Knowledge Graph-Based Glass Product Quality Traceability and Prediction System

Abstract

This invention discloses a glass product quality traceability and prediction system based on a knowledge graph. The system includes a data acquisition module, a data preprocessing module, a knowledge graph construction module, a knowledge update and fusion module, a model construction module, a model correction module, a quality traceability module, a trend prediction and early warning module, and a recycling strategy optimization module. The system collects multi-dimensional data on batches of glass during the production, use, and recycling stages to construct a knowledge graph with glass as the core entity. It integrates materials science theoretical models with real-time data to achieve full life-cycle traceability and future state prediction of glass product quality. The system supports visualized display of quality data, anomaly early warning, and personalized recycling strategy recommendations. It has advantages such as combining data-driven and theoretical models, strong adaptive optimization capabilities, and cross-batch knowledge fusion, thereby improving the intelligence level and resource utilization efficiency of glass product quality management.

Description

Technical Field

[0001] This invention belongs to the field of intelligent manufacturing and industrial knowledge graph application technology, and specifically discloses a glass product quality traceability and prediction system based on knowledge graph. Background Technology

[0002] As a crucial industrial and building material, glass's quality directly impacts building safety, industrial performance, and lifespan. Traditional glass quality management relies heavily on sampling inspections and manual recording during production, resulting in fragmented data with weak correlations, hindering comprehensive lifecycle quality tracking and analysis. Particularly during the use and recycling stages, the lack of quantitative modeling and predictive capabilities regarding the gradual changes in glass performance leads to delayed warnings of quality degradation and inability to optimize recycling strategies, resulting in resource waste and economic losses.

[0003] Currently, some enterprises have attempted to introduce information systems for quality data management, but most systems still have the following limitations: First, the data dimensions are limited, failing to integrate multi-source heterogeneous data from multiple stages such as production, use, and recycling; second, they lack semantic association capabilities, with data existing in isolation, making it difficult to support in-depth causal analysis and knowledge discovery; third, the predictive models mostly rely on pure data-driven or pure theory-driven approaches, with the former having poor interpretability and weak extrapolation capabilities, and the latter being unable to adapt to the complex changes in actual production, resulting in limited prediction accuracy; fourth, they lack decision support for the recycling stage, making it impossible to formulate personalized, cross-batch recycling process optimization plans based on quality degradation trends.

[0004] Therefore, it is necessary to invent a knowledge graph-based glass product quality traceability and prediction system to solve the above problems. Summary of the Invention

[0005] To overcome the aforementioned deficiencies in the prior art, this invention provides a glass product quality traceability and prediction system based on a knowledge graph. By collecting multi-dimensional data on batches of glass during the production, use, and recycling stages, a knowledge graph with batches of glass as the core entity is constructed. This system integrates materials science theoretical models with real-time data to achieve full lifecycle traceability and future state prediction of glass product quality. The system supports visualized display of quality data, anomaly warnings, and personalized recycling strategy recommendations, effectively solving the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a glass product quality traceability and prediction system based on knowledge graphs, specifically including: a data acquisition module, a data preprocessing module, a knowledge graph construction module, a knowledge update and fusion module, a model construction module, a model correction module, a quality traceability module, a trend prediction and early warning module, and a recycling strategy optimization module.

[0007] The data acquisition module is used to collect multi-dimensional data of a single batch of glass during the production, use and recycling stages at preset time intervals; the multi-dimensional data includes production basic data, color data, oxidation degree data, defect data, service life data, hardness data and environmental data; the batch of glass is defined as a collection of glass produced under the same production cycle, the same raw material formula and the same process parameters;

[0008] The data preprocessing module is used to clean, standardize, and align multi-dimensional data according to time series.

[0009] The knowledge graph construction module is used to build a knowledge graph with batch glass as the core entity based on the processed multi-dimensional data, which includes various entity types and relationship types;

[0010] The knowledge update and fusion module is used to realize real-time updates of the knowledge graph and cross-batch knowledge fusion;

[0011] The model building module is used to establish gradual mathematical models based on materials science theory;

[0012] The model correction module is used to trigger a correction mechanism based on the deviation between real-time multi-dimensional data and theoretical values, and to adjust parameters using multiple batches of data;

[0013] The quality traceability module is used to visually display quality data across the entire supply chain;

[0014] The trend prediction and early warning module is used to predict future quality changes and provide early warnings of anomalies based on a gradual mathematical model.

[0015] The recycling strategy optimization module is used to recommend recycling times based on predictions of future quality changes.

[0016] Preferably, the production data includes raw material composition, melting temperature, molding pressure, and annealing time; color data includes color difference value and light transmittance; oxidation degree data includes surface oxygen content and oxide layer thickness; defect data includes defect type, defect quantity, and maximum defect size; service life data includes mechanical stability and chemical stability; hardness data includes Vickers hardness and surface Rockwell hardness; and environmental data includes ambient temperature, humidity, ultraviolet intensity, and pH of the recycling environment.

[0017] Preferably, the knowledge graph construction module performs the following process: taking batch glass as the core entity, constructing a knowledge network containing 6 types of entities and 8 types of core relationships, and adding attributes to each entity.

[0018] Preferably, the knowledge update and fusion module specifically performs the following process:

[0019] Data is collected at preset time intervals, preprocessed, and then incrementally updated through a graph database to complete the mapping of attributes and relationships.

[0020] In cross-batch quality analysis, process parameters and quality indicators of glass from different batches are correlated. The similarity of process parameters is calculated using a similarity algorithm, and the data is combined with the variation patterns of quality indicators to form a general quality impact rule library. The rules in the quality impact rule library contain the correspondence between changes in process parameters and changes in quality indicators. Redundant rules in the rule library are periodically removed.

[0021] Preferably, the model building module constructs a gradual mathematical model based on oxidation kinetics and fatigue damage theory, including an oxidation degree model, a hardness decay model, and a comprehensive quality index model, and performs initial calibration of the model parameters based on historical data.

[0022] Preferably, the model correction module specifically performs the following process:

[0023] Deviation calculation: Calculate the deviation between the real-time quality indicator value and the theoretical value. If the deviation exceeds the set deviation threshold, initiate multi-batch comparison and correction.

[0024] Multi-batch comparison and correction: Extract data of the same batches that meet the preset conditions of process similarity with the target batch from the knowledge graph, calculate the deviation coefficient between the actual mean and the theoretical value, and use it to correct the key parameters of the model;

[0025] Dynamic iteration: Repeated deviation calculation and multiple batch comparison correction processes are performed according to the collection interval to ensure that the model fit meets the preset requirements.

[0026] Preferably, the trend prediction and early warning module specifically performs the following process:

[0027] Trend prediction: Based on a gradual mathematical model, predict the quality indicator values ​​at future time points and output them in a visual form;

[0028] Anomaly warning: Compare the predicted value of the quality indicator with the preset safety threshold. If the value exceeds the threshold, a system warning will be triggered.

[0029] Preferably, the recycling strategy optimization module specifically performs the following process:

[0030] Recovery timing prediction: Based on the prediction results of the comprehensive quality index output by the comprehensive quality index model and the set recovery threshold, the optimal recovery time window is determined;

[0031] Personalized process adjustment: Based on the quality index prediction results obtained from the trend prediction and early warning module, the type, proportion and severity of defects are analyzed and inferred. Related processes are queried from the knowledge graph, and a customized recycling plan is output.

[0032] Cross-batch optimization: Extract historical high-loss batch data similar to the target batch, query optimization records, calculate fit, and recommend or adjust recycling process parameters.

[0033] The technical effects and advantages of this invention are as follows:

[0034] 1. Through multi-dimensional data collection and knowledge graph construction, the entire lifecycle data association and visual traceability from raw materials, production processes, usage environment to recycling and disposal has been realized, breaking down traditional information silos and improving data utilization efficiency;

[0035] 2. A gradual mathematical model is constructed based on materials science theory, and dynamic correction and hybrid-driven optimization are performed in combination with real-time data, which improves the prediction accuracy of changes in glass product quality and the timeliness of anomaly warning.

[0036] 3. By predicting the comprehensive quality index and recommending personalized recycling strategies, the timing and process of recycling are precisely planned, reducing recycling costs and improving resource reuse rate;

[0037] 4. By utilizing knowledge graphs and similarity algorithms, knowledge fusion and automatic mining and updating of quality impact rules across batches of data were achieved, enhancing the system's adaptability and generalization capabilities. Attached Figure Description

[0038] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0039] Figure 1 This is a schematic diagram of the module connection of the present invention.

[0040] Figure 2 This is a schematic diagram of the overall process flow of the present invention. Detailed Implementation

[0041] 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.

[0042] This invention provides, for example Figure 1The knowledge graph-based glass product quality traceability and prediction system shown includes a data acquisition module, a data preprocessing module, a knowledge graph construction module, a knowledge update and fusion module, a model construction module, a model correction module, a quality traceability module, a trend prediction and early warning module, and a recycling strategy optimization module.

[0043] The overall process flow of the present invention is as follows: Figure 2 As shown, it includes:

[0044] Data acquisition module: Collects multi-dimensional data of batches of glass during the production, use and recycling stages at preset time intervals; the multi-dimensional data includes production basic data, color data, oxidation degree data, defect data, service life data, hardness data and environmental data; the batch of glass is defined as a collection of glass produced under the same production cycle, the same raw material formula and the same process parameters;

[0045] Furthermore, in the above technical solution, the production basic data includes raw material composition, melting temperature, molding pressure, and annealing time; color data includes color difference value and light transmittance; oxidation degree data includes surface oxygen content and oxide layer thickness; defect data includes defect type, defect quantity, and maximum defect size; service life data includes mechanical stability and chemical stability; hardness data includes Vickers hardness and surface Rockwell hardness; and environmental data includes ambient temperature, humidity, ultraviolet intensity, and pH of the recycling environment.

[0046] Furthermore, the multi-dimensional data collection process is as follows: Using a single glass batch as the target, the data covers the entire process from production to usage to recycling. Data is collected at preset time intervals: once every 2 hours during the production phase, once every 7 days during the usage phase, and once every hour during the recycling phase. The collection method is as follows:

[0047] Basic production data is obtained directly from the production end, and no further collection is required.

[0048] Color data was collected using a spectrophotometer; surface oxygen content was collected using an X-ray photoelectron spectroscopy (XPS) instrument; oxide layer thickness was collected using an ellipsometry; defect data was collected using a machine vision inspection system; mechanical stability was determined using a falling ball impact tester; chemical stability was obtained by immersing the sample glass in a constant temperature and humidity immersion chamber; hardness data was measured using a hardness tester; temperature, humidity, and ultraviolet intensity were obtained using an integrated temperature, humidity, and ultraviolet sensor; and the pH of the recycling environment was measured using a pH meter.

[0049] It should be further noted that the samples collected by the data acquisition module during the usage phase were unused glass from the same batch, and the collected samples were placed in an outdoor environment.

[0050] Data preprocessing module: performs cleaning, standardization, and time-series alignment on multi-dimensional data;

[0051] Furthermore, in the above technical solution, the cleaning includes filling missing values ​​with the median, removing outliers based on the 3σ principle, and removing duplicate collection records; the standardization is to normalize the collected data through Min-Max; the time alignment method is to align data collected at different frequencies using "batch ID + timestamp" as the primary key, use linear interpolation to fill in missing time point data, and unify the timestamp to ISO 8601 format.

[0052] Knowledge graph construction module: Constructs a knowledge graph with batch glass as the core entity based on the processed multi-dimensional data;

[0053] Furthermore, in the above technical solution, the specific execution process of the knowledge graph construction module is as follows: taking batch glass as the core entity, a knowledge network containing 6 types of entities and 8 types of core relationships is constructed, and attributes are added to each entity.

[0054] Furthermore, the six types of entities are as follows:

[0055] Batch of glass, raw materials: SiO2, Na2O, process parameters: melting temperature, forming pressure, annealing time, quality indicators: color difference value, light transmittance, oxide layer thickness, defect parameters, impact strength attenuation rate, acid and alkali corrosion resistance rate, Vickers hardness, surface Rockwell hardness, environmental factors: ambient temperature, humidity, ultraviolet intensity, pH of the recycling environment, abnormal events: crack propagation, accelerated oxidation.

[0056] The eight core relationships are as follows:

[0057] The relationships are as follows: "Inclusion" of "Batch Glass - Raw Materials", "Association" of "Batch Glass - Process Parameters", "Affected by" of "Quality Indicators - Environmental Factors", "Causing Changes" of "Abnormal Events - Quality Indicators", "Correspondence" of "Batch Glass - Quality Indicators", "Influence" of "Process Parameters - Quality Indicators", "Inducing" of "Environmental Factors - Abnormal Events", and "Adaptation" of "Raw Materials - Process Parameters".

[0058] Batch glass attributes include batch code, production time and quantity; quality indicator attributes include collection time, value and change rate; raw material attributes include component ratio; process parameter attributes include deviation between set value and actual value; environmental factor attributes include collection time and value fluctuation range; and abnormal event attributes include occurrence time and duration of impact.

[0059] Knowledge Update and Fusion Module: Used to enable real-time updates of the knowledge graph and cross-batch knowledge fusion;

[0060] Furthermore, in the above technical solution, the knowledge update and fusion module specifically performs the following process:

[0061] Data is collected at preset time intervals, preprocessed, and then incrementally updated through a graph database to complete the mapping of attributes and relationships.

[0062] Furthermore, the specific implementation method of incremental update of knowledge graph is as follows: after each multi-dimensional data collection is completed, the multi-dimensional data processed by the data preprocessing module is automatically converted into the attributes of the corresponding entities or the relationships between entities in the knowledge graph, and the incremental update of knowledge graph is realized through the Cypher statement of graph database, such as Neo4j.

[0063] In cross-batch quality analysis, process parameters and quality indicators of glass from different batches are correlated. The similarity of process parameters is calculated using a similarity algorithm, and the data is combined with the variation patterns of quality indicators to form a general quality impact rule library. The rules in the quality impact rule library contain the correspondence between changes in process parameters and changes in quality indicators. Redundant rules in the rule library are periodically removed.

[0064] Furthermore, when performing cross-batch glass quality analysis, the "process parameter-quality index" correlation data of different batches of glass is extracted from the knowledge graph, and the process parameter similarity (sim) is calculated using a cosine similarity algorithm. The formula is as follows: , where x i y i These are the standardized values ​​of the i-th process parameter in the two batches, where n is the number of process parameters. The general quality impact rule library is formed by combining the change patterns of quality indicators. Each rule in the rule library contains the corresponding relationship of "change in process parameter - change in quality indicator", such as a 10°C increase in melting temperature → 2% increase in Vickers hardness. Redundant rules with a confidence level of <90% are removed every 50 new rules.

[0065] Model building module: used to build gradual mathematical models based on materials science theory;

[0066] Furthermore, in the above technical solution, the model building module constructs a gradual mathematical model based on oxidation kinetics and fatigue damage theory, including an oxidation degree model, a hardness decay model, and a comprehensive quality index model, and performs initial calibration of the model parameters based on historical data.

[0067] Furthermore, the specific execution process of the knowledge update and fusion module is as follows:

[0068] The specific execution process of the model construction module is as follows: Based on glass materials science theory, including oxidation kinetics and fatigue damage theory, a theoretical gradual mathematical model of multi-dimensional quality indicators changing over time is constructed, specifically including:

[0069] Oxidation degree model: The parabolic oxidation law is adopted, and the formula is as follows: Where d(t) is the oxide layer thickness at time t (unit: nm), k p d0 is the oxidation rate constant (unit: nm² / h), d0 is the initial oxide layer thickness (unit: nm), and t is the usage time (unit: h).

[0070] Hardness decay model: Based on material fatigue theory and considering the effects of environmental corrosion, the formula is as follows: Where HV(t) is the Vickers hardness at time t (unit: HV), HV0 is the initial Vickers hardness (unit: HV), and α is the time decay coefficient (unit: h). -1 ), β is the environmental impact coefficient, and γ is the comprehensive environmental corrosion index at time t, which is calculated by AHP weighting after standardization of environmental factors and takes the value [0, 1];

[0071] Comprehensive Quality Index Model: This model uses a weighted fusion of standardized values ​​from various quality indicators, and the formula is as follows: Where Q(t) is the overall quality index at time t, taking values ​​[0, 1], w i Let q be the weight of the i-th quality indicator. i (t) represents the standardized value of the i-th quality indicator;

[0072] When building the model, k is based on the average data of no less than 50 historical batches. p Initial calibration is performed on α, β, and γ.

[0073] Model correction module: Triggers a correction mechanism based on the deviation between real-time multi-dimensional data and theoretical values, and adjusts parameters using multiple batches of data;

[0074] Furthermore, in the above technical solution, the model correction module specifically performs the following process:

[0075] Deviation calculation: Calculate the deviation between the real-time quality indicator value and the theoretical value. If the deviation exceeds the set deviation threshold, initiate multi-batch comparison and correction.

[0076] Furthermore, the deviation between the real-time quality index value and the theoretical value is expressed by the formula Δq. i =|q i (t)-q it (t)|Calculation, where Δq is in the formula i q represents the difference between the real-time quality index value and the theoretical value. i (t) represents the standardized value of the quality index at time t, q it (t) represents the theoretical value of the quality index output by the gradual mathematical model; the set deviation threshold is the standard deviation of the historical deviation data.

[0077] Multi-batch comparison and correction: Extract data of the same batches that meet the preset conditions of process similarity with the target batch from the knowledge graph, calculate the deviation coefficient between the actual mean and the theoretical value, and use it to correct the key parameters of the model;

[0078] Furthermore, the preset conditions are a similarity of ≥85% to the raw material formula of the target batch and a deviation of ≤10% in process parameters. The correction process specifically involves: calculating the average actual value of the quality indicators at corresponding time points for similar batches. The calculation formula is: In the formula, m represents the quantity of the same batch, and q represents the quantity of the same batch. i，j (t) represents the value of the i-th indicator in the j-th batch; calculate the batch deviation coefficient. q ith When (t)=0, take δ i =0.05; The batch deviation coefficient is substituted into the theoretical model to correct key parameters, such as k. p α, β, the formula is X np X represents the corrected parameters, and X represents the parameters before correction.

[0079] Dynamic iteration: Repeated deviation calculation and multiple batch comparison correction processes are performed according to the collection interval to ensure that the model fit meets the preset requirements.

[0080] Furthermore, the dynamic iteration process is as follows: repeat "deviation calculation - similar batch extraction - parameter correction" at the data collection interval to ensure that the model fits the actual trend R² ≥ 0.9. When the fit R² < 0.9, expand the extraction range of similar batches, for example, raw material similarity ≥ 80% and process deviation ≤ 15%.

[0081] Furthermore, the formula for calculating the goodness of fit R² is as follows: , y in the formula ak Let y be the k-th actual observed value of a certain quality indicator. bk Let y be the model prediction value corresponding to the k-th actual observation value of a certain quality indicator. ck Let M be the average of the actual observed values ​​of M quality indicators, where M is the number of quality indicators collected.

[0082] Quality traceability module: Supports forward and reverse traceability, and provides a visual display of quality data across the entire supply chain;

[0083] Furthermore, in the above technical solution, the specific execution process of the quality traceability module is as follows:

[0084] Forward tracing:

[0085] Receive a batch code input by the user, wherein the batch code is a unique number for a batch of glass;

[0086] The knowledge graph was used to query the source of the raw materials for this batch: supplier, test report number, warehousing time, production process parameters: melt temperature profile, molding pressure fluctuation, annealing time deviation, quality indicators at each time point: color difference value, light transmittance and other change data, and usage environment data: temperature and humidity, ultraviolet curve;

[0087] Visual output using timeline charts, relationship diagrams, and data tables, with support for clicking on nodes to view details;

[0088] Reverse tracing:

[0089] Receive abnormal information input by the user, such as the type of abnormal indicator and the time of occurrence t;

[0090] By querying the relationship between "abnormal events - quality indicators" and "quality indicators - environment / process" through the knowledge graph, the abnormal influencing factors can be located, such as a sudden increase in humidity or insufficient annealing time.

[0091] Query upstream data of influencing factors (equipment number, worker number, raw material batch, environmental monitoring fault record);

[0092] Output a root cause report, including a logical chain such as "insufficient annealing time → low initial hardness → sudden drop in hardness at time t" and data evidence (annealing time deviation record, initial hardness report).

[0093] Trend prediction and early warning module: predicts future quality changes and provides early warnings of anomalies based on a gradual mathematical model;

[0094] Furthermore, in the above technical solution, the trend prediction and early warning module specifically performs the following process:

[0095] Trend prediction: Based on a gradual mathematical model, predict the quality indicator values ​​at future time points and output them in a visual form;

[0096] Furthermore, the specific process of trend prediction is as follows: Obtain the corrected or hybrid optimized gradual mathematical model; input the current time t0 and three future unit time nodes: t1=t0+T, t2=t0+2T, t3=t0+3T, where T is the data collection interval; calculate the predicted value of the quality index for each node; output as a visualized trend chart.

[0097] Anomaly warning: Compare the predicted value of the quality indicator with the preset safety threshold. If the value exceeds the threshold, the system will issue a warning and notify the administrator via pop-up window and SMS. At the same time, the high-risk event will be marked in the knowledge graph.

[0098] Furthermore, the predicted values ​​of the quality indicators are compared with the corresponding preset safety thresholds; if they exceed the thresholds, the prediction is marked as "abnormal"; Δq is calculated. i If Δq iIf the deviation exceeds twice the threshold, it is marked as "real-time anomaly"; triggering an alert: the system pops up an anomaly information (type, current / predicted value, threshold, time), and sends an SMS notification to the administrator (including batch and anomaly level); it is marked as "high-risk anomaly event" in the knowledge graph, and associated with the batch, indicator and alert time.

[0099] Recycling strategy optimization module: Recommends recycling time based on future quality change prediction results.

[0100] Furthermore, in the above technical solution, the recycling strategy optimization module specifically performs the following process:

[0101] Recovery timing prediction: Based on the prediction results of the comprehensive quality index output by the comprehensive quality index model and the set recovery threshold, the optimal recovery time window is determined;

[0102] Furthermore, the optimal recovery time window is [t] rec -T rec / 2,t rec +T rec / 2], where t rec For the first time, the overall quality index Q(t) drops to the set recycling threshold Q. rec Time t rec T rec The historical recycling period, the recycling threshold Q rec (Usually 0.6, but custom versions are supported);

[0103] Personalized process adjustment: Based on the quality index prediction results obtained from the trend prediction and early warning module, the type, proportion, and severity of defects are analyzed and inferred. Related processes are queried from the knowledge graph, and customized recycling solutions are output. For example, when the crack proportion is 15%, it is recommended to "ultrasonic cleaning (500W, 15min) → heat repair (600℃, 30min) → re-inspection"; when the oxide layer exceeds the standard, it is recommended to "pickling (pH=2.5, 10min) → deionized water rinsing".

[0104] Cross-batch optimization: Extract historical high-loss batch data similar to the target batch, query optimization records, calculate fit, and recommend or adjust recycling process parameters.

[0105] The specific implementation method for cross-batch optimization is as follows: Extract historical high-loss batches (energy consumption exceeding the average by 15% or utilization rate <80%) that have "defect similarity ≥90% and raw material similarity ≥85%" with the target batch; query historical batch optimization records (process parameters, changes in energy consumption / utilization rate); calculate the fit (raw materials 0.4, defects 0.3, environment 0.3) with weighted averages. If the fit is ≥80%, the optimized process is reused; if it is <80%, the parameters are adjusted and the effect is predicted.

[0106] Furthermore, the formula for calculating the fitness level is: Fitness Level = 0.4 × S 原 +0.3×S 缺 +0.3×S 环 In the formula, S 原 For raw material similarity; S 缺 Similarity between defect type and proportion; S 环 For similarity of usage environment;

[0107] Furthermore, the similarity S of the raw materials 原 The calculation formula is: In the formula, A i and B i These represent the percentage content of the i-th raw material in the two batches, respectively, where n is the number of types of raw material components; S 原 The closer the value is to 1, the more similar the recipes are;

[0108] The defect similarity S 缺 The calculation formula is: In the formula, D A and D B These are the defect distribution vectors for the target batch and historical batches, respectively. Each element of the defect distribution vector represents the percentage of a certain defect type (e.g., the percentage of cracks in the total number of defects), min(D A D B ) is the minimum value of the proportion of the two vectors for each defect type, max(D) A D B S represents the maximum of the proportions of the two vectors for each defect type. 缺 The closer the value is to 1, the more similar the defect composition;

[0109] The environmental similarity S 环 The calculation formula is: , in the formula and These are the average values ​​of the two batches on the r-th environmental factor, such as average temperature, average humidity, and average UV intensity. r S represents the weight of the r-th environmental factor, allocated based on the magnitude of the environmental factor's influence on the prediction; 环 The closer the value is to 1, the more similar the defect composition;

[0110] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A glass product quality traceability and prediction system based on a knowledge graph, characterized in that, The system comprises a data collection module, a data preprocessing module, a knowledge graph construction module, a knowledge updating and fusion module, a model construction module, a model correction module, a quality tracing module, a trend prediction and early warning module, and a recycling strategy optimization module. The data collection module collects multi-dimensional data of the batch glass at the production, use, and recycling stages at preset time intervals; the multi-dimensional data includes production basic data, color data, oxidation degree data, defect data, service life data, hardness data, and environmental data; the batch glass is defined as a glass set produced under the same production cycle, same raw material formula, and same process parameters; The data preprocessing module performs cleaning, standardization, and time sequence alignment processing on the multi-dimensional data; The knowledge graph construction module constructs a knowledge graph with the batch glass as the core entity based on the processed multi-dimensional data; The knowledge updating and fusion module is used to realize real-time updating and cross-batch knowledge fusion of the knowledge graph; The model construction module is used to establish a gradual mathematical model based on the theory of materials science; The model correction module triggers a correction mechanism through the deviation of real-time multi-dimensional data and theoretical values, and adjusts parameters using multi-batch data; The quality tracing module supports forward and reverse tracing, and visually displays the whole-chain quality data; The trend prediction and early warning module predicts future quality changes and issues abnormality warnings based on the gradual mathematical model; The recycling strategy optimization module recommends recycling time based on the prediction results of future quality changes; The knowledge updating and fusion module specifically performs the following processes: Collect data at preset time intervals, and perform incremental updating of the knowledge graph through a graph database after preprocessing, to complete the updating and mapping of attributes and relationships; When analyzing cross-batch quality, extract the process parameters and quality index correlation data of different batches of glass, calculate the process parameter similarity through a similarity algorithm, and combine the quality index change rule to form a general quality influence rule library; The rules in the quality influence rule library include the corresponding relationship between process parameter changes and quality index changes; the rule library is periodically checked for redundant rules; The model construction module constructs gradual mathematical models including oxidation degree models, hardness decay models, and comprehensive quality index models based on oxidation kinetics and fatigue damage theory, and initially calibrates model parameters based on historical data; The model correction module specifically performs the following processes: Deviation calculation: calculate the deviation of real-time quality index values and theoretical values; if the deviation exceeds the set deviation threshold, start multi-batch comparison correction; Multi-batch comparison correction: extract the same batch data with process similarity meeting the preset conditions from the knowledge graph, calculate the deviation coefficient of the actual mean value and the theoretical value, and use it to correct the key parameters of the model; Dynamic iteration: repeat the deviation calculation and multi-batch comparison correction processes at the collection interval to ensure that the model fitting degree meets the preset requirements.

2. The knowledge graph based glass product quality traceability and prediction system of claim 1, wherein: The production base data includes raw material composition, melting temperature, molding pressure and annealing time; color data includes color difference value and light transmittance; oxidation degree data includes surface oxygen element content and oxidation layer thickness; defect data includes defect type, defect number and maximum defect size; service life data includes mechanical stability and chemical stability; hardness data includes Vickers hardness and surface Rockwell hardness; environmental data includes use environment temperature, humidity, ultraviolet intensity and recovery environment pH.

3. The knowledge graph based glass product quality traceability and prediction system of claim 1, wherein: The knowledge graph construction module specifically performs the following process: taking a batch of glass as the core entity, a knowledge network containing 6 types of entities and 8 core relationships is constructed, and attributes are added to each entity.

4. The knowledge graph based glass product quality traceability and prediction system of claim 1, wherein: The trend prediction and early warning module specifically performs the following process: Trend prediction: based on a gradual change mathematical model, the quality index value at a future time node is predicted, and the output is in a visual form; Abnormal early warning: comparing the predicted value of the quality index value with the preset safety threshold, if it exceeds, the system will trigger an alarm.

5. The knowledge graph based glass product quality traceability and prediction system of claim 1, wherein: The recycling strategy optimization module specifically performs the following process: Recycling time prediction: based on the prediction result of the comprehensive quality index output by the comprehensive quality index model and the set recycling threshold, the best recycling time window is determined; Individualized process adjustment: based on the quality index prediction result obtained by the trend prediction and early warning module, the type, proportion and severity of defects are analyzed and inferred, the associated process is queried from the knowledge graph, and a customized recycling scheme is output; Cross-batch optimization: extract historical high-loss batch data similar to the target batch, query optimization records, calculate the adaptation degree and recommend or adjust the recycling process parameters.

Images