Aircraft part production quality optimization method based on data analysis

Abstract

The invention relates to the technical field of aircraft manufacturing, and discloses an aircraft part production quality optimization method based on data analysis. The method comprises the following steps: collecting multi-process real-time processing parameters and quality inspection data of a production line, and generating a dynamic quality weight matrix according to a process parameter coupling degree; extracting process path difference characteristics of qualified products and unqualified products in historical batches, and encoding the process path difference characteristics as a quality evolution chain matched with adjacent process parameter mutation relevance; optimizing process parameters at a quality evaluation node, driving a parameter combination to iterate to minimize quality fluctuation, generating an adjustment amount, updating an evolution chain constraint coefficient, and synchronously constructing a stability evaluation function of a correlation weight matrix and a defect propagation path; triggering process compensation according to an adjustment amount gradient, and verifying the cohesion through a tolerance rule by taking a reference parameter matched with the target evolution chain as compensation data; and generating a feedback matrix by using the quality fluctuation index and the compensation result, and correcting the mapping relation between the weight matrix and the evolution chain.

Description

Technical Field

[0001] This invention relates to the field of aircraft manufacturing technology, specifically to a method for optimizing the production quality of aircraft parts based on data analysis. Background Technology

[0002] As a core component of aviation equipment, the production quality of aircraft parts is directly related to flight safety and the overall performance of the equipment. The production process of parts involves many complex procedures, and there are close parameter correlations between these procedures, which makes quality control face many challenges.

[0003] Currently, in aircraft component production lines, most quality control methods still rely on parameter monitoring and post-production inspection of single processes, failing to fully consider the coupling effect of parameters between processes. For example, a slight fluctuation in the processing parameters of a certain process may, through the cumulative effect of multiple subsequent processes, eventually lead to dimensional deviations or surface defects in the components. This isolated quality control model not only makes it difficult to identify potential quality risks in advance, but also, when non-conforming products are discovered, the inability to trace the propagation path of quality problems may result in a large amount of rework or scrap, increasing production costs and production cycles.

[0004] Existing technologies for process parameter optimization often employ fixed parameter adjustment strategies, lacking in-depth mining and utilization of historical production data. Although data on qualified and unqualified products is collected in some production stages, the differences in process paths are not effectively extracted, making it impossible to construct a model that reflects the patterns of quality changes. This results in parameter optimization schemes failing to adapt to dynamic changes in production conditions when facing different batches of production tasks, often leading to excessive quality fluctuations. Furthermore, when performing process compensation, existing methods often neglect the connection between compensation parameters and adjacent processes, easily causing new quality problems and further reducing the stability of the production process.

[0005] Current quality assessment systems lack scientific stability evaluation functions, making it impossible to accurately quantify the correlation between quality fluctuation indices and process parameter adjustments. This makes it difficult for production managers to formulate reasonable quality control strategies based on objective data, forcing them to rely on experience for judgment, resulting in insufficient accuracy and reliability of quality control. As the aerospace industry's requirements for component precision continue to increase, traditional quality control methods can no longer meet the needs of modern production. A quality control method that comprehensively considers process coupling, historical data value, and dynamic parameter optimization is needed to improve the stability and consistency of aircraft component production quality. Summary of the Invention

[0006] The purpose of this invention is to provide a data analysis-based method for optimizing the production quality of aircraft parts, in order to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides a data analysis-based method for optimizing the production quality of aircraft components, the method comprising: Real-time processing parameters and quality inspection data of multiple processes on the production line are collected. The quality inspection data includes dimensional deviations, surface defects and material performance indicators. A dynamic quality weight matrix is ​​generated based on the parameter coupling degree between processes. Extract the process path difference features between qualified and unqualified products in historical batches, encode the difference features into a quality evolution chain, and generate the quality evolution chain by matching the parameter mutation correlation of adjacent processes; At the quality assessment node, the process parameters are optimized collaboratively, driving the parameter combination to iterate in the direction of minimizing quality fluctuations. Each iteration generates parameter adjustment amount and updates the process constraint coefficients of the quality evolution chain. A stability assessment function is constructed based on the correlation between the dynamic quality weight matrix and the defect propagation path. The gradient distribution of the parameter adjustment amount triggers process compensation for key processes. The compensation data is a set of benchmark parameters in the current production batch that match the target quality evolution chain. The process connection of the compensation parameters is verified by the tolerance rules in the quality evolution chain. The quality fluctuation index output by the stability evaluation function and the process compensation verification results are used to generate a feedback matrix, which dynamically corrects the mapping relationship between the dynamic quality weight matrix and the quality evolution chain.

[0008] Preferably, the step of performing collaborative optimization of process parameters at the quality assessment node, driving the parameter combination to iterate towards minimizing quality fluctuations, includes: The process parameters are encoded into a multi-dimensional process vector, with each dimension corresponding to the adjustable range of the process constraint coefficient. The process interference degree is calculated based on the quality fluctuation range covered by the current parameter combination. Based on the process interference degree, a fluctuation neighborhood table is constructed, and the differences in process paths and process constraint coefficients that have quality interference with the current parameter combination are recorded to generate candidate adjustment amounts; A two-stage screening mechanism is used to select the target adjustment amount from the candidate adjustment amounts. The first stage screening is based on the effectiveness of the candidate adjustment amounts in reducing fluctuations in historical iterations, and the second stage screening is based on the compatibility threshold between the target adjustment amount and the tolerance rules in the quality evolution chain. The target adjustment amount is applied to the current process parameters to dynamically adjust the coefficient values ​​of the corresponding dimensions in the multidimensional process vector, while updating the process interference threshold of the fluctuation neighborhood table to trigger the elimination of low-stability paths. The quality matching degree between the updated parameters and the baseline parameter set is recalculated, and the change in the quality matching degree is fed back to the coefficient allocation of the dynamic quality weight matrix.

[0009] Preferably, the step of selecting the target adjustment amount from the candidate adjustment amounts using a two-stage screening mechanism includes: Based on the effectiveness of reducing fluctuations in the candidate adjustment quantities described in historical iterations, the ratio of the number of successful adjustments to the number of quality weight improvements for each candidate adjustment quantity within the preset production cycle is statistically analyzed and normalized to become the benchmark weight for the first stage of screening. Extract the process boundary values ​​bound by the tolerance rules in the quality evolution chain, calculate the coverage ratio of the dimension corresponding to the candidate adjustment amount and the allowable interval of the boundary value, and map it as the boundary constraint weight for the second stage of screening; The benchmark weights and boundary constraint weights are dynamically weighted and fused. The weighting coefficients are adjusted according to the density index of low-stability paths in the fluctuation neighborhood table, so that the boundary constraints dominate the weights in the areas of dense interference. Based on the fusion weight, hierarchical screening is performed to verify the non-conflict between the candidate adjustment amount and the core process parameters of the quality evolution chain. If there is a conflict, the process switches to the suboptimal candidate adjustment amount until the boundary constraints are met.

[0010] Preferably, the step of dynamically weighting and fusing the benchmark weights and boundary constraint weights includes: Based on the comparison between the proportion of low-stability paths in the fluctuation neighborhood table and the historical maximum process capacity, the interference density index is generated through piecewise linear transformation. Dynamic fusion coefficients are generated based on the interference density index, and a sliding window mechanism is used to attenuate historical fusion records to suppress abrupt changes in the dynamic fusion coefficients. The benchmark weights are coupled with the dynamic fusion coefficients, and the coupling result is superimposed with the boundary constraint weights. The superposition result is then truncated and normalized. The intensity of attenuation compensation is adjusted based on the deviation between the normalization result and the historical moving average. If the deviation exceeds the tolerance threshold, the compensation intensity is increased and applied to the fusion calculation in the next iteration cycle.

[0011] Preferably, the step of constructing a stability evaluation function based on the correlation between the dynamic quality weight matrix and the defect propagation path includes: Based on the process distribution of the dynamic quality weight matrix, the spatial coupling degree between defect propagation path and parameter sensitivity is calculated; A dynamic correction factor is generated based on the spatial coupling degree, which correlates the parameter fluctuations of high-defect regions with low-sensitivity processes, and the scope of the dynamic correction factor is limited based on the process constraint coefficient of the quality evolution chain. A stability evaluation function is constructed by integrating the dynamic quality weight matrix, dynamic correction factor, and spatial coupling degree, and a penalty term related to the difference in process constraint coefficients is introduced.

[0012] Preferably, generating the dynamic correction factor based on the spatial coupling degree includes: Based on the aforementioned spatial coupling degree and the inverse correlation between defect propagation rate and parameter adjustment gradient, the fluctuation difference index between high-defect regions and low-sensitivity processes is calculated. An initial correction factor is generated based on the fluctuation difference index. The process stability of the high-defect region is correlated with the parameter coverage sparsity of the low-sensitivity process. The rate of change of the initial correction factor is constrained by the process constraint coefficient of the quality evolution chain. Based on the historical correction records of the process constraint coefficients, dynamic attenuation compensation is applied to the initial correction factor to suppress asymmetric fluctuation overload.

[0013] Preferably, the step of generating a feedback matrix by combining the quality fluctuation index output by the stability evaluation function with the process compensation verification results includes: Based on the tolerance compliance rate in the quality fluctuation index and the process compensation verification results, the coupling coefficient between fluctuation and tolerance is calculated. An initial feedback matrix is ​​generated based on the coupling coefficient, the tolerance loss in the high fluctuation region is associated with the parameter adjustment amount, and a dynamic decay constraint is applied to the initial feedback matrix based on the historical correction record of the quality evolution chain. A multidimensional feedback model is constructed by integrating the coupling coefficient, quality fluctuation index, and tolerance compliance rate, and a penalty term related to the difference in the dynamic quality weight matrix is ​​introduced to generate the feedback matrix.

[0014] Preferably, the extraction of process path differences between qualified and unqualified products in historical batches includes: Time-frequency joint analysis was performed on the processing parameter sequences of each process in historical batches to extract the steady-state characteristics of qualified product sequences and the abrupt change characteristics of unqualified product sequences; Calculate the transfer gain of steady-state and abrupt change characteristics between adjacent processes, and mark process pairs with gains exceeding the threshold as critical coupling nodes; Based on the spatial distribution of the key coupling nodes, process path difference features are generated, and these difference features are encoded into a directed quality evolution chain according to the process transmission direction.

[0015] Preferably, the time-frequency joint analysis of the processing parameter sequences of each process in the historical batch includes: The processing parameter sequence is divided into overlapping segments according to the production time window, and time-domain smoothing and frequency-domain power spectrum analysis are performed simultaneously on each segment. Extract the extreme points of the time-domain smooth curve and the peak interval of the frequency-domain power spectrum, and mark the spatiotemporal overlap region of the two as the parameter-sensitive interval; Steady-state features and mutation features are generated based on the distribution differences of the sensitive intervals in the sequences of qualified and unqualified products.

[0016] Preferably, generating the dynamic quality weight matrix based on the parameter coupling degree between processes includes: Calculate the correlation coefficient matrix of the processing parameters of each process, and mark the process pairs with correlation coefficients exceeding the threshold as strongly coupled groups; An initial weight matrix is ​​generated based on the spatial density distribution of the strongly coupled group, and the weight coefficients of the high-density region are correlated with the decay rate of the low-density region. The initial weight matrix is ​​dynamically corrected based on the defect distribution in real-time quality inspection data, so that the weight coefficients are negatively correlated with the defect clustering degree.

[0017] Compared with the prior art, the beneficial effects of the present invention are: By collecting real-time processing parameters and quality inspection data from multiple processes on the production line, and generating a dynamic quality weight matrix based on the parameter coupling degree between processes, the limitations of traditional single-process quality control are broken. This allows for a comprehensive reflection of the impact of each process parameter on the overall quality. This dynamic weight setting based on parameter coupling degree makes the focus of quality control clearer, enabling precise control of key process parameters that affect quality. This avoids overlooking quality risks due to neglecting inter-process relationships, effectively improving the comprehensiveness and relevance of quality control.

[0018] By extracting the process path differences between qualified and unqualified products in historical batches and encoding them into a quality evolution chain, the value of historical production data was fully utilized. The quality evolution chain, generated by matching the correlation of parameter mutations in adjacent processes, clearly presents the pattern of quality problems from their inception to their propagation, providing a strong basis for subsequent process parameter optimization. With the help of the quality evolution chain, production personnel can intuitively understand the impact of different process paths on quality. In subsequent production processes, they can proactively avoid process paths that easily lead to unqualified products, while also learning from the process parameter combinations of qualified products to reduce quality problems caused by inappropriate process path selection, further improving the pass rate of parts production.

[0019] At the quality assessment node, collaborative optimization of process parameters is performed, driving the parameter combination iteratively towards minimizing quality fluctuations. Each iteration generates parameter adjustments and updates the process constraint coefficients of the quality evolution chain. Simultaneously, a stability assessment function is constructed based on the correlation between the dynamic quality weight matrix and the defect propagation path, achieving dynamic optimization of process parameters and scientific quality assessment. Compared to traditional fixed parameter adjustment strategies, this iterative parameter optimization method continuously adjusts parameter combinations according to actual quality changes during production, keeping parameters within a range conducive to stable quality and effectively reducing quality fluctuations. The construction of the stability assessment function provides a quantitative indicator of quality fluctuations, objectively reflecting the impact of process parameter adjustments on quality stability. This helps production personnel promptly grasp quality change trends and provides a reliable reference for subsequent quality control decisions.

[0020] When triggering process compensation for critical processes based on the gradient distribution of parameter adjustments, the benchmark parameter set matching the target quality evolution chain in the current production batch is used as the compensation data. The compensation parameters are then verified for process continuity through tolerance rules within the quality evolution chain, ensuring the accuracy and rationality of the process compensation. This benchmark parameter set-based compensation method avoids parameter disorder caused by blind compensation, while process continuity verification effectively prevents new quality problems arising from mismatches between compensation parameters and adjacent process parameters, further improving the stability of the production process and reducing rework and waste caused by improper process compensation.

[0021] A feedback matrix is ​​generated by combining the quality fluctuation index output by the stability assessment function with the process compensation verification results. This dynamically corrects the mapping relationship between the dynamic quality weight matrix and the quality evolution chain, forming a closed-loop quality optimization system. This dynamic correction mechanism enables the quality control model to continuously adapt to changes in production conditions. As production data accumulates, the accuracy of the dynamic quality weight matrix and the quality evolution chain continuously improves, leading to more precise process parameter optimization and quality assessment. This creates a virtuous cycle, ensuring the long-term stability and consistency of aircraft component production quality, reducing production costs, shortening production cycles, and better meeting the high quality requirements of the aviation industry. Attached Figure Description

[0022] Figure 1 This is a schematic diagram illustrating the working principle of the data analysis-based aircraft component production quality optimization method described in this invention. Figure 2 A flowchart for the collaborative optimization and iterative adjustment of process parameters; Figure 3 A flowchart for constructing a stability evaluation function. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Please see Figure 1 This invention provides a data analysis-based method for optimizing the production quality of aircraft components. The method includes: collecting real-time processing parameters and quality inspection data from multiple processes on the production line, including dimensional deviations, surface defects, and material performance indicators. Based on the parameter coupling degree between processes, a dynamic quality weight matrix is ​​generated, reflecting the weight distribution of the impact of different process parameters on overall quality. Simultaneously, the process path difference characteristics of qualified and unqualified products are extracted from historical batches, and these differences are encoded into a quality evolution chain. This chain is generated by matching the parameter mutation correlations of adjacent processes, thereby capturing the propagation patterns of defects between processes. At the quality assessment node, collaborative optimization of process parameters is performed, driving the parameter combination to iterate towards minimizing quality fluctuations. Each iteration generates parameter adjustment amounts and updates the process constraint coefficients of the quality evolution chain. Based on the correlation between the dynamic quality weight matrix and the defect propagation path, a stability assessment function is constructed to quantify the risk of quality fluctuations. Based on the gradient distribution of parameter adjustment amounts, process compensation for key operations is triggered. The compensation data is a set of benchmark parameters in the current production batch that match the target quality evolution chain. Through tolerance rules in the quality evolution chain, the compensation parameters are verified for process integration to ensure the feasibility of parameter adjustments. A feedback matrix is ​​generated by combining the quality fluctuation index output by the stability assessment function with the process compensation verification results. This dynamically corrects the mapping relationship between the dynamic quality weight matrix and the quality evolution chain, achieving continuous system optimization.

[0025] Example 1: See Figure 2The collaborative optimization of process parameters begins by encoding the processing parameters of each process step on the production line into a multi-dimensional process vector. Each dimension corresponds to an adjustable range of process constraint coefficients, defining the upper and lower limits of parameter adjustments. Based on the parameter combination used in the current production batch, the system calculates the quality fluctuation range covered by this combination and further derives the process interference degree. The process interference degree quantifies the degree of cascading impact on other related processes when adjusting a certain process parameter. The calculation process references fluctuation patterns caused by similar parameter combinations in historical production data, thereby more accurately predicting the potential risks of the current settings. Based on the calculated process interference degree, the system dynamically constructs and maintains a fluctuation neighborhood table. The core function of this table is to record all process paths that have significant quality interference with the current parameter combination, while also detailing the differences in process constraint coefficients involved in these paths. The fluctuation neighborhood table is not static data; it is continuously updated as the production process progresses, constantly incorporating newly discovered interference relationships. Based on the information recorded in this table, the system can generate a series of candidate parameter adjustment amounts, each representing a possible direction for quality fluctuation optimization. When generating candidate adjustment values, the adjustable range of parameters and the expected impact on adjacent processes are taken into account.

[0026] A two-stage screening mechanism is employed to select the final target adjustment from a large pool of candidate adjustments. This mechanism is designed to balance the effectiveness and safety of the adjustment. The first stage of screening primarily examines the effectiveness of candidate adjustments in reducing quality fluctuations throughout historical iterations. The system counts the number of times each candidate adjustment successfully reduces fluctuations within a preset production cycle and compares this number with the number of times it causes a positive update to the quality weight matrix, calculating a ratio. This ratio, after normalization, becomes the baseline weight for the first stage of screening. Candidate adjustments with higher baseline weights indicate more reliable historical performance. The second stage of screening focuses on evaluating the compatibility of candidate adjustments with the tolerance rules in the existing quality evolution chain. The system extracts the process parameter boundary values ​​bound to all tolerance rules from the quality evolution chain and calculates the coverage ratio of each candidate adjustment's adjustment value within the allowed boundary value range for the parameters it affects. This coverage ratio is mapped to the boundary constraint weight for the second stage of screening. Adjustments with higher boundary constraint weights indicate a lower risk of violating process constraints.

[0027] The baseline weights and boundary constraint weights calculated in the two stages are comprehensively considered, and a dynamic weighted fusion strategy is used to determine the overall priority of each candidate adjustment. This weighted fusion is not fixed; its weighting coefficients are dynamically adjusted based on the density index of low-stability paths reflected in the current fluctuation neighborhood table. When the system identifies a large number of low-stability paths—i.e., regions with tight inter-process coupling that are prone to triggering chain fluctuations—the fusion strategy assigns higher priority to boundary constraints to ensure operational safety under complex conditions. Based on the overall priority after fusion, the system performs hierarchical screening, sequentially verifying whether there is a conflict between the highest-priority candidate adjustment and the core process parameters defined in the quality evolution chain. If a conflict is found, the system automatically switches to verifying the next-highest-priority candidate adjustment until a target adjustment that meets both the validity requirements and does not conflict with any core parameters is found.

[0028] Once the target adjustment amount is determined, it is formally applied to the current production process parameters. This is manifested in the dynamic adjustment of the coefficient values ​​of the corresponding dimensions in the multi-dimensional process vector. Simultaneously, the fluctuation neighborhood table is updated, particularly the process interference threshold used to determine path stability, which is recalibrated based on the adjustment results. This calibration process may trigger the elimination of some low-stability paths recorded in the table, meaning these paths with extremely high fluctuation risks are no longer included in subsequent considerations. After the parameter adjustment is implemented, the system recalculates the quality matching degree between the adjusted parameter combination and the preset benchmark parameter set. This matching degree reflects the closeness of the current parameter settings to the ideal state. Finally, the change in quality matching degree caused by this parameter update is used as an important feedback signal and fed back to the coefficient allocation logic of the dynamic quality weight matrix, thereby affecting the weight distribution of the matrix in the next optimization cycle, realizing a closed-loop optimization process from parameter adjustment to effect evaluation to model update.

[0029] At the engineering implementation level, encoding process parameters into multi-dimensional process vectors requires establishing a structured data model. This model must accurately map each process parameter to a specific dimension of the vector and associate it with its adjustable range metadata. Calculating process interference may involve complex multivariate analysis algorithms, requiring the processing of large amounts of real-time sensor data and historical database records. Constructing the fluctuation neighborhood table is similar to a real-time updated topology graph, recording the mutual influence relationships between process nodes. The generation of candidate adjustment quantities can be achieved by searching within a specified solution space using optimization algorithms. The implementation of the two-stage screening mechanism relies on an efficient data query and weight calculation engine. The normalization of baseline weights typically employs standard scaling algorithms to adapt to a uniform evaluation scale, while the calculation of boundary constraint weights requires an inference module capable of parsing and executing tolerance rule logic. Applying target adjustment quantities involves direct intervention in the production line control system. Recalculating the quality matching degree requires defining and calculating an effective similarity metric. The implementation of the entire feedback loop demands a highly reliable data bus and low-latency processing capabilities.

[0030] Example 2: Dynamic weighted fusion is the core decision-making step in the entire parameter selection process. Its goal is to combine the benchmark weight, reflecting the historical effectiveness of the adjustment, with the boundary constraint weight, representing its compliance with process specifications, into a comprehensive evaluation value. This fusion is not a simple fixed-ratio mixing, but a dynamic process adjusted by the real-time status of the production system. The success of the fusion strategy directly affects whether the selected target adjustment can maintain the stability of the process system while improving quality. The calculation of the benchmark weight begins with a backtesting analysis of historical data. The system tracks the historical application history of each candidate adjustment within a configurable statistical time window. For each record, the system determines whether the application successfully reduced quality fluctuations in the production process and records whether the application was accompanied by a positive optimization of the system's quality weight matrix. Records that successfully reduced fluctuations and triggered an increase in the weight matrix are marked as valid events. The benchmark weight is calculated based on the proportion of valid events in all applications of the candidate adjustment. This proportion is normalized and transformed into a benchmark weight value between 0 and 1, whose value directly reflects the reliability of the adjustment in past practice. The calculation process needs to take into account the time decay effect of the data. The reliability of recent application records is usually higher than that of distant records. Therefore, the system may assign different time decay coefficients to records from different periods.

[0031] The calculation of boundary constraint weights relies on the analysis of tolerance rules in the quality evolution chain. The quality evolution chain, as a knowledge base containing inter-process constraint relationships, explicitly defines the permissible operational boundaries of key parameters for each process. The system needs to extract all tolerance rules directly related to the parameter dimensions affected by the candidate adjustment amount from the evolution chain and analyze the parameter safety intervals defined by these rules. The core of boundary constraint weights is calculating the coverage ratio between the suggested value of the candidate adjustment amount and its corresponding parameter safety interval. If the adjustment amount falls entirely within the safety interval, the coverage ratio is 1; if it partially falls outside the interval, it is reduced proportionally to the degree of deviation. This coverage ratio, after being processed by a mapping function, forms the boundary constraint weight, which quantifies the technical compliance risk brought about by executing the adjustment amount. The key to dynamic weighted fusion lies in determining the relative importance of the benchmark weight and boundary constraint weight in the final decision, i.e., their respective weighting coefficients. This decision is not pre-set but dynamically driven by an index called the "interference density index," calculated in real-time from the fluctuation neighborhood table. The fluctuation neighborhood table continuously monitors the stability of different process paths in the production process, counting the proportion of paths marked as "low stability" out of the total number of paths. The system maintains a reference value of "historical maximum process capacity," representing the maximum complexity the system has been able to stably handle in previous operations. The interference density exponent is calculated using a piecewise linear transformation function that compares the current proportion of low-stability paths with the historical maximum capacity. When the current proportion approaches or exceeds the historical capacity, the transformation function outputs a higher density exponent value.

[0032] This interference density index is directly used to generate dynamic fusion coefficients, which in turn determine the proportions of baseline weights and boundary constraint weights in the fusion formula. A high interference density index indicates that the production system is in a sensitive state where inter-process coupling is tight and prone to cascading fluctuations. In this case, decision-making tends to be conservative, and the dynamic fusion coefficients are adjusted to ensure that the boundary constraint weights, representing safety, dominate the overall evaluation, thereby avoiding risky operations that could lead to system instability. To ensure the stability of decisions and prevent drastic changes in weighted coefficients due to minor fluctuations in production status, the system introduces a sliding window mechanism to smooth the dynamic fusion coefficients. This mechanism maintains a time window containing recent historical fusion coefficient records and applies attenuation compensation to the data within this window, giving higher weights to recent data and lower weights to older data. By calculating a weighted average, it suppresses step-like changes in the coefficients, making the changes in the fusion coefficients more gradual.

[0033] The specific fusion operation involves multiplying and coupling the baseline weights with the calculated dynamic fusion coefficients, and then adding the coupling result to the boundary constraint weights. The resulting raw comprehensive score needs to be truncated and normalized to standardize it onto a unified evaluation scale, facilitating comparisons between different candidate adjustment values. Truncating prevents extreme values ​​from having an excessive impact on the overall evaluation, while normalization eliminates dimensional differences. The system continuously monitors the normalized result obtained from this fusion calculation and compares it with a moving average calculated based on historical data, calculating the deviation between the two. If the deviation exceeds the system's preset tolerance threshold, it indicates a potential significant change in the current production state. In the next iteration, the system automatically strengthens the attenuation compensation, placing greater emphasis on recent fusion records to allow the fusion coefficients to adapt to the new production situation more quickly.

[0034] Based on the final generated fusion weights, the system sorts all candidate adjustment quantities and performs hierarchical screening. The screening process checks for conflicts between candidate adjustment quantities and core process parameters defined in the quality evolution chain, in descending order of fusion weight. Core process parameters are typically critical parameters that have a decisive impact on the final product quality and have a very narrow permissible range of variation. The check compares whether the adjustment quantity causes any core parameter to exceed its specified tolerance range. If the candidate adjustment quantity with the highest weight passes the conflict-free verification, it is selected as the target adjustment quantity. If a conflict is found during verification, the adjustment quantity is excluded, and the system automatically switches to the next lower weight candidate for adjustment, repeating the verification process until an adjustment quantity with high comprehensive evaluation value and passing the compliance checks of all core parameters is found as the final execution target. The entire dynamic weighted fusion and screening process constitutes a highly adaptable and fault-tolerant decision-making chain, providing crucial support for the stable optimization of the core production system.

[0035] Example 3: See Figure 3 The stability assessment function is a core tool for measuring the risk of quality fluctuations throughout the entire production process. Its construction begins with an in-depth analysis of the dynamic quality weight matrix, which reflects the relative importance of different process parameters on the final quality. Based on the process distribution of weights in the matrix, it is necessary to calculate the spatial coupling degree between the defect propagation path and parameter sensitivity. The defect propagation path describes the typical route of quality defects transmitted between processes on the production line, usually derived from historical nonconforming product analysis. Parameter sensitivity quantifies the degree of impact of small changes in processing parameters at each process on quality inspection indicators (such as dimensional deviations). The calculation of spatial coupling degree aims to reveal the spatial correlation between the "high-speed channel" of defect propagation and the areas where parameters are prone to fluctuation. High coupling degree means that parameter instability occurs precisely in critical areas where defects are easily generated and transmitted, thus amplifying the quality risk.

[0036] Calculating spatial coupling involves correlation analysis of multidimensional data. The system needs to map the defect propagation path into a space composed of process and parameter dimensions, while also projecting parameter sensitivity into the same space. By analyzing the distribution patterns of both in this space, such as calculating the overlap or correlation of their distribution profiles, a quantitative index characterizing their correlation tightness is obtained. This index is one of the basic inputs for constructing the evaluation function, helping to identify process steps that require key monitoring. Based on the calculated spatial coupling, the system needs to generate a dynamic correction factor. The core function of this factor is to correlate high-defect areas with parameter fluctuations in low-sensitivity processes. High-defect areas refer to processes or combinations of processes that have historically had frequent quality problems, while low-sensitivity processes refer to those steps where parameter changes have a relatively insignificant impact on quality indicators. The logic behind this correlation is that parameter fluctuations in low-sensitivity processes are often easily overlooked, but if such fluctuations occur near high-defect areas, their indirect, cumulative effects may eventually trigger defects through complex process coupling. The dynamic correction factor is precisely designed to capture and amplify this potential risk. When generating the initial correction factor, it is necessary to comprehensively consider spatial coupling and another derived indicator—the inverse correlation between defect propagation rate and parameter adjustment gradient. Defect propagation rate refers to how quickly a defect spreads along a path, while parameter adjustment gradient reflects the magnitude of parameter changes between processes. The inverse correlation means that when defect propagation is rapid, parameter adjustment often appears slow or inconsistent; this mismatch exacerbates system instability. Based on spatial coupling and inverse correlation, a fluctuation difference index can be calculated, reflecting the inconsistency in behavioral patterns between high-defect regions and low-sensitivity processes.

[0037] The initial correction factor is generated based on the fluctuation difference index. Its purpose is to link the inherent weak points in process stability of high-defect regions with the sparse parameter monitoring coverage typically present in low-sensitivity processes. However, the change of the initial correction factor cannot be arbitrary; it must be strictly limited by the process constraint coefficients defined in the quality evolution chain. These constraint coefficients specify the legal boundaries and rates of parameter evolution, thereby constraining the scope and behavior of the correction factor and preventing it from amplifying abnormal noise fluctuations. To further enhance the robustness of the correction factor, the system references the historical correction records of the process constraint coefficients themselves and applies dynamic attenuation compensation to the initial correction factor. The historical correction records reflect the past adjustments of the constraint coefficients. If recent adjustments are frequent, it indicates that the region is in an unstable state. In this case, the attenuation compensation of the correction factor will be enhanced to suppress the asymmetric fluctuation overload that may be caused by frequent adjustments of local parameters, and to avoid violent oscillations in the evaluation function.

[0038] Having obtained the core elements such as the dynamic quality weight matrix, dynamic correction factor, and spatial coupling degree, we can then construct the final stability evaluation function. This function combines these elements through a fusion model, and its general form can be expressed as a weighted sum of each element, introducing a penalty term related to the difference in process constraint coefficients. The penalty term is designed to punish parameter configurations that cause significant deviations in the constraint coefficients of adjacent processes, as such deviations often signify a disruption of process path consistency and increase the risk of runaway. The output of the evaluation function is a quantified quality fluctuation index, which intuitively reflects the overall stability level under the current production state.

[0039] When constructing the evaluation function, a possible fusion model can be considered in the following form: ; in: This represents the final calculated quality fluctuation index. (Symbol) This indicates the total number of processes involved in the production process. It is the weight assigned to the first in the dynamic quality weight matrix. The weighting coefficient of each process reflects the degree of influence of that process on the overall quality. It is in the The spatial coupling degree component calculated in each process reflects the local correlation characteristics between defect propagation and parameter sensitivity at that process. It is applied to the first The dynamic correction factor for each process step is used to adjust the volatility risk assessment for that process step. This represents the maximum value of the dynamic correction factor across all processes, used for... Normalization is performed. It is the first The difference between the current process constraint coefficient and a reference baseline (such as a standard value or the average of adjacent processes) is used as input for the penalty term. (Greek letters) , , These are harmonic parameters, used to adjust the relative contribution weights of spatial coupling, dynamic correction factor, and constraint coefficient difference penalty term in the overall evaluation. These parameters need to be determined through system debugging or learning from historical data.

[0040] The quality fluctuation index calculated by the stability assessment function, along with the verification results of the process compensation operation, is used to generate a feedback matrix. The process compensation verification results mainly include the "tolerance compliance rate," which is the proportion of the compensated parameters that actually meet the tolerance rules in the quality evolution chain. The first step in generating the feedback matrix is ​​to calculate a coupling coefficient between fluctuation and tolerance based on the quality fluctuation index and the tolerance compliance rate. This coefficient reflects the strength of the correlation between quality fluctuation and process compliance. Based on this coupling coefficient, an initial feedback matrix can be generated. This matrix attempts to establish a correlation between high fluctuation areas and tolerance deficiencies, and maps this correlation to specific parameter adjustment amounts. To enable the feedback matrix to adapt to changes in the production environment and to avoid drastic changes in the matrix due to outliers in a single assessment, the system applies a dynamic decay constraint to the initial feedback matrix based on historical correction records of the quality evolution chain. Historical correction records provide trend information on matrix changes, and the decay constraint ensures that matrix updates are smooth and gradual. Finally, by integrating the coupling coefficient, the quality fluctuation index, and the tolerance compliance rate, and introducing a penalty term related to the difference between the current dynamic quality weight matrix and a certain ideal reference matrix, a multidimensional feedback model is constructed. The output of this model is the final feedback matrix used to guide the next round of optimization.

[0041] Example 4: The multi-stage production process of aircraft turbine blades is illustrated using this example. Turbine blade manufacturing typically involves several key processes, including profile milling, heat treatment, profile grinding, and surface treatment. Each process has several monitoring parameters, such as milling speed, feed rate, heat treatment temperature, holding time, and grinding pressure. The system collects time-series data of these parameters for each blade in historical batches, arranged sequentially by process, along with the final quality judgment result (qualified or unqualified). The goal of extracting differential features is to identify systematic deviations in the process path between qualified and unqualified products, thereby identifying process steps prone to defects and their correlations. Time-frequency joint analysis of the processing parameter sequences for each process in historical batches is performed. First, the parameter sequence for each blade needs to be divided into a series of overlapping segments according to production time windows. For example, each time window consists of 10 consecutive processing points, with adjacent time windows overlapping by 50% of their data points to ensure the continuity of the analysis. Taking the milling speed parameter in profile milling as an example, the speed sequence of qualified products may exhibit stable fluctuations, while the sequence of unqualified products may show sudden sharp increases or decreases within a certain time window. For each time window segment, the system simultaneously performs time-domain smoothing and frequency-domain power spectrum analysis. Time-domain smoothing uses a moving average filter to remove random noise and extract the overall trend of the sequence, while frequency-domain power spectrum analysis uses a fast Fourier transform to convert the sequence to the frequency domain and identify its periodic and sudden frequency components. From the time-domain smoothing curve, the algorithm locates extreme points (such as local maxima and minima), and from the frequency-domain power spectrum, it identifies peak intervals with power significantly higher than the background. Those regions that show anomalies in both the time and frequency domains, i.e., the regions where extreme points and peak intervals coincide on the time axis, are marked as parameter-sensitive intervals. The sensitive intervals of qualified product sequences are usually evenly distributed and have small amplitudes, exhibiting steady-state characteristics, while the sensitive intervals of unqualified product sequences may appear densely in specific time periods and have large amplitudes, exhibiting abrupt changes.

[0042] Table 1 shows a comparison of key features of milling speed parameter sequence segments of qualified and unqualified turbine blades in the profile milling process of a certain batch, after time-frequency joint analysis. This table is only used to illustrate the analysis process, and the actual data dimensions are more complex.

[0043] Table 1: Time-Frequency Characteristics of Milling Speed ​​Parameters for Qualified and Unqualified Products

[0044] Calculating the transfer gain between steady-state and abrupt changes in characteristics between adjacent processes is a core step in identifying critical coupling nodes. Transfer gain quantifies the efficiency with which parameter fluctuations in a preceding process affect the quality of a subsequent process. For example, in turbine blade production, the profile milling process (process A) and the heat treatment process (process B) are adjacent critical steps. The system extracts the steady-state characteristics of the output parameters of process A and the abrupt changes in the input parameters of process B from historical data of both qualified and unqualified products. The calculation of transfer gain is based on a transfer function model that simulates the degree of influence of changes in process A characteristics on process B characteristics. A high gain value indicates that fluctuations in the preceding process are easily transferred and amplified to the subsequent process. The system sets a gain threshold, which is usually determined based on the statistical distribution of historical data. When the transfer gain of a process pair exceeds this threshold, the process pair is marked as a critical coupling node. In the example, a high transfer gain might be found between the abrupt change in rotational speed during profile milling and the abnormal fluctuation in heat treatment temperature, thus marking the "milling-heat treatment" process pair as a critical node. Process path difference features are generated based on the spatial distribution of key coupling nodes. Spatial distribution refers to the location and connection relationships of these key nodes in the overall process route diagram. The system abstracts the entire production process into a directed graph, where nodes represent processes, directed edges represent the sequence of processes and material flow, and key coupling nodes correspond to connections with high edge weights (determined by transfer gain). Process paths for qualified products typically show fewer and more dispersed key nodes, while paths for unqualified products may show key nodes clustered in specific areas, such as several consecutive processes exhibiting high coupling. Difference features are extracted from this distribution comparison, including statistics and topological indicators such as the number and position sequence of key nodes, and the cumulative value of inter-node gain. These difference features are then encoded into directed quality evolution chains according to the process transfer direction. An evolution chain is a data structure that records how quality features evolve step by step from raw materials to finished products. During encoding, the evolution chain for qualified products emphasizes the transfer path of steady-state features, while the evolution chain for unqualified products highlights the propagation path of abrupt changes. Directionality ensures that the direction of defect propagation is consistent with the direction of production process flow.

[0045] The overlapping time window segmentation strategy in time-frequency joint analysis needs to be optimized based on the specific production rhythm and parameter sampling frequency. The choice of window size and overlap rate will affect the detection sensitivity of sensitive intervals. The choice of filter in time-domain smoothing is also important. For example, using a Gaussian filter can better preserve trend information, while the power spectrum estimation method in frequency-domain analysis will affect the accuracy of peak detection. The extracted steady-state and abrupt features need to be transformed into quantifiable indicators. For example, steady-state features may include the mean, standard deviation, and autocorrelation time of parameters, while abrupt features may include the maxima of the difference sequence and the number of consecutive abrupt points. The distribution differences of these feature values ​​can be quantified using statistical hypothesis testing methods to ensure that the extracted difference features are statistically significant. The marking of key coupling nodes depends not only on a single calculation but also on the reproducibility in historical batches. A node will only be finally confirmed if it shows high gain in multiple batches of non-conforming products. After generating a directed quality evolution chain, the nodes in the chain will be bound to tolerance rules, and the edges will be weighted. The weight values ​​reflect the strength of the quality influence between processes. This chain will serve as an important input for subsequent parameter optimization and stability assessment.

[0046] Example 5: A matrix is ​​constructed based on the parameter coupling degree between processes to dynamically reflect the weight of each process's quality impact. The matrix generation begins with calculating the correlation coefficient matrix between the processing parameters of each process. The correlation coefficient measures the degree of synergy in the changes of parameters across different processes in historical production data. Taking the multi-process production of aircraft casing structural components as an example, key processes may include rough milling, finish milling, drilling, heat treatment, deburring, etc. Parameters monitored for each process include spindle power for rough milling, tool vibration frequency for finish milling, feed rate for drilling, heating rate for heat treatment, and grinding head pressure for deburring. The system collects data on these process parameters from all production batches within a historical period, calculates the Pearson correlation coefficient between any two parameters, thus forming a symmetrical correlation coefficient matrix. This matrix reveals whether the parameters are positively correlated, negatively correlated, or unrelated. Marking process pairs with correlation coefficients exceeding a preset threshold as strongly coupled groups is a step in identifying key interactions. The threshold setting needs to be based on the characteristics of the specific production process and the strictness of quality control, and is usually determined by analyzing the distribution of correlation coefficients in historical data or combining domain expert experience. For example, in the machining of machine housings, a high positive correlation may be found between the spindle power in the rough milling process and the tool vibration frequency in the finish milling process. This means that unstable power during rough milling often carries over to the finish milling process, causing increased tool vibration. Similarly, the heating rate in the heat treatment process may be negatively correlated with the grinding head pressure in the deburring process, because a large amount of deformation during heat treatment increases the difficulty and pressure requirements of subsequent deburring. These identified process pairs, such as "rough milling-finish milling" and "heat treatment-deburring," constitute strongly coupled groups, representing intrinsic connections in the production process that require special attention.

[0047] The initial weight matrix is ​​generated based on the spatial density distribution of identified strongly coupled groups, which refers to the clustering of these strongly coupled groups on the overall process roadmap. The system treats the entire production process as a network, with processes as nodes and strong coupling relationships as edges. The density of strongly coupled connections around each process node can be calculated using kernel density estimation or simple grid counting methods. For example, if the finish milling process is strongly coupled with multiple preceding processes such as rough milling and drilling, as well as the subsequent process of heat treatment, then the connection density in its region is very high. The principle for allocating the initial weight matrix is ​​to assign higher initial weight coefficients to processes in high-density regions, because parameter fluctuations in these processes can have a wide-ranging cascading effect through the dense connection network, naturally resulting in a greater weight on the final quality. Conversely, the weight coefficients for processes in low-density regions are set lower. Simultaneously, the decay rate of the weight coefficients in low-density regions also needs to be set. The decay rate defines the gradient as the weight decreases with sparser coupling connections, typically using an exponential decay function to simulate this reduction in the range of influence.

[0048] Dynamically adjusting the initial weight matrix based on defect distribution in real-time quality inspection data is crucial for making the matrix adaptable. The initial weights are a static starting point calculated based on historical coupling relationships, while real-time production quality feedback is used to calibrate them. The system continuously receives quality inspection data from the current production batch, including which process, what type of defect, and its severity were found. Defect distribution information is used to calculate the defect clustering degree for each process or process area. The defect clustering degree can be measured by the frequency and spatial proximity of defects occurring within a unit of time or unit of product quantity. The logic of dynamic adjustment is to make the weight coefficients negatively correlated with the defect clustering degree. This may seem counterintuitive, but the underlying principle is preventative adjustment: if a region has a high defect clustering degree, it indicates that the existing process control in that region may have failed or is under great pressure. Blindly increasing its weight in this case may amplify noise interference. Conversely, by appropriately reducing its weight, the system can allocate more optimization resources to potential risk areas where the current defect rate is not high but may cause new problems through strong coupling relationships, thereby achieving a more global and forward-looking risk balance. The adjustment operation usually uses a weighted update rule, such as exponentially smoothing the initial weights based on newly observed defect data.

[0049] Calculating the correlation coefficient matrix requires addressing potential issues of inconsistent dimensions and missing values, necessitating data standardization and appropriate interpolation. After labeling strongly coupled groups, community detection algorithms from graph theory may be needed to identify tightly connected modules inherent in the process network. When generating the initial weight matrix, the bandwidth selection for kernel density estimation affects the smoothness of density calculations and needs adjustment based on the size of the process network. Calculating defect clustering may involve spatial statistical methods, such as calculating the Moran index to determine whether defects exhibit spatial autocorrelation between processes. The update frequency for dynamic corrections needs careful setting; updates that are too rapid may cause the weight matrix to become unstable due to random fluctuations, while updates that are too slow may fail to reflect changes in process status in a timely manner. The entire process of generating the dynamic quality weight matrix constitutes the perceptual foundation of the quality optimization system, quantitatively describing which interactions in the production process are most critical and how these criticalities should be dynamically adjusted based on actual quality performance.

[0050] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0051] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An aircraft component production quality optimization method, characterized by, The method comprises the following steps: Collecting real-time processing parameters and quality inspection data of multiple processes on the production line, wherein the quality inspection data includes size deviation, surface defects and material performance indicators, and generating a dynamic quality weight matrix according to the parameter coupling degree between processes; Extracting the process path difference characteristics of qualified products and unqualified products in historical batches, and encoding the difference characteristics into a quality evolution chain, wherein the quality evolution chain is generated by matching the parameter mutation correlation of adjacent processes; Performing collaborative optimization of process parameters at the quality evaluation node, driving the parameter combination to iterate in the direction of minimizing quality fluctuation, generating parameter adjustment amount and updating the process constraint coefficient of the quality evolution chain each time, and constructing a stability evaluation function based on the correlation between the dynamic quality weight matrix and the defect propagation path; Triggering process compensation of key processes according to the gradient distribution of the parameter adjustment amount, wherein the compensation data is a reference parameter set matched with the target quality evolution chain in the current production batch, and the compensation parameters are verified for process connection through the tolerance rules in the quality evolution chain; Generating a feedback matrix from the quality fluctuation index output by the stability evaluation function and the process compensation verification result, and dynamically correcting the mapping relationship between the dynamic quality weight matrix and the quality evolution chain.

2. The method of claim 1, wherein, The method of performing collaborative optimization of process parameters at the quality evaluation node, driving the parameter combination to iterate in the direction of minimizing quality fluctuation, comprises the following steps: Encoding process parameters into a multi-dimensional process vector, wherein each dimension corresponds to the adjustable interval of the process constraint coefficient, and calculating the process interference degree based on the quality fluctuation range covered by the current parameter combination; Constructing a fluctuation neighbor table according to the process interference degree, recording the process path and process constraint coefficient difference that exist quality interference with the current parameter combination, and generating candidate adjustment amount; Selecting target adjustment amount from the candidate adjustment amount by using a two-stage screening mechanism, wherein the first stage screening is based on the effectiveness of the candidate adjustment amount in reducing fluctuation in historical iterations, and the second stage screening is based on the compatibility threshold of the target adjustment amount and the tolerance rules in the quality evolution chain; Applying the target adjustment amount to the current process parameters, dynamically adjusting the coefficient value of the corresponding dimension in the multi-dimensional process vector, and updating the process interference degree threshold of the fluctuation neighbor table to trigger the elimination of low stability paths; Recomputing the quality matching degree of the parameter updated with the reference parameter set, and feeding back the quality matching degree change to the coefficient distribution of the dynamic quality weight matrix.

3. The method of claim 2, wherein, The method of selecting target adjustment amount from the candidate adjustment amount by using a two-stage screening mechanism comprises the following steps: Based on the effectiveness of the candidate adjustment amount in reducing fluctuation in historical iterations, the ratio of the number of successful times to the number of quality weight promotion times of each candidate adjustment amount in a preset production period is normalized as the reference weight of the first stage screening; Extracting the process boundary value bound by the tolerance rule in the quality evolution chain, calculating the coverage ratio of the corresponding dimension of the candidate adjustment amount to the allowed interval of the boundary value, and mapping it to the boundary constraint weight of the second stage screening; Dynamically weighting and fusing the reference weight and the boundary constraint weight, and adjusting the weighting coefficient according to the density index of the low stability path in the fluctuation neighbor table, so that the boundary constraint occupies the dominant weight in the interference intensive area. Perform hierarchical screening based on fusion weight, verify non-conflict of the candidate adjustment amount and the quality evolution chain core process parameter, if conflict, switch to suboptimal candidate adjustment amount until the boundary constraint is met.

4. The method of claim 3, wherein, The dynamic weighting fusion of the reference weight and the boundary constraint weight comprises: Based on the comparison between the low stability path proportion in the fluctuation neighborhood table and the historical maximum process capacity, an interference density index is generated through piecewise linear conversion; A dynamic fusion coefficient is generated according to the interference density index, a sliding window mechanism is used to attenuate and compensate the historical fusion record, and the step mutation of the dynamic fusion coefficient is inhibited; The reference weight is coupled with the dynamic fusion coefficient, and the coupling result is superimposed with the boundary constraint weight, and the superimposed result is truncated and normalized; According to the deviation amplitude of the normalized result and the historical sliding mean, the intensity of attenuation compensation is adjusted, if it exceeds the tolerance threshold, the compensation intensity is enhanced and applied to the fusion calculation of the next iteration period.

5. The method of claim 1, wherein, The construction of the stability evaluation function based on the correlation between the dynamic quality weight matrix and the defect propagation path comprises: Based on the process distribution of the dynamic quality weight matrix, the spatial coupling degree of the defect propagation path and the parameter sensitivity is calculated; A dynamic correction factor is generated according to the spatial coupling degree, the parameter fluctuation of the high defect area is associated with the low sensitivity process, and the action range of the dynamic correction factor is limited based on the process constraint coefficient of the quality evolution chain; The dynamic quality weight matrix, the dynamic correction factor and the spatial coupling degree are fused to construct the stability evaluation function, and a penalty term related to the difference of the process constraint coefficient is introduced.

6. The method of claim 5, wherein, The generation of the dynamic correction factor according to the spatial coupling degree comprises: Based on the spatial coupling degree and the inverse correlation between the defect propagation rate and the parameter adjustment gradient, the fluctuation difference index of the high defect area and the low sensitivity process is calculated; An initial correction factor is generated according to the fluctuation difference index, the process stability of the high defect area is associated with the parameter coverage sparsity of the low sensitivity process, and the change rate of the initial correction factor is constrained by the process constraint coefficient of the quality evolution chain; Based on the historical correction record of the process constraint coefficient, dynamic attenuation compensation is applied to the initial correction factor to inhibit asymmetric fluctuation overload.

7. The method of claim 1, wherein, The generation of the feedback matrix from the quality fluctuation index and the process compensation verification result output by the stability evaluation function comprises: Based on the tolerance coincidence rate in the quality fluctuation index and the process compensation verification result, the coupling coefficient of fluctuation and tolerance is calculated; An initial feedback matrix is generated according to the coupling coefficient, the tolerance loss of the high fluctuation area is associated with the parameter adjustment amount, and the initial feedback matrix is subjected to dynamic attenuation constraint based on the historical correction record of the quality evolution chain; A multi-dimensional feedback model is constructed by fusing the coupling coefficient, the quality fluctuation index and the tolerance coincidence rate, a penalty term related to the difference of the dynamic quality weight matrix is introduced to generate the feedback matrix.

8. The method of claim 1, wherein, The extraction of the process path difference characteristics of the qualified products and unqualified products in the historical batches comprises: The time-frequency joint analysis of the processing parameter sequence of each process in the historical batches is performed, and the steady-state characteristics of the qualified product sequence and the mutation characteristics of the unqualified product sequence are extracted; The transmission gain of the steady-state feature and the mutation feature between adjacent processes is calculated, and the process pair with the gain exceeding a threshold is marked as a key coupling node; A process path difference feature is generated according to the spatial distribution of the key coupling node, and the difference feature is encoded into a directed quality evolution chain according to the process transmission direction.

9. The method of claim 8, wherein, The time-frequency joint analysis of the processing parameter sequence of each process in the historical batch includes: The processing parameter sequence is divided into overlapping segments according to the production time window, and time domain smoothing and frequency domain power spectrum analysis are simultaneously performed on each segment; The extreme points of the time domain smoothing curve and the peak value interval of the frequency domain power spectrum are extracted, and the time and space overlapping region is marked as a parameter sensitive interval; The steady-state feature and the mutation feature are generated according to the distribution difference of the sensitive interval in the qualified product and unqualified product sequence.

10. The method of claim 1, wherein, The dynamic quality weight matrix is generated according to the parameter coupling degree between processes, including: The correlation coefficient matrix of the processing parameters of each process is calculated, and the process pair with the correlation coefficient exceeding a threshold is marked as a strong coupling group; An initial weight matrix is generated according to the spatial density distribution of the strong coupling group, and the weight coefficient of the high-density region is associated with the decay rate of the low-density region; The initial weight matrix is dynamically corrected based on the defect distribution in the real-time quality inspection data, so that the weight coefficient is negatively correlated with the defect aggregation degree.

Images