to determine main abnormal items, extracts attribution discrimination features of the main abnormal items, determines abnormal attribution categories through upstream and downstream attribution discrimination, reverses cross-batch attribution through cross-batch attribution tracing and positioning, generates production diagnosis conclusions through tracing and verification, integrates the production diagnosis conclusions and the cross-process recurrence rates to generate a process disposal priority set, predicts and arranges output continuous production control instructions, and realizes automatic tracing and disposal priority dynamic arrangement of cross-batch root causes under compensation and cover-up environment.

Description

Technical Field

[0001] This invention relates to the field of quality control technology in rubber product manufacturing, and in particular to a continuous production method and system for rubber-based brake pads. Background Technology

[0002] The production of rubber-based brake pads involves multiple continuous processes, including ingredient formulation, mixing, preforming, hot pressing and curing, heat treatment, and grinding and shaping. The quality of each process is interdependent. Some manufacturers incorporate recycled rubber materials (hereinafter referred to as recycled materials) obtained from the recycling of waste rubber products into their formulations to control raw material costs. The differences in physical properties between batches of recycled materials further exacerbate the batch-to-batch fluctuations in the pass rate of each process. Existing quality monitoring methods, which mainly rely on single-process parameter threshold warnings, are insufficient to systematically identify the quality transmission relationships between processes.

[0003] In multi-process continuous production, downstream process operators often proactively adjust their process parameters to maintain the pass rate after noticing a decline in quality. This cross-process compensation behavior masks the true source of the abnormality in the upstream process, making it difficult to accurately pinpoint the root cause of the quality through single-batch process records. Furthermore, the cause of the same abnormality in different batches may repeatedly switch between the upstream process and the current process. Existing traceability methods rely on manual step-by-step investigation, which has low traceability efficiency when batch switching is frequent or when multiple processes have simultaneous parameter deviations. It is impossible to predict when the abnormality will recur, and there is a lack of quantitative basis for the order of handling multiple abnormal processes. Summary of the Invention

[0004] This invention discloses a continuous production method and system for rubber-based brake pads. Through inter-process deviation response analysis and anomaly co-occurrence correlation comparison, hypothetical deviation is generated by non-disturbance benchmark comparison and continuous deviation verification. Major anomalies are identified by cross-process recurrence rate statistics and node correction merging and screening. Then, production diagnostic conclusions are generated by attribution judgment and cross-batch attribution reversal tracing. Correlation recurrence rate prediction and arrangement output continuous production control instructions are used to realize automatic cross-batch root cause tracing and dynamic priority arrangement of handling under compensation and masking environment.

[0005] The first aspect of this invention provides a continuous production method for rubber-based brake pads, comprising the following steps: Collect process parameter data and quality inspection records for each production process, and perform attribution analysis on the synchronous fluctuation of the pass rate of adjacent processes to generate deviation response characteristics. Based on the aforementioned deviation response characteristics, anomaly co-occurrence correlation analysis is performed to generate candidate anomaly co-occurrence groups, and the candidate anomaly co-occurrence groups are compared with an unperturbed benchmark to generate hypothesis deviation. The cross-process recurrence rate is statistically analyzed by the candidate anomaly co-occurrence group. Based on the hypothesis deviation and the cross-process recurrence rate, the recurrence rate drop node correction is performed on the candidate anomaly co-occurrence group to generate a correction anomaly group. The correction anomaly group is then merged and screened to determine the main anomalies. Extract the attribution discrimination features of the main anomalies, perform upstream and downstream process attribution discrimination based on the attribution discrimination features to determine the attribution category of the anomalies, perform cross-batch attribution traceability comparison based on the attribution category of the anomalies to determine cross-batch attribution reversal items, and perform upstream traceability verification based on the cross-batch attribution reversal items to generate production diagnostic conclusions; Based on the production diagnostic conclusions and the cross-process recurrence rates, a process handling priority set is generated by correlation and integration. The process handling priority set and the production diagnostic conclusions are then used to predict and arrange the recurrence rates and output continuous production control instructions.

[0006] A second aspect of this invention provides a continuous production system for rubber-based brake pads, comprising: The attribution mapping module is used to collect process parameter data and quality inspection records of each production process, and to perform synchronous fluctuation attribution analysis of the pass rate of adjacent processes to generate deviation response characteristics. The benchmark module is used to perform anomaly co-occurrence correlation analysis based on the deviation response characteristics to generate candidate anomaly co-occurrence groups, and to perform unperturbed benchmark comparison on the candidate anomaly co-occurrence groups to generate hypothesis deviation. The convergence correction module is used to statistically analyze the cross-process recurrence rate through the candidate anomaly co-occurrence group, perform recurrence rate drop node correction on the candidate anomaly co-occurrence group based on the hypothesis deviation and the cross-process recurrence rate to generate a correction anomaly group, and perform merging and screening on the correction anomaly group to determine the main anomalies. The attribution diagnosis module is used to extract the attribution discrimination features of the main anomalies, perform upstream and downstream process attribution discrimination based on the attribution discrimination features to determine the attribution category of the anomalies, perform cross-batch attribution traceability comparison based on the attribution category of the anomalies to determine cross-batch attribution reversal items, and perform upstream traceability verification based on the cross-batch attribution reversal items to generate production diagnosis conclusions. The instruction output module is used to generate a process handling priority set by associating and integrating the production diagnosis conclusion with the cross-process recurrence rate, and to output continuous production control instructions by predicting the recurrence rate of the process handling priority set and the production diagnosis conclusion.

[0007] The beneficial effects of this invention are reflected in the following points: 1. Characterized by the reverse linkage of the pass rates of adjacent processes, it identifies the compensation masking process pairs in which downstream processes adjust parameters to maintain the pass rate of their own processes when the pass rate of upstream processes declines. Based on the statistical regularity of parameters during the undisturbed steady-state period of each process, it quantifies the standardized magnitude of the continuous deviation of the parameter of the initiating process within the batch of abnormal co-occurrence. This benchmark also provides a normalized conversion between the downstream compensation intensity and the upstream intervention offset, so that the upstream real deviation smoothed out by the downstream compensation action can still be quantitatively restored, breaking through the limitation that single-process threshold warning cannot penetrate cross-process compensation. 2. Unlike the conventional approach of directly sorting by deviation magnitude, it first defines the screening boundary by the sudden drop nodes of the cross-process and cross-batch recurrence frequency distribution, eliminates high-frequency but extremely small noise co-occurrences, and then merges the common causes of those that pass the screening, ranking them by the intensity of the downstream parameter compensation adjustment they trigger. Those whose compensation response exceeds the normal level are identified as the main abnormal items, achieving the simultaneous completion of noise suppression and key abnormality location. 3. By leveraging the reversal phenomenon of anomaly attribution from upstream to the current process in batches, the intervention batch is masked and locked. In such batches, persistent parameter deviations that precede downstream interventions are examined, and the upstream parameters with the earliest and most stable deviations are identified as the root cause. Then, each process is ranked hierarchically by the product of the root cause confidence level and the cross-process recurrence rate. The key root cause process, which is jointly confirmed by the compensation response intensity, is given the highest confidence weight. Based on the batch switching pattern and the batch dispersion of the downstream compensation response, the correction direction and magnitude opposite to the root cause deviation direction are arranged as pre-judgment triggering or real-time over-threshold triggering instructions before the switching node. This transforms quality control from post-event step-by-step investigation to pre-event intervention based on priority. Attached Figure Description

[0008] Figure 1 This is a schematic flowchart of a continuous production method for rubber-based brake pads according to the present invention.

[0009] Figure 2 This is a schematic diagram of the layout of the continuous production line for rubber-based brake pads of the present invention.

[0010] Figure 3 This is a schematic diagram illustrating the inter-process deviation response transmission and opposite direction co-occurrence identification of the present invention.

[0011] Figure 4 This is a structural block diagram of a continuous production system for rubber-based brake pads according to the present invention.

[0012] Wherein: 1-Ingredient feeding station; 2-Internal mixer; 3-Pre-forming press; 4-Hot pressing and curing machine; 5-Heat treatment furnace; 6-Grinding machine; 7-Inspection instrument table; 8-Upstream process pass rate curve; 9-Downstream process pass rate curve; 10-Initiation time of negative deviation in upstream process; 11-Initiation time of positive deviation in downstream process; 12-Time of sudden increase in correlation; 13-Response lag interval; 14-Direction of deviation transmission; 15-Co-occurrence interval of opposite directions. Detailed Implementation

[0013] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0014] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0015] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0016] The technical solutions of the embodiments of this application are described below.

[0017] like Figure 1 As shown, this embodiment of the invention provides a continuous production method for rubber-based brake pads, including the following steps S11-S15: Step S11: Collect process parameter data and quality inspection records for each production process, and perform attribution analysis on the synchronous fluctuation of the pass rate of adjacent processes to generate deviation response characteristics.

[0018] Specifically, process parameter data and quality inspection records for each production step are collected. For example... Figure 2As shown, the continuous production line is sequentially equipped with a batching and feeding station 1, a mixer 2, a preforming press 3, a hot press curing machine 4, a heat treatment furnace 5, a grinding machine 6, and an inspection instrument table 7. Process parameter data are collected in real time by sensors at each station: the batching and feeding station 1 collects the proportion of recycled material and the weight of each component; the mixer 2 collects the mixing temperature, rotor speed, and torque timing; the preforming press 3 collects the pre-pressure, holding time, and material temperature; the hot press curing machine 4 collects the mold temperature and forming pressure of each section; the heat treatment furnace 5 collects the furnace temperature distribution and holding time; and the grinding machine 6 collects the feed rate and grinding allowance. Among the above parameters, those that have a significant impact on the exit qualification rate of this process are called the key parameters of this process. Each adjustment of the parameter setpoint by the control system is recorded with the adjustment time and the values ​​before and after the adjustment to form a process parameter adjustment record. This record shares a batch number and time index with the process parameter data; only after aligning these two can the source of parameter deviation be distinguished: if the setpoint remains unchanged but the measured value moves downwards, it indicates process state drift; if the setpoint is adjusted and the measured value moves synchronously, it indicates manual intervention. Quality inspection records are collected by the exit inspection stations of each process. Weighing deviation, Mooney viscosity, billet weight, liner density, surface hardness, and thickness uniformity are inspected sequentially along the above workstations. In the quality inspection records, each process is sampled in groups at fixed time intervals within the batch. The pass rate is the ratio of the number of qualified products to the total number of samples, forming a grouped time sequence within the batch and a continuous time sequence as the batch progresses. In addition to the recycled material blending ratio marker, the batch number also includes production node markers such as raw material batch switching, mold replacement, and equipment alarms. Batches with a higher recycled material blending ratio typically experience greater fluctuations in the process parameter data of the internal mixer.

[0019] In some embodiments, the step of performing attribution analysis on the synchronous fluctuation of the pass rate of adjacent processes using the process parameter data and the quality inspection records to generate deviation response characteristics includes: extracting historical pass rate data by process from the process parameter data and the quality inspection records to construct a set of process pass rate curves; checking the fluctuation of adjacent processes on the set of process pass rate curves to form a distribution of process correlation; determining adjacent process pairs at the moment of sudden increase in correlation based on the distribution of process correlation to form a set of impact response process pairs; and analyzing the impact source direction and amplitude based on the set of impact response process pairs to generate deviation response characteristics.

[0020] Historical pass rate data for each process is extracted from process parameter data and quality inspection records to construct process pass rate curve sets. The historical pass rate data for each process, statistically analyzed by batch, is arranged chronologically in the quality inspection records. The batch number and recycled material blending ratio provided by the process parameter data serve as the batch index for this sequence. Successive batches are labeled sequentially with the batch number as the horizontal axis and the pass rate as the vertical axis, thus obtaining the historical pass rate curve for that process. The batch range of the historical pass rate data is defined by the earliest to the latest batch within the current analysis window. Each process uses the same range to ensure horizontal comparability. The number of batches in the window is determined based on the production line's material changeover cycle and the frequency of historical anomalies, typically ranging from 30 to 50 batches. If the window is too narrow, low-frequency anomalies will not have enough time to leave a complete shape on the curve; if the window is too wide, the process status of earlier batches will not be sufficient to represent the current production conditions. In the mixing process, the pass rate of batches with high proportions of recycled materials experienced a phased decline. The curve showed a continuous low-value segment within the corresponding batch range, with the start and end batches corresponding to the proportion markings in the recycled material batch numbers. The mixing temperature and rotor torque fluctuations of the same batch were synchronously larger. The depth and width of the low-value segment reflected the degree of influence of the recycled material batch on the mixing quality. If the hot-pressing curing process showed a synchronous decline within the same batch range, the two curves showed highly similar shapes in that range, indicating that the quality fluctuations in the mixing process had been transmitted downstream. The process pass rate curve group was formed by juxtaposing the curves of each process sequentially along the production line. Processes with overall stability within the group indicated that the parameter settings recorded in the process parameter data were well matched with the product quality requirements. Processes with continuous low-value segments or drastic fluctuations indicated systemic quality risks. For processes with insufficient batch coverage in the quality inspection records, their curves were broken in the process pass rate curve group by missing measurement markings. Correlation checks between adjacent processes were only performed on batch segments where both processes had measured data.

[0021] The process pass rate curve group is checked to examine the fluctuations of adjacent processes, forming a distribution of process correlation. For example... Figure 3 As shown, due to the batch delay in the transfer of the same material between adjacent processes, before pairing the upstream process pass rate curve 8 and the downstream process pass rate curve 9 in the process pass rate curve group, curve 9 must be shifted forward according to the number of transfer batches to align with curve 8 on the same material batch. After alignment, the Pearson correlation coefficient is calculated. v is the aligned batch number, a(v) is the upstream process pass rate, b(v) is the downstream process pass rate, μ(a) and μ(b) are their mean values, and r takes values ​​in the range [-1, 1] with a dimension of one. When r is close to 1, the pass rates of the two processes rise and fall together; when r is close to -1, the fluctuations are in opposite directions. For example, if the pass rate of mixing declines but rises again after pressurization at the hot pressing end, it indicates a compensation-masking relationship. The calculation of r is based on all batches covered by the process pass rate curve group within the analysis window, excluding batches marked as equipment alarms or mold replacements. The distribution of process correlation is given item by item for adjacent process pairs. Each item records the process pair number, the number of batches used for alignment, and the current r value, and retains the aligned batch pass rate pairing sequence. The length of this sequence is equal to the number of batches shared by the two processes. Process segments with generally high r values ​​have tight transmission chains, and upstream anomalies can be reflected in the downstream pass rate in a short time; process pairs with r values ​​close to zero fluctuate independently. The r-value of the hot pressing curing and heat treatment process pair has maintained a high positive value for a long time. The abnormal density uniformity of hot pressing has been reflected in the surface hardness qualification rate of heat treatment. This process pair is marked as highly correlated in the process correlation distribution. The mixing and grinding process pair is usually lower than the adjacent process pair because there are two processes in between.

[0022] Based on the distribution of process correlation, adjacent process pairs at the moment of sudden increase in correlation are identified to form impact response process pair groups. The r-values ​​of each adjacent process pair in the process correlation distribution are not static within the analysis window and need to be recalculated batch by batch using sliding sub-windows. The width of the sub-window is taken as an integer multiple of the normal batch cycle of the production line. A new r-value is obtained for each batch slid. The time sequence of r-values ​​at all sliding positions constitutes the dynamic curve of the correlation of that process pair, which is the criterion for the impact response process pair group. The dynamic curves of each process pair are arranged side-by-side, forming the dynamic form of the process correlation distribution. The moment when the absolute value of r in the dynamic curve increases by more than a certain multiple of the historical standard deviation of the absolute value of r for that process pair within a single sliding step is defined as the moment of sudden increase in correlation (12). Taking the absolute value allows both positive and negative coupling enhancements to be detected. A sudden increase means that the upstream anomaly has changed from sporadic fluctuations to a shift in the physical properties of the entire batch of incoming materials. For example, after material change, the density of each mold blank decreases uniformly, and the downstream pass rate follows batch by batch, suddenly intensifying the linkage between the two processes. The pass rate of the mixing process declined rapidly when switching to batches with a high proportion of recycled material. The pass rate of the preforming process also declined synchronously in the following batches. The dynamic curve of this process pair showed a sharp increase before and after the batch switching. The sharp increase in correlation at time 12 occurred several batches after the ratio switching mark, pointing to the external source of the recycled material batch switching. The sharp increase of another adjacent process pair occurred during the stage of accelerated wear of the hot pressing mold, corresponding to the node when the cumulative number of batches used by the mold exceeded the maintenance cycle, which is an internal source of equipment deterioration. The impact response process pair group includes all adjacent process pairs that showed the sharp increase in correlation at time 12. Each pair is accompanied by the batch in which the sharp increase occurred and the corresponding increase in the absolute value of r. The historical standard deviation of the absolute value of r in the criterion is obtained by statistically analyzing the r values ​​at all sliding positions of the dynamic curve of the process pair. It has the same source as the increase and can be directly compared.

[0023] Based on the analysis of the impact source direction and amplitude of the impact response process pair group, deviation response characteristics are generated. The generation of deviation response characteristics presupposes identifying the source direction of each impact in the impact response process pair group. The pass rate curves of the corresponding upstream and downstream processes are traced back to the time sequence of pass rates before and after the sudden increase in correlation (12). The direction of change is checked pairwise: if the upstream pass rate had already begun to decline several batches before the sudden increase, while the downstream rate declined synchronously or lagging behind, it is determined to be a positive impact transmission, with the source direction pointing to the upstream process, forming deviation transmission direction 14; if the upstream rate declined while the downstream rate rose in the opposite direction, it is determined to be a compensatory reverse response, with deviation transmission direction 14 pointing to the compensatory action of the downstream process. The impact amplitude is represented by the absolute value of the change in pass rate of each process before and after the sudden increase; the larger the change, the more severe the disturbance. The ratio η of the downstream change to the upstream change reflects the transmission efficiency of the impact between the two processes. η has a dimension of one; a value close to 1 indicates that the impact is transmitted to the downstream with almost no attenuation, while a lower value indicates that the impact is partially absorbed or compensated during transmission. In one process pair within the impact response process group, the pass rate of the mixing process declined significantly after the batch switching of recycled materials. The preforming process showed a similar decline in the same batch, with η approaching full value, recorded as positive high-efficiency transmission. The hot-pressing curing process showed a decline in the pass rate of the heat treatment process, while the heat treatment process only followed slightly, with a low η, recorded as positive low-efficiency transmission. The deviation response characteristics are organized into four types of fields using a two-level index of process number and sudden increase batch: the pass rate sequence of each process batch is stored continuously by batch number; the impact source direction is recorded with a binary label pointing upstream or downstream; the transmission efficiency η and impact amplitude are recorded as pairs of process pairs, both taken from the same pass rate time series before and after the same sudden increase moment, and the values ​​can be directly compared; the absolute value increase of r is attached for reference.

[0024] Step S12: Based on the deviation response characteristics, perform anomaly co-occurrence correlation analysis to generate candidate anomaly co-occurrence groups, and perform unperturbed benchmark comparison on the candidate anomaly co-occurrence groups to generate hypothesis deviation.

[0025] In some embodiments, the step of generating candidate anomaly co-occurrence groups based on the deviation response characteristics through anomaly co-occurrence association analysis includes: grouping the deviation response characteristics according to the direction of process deviation to form a positive deviation group and a negative deviation group; statistically analyzing the cross-directional process co-occurrence frequency of the positive deviation group and the negative deviation group to form a direction-opposite co-occurrence distribution table; screening process pairs with co-occurrence frequencies exceeding a threshold from the direction-opposite co-occurrence distribution table to check compensation and masking relationships to form a masking candidate group; and selecting the deviation initiator that occurred first from the masking candidate group to generate candidate anomaly co-occurrence groups.

[0026] The deviation response characteristics are grouped according to the direction of process deviation, forming a positive deviation group and a negative deviation group. Grouping is based on the direction of batch-by-batch pass rate changes for each process as attached to the deviation response characteristics. Batches with pass rates lower than the historical average and whose deviation exceeds a certain multiple of the historical standard deviation for that process are marked as negative deviations; those higher than the historical average by an equal multiple are marked as positive deviations; and batches with absolute deviations lower than the judgment multiple are marked as no deviation. For example, a slight drop in the pass rate caused by one or two slightly out-of-tolerance thicknesses during sampling falls into this category. Only the first two types of batches constitute the grouping results. The historical average and historical standard deviation are statistically analyzed within the analysis window based on the batch-by-batch pass rate sequence for that process attached to the deviation response characteristics. The judgment multiple is typically 1.5 to 2 times. A smaller multiple will result in a large number of normal fluctuations being included in the deviation batches, while a larger multiple will make it difficult for minor but continuous quality declines to be included in the grouping. The set of batches with positive deviations for each process is called the positive deviation group, and the set of batches with negative deviations is called the negative deviation group. "Negative" refers to the deviation relative to "positive." The naming of the two groups corresponds one-to-one with the direction of deviation of the pass rate relative to the historical average. In the mixing process, batches with high proportions of recycled materials exhibiting negative deviations are concentrated in the batch range, with a large number of batches entering the reverse deviation group. Fluctuations in Mooney viscosity among recycled material batches cause the mixing process's pass rate to remain below the historical average, resulting in a dense concentration of deviation records in this range. In the pure virgin material batch range, the pass rate is stable, with no valid deviation records in either the positive or reverse deviation groups. Each deviation record in both the positive and reverse deviation groups includes the process number, batch number, deviation multiple, and the raw material attribute marker for that batch. The deviation multiple is used to sort multiple records by deviation intensity when they are parallel, and the raw material attribute marker is used for merging and statistically analyzing similar batches. Cross-directional pairing is achieved by finding the intersection based on the batch number.

[0027] For both the positive and negative deviation groups, the co-occurrence frequency of cross-directional processes is statistically analyzed to form an opposite-direction co-occurrence distribution table. The normalized co-occurrence frequency C(i,j) = n(i,j) / N, where i and j are the process numbers, and n(i,j) is the number of batches in the negative deviation group where process i and process j in the positive deviation group simultaneously deviate within the same material batch flow window. Figure 3The co-occurrence interval 15 represents the opposite direction. During statistics, the corresponding co-occurrence batch number list is recorded simultaneously. N is the total number of batches within the analysis window. C(i,j) has a dimension of one; a larger value indicates that process j performs compensation operations more frequently to maintain the pass rate when the quality of process i declines, and the degree of compensation masking correlation between the two processes is higher. The reverse deviation group provides row indexes, and the forward deviation group provides column indexes. After alignment, the C(i,j) of all process pairs is expanded with process i as the row and process j as the column, resulting in a two-dimensional table, which is the opposite direction co-occurrence distribution table. Process pairs with high C(i,j) usually have upstream and downstream relationships, and their co-occurrence frequency is higher than that of process pairs without direct correlation. The C(i,j) of the mixing process and the hot-pressing curing process is significantly higher in the recycled material batch interval than in the pure virgin material batch interval, indicating that the hot-pressing curing process frequently maintains the pass rate by increasing the molding pressure when the mixing quality declines. The C(i,j) of the mixing process and the grinding process is lower, and the high-value cluster area corresponds to the process segment with the densest compensation masking relationship. When non-adjacent process pairs in the opposite direction co-occurrence distribution table show abnormally high C(i,j), it indicates the existence of a cross-segment compensation chain. For example, the high frequency co-occurrence of negative deviation in mixing and positive deviation in grinding is actually due to insufficient blank density being absorbed step by step by the preforming and hot pressing processes. The residual thickness deviation is offset by increasing the allowance at the grinding end. The positive deviation of process j comes from the intermediate process rather than being directly compensated by this process.

[0028] The process pairs with co-occurrence frequencies exceeding a threshold are screened from the opposite-direction co-occurrence distribution table to verify compensation and masking relationships, forming a masking candidate group. When the C(i,j) of any process pair in the opposite-direction co-occurrence distribution table exceeds the mean of the entire table plus 1.5 to 2 times the standard deviation, it is determined to be a process pair with a co-occurrence frequency exceeding the threshold. This means that the negative deviation of process i and the positive deviation of process j occur in the same batch with a frequency significantly higher than random. The common manifestation of compensation and masking is that downstream operators, after noticing the deterioration of incoming materials, do not report it but instead increase the process intensity of their own process, causing the export qualification rate of their own process to increase instead of decrease, and upstream anomalies to be invisible in quality reports. The verification is based on the process parameter data of the processes exceeding the threshold within the co-occurring batches. Processes with low transmission efficiency η and similar upstream and downstream impact amplitudes in the deviation response characteristics are prioritized for verification because the impact is absorbed during transmission. If the key parameters of process j in the co-occurring batch are systematically increased or decreased compared to non-co-occurring batches, it indicates that it has indeed implemented active parameter adjustments to offset the decline in upstream quality, and the compensation and masking relationship is verified at the process parameter level. This process is included in the masking candidate group. If the parameters are not systematically adjusted but the pass rate is still abnormally high, the positive deviation comes from the random high fluctuation of this process itself rather than active compensation. Its high C(i,j) in the opposite direction co-occurrence distribution table is judged as statistical coincidence. The hot pressing curing process has systematically increased molding pressure in co-occurring batches with declining mixing quality, and this recurs in multiple batches. The grinding process has no adjustment in feed rate in the same co-occurring batch, and the high pass rate is due to the occasional small thickness deviation of the product in this batch. The masking candidate group includes all verified threshold-exceeding process pairs, sorted by C(i,j) from high to low, and each inherits the corresponding co-occurrence batch number list.

[0029] Candidate anomaly co-occurrence groups are generated based on the screening of candidate groups for deviations that occur earlier. To clarify the causal sequence in the confirmed compensation relationship, each process in the candidate group checks the deviation start time of the pass rate of process i and process j within the co-occurring batch. The start time is the grouping inspection time when the pass rate of the batch first deviates from the normal range in the quality inspection record. The process with the earlier deviation is determined as the initiating item, corresponding to the negative deviation start time 10 of the upstream process; the later deviation is determined as the response item, corresponding to the positive deviation start time 11 of the downstream process. The longer the response lag interval 13 formed by the two start times, the longer the upstream anomaly lasts before triggering downstream compensation, and the clearer the causal sequence. When the response lag interval 13 is extremely short, the impact source direction of the process pair in the deviation response characteristics is introduced as an auxiliary basis. If the source direction points to process i, process i is still determined as the initiating item. In the candidate anomaly masking group, the negative deviation start time of the mixing process was significantly earlier than the positive deviation start time of the hot-pressing curing process, spanning several detection cycles; the mixing process was identified as the initiating item. In another process pair with a very small time difference, the batching process was identified as the initiating item after the impact source direction was introduced. These two scenarios were recorded in the candidate anomaly co-occurrence group using the initiating and responding item process numbers, time difference, and auxiliary judgment markers, and inherited from the corresponding co-occurrence batch number list. When the same initiating item process appeared repeatedly in multiple process pairs, multiple responding item processes simultaneously performed compensation operations, causing the upstream real anomaly to be covered by a multi-layered masking effect. Multiple records corresponding to this initiating item were retained side-by-side in the candidate anomaly co-occurrence group, sorted from high to low deviation factors.

[0030] In some embodiments, the step of generating a hypothesis deviation by performing a non-disturbance benchmark comparison on the candidate anomaly co-occurrence group includes: after removing co-occurrence batches of the candidate anomaly co-occurrence group, extracting process parameters for the steady-state period of each process to construct a non-disturbance benchmark state group; comparing the non-disturbance benchmark state group with the candidate anomaly co-occurrence group item by item to form a state difference record; performing a batch switching node deviation correlation assessment on the state difference record to obtain a continuous deviation distribution; and performing segment-by-segment deviation verification based on the continuous deviation distribution to generate a hypothesis deviation.

[0031] After removing co-occurring batches from the candidate anomaly co-occurrence group, process parameters for the steady-state period of each process are extracted to construct a disturbance-free baseline state group. The steady-state period refers to a continuous batch segment in which the pass rate of each production process is consistently maintained near the historical average and the variation of process parameters is lower than the historical standard deviation of that process within the analysis window. The selection of this period must avoid all co-occurring batches inherited from the candidate anomaly co-occurrence group, and there must be no equipment alarm records or planned parameter adjustments within this batch segment. Both conditions together ensure that the extracted parameters represent the true normal operating state of the process when the same grade of virgin rubber is continuously applied, the mold is not disassembled, and the controller settings are not manually changed. During the steady-state period, the mean μ(base,k) and standard deviation σ(base,k) of each process parameter are calculated and included in the undisturbed baseline state group, where k is the parameter number. Both values ​​have the same dimensions as the corresponding parameter. The number of consecutive batches during the steady-state period should be no less than 30% of the number of batches in the analysis window; otherwise, μ(base,k) and σ(base,k) will drift with individual batch fluctuations. The smaller σ(base,k), the higher the sensitivity to detecting deviations. Processes with large σ(base,k) exhibit wide natural fluctuation bands in process parameters even under conditions without external disturbances; their judgment threshold should be appropriately relaxed to avoid misjudging normal fluctuations as abnormal deviations. In the candidate abnormal co-occurrence group, the initiating item of the mixing process has a stable pass rate and small fluctuation range of process parameters in the pure virgin material batch interval. This interval represents the steady-state period of the mixing process, and the baseline values ​​of mixing temperature, rotor speed, and torque are included in the undisturbed baseline state group accordingly. The pass rate of the recycled material batch interval fluctuates significantly and parameters are frequently adjusted; the baseline value for the mixing process is only taken from the aforementioned pure virgin material batch interval. For processes with a small batch size during the steady-state period, the uncertainty of the baseline value estimation is relatively large, and the deviation judgment threshold of the process parameters is adjusted accordingly.

[0032] For example, the step of comparing the undisturbed baseline state group with the candidate anomaly co-occurrence group item by item to form a state difference record includes: extracting the current process parameter values ​​of each process for the candidate anomaly co-occurrence group; comparing the current process parameter values ​​with the undisturbed baseline state group item by item to form a difference record; identifying parameters with continuously zero differences in the difference record to generate a compensation masking mark group; and performing a correlation trend type check on the difference record based on the compensation masking mark group to generate a state difference record.

[0033] For candidate anomaly co-occurrence groups, the current process parameter values ​​for each process are extracted. Using the original acquisition interval as the accuracy benchmark and without pre-smoothing, process parameter data for all acquisition cycles of each process within the coverage of the co-occurrence batch number list inherited by the candidate anomaly co-occurrence group are extracted, and the original readings are retrieved for each parameter. The fluctuation range within a batch is itself an important indicator of process execution stability; pre-smoothing can mask the impact of short-term fluctuation events on the batch mean. Parameters of multiple processes involved in the same co-occurrence event within the candidate anomaly co-occurrence group must be compared on the same clock benchmark. Acquisition times of different parameters in the same process are aligned with the unified clock of the data system. If the clock deviation exceeds one acquisition interval, the parameter value at the nearest acquisition time is interpolated. When the number of interpolated points exceeds a set upper limit for the total number of acquisition points in the batch, the corresponding parameter value is marked with low confidence. Interpolation is performed linearly between adjacent acquisition points without crossing batch boundaries. The deviation threshold for this parameter is adjusted accordingly to avoid interpolation distortion. When the acquisition intervals of different processes are inconsistent, the process with the longest interval is used as the alignment benchmark, and the arithmetic mean of processes with shorter intervals is taken within the alignment window to avoid distortion of the batch mean due to differences in sampling density. The batch mean and range of each parameter within the co-occurring batch are included in the current process parameter values. The former reflects the overall parameter level, while the latter reflects the stability of execution within the batch. Parameters with large ranges are marked with additional range anomalies for the corresponding batches. In the hot-pressing curing process, the mold temperature range within the co-occurring batch is significantly higher than the range of the same parameter during the steady-state period. The batch mean is only slightly different from the baseline, and the range anomaly mark reveals the periodic fluctuation characteristics of mold heating within this batch. In the mixing process, the rotor torque range within the same batch is close to that during the steady-state period. Both the batch mean and range in the current process parameter values ​​are normal, indicating that the execution stability of this process has not deteriorated within this range.

[0034] The current process parameter values ​​are compared with the undisturbed baseline state group item by item to form a difference record. The difference Δp(k) is obtained by subtracting the baseline mean μ(base,k) of the corresponding parameter in the undisturbed baseline state group from the batch mean of the current process parameter value. k is the parameter number, and its dimension is consistent with that of the parameter. Δp(k) is arranged in the order of co-occurring batches to form a difference time series. The difference time series of each parameter in each process are arranged side by side to form a difference record. The time series length is equal to the number of co-occurring batches. The difference time series of multiple parameters in the same process are arranged adjacently to facilitate joint observation of cross-parameter direction and amplitude. In the difference record, the mixing temperature difference time series in the batch containing recycled material during the mixing process is consistently negative, while the rotor torque difference time series is consistently positive. The rotor speed difference time series shows no directional change. The two parameters are opposite in direction and continuous, indicating that the high viscosity of the recycled material causes a lag in material heating, resulting in the measured temperature consistently falling below the set value while the torque passively increases. The difference time series of another process parameter randomly alternates between positive and negative around the baseline mean in the undisturbed baseline state group, without a continuous direction. The difference record performs the same subtraction on batches with batch-specific extreme difference anomaly markers in the current process parameter values, and synchronously retains the marker on the corresponding difference time series, allowing for separate verification of Δp(k) for that batch during reading. The difference records are archived with two levels of indexing: process and parameter. Each difference time series is accompanied by the parameter name, dimension, co-occurrence batch number and the reference mean μ(base,k) used: the parameter name and batch number determine the position of the time series in the difference record, the dimension is retained along with the parameter to avoid scale confusion when comparing across parameters, and the reference mean ensures that the comparison benchmark used to recalculate Δp(k) in any step is completely consistent with the current step.

[0035] For difference records, a compensation masking flag group is generated by identifying parameters with continuously zero differences. When the absolute value of parameter Δp(k) in the difference record is lower than a certain proportion of the parameter σ(base,k) in several consecutive batches, this batch segment is defined as a continuously zero difference segment. Parameters within the co-occurring batch range of this segment are candidate parameters for the compensation masking flag group. Such a near-zero parameter Δp(k) does not indicate normal operation; rather, it suggests that operators should keep this parameter unchanged according to the work instructions, only compensating with other parameters, or forcibly pull the controller back to the set value in the closed loop, thus masking the manifestation of process deviations in this parameter. In the difference record, the heat treatment process holding time difference is continuously zero in the co-occurring batches, while the concurrent pass rate shows a negative deviation, indicating that the operator maintained the holding time unchanged when the abnormality occurred, and the quality decline came from other unlocked parameters. The furnace temperature distribution difference shows a continuous positive deviation and is not locked, indicating that the operator chose an operation strategy of maintaining a stable holding time while actively adjusting the furnace temperature. The starting batch and the number of consecutive batches in the segment with a persistent zero difference are recorded simultaneously. For candidate parameters whose starting batch is significantly later than the start time of co-occurrence, the parameter locking stems from the operator's delayed perception of the anomaly rather than preventative locking. Candidate parameters whose segment coverage reaches a certain proportion of the total co-occurring batches and whose absolute value is stable are marked with a high-confidence masking label, while those with a low coverage proportion are marked with a low-confidence label. The former has sufficient evidence of active locking, while the latter may stem from natural parameter stabilization. The compensation masking label group covers all compensation masking candidate parameters and their confidence labels. Confidence levels are categorized based on the segment coverage accounting for 60% of the total co-occurring batches. High-confidence masking candidate parameters are listed separately with a locking label. It is assumed that deviation verification is performed only on low-confidence candidate parameters and non-masked parameters, both of which are included in trend verification.

[0036] Based on the compensation masking mark group, the difference records are used to perform correlation trend type verification to generate state difference records. Each process and parameter in the difference records is first divided into masked parameters and non-masked parameters according to the marking status of the compensation masking mark group. Masked parameters have a masking attribute label added to their time series in the difference records. Non-masked parameters not included in the compensation masking mark group have their difference time series entered into correlation trend type verification. The verification uses directional persistence and amplitude consistency as evaluation dimensions. Directional persistence is measured by the proportion of batches with positive or negative values, and amplitude consistency is measured by the difference time series coefficient of variation CV(k), which is the ratio of the absolute value of the difference time series standard deviation to the mean, with a dimension of one. A persistent direction with a low CV(k) is judged as a monotonic deviation type; a persistent direction but with increasing amplitude batch by batch and a high CV(k) is judged as an aggravated deviation type; and a random direction but with an overall large amplitude is judged as a random high amplitude type. These three trend types are labeled parameter by parameter in the state difference records. The aggravated deviation type corresponds to a continuous worsening of the deviation as the co-occurring batches progress. For example, the gradual addition of recycled material causes the heat absorption of the rubber compound to accumulate batch by batch, and the mixing temperature decreases batch by batch. This indicates that the parameter has not been effectively pulled back after the switch, and its weight in the determination of the assumed deviation amount is higher than that of the monotonic deviation type. The random high-amplitude deviation is dominated by random disturbances, and its causal relationship with the batch switching event is weaker than the first two types. It is distinguished by a weak correlation mark in the state difference record. The mixing temperature difference in the recycled material batch range during the mixing process is consistently low, and the deviation amplitude deepens batch by batch. This is determined to be an aggravated deviation type, which contrasts with the monotonic deviation type of rotor torque in the same process. The former indicates that the response capability of the temperature control system has failed to keep up with the rhythm of the cumulative change in the viscosity of the recycled material with each batch, while the latter indicates that the viscosity increase has remained at a stable level throughout the batch range without further aggravation.

[0037] A persistent deviation distribution was obtained by evaluating the batch switching node deviation correlation of state difference records. The time sequence of Δp(k) for each parameter in the state difference records was compared one by one with the batch switching markers in the batch numbers. Switching markers included production nodes such as raw material batch switching, mold replacement, equipment maintenance, and adjustments to the recycled material blending ratio. Parameters whose Δp(k) showed abrupt changes in sign or amplitude before and after a switching marker were defined as switching-related parameters, indicating a temporal causal relationship between the parameter's deviation from the baseline and the corresponding batch switching event. Among the switching-related parameters, those whose Δp(k) continuously deviated after batch switching and did not naturally return to the baseline level were classified as persistent deviations. Physically, this means that the parameter was continuously pulled away from the baseline after switching, and this deviation was the true process deviation subsequently masked by downstream compensation. Δp(k) that deviated briefly and then returned was classified as a transient response, such as higher molding pressure in the first few molds after mold replacement, which returned to the baseline as the mold temperature field stabilized. Parameters marked as aggravating deviations in the state difference records were prioritized as persistent deviations because their amplitude deepened batch by batch. The persistent deviation distribution is compiled from all persistent deviation items by process number. Each item records the parameter name, the batch switching marker that triggered the deviation, the starting and ending batches of the deviation, the number of batches the deviation spans, and the direction and mean of Δp(k) within that segment. The more batches spanned, the longer the impact time of the switching event on the process. In the mixing process, after the switching marker for increased recycled material ratio, both mixing temperature and rotor torque are registered as persistent deviation items, and their deviation batch intervals in the persistent deviation distribution highly overlap. In the hot pressing curing process, only molding pressure is registered as a persistent deviation item after the same marker.

[0038] Based on the continuous deviation distribution, a segmented deviation verification is performed to generate the hypothetical deviation amount. The standardized hypothetical deviation amount δ(k) = Δp(avg,k) / σ(base,k), where k is the parameter index, Δp(avg,k) is the segmented mean of the Δp(k) time sequence corresponding to each initiating process continuous deviation item in the continuous deviation distribution after being divided into several deviation segments according to the batch switching mark, and σ(base,k) is the benchmark standard deviation of this parameter in the undisturbed benchmark state group. The numerator and denominator have the same dimensions, and δ(k) has a dimension of one. The larger the absolute value of δ(k), the more significant the continuous deviation of this parameter is relative to the benchmark fluctuation level, and the deeper the degree to which its process execution state deviates from the normal operating range. The segmented deviation assessment of each item in the continuous deviation distribution uses a threshold value for δ(k) exceeding a certain multiple, typically between 2 and 3. For parameters marked as aggravating deviation in the state difference record, the threshold value is lowered accordingly; for parameters with random high amplitude, the threshold value is raised accordingly. Parameters exceeding the threshold are considered significantly deviating, while those below the threshold are considered slightly deviating. Only significantly deviating parameters are included in the hypothesized deviation amount; slightly deviating parameters and locked-mark parameters are retained as notes for future reference. In the mixing process, after the batch switch from recycled materials, the absolute values ​​of δ(k) for both mixing temperature and rotor torque are significantly higher, and both parameters simultaneously enter the significantly deviating range. In the hot-pressing curing process, the absolute value of molding pressure δ(k) is higher, while the mold temperature δ(k) is close to zero. It is assumed that only the molding pressure in the hot-pressing process shows a significant deviation, while the mold temperature is retained as a note. The simultaneous deviation of multiple parameters from the baseline in the mixing process is greater than that in the hot-pressing process, representing the most severe source of process state deviation in the current batch switch. Therefore, parameter deviations in the mixing process are the primary target for targeted tracing within this batch range.

[0039] Step S13: Calculate the cross-process recurrence rate through candidate anomaly co-occurrence groups, and perform recurrence rate drop node correction on candidate anomaly co-occurrence groups based on hypothesis deviation and cross-process recurrence rate to generate correction anomaly groups. Merge and screen the correction anomaly groups to determine the main anomalies.

[0040] Specifically, the cross-process recurrence rate is calculated using candidate anomaly co-occurrence groups. The cross-process recurrence rate measures the degree to which the same anomaly repeatedly occurs in different batches and affects multiple processes along the production line; its value is governed by both the number of repetitions and the scope of impact. Each co-occurring event is jointly identified by the initiating process number in the candidate anomaly co-occurrence group and the name of the significant deviation parameter for that process in the hypothesis deviation. The cross-process recurrence rate R(e) = (n(e) × p(e)) / (N × P), where e is the co-occurrence event number, n(e) is the number of batches in the analysis window where the co-occurrence event occurs, p(e) is the number of processes affected by the event (i.e., all co-occurring processes originating from the same initiating event, counted after deduplication), N is the total number of batches in the analysis window, and P is the total number of processes participating in the co-occurrence statistics. Inspection instrument 7 is not included in P. The dimension of R(e) is one. Higher R(e) indicates a wider spread of co-occurring events across both batch and process dimensions, typically corresponding to systemic anomalies involving synchronized responses across multiple processes during recycled material blending changes. Lower R(e) often originates from occasional disturbances in a single process. Events with high n(e) and low p(e) occur almost in every batch but only affect the current process and adjacent processes; R(e) is suppressed due to the small p(e) multiplier. Events with low n(e) and high p(e) occur infrequently but affect multiple processes each time. Since R(e) may be similar in both cases but have different diffusion patterns, both components must be stored and recorded separately along with R(e): n(e) records the frequency of the event in the batch dimension, and p(e) records its breadth of impact in the process dimension. The R(e) values ​​of each co-occurring event in the candidate anomaly co-occurrence group are listed one by one according to the initiating process number. Multiple co-occurring events of the same initiating process are compared in parallel to determine the cross-process recurrence rate. When the R(e) of the initiating process is generally high, its anomaly has a strong systemic transmission effect on the downstream.

[0041] In some embodiments, the step of correcting the recurrence rate drop node of the candidate anomaly co-occurrence group based on the assumed deviation and the cross-process recurrence rate to generate a corrected anomaly group includes: classifying the candidate anomaly co-occurrence group according to anomaly type to form a classified anomaly subgroup; checking the recurrence rate level of each classified anomaly subgroup based on the cross-process recurrence rate to form a recurrence rate level distribution; determining the recurrence rate drop node based on the recurrence rate level distribution and the assumed deviation to form a correction screening threshold; and screening the candidate anomaly co-occurrence group according to the correction screening threshold to generate a corrected anomaly group.

[0042] Candidate anomaly co-occurrence groups are categorized into classification anomaly subgroups based on anomaly type. The classification unfolds along two dimensions: the sign of the abnormal parameter δ(k) of the initiating process of each co-occurring event in the candidate anomaly co-occurrence group determines the direction of parameter rise / fall; the quantile interval of the absolute value of δ(k) within the distribution of all hypothetical significant deviations from the parameter determines the deviation magnitude level, typically divided into mild, moderate, and severe levels using tertiles. Each co-occurring event is assigned to its corresponding anomaly type based on the two-dimensional combination of rise / fall direction and magnitude level, forming six anomaly subgroups: mild rising deviation, moderate rising deviation, severe rising deviation, mild falling deviation, moderate falling deviation, and severe falling deviation. The set of co-occurring events corresponding to each of the six anomaly types constitutes the six member subgroups of the classification anomaly subgroups. Within the same type, member co-occurring events exhibit homogeneity in both rise / fall direction and magnitude interval. For types with low dispersion of R(e) values, the cross-process diffusion pattern of this type of anomaly is more consistent across different batches. Co-occurring events whose absolute value of δ(k) falls exactly near the amplitude quantile boundary are simultaneously classified into two adjacent categories and given boundary markers. When checking the recurrence rate level, if the influence of the boundary event on the mean R(e) of both types does not exceed a certain proportion, the current classification result remains unchanged to prevent the classification conclusion of this type of event from having an unrobust impact on the location of the sudden drop in recurrence rate due to the error in setting the quantile boundary. Co-occurring events caused by a sudden increase in the proportion of recycled material in the candidate anomaly co-occurrence group are concentrated in the moderate anomaly subgroup with a decrease in the ratio. The initiating items are mostly due to the low temperature in the mixing process. In contrast to the mild anomaly subgroup with an increase in the ratio caused by occasional equipment disturbances during the period of stable ratio, the former usually has a significantly larger number of events than the latter. Each member subgroup, together with its boundary markers, constitutes a complete classification anomaly subgroup. Each subgroup registers a list of type names, member events, and member event numbers.

[0043] Based on the cross-process recurrence rate, the recurrence rate levels of each type of anomaly subgroup were examined to form a recurrence rate level distribution. The R(e) values ​​of all co-occurring events in each type of anomaly subgroup were statistically analyzed by type. The mean R(e) of a type of subgroup represents the average cross-process recurrence rate of that type of anomaly within the analysis window; a higher mean indicates a stronger overall diffusion ability across batch and process dimensions. The standard deviation of R(e) reflects the consistency of the diffusion ability of co-occurring events within the same type; a smaller standard deviation indicates that the diffusion pattern of that type of anomaly is highly stable across different batches, while a larger standard deviation indicates that its diffusion range varies significantly across different batches. When observing both statistics together, type subgroups with high mean and low standard deviation correspond to systematic anomalies with strong diffusion ability and batch-to-batch stability, while those with low mean and low standard deviation correspond to occasional local anomalies with weak diffusion ability and low stability. Before statistical analysis, subgroups containing only a single co-occurring event were removed from the abnormal subgroups. Their means and standard deviations were not statistically significant, and these subgroups were marked as having insufficient samples in the recurrence rate distribution. Further verification was conducted after the analysis window was expanded. The number of events, mean R(e), and standard deviation of R(e) for each subgroup were recorded together. The number of events was used to measure whether the samples on both sides of the fault were balanced, and the standard deviation was used to measure whether the fault drop was significantly greater than the dispersion level within the type. The mean R(e) was given for each type. The statistically obtained mean values ​​for each type were listed side-by-side according to type number, representing the recurrence rate distribution. Types with insufficient samples were marked as placeholders. Among these, the moderately abnormal subgroup (consistently low mixing temperature) had a significantly higher mean recurrence rate across processes than other subgroups, indicating that this type of anomaly triggered synchronous responses in multiple processes across multiple batches. The mildly abnormal subgroup (temporarily high hot-pressing pressure) had a low mean R(e), indicating that the intermittent high hot-pressing pressure event had limited cross-process transmission and was insufficient to form a full-line linkage.

[0044] The recurrence rate drop node is determined based on the horizontal distribution of recurrence rate and the assumption deviation to form a correction screening threshold. After sorting the mean R(e) of each type subgroup in the horizontal distribution of recurrence rate from high to low, the difference between the means of adjacent two type subgroups is calculated one by one. The adjacent type pair with the largest difference and the difference exceeding the sum of the standard deviations of R(e) of the type subgroups on both sides of the fault corresponds to the fault with the largest diffusion capacity. The location of this fault is defined as the recurrence rate drop node. Above the node, it is a systemic anomaly caused by changes in the entire batch of incoming materials or formula, which can be transmitted to each process along the entire line. Below the node, it is a local disturbance with occasional vibration of a single piece of equipment, which disappears on its own in the next batch. The nature of the two sides is essentially different. When the largest and second largest differences are close, there are two potential faults in the horizontal distribution of the recurrence rate. In this case, a hypothesis deviation is introduced for auxiliary correction: the mean absolute value of δ(k) of the corresponding subgroups of the two candidate fault locations is used as an auxiliary criterion. Candidate locations with a larger difference in the mean absolute value of δ(k) are preferentially identified as nodes of sudden drop in recurrence rate, ensuring that the node location has the greatest distinguishing significance in both R(e) and δ(k) dimensions. The correction screening threshold is determined by using the mean R(e) corresponding to the sudden drop in recurrence rate node as the lower limit, combined with the quantile threshold of the parameter deviation amplitude in the hypothesis deviation. For co-occurring events in the candidate abnormal co-occurrence group, both conditions must be met: R(e) must be higher than the threshold of the sudden drop node, and the corresponding absolute value of δ(k) must be higher than the amplitude quantile threshold, in order to pass the joint judgment of the correction screening threshold. For co-occurrence events of low mixing temperature in batches with recycled material blending ratio switching, both conditions are met, and joint screening is passed; for events of slight fluctuation in surface hardness during heat treatment, although R(e) is high, the absolute value of δ(k) is close to the slight deviation boundary, and these events are excluded by the joint screening conditions.

[0045] Candidate anomaly co-occurrence groups are screened based on correction screening thresholds to generate correction anomaly groups. Co-occurrence events that simultaneously meet both the lower limit of R(e) and the amplitude threshold of δ(k) in the correction screening threshold are screened and enter the correction anomaly group. Co-occurrence events that do not meet the criteria are retained in the candidate anomaly co-occurrence group as notes. Co-occurrence events retained in notes are re-involved in cross-process reproducibility statistics when the analysis window is updated batch by batch. The rolling update is based on a batch as the progress step to avoid the incorrect exclusion of real systemic anomalies due to the small batch size in the current batch window. Screening is performed by comparing each co-occurrence event against the correction screening threshold. The two criteria, R(e) and δ(k), are not interchangeable. Only events that meet both criteria can enter the correction anomaly group to prevent high-frequency low-amplitude events or low-frequency high-amplitude events from passing the screening based on only one indicator. Co-occurrence events with R(e) values ​​higher than the drop node threshold but absolute δ(k) values ​​slightly lower than the amplitude threshold are defined as boundary candidate events. They are temporarily listed as observation items in the current analysis window. If the absolute value of δ(k) increases with the accumulation of batches after the window is scrolled forward, a re-evaluation is automatically triggered. The existence of boundary candidate events indicates that the parameter anomaly is in the transitional stage from slight deviation to significant deviation. The evolution trend of its deviation magnitude needs to be continuously observed in the rolling window. The re-evaluation trigger is based on a continuous monotonically increasing absolute value of δ(k) over several batches. The number of remaining co-occurring events in the corrected anomaly group is usually significantly less than that in the candidate anomaly co-occurrence group: several high-frequency but extremely small surface hardness fluctuation events in the heat treatment process are excluded from the analysis window; the co-occurring event of the mixing temperature in the mixing process has a high R(e) value and its absolute value of δ(k) far exceeds the amplitude threshold, successfully passing the correction screening and entering the corrected anomaly group.

[0046] In some embodiments, the step of merging and screening the corrected anomaly groups to determine the main anomalies includes: performing common cause correlation checks on each anomaly in the corrected anomaly group to obtain common cause correlation groups; extracting the anomaly with the highest cross-process recurrence rate within each common cause correlation group as a representative item; performing downstream process compensation response intensity assessment on each representative item to generate a compensation response index; and screening representative items whose compensation response exceeds a threshold based on the compensation response index to determine the main anomalies.

[0047] Common cause association checks were performed on each anomaly in the corrected anomaly group to obtain common cause association groups. Co-occurring events with the same origin or highly correlated origin were grouped into the same common cause association group within the corrected anomaly group. Highly correlated origins refer to parameter anomalies in different processes that, although manifesting differently, can be traced back to the same material change or mold change. For example, slow mixing temperature and low preforming material temperature after material change are actually caused by the same batch of rubber. Co-occurring events with the same initiating process and consistent abnormal parameter δ(k) direction are grouped into the same common cause association group regardless of batch switching markers. The check also introduced cross-validation of batch switching marker association and δ(k) direction consistency. Grouping conclusions satisfying both conditions have the highest reliability; those satisfying only one condition are labeled with medium reliability. Representative items in medium reliability common cause association groups must simultaneously satisfy the highest values ​​for both R(e) and δ(k) within the group to be established, thus constraining the interference of grouping uncertainty on the selection of representative items. The verification was carried out pairwise within the correction anomaly group. Events with different parameters in the same initiating process were compared first, while cross-process events were mainly based on the overlap of batch switching marker times. Within this batch interval, the co-occurrence of low mixing temperature and high rotor torque in the mixing process and the co-occurrence of low material temperature in the preforming process both pointed to insufficient heat supply caused by the change in the recycled material blending ratio. All three met the batch switching marker correlation conditions. In addition, the mixing temperature and material temperature also met the δ(k) direction consistency condition, and were classified into a high-confidence common cause correlation group. The occasional high molding pressure event that occurred independently in the hot pressing process had no batch switching marker correlation and the δ(k) direction was inconsistent with the mixing group. It was retained as an independent common cause correlation group, and its representative item was the event itself.

[0048] For each co-causal association group, the anomaly with the highest cross-process recurrence rate within the group is extracted as the representative item. The co-occurring event with the highest R(e) value within the group is defined as the representative event, which best represents the transmission influence capability of the co-causal association group in the cross-process diffusion dimension. The initiating process and anomaly parameter of the representative event are used as the representative items of the co-causal association group. When the R(e) values ​​of member co-occurring events are similar, it indicates that multiple anomaly parameters within the group are highly similar in cross-process diffusion capability. In this case, the member co-occurring event with the largest absolute value of δ(k) is used as the representative item. The one with the largest absolute value of δ(k) deviates the most from the baseline and has the most significant substantial impact on the process operation status. The cross-process recurrence rate of co-occurring events of each member within the group is taken from the aforementioned statistical results and does not need to be recalculated. The extraction is performed independently for each co-causal association group. The relative importance between groups is left to be uniformly measured in the compensation response intensity assessment stage. This step only solves the problem of the uniqueness of the representative items within the group. After the representative item is determined, its common cause association group number, initiating process number, abnormal parameter name, R(e) value, and δ(k) value are recorded as the data basis for the compensation response intensity assessment. When the difference between the highest and second highest R(e) values ​​within the group is extremely small and the absolute values ​​of δ(k) of the two parameters are also close, both co-occurring events of the two members enter the compensation response intensity assessment as joint representative items. Their compensation response intensity is calculated independently, and the one with the higher compensation response intensity is selected as the final representative item. This ensures that the selection of representative items can still reach a unique and definite conclusion when the cross-process recurrence rate and δ(k) dimensions cannot be distinguished by a single factor. In the common cause association group, the R(e) value of the co-occurring event of mixing temperature is significantly higher than that of the co-occurring event of rotor torque in the same group. The co-occurring event of mixing temperature enters the compensation response intensity assessment as a representative item.

[0049] For each representative item, a downstream process compensation response intensity assessment is performed to generate a compensation response index. The compensation response index is quantified by the compensation response intensity I(c), where I(c) = Σ(d=1→M)|ΔS(d) / σ(base,d)| / M, c is the representative item number, d is the associated downstream process number, ΔS(d) is the difference between the parameter setting value of the d-th associated downstream process before and after adjustment in the corresponding batch, and this setting value adjustment is reflected in the measured value with the same magnitude. Therefore, the baseline standard deviation σ(base,d) of this parameter in the undisturbed baseline state group is taken as the normalization scale, and the two have the same dimension. M is the number of associated downstream processes, and I(c) has a dimension of one. The larger the value, the more widespread and intense the compensation response triggered by the anomaly corresponding to the representative item in the downstream, and the more significant the masking effect. Positive deviation events in downstream processes are obtained from the candidate anomaly co-occurrence group by extracting the response item process number according to the representative item. The parameter adjustment magnitude of the downstream process in the batch of the initiating item anomaly is read from the process parameter adjustment record. The batch-to-batch variance of I(c) also carries information: a smaller variance indicates that the downstream compensation for the anomaly is highly consistent across batches. For example, if the hot pressing temperature is too low, the molding pressure will be increased to the same target value, which is a fixed intervention habit formed by operators. A larger variance indicates that the compensation magnitude varies significantly from batch to batch. One representative item is the event of low mixing temperature in the mixing process. In the batch with the change in the proportion of recycled materials, it is associated with two downstream compensation responses: the increase in molding pressure in the hot pressing process and the extension of the holding time in the preforming process. The I(c) value ranks relatively high among all representative items and has a small batch-to-batch variance. The compensation response index consists of two parts: the I(c) value of all representative items and its batch-to-batch variance. The former is a strength scalar that can be directly compared between representative items, and the latter is obtained from the dispersion of the I(c) component value of the representative item in each co-occurring batch.

[0050] Based on the compensation response index, representative items with compensation responses exceeding the threshold were selected to determine the main anomalies. Using the mean and standard deviation of I(c) among all representative items as the screening benchmark, representative items whose I(c) exceeds the mean plus one standard deviation were identified as having compensation responses exceeding the threshold. The corresponding initiating process anomalies generated a strong compensation drive beyond normal levels in downstream processes, representing the most significant source of anomalies with a masking effect in the production line, and were thus identified as main anomalies. Representative items whose I(c) did not exceed the threshold, although passing the dual screening of R(e) and δ(k) in the correction screening, still had downstream compensation responses within the normal fluctuation range of the production line in actual production. These were considered secondary sources of quality masking in the current batch range and were retained as secondary anomalies for rolling analysis reference. The distribution pattern of I(c) among all representative items in the compensation response index also reflects the overall masking intensity of the production line in the current batch range: when the overall mean is high, the various processes in the current batch range generally implement compensation interventions with high intensity, and the abnormal masking behavior occurs systematically throughout the entire line, usually corresponding to the concentrated adjustment stage after a significant change in the recycled material blending ratio; when the overall mean is low but individual representative items I(c) are prominently high, only a few process anomalies have a strong downstream compensation driving ability, and the masking effect is limited to a specific transmission path. The main anomalies finally identified within the analysis window are concentrated in the low mixing temperature event in the mixing process. This event has the highest R(e) value, the largest absolute value of δ(k), a clear representative item status, and I(c) ranked first, with all four indicators pointing to the same source. After the main anomalies are identified, their corresponding co-occurring batch number list and associated downstream process list are recorded together, which respectively define the batch range and process range of the anomaly.

[0051] Step S14: Extract the attribution discrimination features of the main anomalies, perform upstream and downstream process attribution discrimination based on the attribution discrimination features to determine the attribution category of the anomalies, perform cross-batch attribution traceability comparison based on the attribution category of the anomalies to determine the cross-batch attribution reversal items, and perform upstream traceability verification based on the cross-batch attribution reversal items to generate production diagnostic conclusions.

[0052] Specifically, the attribution discrimination features of major anomalies are extracted. The feature extraction window extends forward several batches from the batch of co-occurring events corresponding to the major anomalies. The anomaly parameter values ​​of each batch within the window and the key parameter values ​​of the upstream processes are read from the process parameter data, and the corresponding batch pass rate is read from the quality inspection records. The two are synchronized according to the batch number. The attribution discrimination features are extracted from three dimensions. The temporal offset feature is quantified by the batch time difference between the deviation of the main anomaly parameter and the upstream process parameter from the starting time. The method for determining this time difference is consistent with the time difference registered in the candidate anomaly co-occurrence group. f1(k,b)=1-Δb / Δb(max), where k is the anomaly parameter number, b is the batch number, Δb is the number of batches spanned by the time difference, and Δb(max) is the batch span of the feature extraction window. The dimension is one. When Δb exceeds the window, it is taken as 0. The smaller the time difference, the closer the value is to 1, and the stronger the upstream causal transmission. The parameter correlation feature is quantified by the Pearson correlation coefficient between the anomaly parameter deviation sequence within the window and the deviation sequence of each upstream process parameter. Its absolute value is denoted as f2(k,b). The dimension is one. A larger value indicates a stronger inter-batch fluctuation correlation. The batch consistency feature is quantified by the consistency rate of the deviation direction of the anomaly parameter within the same raw material attribute batch group, f3(k,b). A high consistency rate indicates a stable correlation with the raw material batch attribute. The main anomaly of the mixing process with low mixing temperature is f1(k,b), which approaches 1 because it deviates from the starting time and is very close to the time of adjustment of the recycled material ratio in the upstream batching process. f2(k,b) is much higher when paired with the batching ratio than when paired with the equipment status parameters of this process. f3(k,b) approaches 1 in the high-ratio batch group. The three components are registered with parameter and batch two-level indexes, which together constitute the classification characteristics of this main anomaly.

[0053] Based on the attribution discrimination features, the attribution of upstream and downstream processes is determined to identify the category of anomalies. The weighted score G(k,b) = w1×f1(k,b) + w2×f2(k,b) + w3×f3(k,b) is a weighted composite of the three components of the attribution discrimination features. k is the anomaly parameter number, b is the batch number, f1(k,b), f2(k,b), and f3(k,b) are, respectively, the normalized value of the time-series offset feature in the attribution discrimination features, the consistency rate of the batch with the largest absolute value of the upstream correlation coefficient, and w1, w2, and w3 are the weight coefficients. The sum of the three is 1, and G(k,b) has one dimension. The weights are determined by taking historical batch samples that have been manually checked, confirmed, and archived. The values ​​of each component are statistically analyzed according to two categories, and the separation degree obtained by dividing the difference between the two means by the sum of the standard deviations is normalized to obtain the weight. The two attribution judgment thresholds are also determined by the distribution of historical samples G(k,b), usually 0.7 and 0.4. Batches with G(k,b) exceeding the upstream attribution threshold are classified as upstream-transferred attribution categories; those below the current process attribution threshold are classified as current process attribution categories; and those in between are classified as mixed attribution categories. Each batch thus obtains an abnormal attribution category. The abnormal attribution categories of each batch, along with the batch's G(k,b) and the values ​​of the three components, are arranged into a sequence indexed by the batch number. This sequence is called the attribution category batch sequence. Mixed attribution corresponds to the bias caused by the superposition of upstream incoming material and the current process equipment status. The upstream contribution ratio is calculated by subtracting the current process threshold from G(k,b) and dividing by the difference between the two thresholds. Batches with this ratio greater than half are included in the upstream-transferred attribution category, and the rest are included in the current process attribution category. The main abnormality of low mixing temperature is that in batches with a high proportion of recycled material, G(k,b) is significantly higher than the upstream attribution threshold, and the abnormal attribution category is determined to be upstream-transferred attribution.

[0054] Based on the abnormal attribution category, cross-batch attribution tracing comparison is performed to determine the cross-batch attribution reversal item. Before comparison, batches with mixed attribution categories in the batch time series are first merged into the upstream transmission attribution or the current process attribution according to the upstream contribution ratio, so that the abnormal attribution category retains only two values ​​in the time series. The attribution reversal segment is the batch segment in the time series where the upstream transmission attribution and the current process attribution alternate. Its physical meaning is that the phenomenon of the same low mixing temperature is judged as being caused by heating element aging in batches with stable blending ratio, and as being caused by upstream incoming materials in batches with blending ratio switching. The cause swings back and forth between batches. The determination of the attribution reversal segment requires that the abnormal attribution category changes within a number of consecutive batches. Occasional category changes in a single batch do not trigger the determination, so as to exclude misjudgments caused by random fluctuations of G(k,b) around the attribution discrimination threshold. When comparing the batch time of the attribution reversal segment with the recycled material blending ratio switching mark in the batch number, if the start time highly overlaps with the blending ratio switching mark, the upstream and downstream attribution conversion of this anomaly is driven by the raw material batch switching. Before the switching, this anomaly was mainly attributed to internal factors of this process, but after the switching, it became an upstream transmission, indicating that the change in the properties of upstream raw materials became the main driving force of this parameter anomaly after the switching. Any anomaly parameter with an attribution reversal segment among all major anomalies is considered a cross-batch attribution reversal item. The existence of cross-batch attribution reversal items indicates that the cause of the corresponding major anomaly involves a dynamic switching of upstream and downstream attribution during the production process. Relying solely on the attribution judgment result of a single batch is insufficient to accurately locate the true root cause of the anomaly; further root cause location needs to be pinpointed through cross-batch attribution tracing.

[0055] In some embodiments, the step of generating a production diagnostic conclusion by performing upstream tracing verification based on the cross-batch attribution reversal item includes: intervening in the batch attribution location of the cross-batch attribution reversal item to form a masked intervention batch group; extracting intervention offsets from the upstream processes associated with the masked intervention batch group to form a root cause location marker group; verifying the process attribution consistency of each marker item in the root cause location marker group to form an attribution comparison result; and integrating the attribution results of each process based on the attribution comparison result to generate a production diagnostic conclusion.

[0056] Intervention batches for cross-batch attribution reversal items are used to locate the attribution of batches that have undergone intervention, forming a masked intervention batch group. The location is based on the attribution reversal segment of the cross-batch attribution reversal item. Each item in the cross-batch attribution reversal item is processed segment by segment: first, batches whose abnormal attribution category has changed from upstream transmission to the current process attribution are extracted; then, the process parameter adjustment records of each downstream process within that batch segment are retrieved, and the set values ​​of each parameter are checked to see if there are unidirectional and repetitive adjustments across batches within that segment. Batches with such systematic adjustments are identified as masked intervention batches because their attribution in this process originates from active intervention in the downstream process rather than from the actual cause in this process. The batch number, the downstream process number that performed the intervention, the name of the adjusted parameter, and the direction of adjustment are recorded. All such batches are grouped together to form a masked intervention batch group. In these batches, the upstream blending ratio is still too high, but once pressure is applied at the hot pressing end, the mixing outlet qualification rate on the report is brought back to normal. The anomaly appears to be caused by changes in the hot pressing parameters themselves, thus masking the true upstream source. To avoid misjudgments caused by random fluctuations in the attribution score G(k,b) around the threshold, the above-mentioned positioning process is only initiated after attribution switching occurs for several batches or more. Events where a single batch switch immediately reverts to upstream attribution are defined as attribution boundary fluctuations. In the mixing process, the main anomaly of low mixing temperature switched its attribution category to this process in several batches. Upon investigation, it was found that the molding pressure in the hot pressing process was systematically increased in these batches. The corresponding batches were identified as masking intervention batches and included in the masking intervention batch group. The registration information includes the batch number, hot pressing process number, molding pressure name, and its upward direction. Based on this, the intervention subject and object can be retrieved by batch within the same group.

[0057] For example, the step of extracting intervention offsets from upstream processes associated with the masking intervention batch group to form a root cause localization marker group includes: extracting the intervention offsets and intervention sequence of upstream processes for each batch in the masking intervention batch group; filtering low-amplitude offset process segments based on the intervention offsets and statistically analyzing the offset magnitudes by process; comparing the low-amplitude offset process segments with the intervention sequence by batch to generate suspected root cause items; and tracing the source of the suspected root cause items across batches of upstream processes to determine the root cause localization marker group.

[0058] For each batch in the masking intervention batch group, the intervention offset and intervention timing of the upstream process are extracted. Using the execution time period of each intervention batch in the masking intervention batch group as the reading range, the actual measured values ​​of all key parameters of the upstream process are extracted from the process parameter data at the original acquisition intervals, time by time. No pre-smoothing is performed here to preserve the complete information of fluctuations within the batch. δ(int,m)(t)=(x(m)(t)-μ(base,m)) / σ(base,m) constitutes the intervention offset timing of this parameter within the batch, where m is the upstream process parameter number, t is the acquisition time, x(m)(t) is the measured value of parameter m at time t, and μ(base,m) and σ(base,m) are the baseline mean and baseline standard deviation of this parameter in the unperturbed baseline state group, respectively. The numerator and denominator have the same dimensions, while δ(int,m)(t) has a dimension of one. In the intervention offset time series, the period during which δ(int,m)(t) maintains the same direction within a batch corresponds to the period of systematic deviation of this parameter within that batch; the longer the duration, the wider the time span of deviation from the baseline. A time series pattern where occasional, brief deviations within a batch are followed by a return to the baseline indicates random parameter fluctuation rather than systematic deviation. The moment when a systematic parameter adjustment first occurs in the downstream processes of each batch within the masked intervention batch group is defined as the intervention time series. This is marked by the earliest moment when the parameter setpoints of the downstream processes in the process parameter adjustment records show significant changes. If multiple downstream processes within the same batch experience systematic parameter adjustments, the moment corresponding to the earliest triggered adjustment is taken. The early and late distribution of intervention times across batches reflects the batch-to-batch differences in the intervention response speed of downstream processes. In batches with high recycled material blending ratios, the intervention time for increasing the molding pressure in the hot pressing process is significantly earlier than in batches with low blending ratios, indicating that downstream processes are more sensitive to upstream mixing anomalies under high blending ratio conditions.

[0059] Based on the intervention offset, low-amplitude offset process segments are screened by statistically analyzing the offset magnitude of each process step. The timing sequence of intervention offsets for each key parameter in each upstream process is merged across all batches in the masked intervention batch group. The absolute average of the merged cross-batch sequences is taken as the overall offset magnitude of the parameter. The lower the magnitude, the smaller the deviation of the parameter from the benchmark in all masked intervention batches. The absolute average is used here instead of the arithmetic mean to avoid the positive and negative deviations of the same parameter in different batches canceling each other out and underestimating its actual deviation. The median of the overall magnitude distribution of intervention offsets for all parameters in all upstream processes is used as the upper limit threshold for screening. Parameters below this threshold are defined as low-amplitude offset parameters. Low-amplitude offset parameters in the same process are grouped into one segment, and the entire process segment is the low-amplitude offset process segment. When the recycled material blending ratio is only two or three percentage points higher than the benchmark, the mixing temperature continues to decrease, forcing an increase in the molding pressure at the hot pressing end. This shows that low-amplitude offset parameters are always close to the benchmark in batches with strong downstream intervention, and their small but continuous deviation is itself an abnormal signal. Setting the upper limit threshold too low may misclassify boundary conditions of truly high-amplitude deviation process parameters as low-amplitude deviation candidates, while setting it too high may miss parameters with moderate amplitude but consistent direction. Therefore, it is necessary to combine the consistency of the deviation direction of each parameter to assist in the screening, so as to enhance the robustness of the low-amplitude deviation candidate parameter identification results. The overall deviation amplitude of the recycled material blending ratio parameter in the batching process is slightly lower than the median, while the overall deviation amplitude of the weighing parameters of each component in the same process is extremely low. Both of them enter the low-amplitude deviation process segment candidate. The former has high directional consistency and strong directionality, while the latter has random directionality and weak directionality. The difference in the level of directional consistency among candidate parameters is the key distinguishing criterion for the identification of suspected root causes.

[0060] The low-amplitude offset process segment and the intervention sequence were compared batch by batch to generate suspected root cause items. To verify the temporal sequence between upstream deviation and downstream intervention, the δ(int,m)(t) time series of each parameter in the low-amplitude offset process segment was compared batch by batch with the intervention sequence of the corresponding batch in the masked intervention batch group to check whether δ(int,m)(t) had a continuous small deviation in several collection intervals before the intervention time series. The period in which the low-amplitude offset parameter has continuously deviated before the intervention time series but the downstream process has not yet triggered intervention is defined as the early deviation period. Its existence indicates that the deviation of the parameter is earlier in time than the triggering of the downstream process intervention, which is consistent with the causal timing logic of upstream root cause first and downstream response later. The determination of the early deviation period requires that δ(int,m)(t) continuously maintains the deviation in the same direction before the intervention time series and the deviation amount at each collection time exceeds the noise level threshold. The pattern of falling back immediately after crossing zero at a single moment does not meet the determination condition, so as to exclude random fluctuations from being misjudged as early deviations. When the deviation direction within the initial deviation period is consistent with the majority of deviation directions of the parameter in the masking intervention batch, the low-amplitude deviation parameter is marked as a candidate for suspected root cause. The longer the initial deviation period, the clearer the causal sequence. The initial deviation period of the recycled material blending ratio parameter in the batching process appeared in multiple masking intervention batches, and the initial deviation direction was consistently positive. The intervention sequence of the downstream hot pressing process was triggered several sampling intervals after the start of the blending ratio deviation, and the timing sequence was stable. This blending ratio parameter was confirmed as a suspected root cause. The initial deviation period of the weighing parameters of each component in the same process in the low-amplitude deviation process segment appeared with extremely low frequency and did not meet the judgment criteria.

[0061] For suspected root causes, trace their origin across batches to upstream processes to determine root cause localization marker groups. Traceability proceeds in reverse chronological order along the production line, starting with the deviation time of each suspected parameter in the suspected root cause and checking the deviation of process parameters at earlier times upstream. If an upstream process parameter shows a persistent deviation earlier than the suspected parameter, the trace continues upstream until the most upstream process within the analysis window is reached or no earlier source of deviation is found. For each upstream process node, the number of material batches transferred between that node and downstream nodes must be verified to ensure that the time-sequence comparison is based on the actual flow sequence of the same material. During tracing, if multiple key parameters at an upstream process node show deviations earlier than the current suspected parameter, that process node is marked in the tracing chain as a multi-parameter collaborative first-order marker, which has higher root cause localization reliability compared to single-parameter first-order markers. Tracing suspected root causes prioritizes terminating at the upstream node of the multi-parameter collaborative process. If such a node is absent, tracing terminates at the upstream process parameter where the earliest continuous deviation occurred and whose deviation time predates the current suspected parameter. The process parameter corresponding to the termination node is the root source of the cross-batch tracing chain for the current batch interval. Tracing the recycled material blending ratio parameter revealed that its high starting time highly corresponded to the recycled material batch switching time in the feeding record of batching and feeding station 1. Tracing terminated at the feeding batch switching node, and the feeding batch switching was confirmed as the original source of the high recycled material blending ratio. The time difference of the preceding deviation, the deviation magnitude, and the direction of deviation for each process node in the chain are marked node by node in the root cause localization marker group. The root cause localization marker group thus completely preserves the tracing path from the feeding node to the hot pressing process intervention trigger node.

[0062] The consistency of process attribution for each marker item in the root cause localization marker group is verified to form the attribution comparison results. The directional consistency rate AR(m) = n(agr,m) / n(int), where m is the upstream process parameter number, n(agr,m) is the number of batches in the root cause localization marker group whose δ(int,m) deviation direction is the same as its dominant deviation direction, and n(int) is the total number of masking intervention batches in the masking intervention batch group. Both are batch counts. AR(m) has a dimension of one; a higher value indicates a more stable deviation direction of the marker item's process parameter in the masking intervention batches, and a more reliable association with downstream masking intervention behavior. Marker items with AR(m) higher than the attribution consistency judgment threshold are confirmed as attribution-consistent items, while those below the threshold are considered attribution-inconsistent items. Items near the judgment threshold are defined as attribution boundary items, whose attribution consistency evidence is insufficient. These are retained as notes in the attribution comparison results and will be re-verified as the number of batches in the masking intervention batch group expands with the rolling analysis window to avoid an underestimation of AR(m) due to insufficient batches within the window. Although the attribution inconsistencies in the root cause localization marker group showed prior deviations in the tracing chain, their deviation directions changed repeatedly across different masking intervention batches, lacking a stable correspondence with downstream intervention behaviors. Therefore, they were marked with low confidence in the attribution comparison results. Attribution consistent items, on the other hand, indicate a stable inter-batch correlation between the systematic deviations of the corresponding upstream process parameters in the masking intervention batches and downstream masking intervention behaviors. They possess high confidence as a root cause source and are marked with high confidence in the attribution comparison results. The recycled material blending ratio marker in the batching process showed an AR(m) far exceeding the judgment threshold across all masking intervention batches, confirming it as an attribution consistent item. Its stable inter-batch co-occurrence with downstream intervention behaviors provides direct evidence that the batching blending ratio can serve as direct evidence of the low mixing temperature as a root cause.

[0063] Production diagnostic conclusions are generated by integrating the attribution results of each process based on the attribution comparison results. Each attribution consistency item is summarized by upstream process number. If multiple attribution consistency items appear in the same upstream process, this process exhibits multi-parameter systematic deviations in the masked intervention batch, and is marked as a high-confidence root cause source in the production diagnostic conclusion. Deviations of only a single parameter are marked as medium-confidence root causes. When multiple upstream processes simultaneously obtain high-confidence root cause source markings, for example, if the recycled material blending ratio in the batching process and the mixing temperature in the mixing process both reach high-confidence levels in the same batch, they are recorded side-by-side in the production diagnostic conclusion as root cause co-occurrences. The order of the coordinated deviations of the two processes is checked according to the earlier deviation time difference in the root cause localization marker group. The earlier deviation is more likely to be the original root cause, while the later deviation is more likely to be an intermediate transmission node. The deviation with the larger magnitude is identified as the dominant root cause. The attribution comparison results of major anomalies in the cross-batch attribution reversal items also need to be combined with the compensation response index of the major anomalies: major anomalies with high I(c) and corresponding upstream process attribution consistency items with high confidence levels are marked as key root causes in the production diagnosis conclusion, indicating that the major anomaly not only triggered a strong downstream compensation masking effect, but its true root cause was also confirmed with high confidence through cross-batch attribution tracing. The attribution comparison results of the major anomaly of low mixing temperature within the analysis window confirmed that the high proportion of recycled material in the batching process was a high-confidence root cause, and its I(c) ranked first among all major anomalies. Based on this, the production diagnosis conclusion marked the high proportion of recycled material in the batching process as the key root cause of the current batch interval, and recorded the difference in the manifestation of this root cause in the masked intervention batch and the non-masked intervention batch. In the former, the deviation was offset by downstream compensation, while in the latter, the deviation was directly reflected in the pass rate of this process.

[0064] Step S15: Based on the production diagnosis conclusion and the cross-process recurrence rate, the process handling priority set is generated by correlation and integration. The process handling priority set and the production diagnosis conclusion are recurrence rate predicted and arranged to output continuous production control instructions.

[0065] Specifically, a process handling priority set is generated by integrating the production diagnostic conclusions with cross-process recurrence rates. The handling priority score PS(q) = W(q) × RA(q), where q is the process number, W(q) is the root cause confidence quantification value of the process in the production diagnostic conclusion, and its value directly corresponds to the three-level labeling of the production diagnostic conclusion: 1.2 for critical root causes, 1 for high confidence, 0.6 for medium confidence, and 0 for processes without root cause labeling. RA(q) is the arithmetic mean of the cross-process recurrence rates R(e) of all co-occurring events associated with the process. Both are dimensionless, while PS(q) has a dimension of one. A higher value indicates that the process has both a high confidence level of root cause qualitative analysis and a relatively high cross-process recurrence rate, and the higher the priority of process handling. For processes with low RA(q) but only root cause labeling, although the cause has been identified, it only occurs in a few batches of material changes. A one-time rectification, such as replacing aging heating elements, should be arranged, with the treatment cycle based on the equipment maintenance cycle. For processes with high RA(q) but no root cause labeling in the production diagnostic conclusion, W(q) is set to 0, making PS(q) zero. These processes are listed separately as an observation layer in the process treatment priority set and labeled with observation priority, indicating frequent cross-process diffusion but the anomaly cause has not yet been traced and identified. The remaining processes are sorted by PS(q) from high to low. Processes with PS(q) exceeding a certain multiple of the overall process mean by a certain standard deviation enter the main treatment layer of the process treatment priority set. Processes above the mean but not exceeding that multiple of the standard deviation enter the secondary treatment layer. The two layers of processes, along with their corresponding production diagnostic conclusion root cause parameter labels, are clearly distinguished in the process treatment priority set by hierarchical numbering. The batching process was identified as a critical root cause in the production diagnosis conclusion. The corresponding co-occurrence event RA(q) ranked among the top in the entire process, and PS(q) was the highest. It was then included in the process priority set master disposal layer.

[0066] The recurrence rate prediction and output of continuous production control instructions are based on the priority set of process handling and the production diagnosis conclusions. For each process in the priority set of process handling, the batches with concentrated high R(e) co-occurrence events are retrieved from the cross-process recurrence rate statistics and compared with the switching marker attached to the batch number. The root cause parameters are retrieved from the production diagnosis conclusions as the correction targets. If the batches with high abnormality are mostly concentrated at the recycled material blending ratio adjustment node, the next arriving batch is extrapolated according to the historical average recurrence interval of the marker, and then several collection intervals are pushed forward to predict the early intervention time. The continuous production control instruction consists of five elements: process number, trigger condition type, correction parameter name, correction direction, and correction magnitude. The magnitude is the product of the absolute value of the abnormal parameter δ(k) of the main abnormal item of the corresponding process and the control sensitivity coefficient K(s), where s is the sequence number of the root cause parameter to be corrected, and K(s) represents the downstream standardized deviation that can be offset by each unit adjustment of the parameter, with the same dimension as the parameter. The triggering condition is selected based on the batch-to-batch variance of the representative item I(c) in the compensation response index: those with smaller variances exhibit stronger regularity and are triggered by batch switching prediction, automatically executed several acquisition intervals before the switching marker arrives; those with larger variances are triggered by real-time deviation exceeding the threshold, and corrected when the real-time deviation exceeds δ(k) by a certain proportion. In the process priority set, after the batching process in the main processing layer is identified as a critical root cause in the production diagnosis conclusion, its continuous production control instruction has five elements in the following order: process number 01, batch switching prediction trigger, recycled material blending ratio, negative, 3 percentage points, meaning that the blending ratio setting value is smoothly reduced from 30% to 27% three acquisition intervals before the switching marker arrives; the hot pressing process instruction is process number 04, real-time deviation exceeding the threshold trigger, molding pressure, positive, 0.8 MPa; the root cause of the process in the observation layer has not yet been locked, and only an alarm instruction for real-time deviation exceeding the threshold is output without a correction range.

[0067] In summary, steps S11 to S15 illustrate the processing path using the switching of recycled material blending ratio as an example: Starting with a certain batch, the recycled material blending ratio increases, the pass rate of the mixing process declines, and the pass rate of the hot-pressing curing process rises simultaneously—S11 identifies this transmission and compensation phenomenon based on the linkage between the pass rates of the two processes, locking the mixing process as the process initiating the deviation; S12 uses the parameter statistical regularity of the undisturbed steady-state period of this process as a benchmark to restore the true deviation of the mixing temperature smoothed out by the compensation action at the hot-pressing end; S13 calculates the cross-process recurrence rate of this anomaly. Intermittent noise is filtered out, and the hot-pressing end compensation response intensity is used to confirm it as the main anomaly. S14 uses the fact that this anomaly is sometimes judged as the current process and sometimes as the upstream material reversal between batches to lock the masking intervention batch at the hot-pressing end, and traces back step by step along the previous deviation sequence to the blending ratio adjustment node of the batching station to generate a production diagnosis conclusion. S15 Based on this, the batching station is given the highest handling priority, and the blending ratio setting value is automatically lowered before the next material change arrives according to the blending ratio switching rule, realizing the transformation from post-event investigation to pre-event intervention based on priority.

[0068] To implement the continuous production method for rubber-based brake pads corresponding to the above method embodiments, in order to achieve the corresponding functions and technical effects. See also Figure 4 , Figure 4 This diagram illustrates a structural block diagram of a continuous production system 400 for rubber-based brake pads according to an embodiment of this application. For ease of explanation, only the parts relevant to this embodiment are shown. The continuous production system 400 for rubber-based brake pads provided in this embodiment includes: The attribution mapping module 401 is used to collect process parameter data and quality inspection records of each production process, and to perform synchronous fluctuation attribution analysis of the pass rate of adjacent processes on the process parameter data and the quality inspection records to generate deviation response characteristics. The benchmark comparison module 402 is used to perform anomaly co-occurrence correlation analysis based on the deviation response characteristics to generate candidate anomaly co-occurrence groups, and perform unperturbed benchmark comparison on the candidate anomaly co-occurrence groups to generate hypothesis deviation. The convergence correction module 403 is used to statistically analyze the cross-process recurrence rate through the candidate anomaly co-occurrence group, perform recurrence rate drop node correction on the candidate anomaly co-occurrence group based on the hypothesis deviation and the cross-process recurrence rate to generate a correction anomaly group, and perform merging and screening on the correction anomaly group to determine the main anomalies. The attribution diagnosis module 404 is used to extract the attribution discrimination features of the main anomalies, perform upstream and downstream process attribution discrimination based on the attribution discrimination features to determine the attribution category of the anomalies, perform cross-batch attribution traceability comparison based on the attribution category of the anomalies to determine cross-batch attribution reversal items, and perform upstream traceability verification based on the cross-batch attribution reversal items to generate production diagnosis conclusions. The instruction output module 405 is used to generate a process handling priority set by associating and integrating the production diagnosis conclusion and the cross-process recurrence rate, and to output continuous production control instructions by predicting the recurrence rate of the process handling priority set and the production diagnosis conclusion.

[0069] The continuous production system 400 for rubber-based brake pads described above can implement the continuous production method for rubber-based brake pads described in the above method embodiments. The options in the above method embodiments are also applicable to this embodiment, and will not be detailed here. The remaining contents of this application embodiment can be referred to the contents of the above method embodiments, and will not be repeated in this embodiment.

[0070] The purpose of the above embodiments is to reproduce and derive the technical solution of the present invention by way of example, and to fully describe the technical solution, purpose and effect of the present invention. The purpose is to enable the public to have a more thorough and comprehensive understanding of the disclosure of the present invention, and not to limit the scope of protection of the present invention.

Claims

1. A continuous production method for rubber-based brake pads, characterized in that, include: Collect process parameter data and quality inspection records for each production process, and perform attribution analysis on the synchronous fluctuation of the pass rate of adjacent processes to generate deviation response characteristics. Based on the aforementioned deviation response characteristics, anomaly co-occurrence correlation analysis is performed to generate candidate anomaly co-occurrence groups, and the candidate anomaly co-occurrence groups are compared with an unperturbed benchmark to generate hypothesis deviation. The cross-process recurrence rate is statistically analyzed by the candidate anomaly co-occurrence group. Based on the hypothesis deviation and the cross-process recurrence rate, the recurrence rate drop node correction is performed on the candidate anomaly co-occurrence group to generate a correction anomaly group. The correction anomaly group is then merged and screened to determine the main anomalies. Extract the attribution discrimination features of the main anomalies, perform upstream and downstream process attribution discrimination based on the attribution discrimination features to determine the attribution category of the anomalies, perform cross-batch attribution traceability comparison based on the attribution category of the anomalies to determine cross-batch attribution reversal items, and perform upstream traceability verification based on the cross-batch attribution reversal items to generate production diagnostic conclusions; Based on the production diagnostic conclusions and the cross-process recurrence rates, a process handling priority set is generated by correlation and integration. The process handling priority set and the production diagnostic conclusions are then used to predict and arrange the recurrence rates and output continuous production control instructions.

2. The method according to claim 1, characterized in that, The step of performing synchronous fluctuation attribution analysis on the pass rate of adjacent processes using the process parameter data and the quality inspection records to generate deviation response characteristics includes: Construct a set of process pass rate curves by extracting historical pass rate data for each process from the process parameter data and the quality inspection records; The fluctuation of adjacent processes in the process pass rate curve set is checked to form a distribution of process correlation. Based on the distribution of correlation degree of the processes, adjacent process pairs at the moment of sudden increase in correlation degree are determined to form a process pair group for impact response; Based on the aforementioned impact response process, the group analyzes the impact source direction and amplitude to generate deviation response characteristics.

3. The method according to claim 1, characterized in that, The step of generating candidate anomaly co-occurrence groups based on the deviation response characteristics includes: The deviation response characteristics are grouped according to the direction of process deviation to form a positive deviation group and a negative deviation group; The co-occurrence frequency of cross-directional processes in the positive deviation group and the negative deviation group is statistically analyzed to form a co-occurrence distribution table with opposite directions; The process of screening co-occurrence frequencies exceeding a threshold from the co-occurrence distribution table with opposite directions forms a masking candidate group for verifying compensation masking relationships; Based on the screening of the masked candidate group, candidate anomaly co-occurrence groups are generated by prior deviation initiation items.

4. The method according to claim 1, characterized in that, The step of generating hypothesis deviation by performing a perturbation-free benchmark comparison on the candidate anomaly co-occurrence group includes: After removing the co-occurrence batches of the candidate abnormal co-occurrence group, the process parameters of each process during the steady-state period are extracted to construct a undisturbed baseline state group; The state difference record is formed by comparing the unperturbed baseline state group with the candidate anomaly co-occurrence group one by one. A continuous deviation distribution is obtained by performing batch switching node deviation correlation evaluation on the state difference records; Based on the continuous deviation distribution, a segmented deviation verification is performed to generate the hypothetical deviation amount.

5. The method according to claim 1, characterized in that, The step of generating a corrected anomaly group by correcting the recurrence rate drop node of the candidate anomaly co-occurrence group based on the assumed deviation and the cross-process recurrence rate includes: The candidate anomaly co-occurrence groups are classified into classification anomaly subgroups according to anomaly type; Based on the cross-process recurrence rate, the recurrence rate levels of each of the classified anomaly subgroups are checked to form a recurrence rate level distribution; Based on the horizontal distribution of the recurrence rate and the assumed deviation, a correction screening threshold is formed by determining the recurrence rate drop node. Based on the correction screening threshold, the candidate anomaly co-occurrence group is screened to generate a correction anomaly group.

6. The method according to claim 1, characterized in that, The step of performing a merge screening on the corrected anomaly group to determine the main anomalies includes: Perform a common cause correlation check on each abnormal item in the corrected anomaly group to obtain the common cause correlation group; For each common cause association group, the anomaly with the highest cross-process recurrence rate within the group is extracted as the representative item; For each of the representative items, downstream process compensation response intensity assessment is performed to generate compensation response indicators; Based on the aforementioned compensation response indicators, representative items that exceed the threshold in the compensation response are selected to determine the main anomalies.

7. The method according to claim 1, characterized in that, The process of generating production diagnostic conclusions by performing upstream traceability verification based on the cross-batch attribution reversal item includes: Intervene batch attribution positioning for the cross-batch attribution reversal items to form masked intervention batch groups; Intervention offsets are extracted from upstream processes associated with the masking intervention batch group to form a root cause localization marker group; The consistency of process attribution for each marker item in the root cause localization marker group is checked to generate attribution comparison results; Based on the attribution comparison results, the attribution results of each process are integrated to generate a production diagnostic conclusion.

8. The method according to claim 4, characterized in that, The step of comparing the unperturbed baseline state group with the candidate anomaly co-occurrence group state by state to form a state difference record includes: Extract the current process parameter values ​​for each process step for the candidate anomaly co-occurrence group; The current process parameter values ​​are compared with the undisturbed reference state group item by item to form a difference record; The difference records are used to identify parameters where the difference is consistently zero, and a compensation masking marker group is generated. Based on the compensation masking marker group, the difference records are correlated with trend type verification to generate state difference records.

9. The method according to claim 7, characterized in that, The step of extracting intervention offsets from the upstream processes associated with the masking intervention batch group to form a root cause localization marker group includes: For each batch in the masking intervention batch group, extract the intervention offset and intervention sequence of the upstream process; Based on the aforementioned intervention offset amount, low-amplitude offset process segments are selected by statistically analyzing the offset magnitude of each process. The low-amplitude offset process segment and the intervention sequence are compared and verified batch by batch to generate suspected root cause items; The suspected root cause items are traced back to their source in upstream processes across batches to determine the root cause localization marker group.

10. A continuous production system for rubber-based brake pads, characterized in that, include: The attribution mapping module is used to collect process parameter data and quality inspection records of each production process, and to perform synchronous fluctuation attribution analysis of the pass rate of adjacent processes to generate deviation response characteristics. The benchmark module is used to perform anomaly co-occurrence correlation analysis based on the deviation response characteristics to generate candidate anomaly co-occurrence groups, and to perform unperturbed benchmark comparison on the candidate anomaly co-occurrence groups to generate hypothesis deviation. The convergence correction module is used to statistically analyze the cross-process recurrence rate through the candidate anomaly co-occurrence group, perform recurrence rate drop node correction on the candidate anomaly co-occurrence group based on the hypothesis deviation and the cross-process recurrence rate to generate a correction anomaly group, and perform merging and screening on the correction anomaly group to determine the main anomalies. The attribution diagnosis module is used to extract the attribution discrimination features of the main anomalies, perform upstream and downstream process attribution discrimination based on the attribution discrimination features to determine the attribution category of the anomalies, perform cross-batch attribution traceability comparison based on the attribution category of the anomalies to determine cross-batch attribution reversal items, and perform upstream traceability verification based on the cross-batch attribution reversal items to generate production diagnosis conclusions. The instruction output module is used to generate a process handling priority set by associating and integrating the production diagnosis conclusion with the cross-process recurrence rate, and to output continuous production control instructions by predicting the recurrence rate of the process handling priority set and the production diagnosis conclusion.

Images