Product life cycle coding traceability and quality safety management system based on big data linkage

Abstract

The application discloses a product life cycle coding traceability and quality safety management and control system based on big data linkage, relates to the technical field of product management and control, and comprises a big data fusion construction module, a quality risk monitoring module, a management and control decision execution module and a visual display module. The big data fusion construction module collects data at various stages and interactive data, divides stage clusters by using a graph neural network, and constructs a sub-digital model and an overall digital model by using a DBSCAN clustering algorithm. The quality risk monitoring module monitors running conditions in real time, identifies risk stages and evaluates grades, determines associated stages and close stages, the management and control decision execution module formulates a management and control scheme according to risk characteristics and executes the scheme, and the visual display module displays the overall digital model and the management and control situation. The system realizes product life cycle data integration and analysis, improves quality risk monitoring precision, makes management and control decisions more targeted and effective, and guarantees product quality safety.

Description

Technical Field

[0001] This invention relates to the field of product management technology, and more specifically, to a product lifecycle coding traceability and quality and safety management system based on big data linkage. Background Technology

[0002] In today's complex and ever-changing market environment, a vast amount of diverse data is generated throughout the product lifecycle, from design, raw material procurement, manufacturing, warehousing and logistics, sales to after-sales feedback. This data is both independent and closely interconnected. Traditional product data management models struggle to effectively integrate and analyze such massive amounts of lifecycle data, leading to disjointed product information and difficulties in traceability. In sensitive and safety-critical industries such as food and pharmaceuticals, once quality issues arise, it is difficult to quickly and accurately pinpoint the problem and trace its source. This not only poses health risks to consumers but also causes significant reputational and economic losses for companies. Furthermore, the operational monitoring methods employed by companies at each stage... The current system is relatively lagging behind, making it impossible to identify potential quality risks in a real-time and accurate manner. For example, in the manufacturing process, relying solely on manual sampling and routine statistical process control makes it difficult to detect hidden and sudden quality problems. These problems may continue to escalate until they cause serious consequences. Moreover, existing control decisions are often one-sided, only addressing the specific stage where the problem occurs and ignoring the interrelationships between different stages of the product lifecycle. For instance, when adjusting sales strategies, the adaptability to the production end is not considered, and after-sales improvement measures are not fed back to the design front-end for optimization, leading to repeated problems and failing to fundamentally solve quality issues or improve product quality and enterprise operational efficiency.

[0003] Therefore, existing technologies suffer from difficulties in data integration, inaccurate risk monitoring, and one-sided control decisions, making it difficult to effectively guarantee the quality, safety, and stable operation of products throughout their entire lifecycle. Summary of the Invention

[0004] To overcome the problems of existing technologies, such as difficulties in data integration, inaccurate risk monitoring, and one-sided control decisions, which make it difficult to effectively ensure the quality, safety, and stable operation of products throughout their entire life cycle, this invention discloses a product life cycle coding traceability and quality and safety control system based on big data linkage, which can effectively solve the above-mentioned technical problems.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0006] A product lifecycle coding traceability and quality and safety control system based on big data linkage includes:

[0007] The big data fusion module is used to collect comprehensive data from all stages of the product lifecycle, including product design data, raw material procurement data, production and manufacturing data, warehousing and logistics data, sales data, and after-sales feedback data, as well as interaction data between stages, including information transmission records and collaborative operation processes. Through graph neural networks, the product lifecycle is divided into multiple stage clusters based on stage data. Using the DBSCAN clustering algorithm, sub-digital models corresponding to each stage cluster are constructed based on the interaction data, and then integrated to construct an overall digital model of the product lifecycle.

[0008] The quality risk monitoring module is used to monitor the operational status of each stage of the product lifecycle in real time, including design progress, production quality indicators, logistics and transportation status, and sales feedback; it identifies stages where the operational status deviates from the normal range as risk stages and assesses the risk level of risk stages; based on the overall digital model of the product lifecycle and the risk level, it determines multiple related stages of the risk stage and the degree of correlation between each related stage.

[0009] The control and management decision execution module formulates and executes control and management decision schemes for the risk stage and its associated stages based on the characteristic data and risk level of the risk stage, as well as the characteristic data and degree of correlation of the associated stages. The schemes include design modifications, production process adjustments, logistics route changes, sales strategy optimization, and after-sales measure improvements.

[0010] The visualization module is used to display the overall digital model of the product lifecycle, as well as the implementation status and effect feedback of the control and management decision-making scheme.

[0011] Preferably, the big data fusion construction module includes:

[0012] The data acquisition unit acquires static and dynamic data at each stage of the product lifecycle, as well as interaction data between stages, through multiple channels.

[0013] The stage cluster division unit transforms the data of each stage into feature vectors, uses graph neural networks to learn the correlation weights between feature vectors, and divides similar stages into the same stage cluster based on the weights.

[0014] The model building unit constructs interaction feature vectors based on interaction data, uses the DBSCAN clustering algorithm to perform cluster analysis on the interaction feature vectors within the stage clusters, constructs sub-digital models, and finally merges them to form the overall digital model of the product lifecycle.

[0015] Preferably, the quality risk monitoring module includes:

[0016] The monitoring unit continuously collects real-time operational data at each stage of the product lifecycle;

[0017] The risk identification unit trains a risk identification model based on historical operational data and a preset risk indicator system. Real-time monitoring data is input into the model to identify risk stages and determine risk levels.

[0018] The correlation analysis unit determines the correlation stage based on the overall digital model of the product lifecycle and the risk level of the risk stage, and determines the degree of correlation according to pre-set rules.

[0019] Preferably, the control decision execution module includes:

[0020] The correlation degree conversion unit, according to specific conversion rules, converts the degree of correlation at the correlation stage into the corresponding degree of risk impact;

[0021] The decision generation unit uses the random forest algorithm to train and generate decision models by taking the feature data, risk level and risk impact degree of the risk stage and the associated stage as input, and outputs control decision schemes for each stage.

[0022] The execution control unit is used to implement the control and management decision schemes output by the decision generation unit, regulate the risk stages and related stages of the product life cycle, and track the execution effect.

[0023] Preferably, the process of constructing feature vectors and dividing stage clusters by the stage cluster division unit is as follows: extracting and encoding features from the data of each stage to form stage feature vectors; constructing a graph structure through a graph neural network, wherein nodes are stage feature vectors, learning the association weights between nodes, and clustering stages with high weights into the same stage cluster;

[0024] The process of constructing a sub-digital model by the model building unit is as follows: constructing interactive feature vectors based on interactive data, calculating the similarity of interactive feature vectors within the stage cluster, and merging stages according to density connectivity using the DBSCAN clustering algorithm to form clusters and construct the sub-digital model.

[0025] Preferably, the process of the risk identification unit training the risk identification model is as follows: using historical operating data as input and the corresponding normal or risk label as output, a convolutional neural network model is trained; the rules for the association analysis unit to determine the association stage and the degree of association are as follows: if the risk level of the risk stage is high, the directly preceding and following stages are determined to be highly associated, and the indirectly preceding and following stages are determined to be moderately associated; if the risk level of the risk stage is low, the directly preceding and following stages are determined to be moderately associated, and the indirectly preceding and following stages are determined to be lowly associated.

[0026] Preferably, the conversion rule of the correlation tightness conversion unit is as follows: high correlation corresponds to high risk impact, medium correlation corresponds to medium risk impact, and low correlation corresponds to low risk impact.

[0027] The training data for the decision generation unit to train the decision model consists of various historical control and management decision schemes implemented at each stage of the product lifecycle, as well as corresponding stage characteristic data and risk levels.

[0028] Preferably, the stage cluster partitioning unit learns the association weights using a graph neural network by calculating the association weights between the feature vectors of each stage through a message passing mechanism in the graph structure.

[0029] The method for calculating the similarity of interaction feature vectors in the model building unit is as follows: the Jaccard similarity algorithm is used to calculate the similarity between interaction feature vectors at each stage.

[0030] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention collects and integrates data from all stages and interactions of the product lifecycle through a big data fusion construction module, and constructs an overall digital model using graph neural networks and DBSCAN clustering algorithms, effectively solving the problem of difficult data integration and achieving comprehensive data integration and accurate analysis; the quality risk monitoring module uses a risk identification model trained with historical data to monitor the operational status of each stage in real time, accurately identify risk stages and levels, and determine the related stages and their degree of closeness based on the overall digital model, overcoming the shortcomings of inaccurate risk monitoring in existing technologies; the control decision execution module formulates and executes control decision schemes using random forest algorithms based on the characteristic data of risk stages and related stages, so that control decisions cover multiple related stages, avoiding the problem of one-sided control decisions, thereby comprehensively ensuring the quality safety and stable operation of the product throughout its entire lifecycle, reducing quality risks, and improving product quality control efficiency and enterprise operational benefits. Attached Figure Description

[0031] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other embodiments can be derived from the provided drawings without creative effort.

[0032] Figure 1 This is a system structure diagram of the present invention. Detailed Implementation

[0033] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the scope of this patent.

[0034] To better illustrate this embodiment, some parts in the accompanying drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions.

[0035] It will be understood by those skilled in the art that certain well-known structures and their descriptions may be omitted in the accompanying drawings.

[0036] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0037] Example

[0038] A product lifecycle coding traceability and quality and safety control system based on big data linkage includes:

[0039] The big data fusion module is used to collect comprehensive data from all stages of the product lifecycle, including product design data, raw material procurement data, production and manufacturing data, warehousing and logistics data, sales data, and after-sales feedback data, as well as interaction data between stages, including information transmission records and collaborative operation processes. Through graph neural networks, the product lifecycle is divided into multiple stage clusters based on stage data. Using the DBSCAN clustering algorithm, sub-digital models corresponding to each stage cluster are constructed based on the interaction data, and then integrated to construct an overall digital model of the product lifecycle.

[0040] The quality risk monitoring module is used to monitor the operational status of each stage of the product lifecycle in real time, including design progress, production quality indicators, logistics and transportation status, and sales feedback; it identifies stages where the operational status deviates from the normal range as risk stages and assesses the risk level of risk stages; based on the overall digital model of the product lifecycle and the risk level, it determines multiple related stages of the risk stage and the degree of correlation between each related stage.

[0041] The control and management decision execution module formulates and executes control and management decision schemes for the risk stage and its associated stages based on the characteristic data and risk level of the risk stage, as well as the characteristic data and degree of correlation of the associated stages. The schemes include design modifications, production process adjustments, logistics route changes, sales strategy optimization, and after-sales measure improvements.

[0042] The visualization module is used to display the overall digital model of the product lifecycle, as well as the implementation status and effect feedback of the control and management decision-making scheme.

[0043] The big data fusion construction module includes:

[0044] The data acquisition unit acquires static and dynamic data at each stage of the product lifecycle, as well as interaction data between stages, through multiple channels.

[0045] The stage cluster division unit transforms the data of each stage into feature vectors, uses graph neural networks to learn the correlation weights between feature vectors, and divides similar stages into the same stage cluster based on the weights.

[0046] The model building unit constructs interaction feature vectors based on interaction data, uses the DBSCAN clustering algorithm to perform cluster analysis on the interaction feature vectors within the stage clusters, constructs sub-digital models, and finally merges them to form the overall digital model of the product lifecycle.

[0047] The quality risk monitoring module includes:

[0048] The monitoring unit continuously collects real-time operational data at each stage of the product lifecycle;

[0049] The risk identification unit trains a risk identification model based on historical operational data and a preset risk indicator system. Real-time monitoring data is input into the model to identify risk stages and determine risk levels.

[0050] The correlation analysis unit determines the correlation stage based on the overall digital model of the product lifecycle and the risk level of the risk stage, and determines the degree of correlation according to pre-set rules.

[0051] The control decision execution module includes:

[0052] The correlation degree conversion unit, according to specific conversion rules, converts the degree of correlation at the correlation stage into the corresponding degree of risk impact;

[0053] The decision generation unit uses the random forest algorithm to train and generate decision models by taking the feature data, risk level and risk impact degree of the risk stage and the associated stage as input, and outputs control decision schemes for each stage.

[0054] The execution control unit is used to implement the control and management decision schemes output by the decision generation unit, regulate the risk stages and related stages of the product life cycle, and track the execution effect.

[0055] The process of constructing feature vectors and dividing stage clusters by the stage cluster division unit is as follows: extracting and encoding features from the data of each stage to form stage feature vectors; constructing a graph structure through a graph neural network, where nodes are stage feature vectors; learning the association weights between nodes; and clustering stages with high weights into the same stage cluster.

[0056] The process of constructing a sub-digital model by the model building unit is as follows: constructing interactive feature vectors based on interactive data, calculating the similarity of interactive feature vectors within the stage cluster, and merging stages according to density connectivity using the DBSCAN clustering algorithm to form clusters and construct the sub-digital model.

[0057] The process of training the risk identification model by the risk identification unit is as follows: using historical operating data as input and the corresponding normal or risk label as output, a convolutional neural network model is trained; the rules for determining the correlation stage and the degree of correlation by the correlation analysis unit are as follows: if the risk level of the risk stage is high, the directly preceding and following stages are determined to be highly correlated, and the indirectly preceding and following stages are determined to be moderately correlated; if the risk level of the risk stage is low, the directly preceding and following stages are determined to be moderately correlated, and the indirectly preceding and following stages are determined to be lowly correlated.

[0058] The conversion rule for the correlation tightness conversion unit is as follows: high correlation corresponds to high risk impact, medium correlation corresponds to medium risk impact, and low correlation corresponds to low risk impact.

[0059] The training data for the decision generation unit to train the decision model consists of various historical control and management decision schemes implemented at each stage of the product lifecycle, as well as corresponding stage characteristic data and risk levels.

[0060] The stage cluster partitioning unit learns the association weights using a graph neural network by calculating the association weights between the feature vectors of each stage through a message passing mechanism in the graph structure.

[0061] The method for calculating the similarity of interaction feature vectors in the model building unit is as follows: the Jaccard similarity algorithm is used to calculate the similarity between interaction feature vectors at each stage.

[0062] Please see Figure 1 During the product design phase, the data acquisition unit, through interfaces with the company's internal computer-aided design (CAD) software and product data management (PDM) system, collects static data such as product design drawings, design parameters, and design change records in real time, as well as dynamic data on the design progress. For example, when a mechanical manufacturing company is designing a new product, the system collects the updated information of the design model from the CAD software every hour, including key parameters such as part dimensions and material selection.

[0063] During the raw material procurement stage, the system interfaces with the Supplier Relationship Management (SRM) and Enterprise Resource Planning (ERP) systems to obtain static data such as raw material purchase order information, supplier information, and raw material inspection reports, as well as dynamic data such as raw material price fluctuations and supply delivery dates. Taking an electronics manufacturing company as an example, the system collects the execution status of raw material purchase orders from the SRM system every day, including information such as order quantity and delivery date.

[0064] During the manufacturing phase, sensors installed on production equipment collect real-time data during the production process, such as dynamic data like equipment operating parameters, ambient temperature and humidity, and production cycle time. At the same time, static data such as production task allocation, production process documents, and quality inspection results are obtained from the Manufacturing Execution System (MES). For example, in the welding workshop of an automobile manufacturing plant, sensors collect operating parameters such as current and voltage of the welding equipment once per second and transmit the data to the system.

[0065] During the warehousing and logistics phase, static data such as inventory quantity, inventory location, and goods entry and exit records are obtained through the warehouse management system (WMS), as well as dynamic data such as the real-time location, transportation route, and transportation time of logistics vehicles. For example, when a logistics company is transporting electronic products, the system collects vehicle location information every minute through the vehicle-mounted GPS device and monitors the transportation status of goods in real time in conjunction with the logistics transportation plan.

[0066] During the sales phase, static data such as sales order information, sales channel information, and sales prices are obtained from the company's internal customer relationship management system (CRM) and e-commerce platform, as well as dynamic data such as sales trends and changes in market demand. For example, a home appliance manufacturer uses its CRM system to collect customer information, product models, sales regions, and other data on sales orders every day to analyze market demand dynamics.

[0067] During the after-sales feedback phase, static data such as customer complaint information, product failure feedback, and product repair records are obtained from after-sales management systems, such as call center systems and online customer service systems, as well as dynamic data such as customer satisfaction survey results and after-sales processing progress. For example, after receiving a customer complaint, a mobile phone manufacturer records the complaint content and processing process in real time and collects customer satisfaction survey results regularly to analyze product quality issues later.

[0068] The collected data from each stage are preprocessed, including data cleaning and data standardization. Then, feature extraction algorithms, such as principal component analysis (PCA), are used to extract features from the data at each stage to form a stage feature vector. For example, for the data in the production and manufacturing stage, after standardizing the data such as equipment operating parameters, production cycle time, and quality inspection results, the main feature components are extracted by PCA to form a stage feature vector with a length of 10.

[0069] A graph neural network (GNN) model is used to construct a relationship graph of each stage of the product lifecycle. The feature vectors of each stage are used as nodes in the graph to build an initial graph structure. During training, a message passing mechanism allows nodes to exchange feature information and learn the relationship weights between nodes. For example, in the system of a machinery manufacturing company, the product design stage, raw material procurement stage, and production manufacturing stage are used as nodes. Through GNN model training, it is found that the relationship weight between the product design stage and the raw material procurement stage is high, indicating that these two stages are closely related in the product lifecycle. Based on the relationship weight, similar stages are divided into the same stage cluster. For example, the product design stage, raw material procurement stage, and production process design stage are divided into one stage cluster because the relationship weights between them are all higher than a set threshold.

[0070] Based on the interaction data between each stage, such as information transmission records and collaborative operation processes, an interaction feature vector is constructed. For example, between the product design and raw material procurement stages, information such as the raw material specification requirements sent by the design department to the procurement department and the raw material availability information fed back by the procurement department to the design department are recorded. This interaction information is quantified and encoded to form an interaction feature vector.

[0071] The DBSCAN clustering algorithm is used to perform cluster analysis on the interaction feature vectors within stage clusters. First, the similarity between interaction feature vectors is calculated, which can be done using the Jaccard similarity algorithm. Then, based on the principles of similarity and density connectivity, the interaction feature vectors are clustered into different subclasses. For example, within a certain stage cluster, the DBSCAN clustering algorithm divides the interaction feature vectors into two categories: one representing closely collaborative interaction feature vectors and the other representing general collaboration interaction feature vectors. Based on the clustering results, sub-digital models are constructed, and finally, they are merged to form an overall digital model of the product lifecycle. The overall digital model presents the various stages of the product lifecycle and their interaction relationships in a graphical way, providing a basis for quality risk monitoring and control decisions.

[0072] The monitoring unit acquires operational data at each stage of the product lifecycle in real time through the data interface of the big data integration module. For example, in the manufacturing stage, it collects key quality indicator data such as product dimensional accuracy and surface roughness from the sensors of the production equipment every minute; in the logistics and transportation stage, it acquires environmental data such as the temperature and humidity of the goods from the monitoring system of the logistics vehicles every half hour.

[0073] The collected real-time data is stored in a distributed database and preprocessed, including data cleaning and data completion. For example, for temperature data during the logistics and transportation phase, if data is missing at a certain point in time, linear interpolation is used to complete the data to ensure data integrity.

[0074] Based on historical operational data and a pre-defined risk indicator system, a risk identification model is trained. The historical operational data includes normal operation data and data on risk situations at each stage. The risk indicator system is customized according to the characteristics of different products and industries. For example, in the food production industry, key risk indicators include microbial content and additive content. Taking a food production company as an example, production data from the past year is collected, including data on normal production batches and batches recalled due to excessive microbial levels. This data is used as input, and the corresponding normal or risk label is used as output to train a convolutional neural network (CNN) model. During the training process, by adjusting the model parameters, such as the size of the convolutional kernel and the number of neurons, the model can accurately identify the risk stage and risk level.

[0075] Real-time monitoring data is input into a trained risk identification model, which outputs the risk status and risk level for each stage. For example, on the production line of an electronics manufacturing company, when real-time production data is input into the risk identification model, the model determines that a certain production stage is a high-risk stage with a risk level of 3 (the highest level is 5). At the same time, the system records the risk identification results, including the time of risk occurrence, the risk stage, and the risk level, for subsequent analysis and processing.

[0076] Based on the overall digital model of the product lifecycle and the risk level of each risk stage, the associated stages of the risk stage are determined. For example, when a certain production and manufacturing stage is identified as a high-risk stage, the raw material procurement stage, the previous production process stage, and the subsequent warehousing stage, which are directly related to this production stage, are determined as associated stages based on the relationship between each stage in the overall digital model.

[0077] The degree of correlation between related stages is determined according to pre-set rules. For example, for a high-risk production stage, the directly preceding and following stages are determined to be highly correlated (correlation degree value of 0.9), and the indirectly preceding and following stages are determined to be moderately correlated (correlation degree value of 0.6); for a low-risk production stage, the directly preceding and following stages are determined to be moderately correlated (correlation degree value of 0.6), and the indirectly preceding and following stages are determined to be poorly correlated (correlation degree value of 0.3). The correlation degree value is used for control decision making, indicating the degree to which the related stage is affected by the risk stage.

[0078] According to the pre-set conversion rules, the degree of association of the association stage is converted into the corresponding degree of risk impact. For example, high association corresponds to high risk impact (risk impact value is 0.8), medium association corresponds to medium risk impact (risk impact value is 0.5), and low association corresponds to low risk impact (risk impact value is 0.2). The risk impact value indicates the magnitude of the impact that the association stage may be affected by during the risk propagation process.

[0079] Using the random forest algorithm, a decision-making model is trained with the characteristic data, risk level, and risk impact degree of the risk stage and related stages as input. The training data includes various historical control decision schemes implemented at each stage of the product life cycle, as well as the corresponding stage characteristic data and risk level. For example, in a machinery manufacturing company, control decision schemes adopted for different risk situations in the past two years are collected, such as process adjustment schemes and raw material procurement strategy change schemes, as well as the corresponding risk stage characteristic data, such as production quality indicators, raw material quality data, and risk level. These data are used as input, and the effectiveness evaluation of the control decision schemes, such as whether they effectively reduce risks and improve product quality, is used as output to train the random forest model. By adjusting the model parameters, such as the number of decision trees and the depth of the trees, the model can accurately output control decision schemes for different input situations.

[0080] The characteristic data and risk level of the current risk stage, as well as the characteristic data and risk impact of related stages, are input into the trained decision model. The model outputs control and management decision schemes for each stage. For example, when a high-risk situation occurs in a certain production stage, and the characteristic data and risk impact of related stages are input into the model, the control and management decision scheme output by the model includes: adjusting the production process in the risky production stage, such as adjusting processing parameters and replacing processing equipment; strengthening the supplier quality audit in the directly related raw material procurement stage; and increasing the frequency of inventory quality spot checks in the subsequent warehousing stage.

[0081] The execution control unit regulates the risk stages and related stages of the product lifecycle based on the control decision scheme output by the decision generation unit. For example, in the manufacturing stage, the system sends control commands to the production equipment to adjust production parameters; in the raw material procurement stage, it sends a notification to the procurement department to require strengthened supplier quality audits. At the same time, the system tracks the execution effect and collects data in real time during the execution process, such as changes in production quality indicators and inventory quality sampling results. For example, after implementing the production process adjustment plan, the system collects the quality inspection data of the produced products every hour to analyze whether the production quality has been improved after the adjustment.

[0082] Based on the execution results data, the effectiveness of the control and management decision-making plan is evaluated. If the execution results do not achieve the expected goals, such as the risk not being effectively reduced or the product quality not being significantly improved, the system will call back the decision generation unit to generate an adjusted control and management decision-making plan based on the new data and situation, and execute it again. If the product quality indicators still do not meet the standards after the first production process adjustment, the system will combine the new quality inspection data and production operation data to retrain the decision model, generate a new production process adjustment plan, such as further optimizing the processing parameters or replacing the processing equipment with higher precision equipment, implement it again, and track the effect until the risk is effectively controlled and the product quality meets the requirements.

[0083] The visualization module presents a digital model of the entire product lifecycle through an intuitive graphical interface. Using 3D modeling technology, it presents each stage of the product lifecycle and their interactions in 3D graphics. For example, in the system of an automobile manufacturing company, it displays the complete lifecycle process of a car from design, parts procurement, production assembly, vehicle testing, sales delivery to after-sales service. The degree of connection between each stage is indicated by lines of different colors and line widths. Users can rotate and zoom the graphics with the mouse to view detailed information about each stage.

[0084] The system provides real-time updates on the implementation and feedback of control decisions, presenting information such as the progress of control measures and changes in performance indicators at each stage through charts, graphs, and reports. For example, in the production and manufacturing stage, a bar chart compares production quality indicators before and after process adjustments, while a line chart shows the trend of inventory quality sampling pass rate over time. Similarly, for the after-sales feedback stage, the system displays changes in customer satisfaction survey results, such as using a pie chart to show the percentage of customers with different satisfaction levels. This visually reflects the impact of control decisions on product quality and customer satisfaction. Users can filter by time period, product model, stage, and other criteria to view the implementation and feedback of corresponding control decisions, providing a basis for decision adjustments and continuous product quality improvement.

[0085] The same or similar labels correspond to the same or similar parts;

[0086] The terms used to describe positional relationships in the accompanying drawings are for illustrative purposes only and should not be construed as limiting this patent.

[0087] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. For those skilled in the art, other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all implementation methods here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the claims of the present invention.

Claims

1. A product lifecycle coding traceability and quality and safety control system based on big data linkage, characterized in that: include: The big data fusion module is used to collect comprehensive data from all stages of the product lifecycle, including product design data, raw material procurement data, production and manufacturing data, warehousing and logistics data, sales data, and after-sales feedback data, as well as interaction data between stages, including information transmission records and collaborative operation processes. Through graph neural networks, the product lifecycle is divided into multiple stage clusters based on stage data. Using the DBSCAN clustering algorithm, sub-digital models corresponding to each stage cluster are constructed based on the interaction data, and then integrated to construct an overall digital model of the product lifecycle. The quality risk monitoring module is used to monitor the operational status of each stage of the product lifecycle in real time, including design progress, production quality indicators, logistics and transportation status, and sales feedback; it identifies stages where the operational status deviates from the normal range as risk stages and assesses the risk level of risk stages. Based on the overall digital model of the product lifecycle and the risk level, identify multiple related stages of the risk stage, as well as the degree of correlation between each related stage; The control and management decision execution module formulates and executes control and management decision schemes for the risk stage and its associated stages based on the characteristic data and risk level of the risk stage, as well as the characteristic data and degree of correlation of the associated stages. The schemes include design modifications, production process adjustments, logistics route changes, sales strategy optimization, and after-sales measure improvements. The visualization module is used to display the overall digital model of the product lifecycle, as well as the execution status and effect feedback of the control and management decision-making scheme; The big data fusion construction module includes: The data acquisition unit acquires static and dynamic data at each stage of the product lifecycle, as well as interaction data between stages, through multiple channels. The stage cluster division unit transforms the data of each stage into feature vectors, uses graph neural networks to learn the correlation weights between feature vectors, and divides similar stages into the same stage cluster based on the weights. The model building unit constructs interaction feature vectors based on interaction data, uses the DBSCAN clustering algorithm to perform cluster analysis on the interaction feature vectors within the stage clusters, constructs sub-digital models, and finally integrates them to form the overall digital model of the product lifecycle. The process of constructing feature vectors and dividing stage clusters by the stage cluster division unit is as follows: extracting and encoding features from the data of each stage to form stage feature vectors; constructing a graph structure through a graph neural network, where nodes are stage feature vectors; learning the association weights between nodes; and clustering stages with high weights into the same stage cluster. The process of constructing a sub-digital model by the model building unit is as follows: constructing interactive feature vectors based on interactive data, calculating the similarity of interactive feature vectors within the stage cluster, and merging stages according to density connectivity using the DBSCAN clustering algorithm to form clusters and construct the sub-digital model.

2. The control system according to claim 1, characterized in that, The quality risk monitoring module includes: The monitoring unit continuously collects real-time operational data at each stage of the product lifecycle; The risk identification unit trains a risk identification model based on historical operational data and a preset risk indicator system. Real-time monitoring data is input into the model to identify risk stages and determine risk levels. The correlation analysis unit determines the correlation stage based on the overall digital model of the product lifecycle and the risk level of the risk stage, and determines the degree of correlation according to pre-set rules.

3. The control system according to claim 2, characterized in that, The control decision execution module includes: The correlation degree conversion unit, according to specific conversion rules, converts the degree of correlation at the correlation stage into the corresponding degree of risk impact; The decision generation unit uses the random forest algorithm to train and generate decision models by taking the feature data, risk level and risk impact degree of the risk stage and the associated stage as input, and outputs control decision schemes for each stage. The execution control unit is used to implement the control and management decision schemes output by the decision generation unit, regulate the risk stages and related stages of the product life cycle, and track the execution effect.

4. The control system according to claim 2, characterized in that, The process of training the risk identification model by the risk identification unit is as follows: using historical operating data as input and the corresponding normal or risk label as output, a convolutional neural network model is trained; the rule for determining the correlation stage and the degree of correlation by the correlation analysis unit is as follows: if the risk level of the risk stage is high, the directly preceding and following stages are determined to be highly correlated, and the indirectly preceding and following stages are determined to be moderately correlated. If the risk level of a risk stage is low, the stages directly preceding and following it will be classified as moderately correlated, and the stages indirectly preceding and following it will be classified as lowly correlated.

5. The control system according to claim 3, characterized in that, The conversion rule of the correlation degree conversion unit is as follows: high correlation corresponds to high risk impact, medium correlation corresponds to medium risk impact, and low correlation corresponds to low risk impact. The training data for the decision generation unit to train the decision model consists of various historical control and management decision schemes implemented at each stage of the product lifecycle, as well as corresponding stage characteristic data and risk levels.

6. The control system according to claim 5, characterized in that, The stage cluster partitioning unit learns the association weights using a graph neural network by calculating the association weights between the feature vectors of each stage through a message passing mechanism in the graph structure. The method for calculating the similarity of interaction feature vectors in the model building unit is as follows: the Jaccard similarity algorithm is used to calculate the similarity between interaction feature vectors at each stage.

Images