A quality management method based on standard data

Abstract

The present invention relates to the technical field of quality management, and specifically provides a quality management method based on standard data. Specifically, it includes: constructing a dynamic standard database; collecting production data in real time; adopting a hybrid anomaly detection and dynamic interpolation strategy to handle data missing and noise, generating a standardized matching code based on hash and polynomial ring operations to ensure data consistency; determining quality deviations based on a dynamic threshold model and fuzzy logic, quantifying the deviation amount by combining the weighted Euclidean distance, realizing hierarchical early warning through a membership function, and triggering response measures according to the early warning level; using a scoring model and non-linear prediction to evaluate the effectiveness of measures and screening optimization plans; a closed-loop feedback mechanism continuously updates model parameters and the knowledge base. Through dynamic data integration, hybrid detection algorithms and matching code design, the present invention effectively solves the problems of inflexible static thresholds, single anomaly detection, and poor data consistency in traditional quality management, and significantly improves the detection accuracy and system robustness.

Description

Technical Field

[0001] The present invention relates to the technical field of quality management, and in particular to a quality management method based on standard data. Background Art

[0002] In modern manufacturing and service industries, quality management is a key link to ensure that products and services meet user needs and improve market competitiveness. However, traditional quality management methods often rely on experience and local data, which can hardly adapt to complex and changing production environments, resulting in unstable quality control, delayed defect detection and inefficient quality improvement. With the improvement of information technology and data analysis capabilities, quality management methods based on standard data have gradually become an important means to improve the accuracy of quality control and improve production efficiency. Standard data refers to data that has been strictly verified and accumulated over a long period of time under certain conditions. It is highly reliable and stable and can provide a scientific basis for quality management.

[0003] Publication No. CN118569888A discloses a standardized quality management early warning method and system based on big data analysis, the method comprising: obtaining quality inspection instructions and locating targeted products; obtaining and analyzing targeted product parameters and generating product control ratios; generating quantitative detection feedback data based on the product control ratios; obtaining and analyzing quantitative detection feedback data and generating production control early warning instructions; generating traceability analysis results based on production control early warning instructions; the method and system can perform proportional control and analysis on production line products, conduct diversified quality analysis on products, perform quantitative detection based on the number of types of problems found in quality analysis, and further generate production control early warning instructions and traceability analysis results based on the quantitative detection results, thereby improving product standardization quality and quality control management improvement efficiency.

[0004] Standard data, as the core basis of quality management, can provide unified and standardized quality evaluation standards for different production links to ensure the accuracy and consistency of various data. With the deepening of informatization construction, more and more companies have begun to adopt quality management methods based on standard data, and use data-driven methods to monitor and optimize production processes in real time to improve production efficiency, reduce costs and ensure the stability of product quality. This method can not only accurately identify potential quality problems, but also achieve early warning and timely correction, thereby reducing waste and rework in the production process and improving overall resource utilization. In this context, how to use standard data to build an efficient and scientific quality management system has become an important issue that needs to be solved urgently. Summary of the invention

[0005] The purpose of the present invention is to propose a quality management method based on standard data in view of the problems existing in the background technology.

[0006] Technical solution of the present invention: A quality management method based on standard data, comprising the following specific implementation steps:

[0007] S1. Construct standard data, perform hybrid anomaly detection, process missing values using a dynamic interpolation strategy, generate standardized data after normalization, and generate a first-order standardized matching code using polynomial ring operations;

[0008] S2. Collect real-time data and perform data preprocessing: filter outlier values, noise data, perform data format conversion, and generate a second-order standardized matching code;

[0009] S3. Match the standardized data and real-time data, verify the data standard consistency using polynomial ring operations, quantify the quality deviation by combining the weighted Euclidean distance, and construct a dynamic threshold model and a fuzzy membership function for deviation classification determination;

[0010] S4. By combining the fuzzy membership degree and change trend of the current deviation, construct a dynamic warning index, dynamically adjust the classification threshold according to the mean, standard deviation, and sensitivity adjustment coefficient of the historical warning index, update the warning level boundary value, and trigger classification response measures according to the index value;

[0011] S5. Based on the intelligent optimization suggestion generation method integrating multi-source data, construct an intelligent quality optimization suggestion generation model, generate scores for each potential optimization measure, and quantify the expected improvement effect using a non-linear prediction model, screen the best measures and generate an optimization suggestion report;

[0012] S6. Record the effect data of the optimization measures through a closed-loop feedback mechanism, and update the model parameters and knowledge base by comparing the prediction and actual results.

[0013] Preferably, the detection process of the hybrid anomaly detection is as follows:

[0014] S21. Obtain the standardized data Sdata = X = {x1, x2, …, x i , …, x N}, calculate the mean μ and the standard deviation σ:

[0015] ;

[0016] ;

[0017] where N represents the total number of element data in the standardized data; x i represents the i-th element data in the standardized data;

[0018] S22. According to the 3σ principle, define the normal data range: X f = {x i || x i-μ|≤3σ};

[0019] where x i represents the i-th data sample; μ represents the mean; σ represents the standard deviation; X f represents the data after removing outliers;

[0020] S23. Use the Isolation Forest algorithm to detect high-dimensional outliers: ;

[0021] In the formula, S(x i ) represents the outlier score; E(h(x i )) represents the average path length of the data point x i in the decision tree; c(n) represents the normalization factor of the data scale; h(x i ) represents the path length of the data point x i in the isolation tree of the Isolation Forest.

[0022] Preferably, the generation process of the first-order standardized matching code is as follows:

[0023] S31. Convert the standardized data SⅠdata into binary data Ⅰdata;

[0024] S32. Select a random number αⅠ and calculate the standard first-order element UⅠ = H(Ⅰdata||αⅠ) ∈ IR;

[0025] where IR is a predefined ring, IR = Z q [x] / (x τ + 1); q is a predefined large prime number; τ is a predefined parameter; || is the concatenation operation; H is a predefined hash function; the random number αⅠ is 128 bits;

[0026] S33. Calculate the first-order matching element code CMⅠ = SpⅠ·UⅠ mod (q, x τ + 1);

[0027] where SpⅠ is a randomly generated short polynomial, and its coefficients are randomly selected in ;

[0028] S34. Generate the first-order standardized matching code CSⅠ = (random number αⅠ, first-order matching element code CMⅠ).

[0029] Preferably, the generation process of the second-order standardized matching code is as follows:

[0030] S41. Convert the real-time data Rdata into binary data Ⅱdata;

[0031] S42. Select a random number αⅡ and calculate the standard second-order element UⅡ = H(Ⅱdata||αⅡ) ∈ IR;

[0032] Among them, the random number αⅡ is 128 bits;

[0033] S43. Calculate the second-order matching element code CMⅡ = SpⅡ·UⅡ mod (q,x τ +1);

[0034] Among them, SpⅡ is a randomly generated short polynomial, and its coefficients are randomly selected from ;

[0035] S44. Generate the second-order standardized matching code CSⅡ = (random number αⅡ, second-order matching element code CMⅡ).

[0036] Preferably, the matching process of matching the standardized data and the real-time data is as follows:

[0037] S51. Respectively convert the standardized data SⅠdata and the real-time data Rdata into binary string data dataⅢ and dataⅣ;

[0038] S52. Calculate the matching first-order element MⅠ = H(dataⅢ||αⅠ) and the matching second-order element MⅡ = H(dataⅣ||αⅡ);

[0039] S53. Calculate the first-order parsing code AⅠ = CMⅠ·CMⅡ mod (q,x τ +1);

[0040] S54. Calculate the second-order parsing code AⅡ = AF·MⅠ·MⅡ mod (q,x τ +1);

[0041] Among them, AF is a predefined matching parsing factor, AF = SpⅠ·SpⅡ mod (q,x τ +1);

[0042] S55. If AⅠ = AⅡ mod (q,x τ +1), the matching passes, indicating that there is standard consistency between the standardized data SⅠdata and the real-time data Rdata; otherwise, an alarm is given.

[0043] Preferably, the determination process of deviation grading determination is as follows:

[0044] S61. Obtain the standard data vector Qs = [s1, s2,..., s i ,…, s n , and obtain the real-time quality data vector Qr = [r1, r2,..., r i ,…, r n ;

[0045] Among them, ri represents the real-time measured value of the i-th quality parameter; s i represents the corresponding standard index value; n represents the number of quality indicators;

[0046] S62. Use the weighted Euclidean distance as the deviation metric: ;

[0047] where, ΔQ represents the overall quality deviation; w i represents the weight of the i-th quality parameter, satisfying the normalization condition: ;

[0048] S63. Construct a dynamic threshold model using historical quality deviation data:

[0049] ;

[0050] In the formula, T represents the dynamic judgment threshold of the current quality deviation; represents the mean value of the historical deviation data; represents the standard deviation of the historical deviation data; λ represents the sensitivity adjustment coefficient;

[0051] S64. Use fuzzy logic to construct a membership function to classify and judge the real-time deviation:

[0052] ;

[0053] In the formula, μ(Q) represents the membership degree of the quality deviation, and the value range is between [0, 1]; k represents the slope parameter of the membership function.

[0054] Preferably, the adjustment process of dynamically adjusting the classification threshold is as follows:

[0055] S71. Based on the fuzzy membership degree μ(Q) and the change trend R of the quality deviation over time, define the warning index I warn as: I warn = μ(Q)·(1 + β·R);

[0056] ;

[0057] where, μ(Q) represents the fuzzy membership degree of the current quality deviation; β represents the trend weight coefficient; R represents the change rate of the quality deviation; ΔQ(t) represents the quality deviation at the current moment; ΔQ(t - Δt) represents the quality deviation at the previous moment; Δt represents the sampling time interval;

[0058] S72. According to the value of the warning index I warn set the warning level, and use the classification formula:

[0059] ;

[0060] Among them, η1, η2, and η3 represent the early warning demarcation thresholds, and the initial values are determined based on the distribution of historical early warning data and satisfy: 0 ≤ η1 < η2 < η3 ≤ 1;

[0061] S73. Introduce an adaptive threshold adjustment mechanism, and based on the statistics of the early warning indicator I over a period of history, dynamically update the early warning demarcation value, that is, set the new threshold as: warn ; ;

[0062] Among them, represents the i-th new threshold; represents the mean value of the historical early warning indicators; represents the sensitivity adjustment coefficient, and different values are set according to different early warning levels i; represents the standard deviation of the historical early warning indicators.

[0063] Preferably, the implementation process of the intelligent optimization suggestion generation method based on multi-source data integration is as follows:

[0064] S81. Collect and integrate the following data: the comparison result of real-time quality data and standard data, the intelligent early warning output result, historical optimization data, and empirical rules;

[0065] S82. Construct an intelligent quality optimization suggestion generation model based on the deviation of each key parameter and the early warning level, and generate a score S for each potential optimization measure j j : ;

[0066] In the formula, S j represents the comprehensive optimization suggestion score for measure j; γ ij represents the sensitivity weight of the i-th parameter to optimization measure j; represents the adjustment coefficient associated with the early warning indicator; r i represents the real-time measured value of the i-th quality parameter; s i represents the corresponding standard index value; n represents the number of quality indicators;

[0067] S83. Use a non-linear prediction model to estimate the possible quality improvement brought by each measure, and calculate the expected improvement effect ΔAQ for measure j j : ;

[0068] In the formula, ΔAQ j represents the expected quality improvement amplitude after the implementation of optimization measure j; θ j represents the maximum improvement potential of measure j; κ j represents the improvement sensitivity coefficient of measure j;

[0069] S84. Sort according to the score S of each measure j and the predicted improvement effect ΔAQ j , select the best optimization measure after sorting, and generate a detailed optimization suggestion report.

[0070] Compared with the prior art, the above technical solution of the present invention has the following beneficial technical effects:

[0071] The present invention designs a quality management method based on standard data. Through technology integration and dynamic closed-loop design, significant breakthroughs have been achieved in the real-time, accuracy, security, and self-adaptability of quality management, and it has broad industrial application value:

[0072] (1) Improved dynamic adaptability: By integrating international standards, historical high-quality data, and real-time feedback to construct a dynamic standard database, the problem that the traditional static standard library cannot adapt to changes in the production environment is solved, and the timeliness and applicability of standard data are significantly improved;

[0073] (2) Optimized anomaly detection accuracy: A hybrid detection method combining the 3σ principle and the Isolation Forest algorithm to make up for the detection limitations of a single method for non-normal distribution or high-dimensional data;

[0074] (3) Ensure data integrity and standard consistency: Adopt dynamic interpolation strategies (mean filling / KNN) and matching code generation mechanisms (hash and polynomial ring operations) to effectively handle data missing and noise, and at the same time ensure data consistency and anti-tampering ability through matching codes;

[0075] (4) Intelligent early warning and optimization: A hierarchical early warning mechanism based on a dynamic threshold model, fuzzy logic determination, and trend analysis to achieve multi-level responses from mild prompts to emergency interventions; The optimization suggestion generation module screens the best measures through a scoring model and non-linear prediction, and continuously updates the model parameters in combination with closed-loop feedback to gradually improve the effectiveness of the optimization measures;

[0076] (5) Enhanced self-learning ability: Dynamically adjust sensitivity weights, adjustment coefficients, and knowledge bases through execution feedback, have the ability of continuous iterative optimization, reduce the need for manual intervention, and are applicable to complex and changing manufacturing scenarios. Description of the Drawings

[0077] Figure 1 It is a method flow chart of a quality management method based on standard data proposed by the present invention. Detailed Embodiment

[0078] A quality management method based on standard data proposed by the present invention, as Figure 1 shown, its specific implementation steps are as follows:

[0079] S1. Adopt big data processing, statistical analysis, and machine learning technologies, and combine industry standards, historical high-quality data, and expert experience to form standard data applicable to intelligent quality management. The specific implementation process is as follows:

[0080] S11. The establishment of standard data needs to be based on several credible sources to ensure its representativeness and authority. The data sources include but are not limited to:

[0081] International standards (ISO), industry standards (GB, ASTM): Provide general standards for quality indicators;

[0082] Historical high-quality production data: Extract the optimal process parameters from the production records of high-quality products;

[0083] Expert experience and knowledge base: Use expert judgment and statistical methods to correct the standard data;

[0084] Real-time production data feedback: Update the standard data through a data closed-loop to improve its dynamic adaptability;

[0085] Accordingly: Store the credible standard data Sdata for subsequent data cleaning and feature extraction;

[0086] S12. Adopt an anomaly detection method that combines statistical methods and machine learning to detect outliers far from the normal distribution range to improve robustness. Specifically:

[0087] S1201. For the credible standardized data Sdata = X = {x1, x2, …, x i , …, x N}, calculate the mean and standard deviation:

[0088] ;

[0089] ;

[0090] where N represents the total number of element data in the standardized data; x i represents the i-th element data in the standardized data;

[0091] According to the 3σ principle, define the normal data range: X f = {x i || |x i - μ| ≤ 3σ};

[0092] where x i represents the i-th data sample; μ represents the mean; σ represents the standard deviation; X f represents the data after removing outliers;

[0093] S1202. The statistical method has limitations when the data distribution is non - normal. Therefore, the IsolationForest (IF) algorithm is further used to detect high - dimensional outliers: ;

[0094] In the formula, S(x i ) represents the outlier score; E(h(x i )) represents the average path length of the data point x i in the decision tree; c(n) represents the normalization factor of the data scale; h(x i ) represents the path length of the data point x i in the Isolation Tree, that is, the number of splits experienced by the data point before being isolated;

[0095] Accordingly: Combining the 3σ method and the IF method to improve the accuracy and generalization ability of outlier detection;

[0096] S13. There may be missing values (NaN) during the data collection process. If not processed, it will affect subsequent analysis. Therefore, a dynamic interpolation strategy is adopted to select the optimal filling method according to the data type:

[0097] S1301. Filling of numerical data:

[0098] Mean filling (applicable to data with normal distribution): x miss = μ;

[0099] K - Nearest Neighbor Interpolation (KNN) (applicable to non - normal data): ;

[0100] Among them, x miss represents the missing value; represents the j - th K - nearest neighbor data point; K represents the number of nearest neighbor data points considered during interpolation;

[0101] S1302. Filling of categorical data:

[0102] Mode filling (applicable to categorical variables): ;

[0103] In the formula, P(y i |X) represents the probability that the missing value belongs to the category y i ; P(X|y i ) represents the feature probability under the known category y i ; P(y i ) represents the prior probability; P(X) represents the probability of the occurrence of the observed data X;

[0104] Accordingly: Adopting a dynamic selection strategy to automatically select the filling method according to the data type to improve flexibility;

[0105] S14. Use the Min - Max normalization and Z - Score normalization methods to perform data normalization processing, and output the standardized data SⅠdata;

[0106] S15. Generate a first - order standardized matching code CSⅠ for the standardized data SⅠdata, and its generation process is as follows:

[0107] S1501. Convert the standardized data SⅠdata into binary data Ⅰdata;

[0108] S1502. Select a random number αⅠ, and calculate the standard first - order element UⅠ = H(Ⅰdata||αⅠ) ∈ IR;

[0109] Among them, IR is a predefined ring, IR = Z q [x] / (x τ + 1); q is a predefined large prime number (for example, q = 12289); τ is a predefined parameter; || is a concatenation operation; H is a predefined hash function; the random number αⅠ is 128 - bit;

[0110] S1503. Calculate the first - order matching element code CMⅠ = SpⅠ·UⅠ mod (q,x τ + 1);

[0111] Among them, SpⅠ is a randomly generated short polynomial, and its coefficients are randomly selected from ;

[0112] It should be noted that in the polynomial ring IR = Z q [x] / (x τ + 1), if the coefficients a of the polynomial i satisfy |a i | ≤ B, then f(x) is called a short polynomial;

[0113] S1504. Generate the first - order standardized matching code CSⅠ = (random number αⅠ, first - order matching element code CMⅠ).

[0114] S2. Use sensor networks, Internet of Things (IoT) devices, combined with the ERP / MES system to collect production data, and use edge computing for data pre - processing: filter out outliers and noise data, improve data quality, and perform data format conversion to ensure consistent data structure for easy analysis, obtain real - time data Rdata, and generate a second - order standardized matching code CSⅡ for the real - time data Rdata, and its generation process is as follows:

[0115] S21. Convert the real - time data Rdata into binary data Ⅱdata;

[0116] S22. Select a random number αⅡ, and calculate the standard second-order element UⅡ = H(Ⅱdata||αⅡ) ∈ IR;

[0117] Among them, the random number αⅡ is 128 bits;

[0118] S23. Calculate the second-order matching element code CMⅡ = SpⅡ·UⅡ mod (q,x τ +1);

[0119] Among them, SpⅡ is a randomly generated short polynomial, and its coefficients are randomly selected from ; Accordingly: Calculate the predefined matching analysis factor AF, AF = SpⅠ·SpⅡ mod (q,x τ +1);

[0120] S24. Generate the second-order standardized matching code CSⅡ = (random number αⅡ, second-order matching element code CMⅡ).

[0121] S3. The comparison and analysis module combines the weighted Euclidean distance to quantify the quality deviation, constructs a dynamic threshold model and a fuzzy membership function for deviation classification and determination. The specific implementation process is as follows:

[0122] S31. Extract {CSⅡ = (αⅡ, CMⅡ), Rdata} and {CSⅠ = (αⅠ, CMⅠ), SⅠdata}, and perform matching on the standardized data SⅠdata and the real-time data Rdata. The matching process is as follows:

[0123] S3101. Convert the standardized data SⅠdata and the real-time data Rdata into binary string data dataⅢ and dataⅣ respectively;

[0124] S3102. Calculate the matching first-order element MⅠ = H(dataⅢ||αⅠ) and the matching second-order element MⅡ = H(dataⅣ||αⅡ);

[0125] S3103. Calculate the first-order analysis code AⅠ = CMⅠ·CMⅡ mod (q,x τ +1);

[0126] S3104. Calculate the second-order analysis code AⅡ = AF·MⅠ·MⅡ mod (q,x τ +1);

[0127] Among them, AF is the predefined matching analysis factor, AF = SpⅠ·SpⅡ mod (q,x τ +1);

[0128] S3105. If AⅠ = AⅡ mod (q,x τIf it matches (i.e., +1), the match is successful, indicating that there is standard consistency between the standardized data SⅠdata and the real-time data Rdata; otherwise, an alarm is issued.

[0129] S32. Obtain the standard data vector Qs = [s1, s2, …, s i , …, s n , and obtain the real-time quality data vector Qr = [r1, r2, …, r i , …, r n ;

[0130] Among them, r i represents the real-time measurement value of the i-th quality parameter; s i represents the corresponding standard index value; n represents the number of quality indicators;

[0131] S33. In order to quantify the difference between the real-time data and the standard data, the weighted Euclidean distance is used as the deviation metric: ;

[0132] Among them, ΔQ represents the overall quality deviation amount, and the larger the value, the more serious the deviation from the standard; w i represents the weight of the i-th quality parameter, reflecting the importance of this parameter in the overall quality, and satisfies the normalization condition: ;

[0133] S34. Considering the inherent fluctuations in the production process, a fixed threshold is likely to lead to misjudgment or missed judgment. Therefore, a dynamic threshold model is constructed using historical quality deviation data, and the formula is as follows:

[0134] ;

[0135] In the formula, T represents the dynamic judgment threshold of the current quality deviation; represents the mean value of the historical deviation data; represents the standard deviation of the historical deviation data, reflecting the fluctuation range of the deviation value; λ represents the sensitivity adjustment coefficient, and its value is set by the quality management personnel according to the actual risk tolerance;

[0136] S35. Use fuzzy logic to construct a membership function to classify and judge the real-time deviation:

[0137] ;

[0138] In the formula, μ(Q) represents the membership degree of the quality deviation, and its value range is between [0, 1]; k represents the slope parameter of the membership function, which is used to adjust the steepness of the function.

[0139] S4. Based on the quality deviation ΔQ and its corresponding fuzzy membership μ(Q) (ranging from 0 to 1), combined with trend information, an early warning index that integrates static deviation and dynamic trend is constructed, and an intelligent early warning is realized by using the methods of adaptive threshold adjustment and fuzzy inference. The specific implementation process is as follows:

[0140] S41. Based on the fuzzy membership μ(Q) (reflecting the current quality deviation degree) and the change trend R of the quality deviation over time, an early warning index I is defined warn as:

[0141] I warn = μ(Q)·(1 + β·R);

[0142] ;

[0143] where μ(Q) represents the fuzzy membership of the current quality deviation; β represents the trend weight coefficient (dimensionless), which is used to adjust the influence degree of the trend factor in the overall early warning index; R represents the change rate of the quality deviation, that is, the trend of the quality deviation over time; ΔQ(t) represents the quality deviation at the current moment; ΔQ(t - Δt) represents the quality deviation at the previous moment; Δt represents the sampling time interval;

[0144] S42. According to the value of the early warning index I warn , set the early warning level, and use the grading formula:

[0145] ;

[0146] where η1, η2, η3 represent the early warning demarcation thresholds, and the initial values are determined according to the distribution of historical early warning data and satisfy: 0 ≤ η1 < η2 < η3 ≤ 1;

[0147] S43. In order to adapt to the dynamic changes in the production process, an adaptive threshold adjustment mechanism is introduced, and the early warning demarcation value is dynamically updated according to the statistics of the early warning index I warn over a period of history;

[0148] Set the new threshold as: ;

[0149] where represents the i-th new threshold; represents the mean value of the historical early warning index, that is, the average early warning level under normal production conditions; represents the sensitivity adjustment coefficient, which is set to different values according to the different early warning levels i to adjust the response sensitivity to deviation fluctuations; represents the standard deviation of the historical early warning index, reflecting the index fluctuation range;

[0150] Accordingly, this mechanism ensures that when long-term stability or abnormal fluctuations are detected, the warning level boundary value can be automatically updated, avoiding misjudgment or missed judgment problems that may be caused by fixed thresholds;

[0151] S44. Output results according to the warning level and automatically trigger corresponding response measures:

[0152] Normal state (I warn <η1): No action is required, and normal monitoring continues;

[0153] Minor warning (η1 ≤ I warn <η2): Record the warning information, send a prompt to the operator, and suggest paying attention to the production parameters;

[0154] Medium warning (η2 ≤ I warn <η3): Issue a clear warning signal and require the initiation of an auxiliary detection process;

[0155] Severe warning (Iwarn ≥ η3): Immediately trigger an emergency warning, automatically interrupt the process, and notify the quality control and maintenance teams to intervene quickly.

[0156] S5. An intelligent optimization suggestion generation method based on multi-source data integration. Through real-time data and warning information, an intelligent quality optimization suggestion generation model is constructed to provide accurate and dynamic optimization measures for the production process, and continuously improve the decision-making quality through closed-loop feedback. The specific implementation steps are as follows:

[0157] S51. Collect and integrate the following data:

[0158] The comparison results of real-time quality data and standard data: Obtain the real-time values r i of each quality parameter and the standard values s i (after normalization and weighting), as well as the overall deviation amount ΔQ and the fuzzy membership degree μ(Q);

[0159] The output results of intelligent warning: Obtain the comprehensive warning index I warn (including trend information R and warning level);

[0160] Historical optimization data and experience rules: Including but not limited to historical optimization measures, the parameter adjustment ranges corresponding to each measure, the quality improvement effects brought about after implementation, and the experience rules summarized by experts (constituting the knowledge base);

[0161] S52. To generate optimization suggestions, construct an intelligent quality optimization suggestion generation model based on the deviations of each key parameter and the warning level. Let a score S j be generated for each potential optimization measure j:

[0162] ;

[0163] In the formula, S j represents the comprehensive optimization suggestion score for measure j. The higher the score, the more suitable the measure is as an optimization intervention; γ ij represents the sensitivity weight of the i-th parameter to optimization measure j. This parameter is determined based on historical data and expert evaluation, reflecting the contribution of parameter i to the improvement effect of optimization measure j; represents the adjustment coefficient associated with the warning indicator, reflecting the response ability of measure j to the overall abnormal state. This parameter is determined through historical data and expert knowledge;

[0164] S53. To further verify and quantify the potential improvement effects of each optimization measure, a non-linear prediction model is used to estimate the quality improvement that each measure may bring. Let the expected improvement effect for measure j be ΔAQ j :

[0165] ;

[0166] In the formula, ΔAQ j represents the expected quality improvement amplitude after the implementation of optimization measure j. This indicator can help determine whether the overall deviation ΔQ can be reduced to a reasonable range after the implementation of the measure; θ j represents the maximum improvement potential of measure j, which is obtained by statistical analysis of historical optimization records. That is, under ideal conditions, the maximum improvement amount that measure j can bring; κ j represents the improvement sensitivity coefficient of measure j, reflecting the speed of response of the measure to score changes. This parameter can be obtained by fitting historical data. The larger the value, the more sensitive the effect is to score changes;

[0167] Accordingly, an exponential function is used to model the relationship between the score and the improvement effect, ensuring that the improvement effect increases slowly when the score is low, while the improvement effect tends to saturate when the score is high. This non-linear characteristic is more in line with the marginal effect law of actual optimization measures;

[0168] S54. According to the scores S j of each measure and the predicted improvement effect ΔAQ j , one or more best optimization measures are selected after sorting, and a detailed optimization suggestion report is generated. The content of the report includes but is not limited to:

[0169] Suggested optimization measures;

[0170] The specific adjustment amplitude for each measure (obtained by synthesizing ΔAQ j and historical data);

[0171] Expected improvement effect and possible risk warnings.

[0172] S6. To continuously improve the effectiveness of optimization suggestions, a closed-loop feedback mechanism is introduced to ensure the effective implementation of optimization measures and continuously improve the quality management strategy:

[0173] Implementation feedback: Record the actual effect data after each implementation of the optimization measure;

[0174] Model update: Compare the actual effect with the predicted effect, and continuously update the sensitivity weight γ using the feedback data ij , adjust the coefficient δ j , the maximum improvement potential θ j and the sensitivity coefficient κ j ;

[0175] Knowledge base expansion: Accumulate new optimization cases and experiences into the knowledge base to provide richer data support for subsequent decision-making;

[0176] Accordingly: Through closed-loop feedback, self-learning and self-adaptation are realized, and the model parameters are continuously optimized to improve the accuracy and response efficiency of suggestions.

[0177] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made without departing from the spirit of the present invention within the scope of knowledge possessed by those skilled in the art.

Claims

1. A quality management method based on standard data, characterized in that: The specific implementation steps include the following: S1. Construct standard data and perform mixed anomaly detection. Use dynamic interpolation strategy to handle missing values. Generate standardized data after normalization, and use polynomial ring operation to generate first-order standardized matching code. S2. Collect real-time data and perform data preprocessing: filter outliers, noise data, convert data formats, and generate second-order standardized matching codes; S3, matching the standardized data with the real-time data, using polynomial ring operations to verify the consistency of data standards, combining weighted Euclidean distance to quantify quality deviations, and constructing a dynamic threshold model and fuzzy membership function to perform deviation classification judgment; The determination process of deviation classification is as follows: A1. Obtain the standard data vector Qs=[s1,s2,…,s i ,…,s n ], obtain the real-time quality data vector Qr=[r1,r2,…,r i ,…,r n ]; Among them, r i represents the real-time measurement value of the i-th quality parameter; s i Indicates the corresponding standard index value; n indicates the number of quality indicators; A2. Using weighted Euclidean distance as deviation measure: ； Where ΔQ represents the overall quality deviation; w i Represents the weight of the i-th quality parameter, satisfying the normalization condition: ; A3. Use historical quality deviation data to build a dynamic threshold model: ； In the formula, T represents the dynamic judgment threshold of the current quality deviation; Represents the mean of historical deviation data; represents the standard deviation of historical deviation data; λ represents the sensitivity adjustment coefficient; A4. Use fuzzy logic to construct a membership function and make graded judgments on real-time deviations: ； Wherein, μ(Q) represents the degree of membership of mass deviation, and its value range is between [0,1]; k represents the slope parameter of the membership function; S4. By combining the fuzzy membership and change trend of the current deviation, a dynamic early warning indicator is constructed. The classification threshold is dynamically adjusted according to the mean, standard deviation and sensitivity adjustment coefficient of the historical early warning indicator, the early warning level boundary value is updated, and the graded response measures are triggered according to the indicator value; S5. Based on the intelligent optimization suggestion generation method of multi-source data integration, an intelligent quality optimization suggestion generation model is constructed to generate scores for each potential optimization measure, and a nonlinear prediction model is used to quantify the expected improvement effect, screen the best measures and generate an optimization suggestion report; S6. Record the effect data of optimization measures through a closed-loop feedback mechanism, and compare the predicted and actual results to update the model parameters and knowledge base.

2. A quality management method based on standard data according to claim 1, characterized in that: The detection process of hybrid anomaly detection is as follows: S21, obtain standardized data Sdata=X={x1,x2,…,x i ,…,x N }, calculate the mean μ and standard deviation σ: ； ； Where N represents the total number of element data in the standardized data; x i Represents the i-th element data in the standardized data; S22. According to the 3σ principle, define the normal data range: X f ={x i ||x i -μ|≤3σ}; Among them, x i represents the i-th data sample; μ represents the mean; σ represents the standard deviation; X f Represents the data after removing outliers; S23, using the isolation forest algorithm to detect high-dimensional outliers: ; In the formula, S(x i ) represents the anomaly score; E(h(x i )) represents the data point x i The average path length in the decision tree; c(n) represents the normalization factor for the data size; h(x i ) represents the data point x i Path length among isolated trees in an isolation forest.

3. A quality management method based on standard data according to claim 1, characterized in that: The generation process of the first-order standardized matching code is as follows: S31, converting the standardized data SⅠdata into binary data Ⅰdata; S32, select a random number αⅠ, and calculate the standard first-order element UⅠ=H(Ⅰdata||αⅠ)∈IR; Among them, IR is a predefined ring, IR=Z q [x] / (x τ +1); q is a predefined large prime number; τ is a predefined parameter; || is a concatenation operation; H is a predefined hash function; the random number αⅠ is 128 bits; S33, calculate the first-order matching element code CMⅠ=SpⅠ·UⅠ mod (q,x τ +1); Among them, SpⅠ is a randomly generated short polynomial with coefficients in Randomly selected from S34, generate a first-order standardized matching code CSⅠ=(random number αⅠ, first-order matching meta-code CMⅠ).

4. A quality management method based on standard data according to claim 3, characterized in that: The generation process of the second-order standardized matching code is as follows: S41, converting the real-time data Rdata into binary data IIdata; S42, select a random number αⅡ, and calculate the standard second-order element UⅡ=H(Ⅱdata||αⅡ)∈IR; Among them, the random number αⅡ is 128 bits; S43, calculate the second-order matching element code CMⅡ=SpⅡ·UⅡ mod (q,x τ +1); Among them, SpⅡ is a randomly generated short polynomial with coefficients in Randomly selected from S44, generate a second-order standardized matching code CSⅡ=(random number αⅡ, second-order matching meta-code CMⅡ).

5. A quality management method based on standard data according to claim 4, characterized in that: The matching process for matching standardized data with real-time data is as follows: S51, converting the standardized data SⅠdata and the real-time data Rdata into binary character string data dataⅢ and dataⅣ respectively; S52, calculate the matching first-order element MⅠ=H(dataⅢ||αⅠ) and the matching second-order element MⅡ=H(dataⅣ||αⅡ); S53, calculate the first-order analytical code AⅠ=CMⅠ·CMⅡ mod (q,x τ +1); S54, calculate the second-order analytical code AⅡ=AF·MⅠ·MⅡ mod (q,x τ +1); Where AF is the predefined matching resolution factor, AF=SpⅠ·SpⅡ mod (q,x τ +1); S55, if AⅠ=AⅡ mod (q,x τ +1), the match is successful, indicating that the standardized data SⅠdata and the real-time data Rdata are consistent with the standard; otherwise, an alarm is issued.

6. A quality management method based on standard data according to claim 1, characterized in that: The process of dynamically adjusting the classification threshold is as follows: S71. Based on the fuzzy membership μ(Q) and the changing trend R of quality deviation over time, define the early warning indicator I warn For: I warn =μ(Q)·(1+β·R); ； Among them, μ(Q) represents the fuzzy membership of the current mass deviation; β represents the trend weight coefficient; R represents the rate of change of the mass deviation; ΔQ(t) represents the mass deviation at the current moment; ΔQ(t-Δt) represents the mass deviation at the previous moment; Δt represents the sampling time interval; S72. According to the early warning indicator I warn The value of is used to set the warning level, and the classification formula is used: ； Among them, η1, η2, and η3 represent the warning demarcation thresholds, and the initial values ​​are determined based on the distribution of historical warning data and meet the following conditions: 0≤η1<η2<η3≤1; S73, introduce adaptive threshold adjustment mechanism, based on the warning indicators in a certain period of history warn Statistics, dynamically update the warning threshold value, that is, set the new threshold value as: ; in, represents the i-th new threshold; Represents the mean value of historical early warning indicators; It represents the sensitivity adjustment coefficient, and different values ​​are set according to different warning levels i; Represents the standard deviation of the historical early warning indicator.

7. A quality management method based on standard data according to claim 1, characterized in that: The implementation process of the intelligent optimization suggestion generation method based on multi-source data integration is as follows: S81. Collect and integrate the following data: comparison results of real-time quality data and standard data, intelligent early warning output results, historical optimization data and empirical rules; S82. Build an intelligent quality optimization suggestion generation model based on the deviation and warning level of each key parameter, and generate a score S for each potential optimization measure j. j : ; In the formula, S j represents the comprehensive optimization recommendation score for measure j; γ ij represents the sensitivity weight of the i-th parameter to the optimization measure j; represents the adjustment coefficient associated with the early warning indicator; r i represents the real-time measurement value of the i-th quality parameter; s i Indicates the corresponding standard index value; n indicates the number of quality indicators; S83. Use a nonlinear prediction model to estimate the quality improvement that each measure may bring, and calculate the expected improvement effect ΔAQ for measure j j : ; In the formula, ΔAQ j It represents the expected quality improvement after the implementation of optimization measure j; θ j represents the maximum improvement potential of measure j; κ j represents the improvement sensitivity coefficient of measure j; S84. According to the score of each measure S j And predict the improvement effect ΔAQ j , after sorting, select the best optimization measures and generate a detailed optimization recommendation report.

Images