Production process real-time optimization and quality tracing method and system based on multi-source equipment data fusion

Abstract

The invention discloses a production process real-time optimization and quality tracing method and system based on multi-source equipment data fusion, and relates to the technical field of intelligent manufacturing and quality control. Real-time standardized fusion of multi-source heterogeneous data is realized through edge calculation, quality risks are accurately evaluated by using a process-level digital twin model and a multi-layer graph neural network, real-time adjustment and optimization of a production process are realized by using a multi-objective optimization algorithm, a complete and credible quality tracing system is constructed in combination with a causal graph model and a block chain technology, and the quality tracing efficiency is improved. The problems of data islands, optimization lag and tracing fracture existing in a traditional manufacturing execution system are fundamentally solved, and transparency, intelligent real-time optimization and full-chain credible quality tracing of the production process are achieved.

Description

Technical Field

[0001] This invention relates to the field of underwater vision technology, specifically to a method and system for real-time optimization and quality traceability of production processes based on multi-source device data fusion. Background Technology

[0002] Against the backdrop of the current transformation of the manufacturing industry towards intelligence, precision, and compliance, especially for highly regulated industries such as pharmaceuticals, food, and chemicals, their production process management and quality control face systemic challenges. While typical WMES systems and their supporting hardware have initially established a software functional system covering "basic information, warehouse operations, inspection operations, production operations, equipment management, and statistical reports," and have achieved partial data collection and business operations through hardware such as PDAs, tablets, barcode printers, and sensor networks, they still suffer from the following fundamental defects: First, at the data collection and fusion level, the data generated by system functions (such as equipment monitoring, production reporting, and inspection entry) and hardware (such as sensors, PLCs, and electronic scales) are often isolated and lagging, lacking unified access, standardization, and deep integration of multi-source, heterogeneous, and real-time data. This results in the inability to form a real-time data flow that runs through "people, machines, materials, methods, and environment," leading to numerous "black box" links in the production process. Second, at the process optimization level, existing systems mainly achieve post-event recording, querying, and report statistics (such as process dashboards and batch number traceability). Their "optimization" is mostly static parameter settings and empirical adjustments, lacking dynamic analysis, prediction, and adaptive optimization capabilities based on real-time fused data. They cannot immediately detect efficiency bottlenecks and quality deviations during the production process and automatically adjust process parameters. Furthermore, in terms of quality traceability, existing traceability functions (such as batch number traceability) mostly rely on manual post-event association of documents and records. The traceability chain is static, broken, and its reliability is questionable. It is impossible to accurately associate the final product quality with dynamic information such as equipment operating parameters, environmental data, and operational actions at every moment during the production process, resulting in incomplete traceability information and difficulty in locating the root cause of the problem. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for real-time optimization and quality traceability of production processes based on multi-source device data fusion, to solve the pain points of data barriers, optimization lag and insufficient traceability in traditional production management, and to achieve dynamic adaptive optimization of the production process and high-quality and reliable traceability throughout the entire life cycle.

[0004] The objective of this invention can be achieved through the following technical solutions: This application provides a method for real-time optimization and quality traceability of production processes based on multi-source equipment data fusion, including the following steps: S1. Deploy edge computing nodes on the production site to collect heterogeneous data from programmable logic controllers, sensor devices, electronic scales and mobile terminals. Based on the mapping table between equipment codes and process positions, perform timestamp synchronization and semantic annotation on the data to generate a structured data stream containing material batch numbers, process codes, timestamps and parameter values. S2. Input the structured data stream into the process-level digital twin model. The model calculates the cumulative deviation of process parameters between processes according to the process specifications, evaluates the quality status of intermediate products based on the process quality correlation model, and outputs the quality risk assessment results. S3. When the quality risk assessment result exceeds the preset threshold, the process-level digital twin model is used as a simulator, and material balance, equipment capacity, and process specifications are used as constraints. A multi-objective optimization algorithm is used to solve the process parameter adjustment value or process scheduling scheme, and the optimization result is sent to the production execution unit. S4. Record the complete process data generated by the optimization results, and associate the process data with the structured data stream to form a chain of decision evidence; S5. For each production batch, aggregate the structured data stream, quality risk assessment results, optimization results, and decision evidence chain to construct a batch data cube. Extract the structured features of the data cube, calculate the feature hash value, and store it in the blockchain node to generate a batch traceability identifier.

[0005] Further, step S1 specifically includes: Edge computing nodes synchronize with the factory clock server via the IEEE 1588 protocol to generate a unified time base; the arrival time of each data packet is recorded by dedicated hardware circuitry, and combined with pipeline offset and dynamic network latency compensation, to generate accurate time stamps. Extract the device identifier from the data packet, and use the device identifier as the key to query the pre-set device code and process position mapping table in memory to obtain the process number, parameter semantic definition and numerical conversion rules corresponding to the device. According to the rules, the original values ​​are converted into standardized parameter values, and three-level semantic tags of equipment function, business meaning and process attribute are added to the data points. The standardized parameter values, time tags, semantic tags and current material batch information are encapsulated into a structured data record in a unified format according to a predetermined structure.

[0006] Furthermore, step S1 also includes: a dynamic mapping table of pre-set equipment codes and process positions within the edge computing node; when the equipment layout of the production line changes, the mapping table is updated through the equipment management module of the manufacturing execution system; and the edge computing node automatically adjusts the semantic labeling rules of the data acquisition channel according to the updated mapping table.

[0007] Furthermore, the calculation of the cumulative deviation of process parameters between processes in step S2 specifically includes: S21. Extract the real-time process parameters of the current process from the structured data stream, calculate the absolute deviation between the measured value of each parameter and the preset target value in the process specification, and then divide the absolute deviation by the allowable fluctuation range defined in the process specification to generate a normalized instantaneous deviation coefficient. S22. Based on the predefined inter-process deviation transmission model, the normalized instantaneous deviation coefficient vector output by the upstream process is multiplied by its corresponding deviation transmission coefficient matrix to obtain the transmission influence of the upstream process deviation on the current process. The deviation transmission coefficient matrix is ​​determined by analyzing the statistical correlation between fluctuations in upstream and downstream process parameters in historical production data. S23. Based on the process sequence defined in the production process route, recursively calculate starting from the first process: the total cumulative deviation vector of each process is equal to the sum of the influence of all upstream processes transmitted to this process, and then superimposed with the normalized instantaneous deviation coefficient vector newly generated in this process. S24. For the total cumulative deviation vector output by any process, calculate its Euclidean norm or extract the preset key dimension components to generate a scalar deviation index that characterizes the overall deviation of the process.

[0008] Furthermore, the output of the quality risk assessment result in step S2 specifically includes: inputting the scalar deviation index and the cumulative deviation vector of related process parameters into a trained process quality correlation model. The process quality correlation model is a multi-layer graph neural network, where network nodes correspond to production processes and edges correspond to the quality influence relationships between processes. Based on the input deviation data, the process quality correlation model performs information propagation and feature aggregation between nodes through the graph neural network layers, deduces the probability distribution of the quality status of the current intermediate product and the predicted final product, calculates the probability of the occurrence of defect categories based on the probability distribution of quality status, maps the probability values ​​to a predefined risk level range, and outputs the corresponding quality risk level or numerical score.

[0009] Furthermore, step S3 employs a multi-objective optimization algorithm to solve for process parameter adjustment values ​​or process scheduling schemes, specifically including: S31. Based on the process that needs adjustment indicated by the quality risk assessment results, delineate the system boundary to be optimized in the digital twin model and establish a mathematical model for a multi-objective optimization problem. S32. Transform the three major categories of constraints—material balance, equipment capacity, and process specifications—into mathematical expressions, then dynamically adjust the constraint boundaries of the corresponding process parameters or operating speeds based on the equipment health score calculated from real-time monitoring data, and initialize the constrained particle swarm optimization algorithm. S33. In each iteration cycle, the particle position vector representing the candidate process parameter adjustment value or process scheduling scheme is input into the digital twin model for rapid simulation, which is used to calculate the corresponding objective function value and verify the constraint satisfaction. S34. When the algorithm reaches the termination condition, select all non-dominated solutions from the final generation of particle swarm to form the Pareto optimal frontier, and select the final compromise solution according to the predefined decision strategy. S35. If the final compromise is a process parameter adjustment scheme, then the specific setting value correction instruction to be sent to the equipment controller will be output. If it is a process scheduling scheme, then the adjusted process start time, equipment allocation or work sequence plan will be output and sent to the production scheduling system for execution.

[0010] Furthermore, after using multi-objective optimization algorithms to solve for process parameter adjustment values ​​or process scheduling schemes, cross-batch collaborative optimization is also included. Specifically, this includes: incorporating the current batch and several subsequent batches of the same type to be produced into a unified extended optimization time domain, establishing a multi-batch joint optimization model, using a long sequence composed of multiple batch process parameter sequences as decision variables, and minimizing the overall quality variance, total production time, or total energy consumption of multiple batches as global objectives, and solving for the globally optimal process parameter configuration series under cross-batch constraints.

[0011] Furthermore, the decision evidence chain construction in step S4 adopts a causal graph model, with the quality risk assessment results, the final compromise solution, and the actual process parameters and states as nodes of the graph model, and the logical causality derived from the process quality correlation model and the process specifications and equipment capability constraints as directed edges connecting the above nodes. By automatically associating the triggering, decision and execution events in the optimization process, a complete directed causal evidence network is constructed.

[0012] The causal graph model learns online based on historical production data. When a new process event association pattern is detected, it automatically updates the weights and directions of the edges and adjusts the confidence parameters of the causal relationship.

[0013] Further, step S5 specifically includes: Using the unique identifier of the production batch as an index, the structured data stream, quality risk assessment results, optimization results and decision evidence chain data of the batch are aggregated. Based on the unified timestamp and process code, the data are aligned and associated in the spatiotemporal dimension to construct a batch data cube with process time and data category as dimensions. According to the predefined quality traceability and analysis rules, the statistical characteristics of process parameters, risk event distribution characteristics, optimization action sequence characteristics and causal evidence chain structure characteristics are automatically extracted from the data cube, and the extracted features are normalized and organized into a fixed-dimensional structured feature vector. The structured feature vector is calculated using an encrypted hash function to generate a unique feature hash value. This hash value, the batch unique identifier, the production time range, and the feature vector version metadata are then encapsulated into a data storage package. By calling the smart contract interface deployed on the permissioned blockchain network, the evidence data packet is submitted as a transaction. After being verified by network consensus, it is permanently recorded in the blockchain. The generated blockchain transaction hash and the block height together constitute the traceability identifier for this batch.

[0014] This application provides a real-time optimization and quality traceability system for production processes based on multi-source equipment data fusion, applied to methods for real-time optimization and quality traceability of production processes based on multi-source equipment data fusion, including: The data fusion and preprocessing module is used to collect multi-source heterogeneous raw data, perform millisecond-level timestamp synchronization and three-level semantic annotation according to the preset equipment code and process position mapping table, and output a structured real-time data stream. The digital twin and quality assessment module is used to receive structured data streams, drive the twin model to run synchronously with the production line, calculate the cumulative deviation of process parameter transmission, and deduce the probability distribution of the quality status of intermediate products and finished products in real time, and output quantitative quality risk assessment results. The online optimization decision module uses a digital twin model as a simulator to solve for the optimal process parameter adjustment or production scheduling scheme under the condition of satisfying dynamic constraints, and supports cross-batch collaborative optimization when the quality risk exceeds the threshold. The causal evidence chain construction module, based on the causal graph model, automatically uses optimization trigger events, decision events, and execution events as nodes, establishes directed edges according to the influence relationship between processes and production logic, constructs a traceable decision evidence chain, and has the ability to learn online based on historical data to update causal relationships. The trusted traceability archive generation module constructs a multi-dimensional data cube by aggregating batch full lifecycle data, extracts structured feature vectors and calculates hash values, and submits the evidence data package to the permissioned blockchain through a smart contract to generate an immutable batch traceability identifier.

[0015] The beneficial effects of this invention are as follows: To address the issues of isolated and lagging multi-source data, lack of unified access and deep integration, this invention deploys edge computing nodes on the production site, achieves millisecond-level timestamp synchronization and three-level semantic annotation based on the IEEE 1588 protocol, and completes standardized processing of heterogeneous data by combining a dynamic equipment-process mapping table. This constructs a real-time structured data stream that runs through people, machines, materials, methods, and environment, effectively breaking down data barriers between different equipment and systems, eliminating black-box links in the production process, and enabling scattered equipment operation data, operational data, material data, etc., to form a unified data carrier that can be correlated and analyzed, laying a solid data foundation for subsequent real-time optimization and accurate traceability. To address the issues of production process optimization relying on post-event records and experience-based adjustments, and lacking dynamic adaptive optimization capabilities, this invention uses a process-level digital twin model to calculate the cumulative deviation of process parameters in real time. It utilizes a process quality correlation model constructed with a multi-layer graph neural network to accurately assess quality risks. When the risk exceeds a threshold, the twin model is used as a simulator to solve for the optimal process parameters or scheduling scheme through a constrained multi-objective optimization algorithm and automatically issue it for execution. It also supports cross-batch collaborative optimization to achieve global optimization, upgrading optimization from post-event traceability to in-process prediction and real-time tuning. It can instantly detect efficiency bottlenecks and quality deviations and automatically correct them, significantly improving the stability of the production process, the accuracy of parameter control, and the consistency of multi-batch production, while reducing the quality defect rate and production energy consumption. To address the issues of reliance on manual correlation, statically broken chains, and insufficient credibility and completeness of traceability information in quality traceability, this invention constructs a complete chain of decision-making evidence using a causal graph model. It aggregates full lifecycle data into a multi-dimensional data cube, extracts structured features, generates unique hash values, and stores them on the blockchain, creating an immutable batch traceability identifier. This achieves high-fidelity correlation between the final product quality and dynamic information such as equipment parameters, environmental data, operational actions, and optimization decisions at every moment during the production process. This transforms the traceability chain from statically broken to dynamically complete, and the blockchain's evidentiary characteristics ensure the immutability and verifiability of traceability information, significantly improving the efficiency and accuracy of root cause analysis and fully meeting the stringent requirements of heavily regulated industries for compliant traceability. Attached Figure Description

[0016] To better understand and implement this application, the technical solution is described in detail below with reference to the accompanying drawings.

[0017] Figure 1 This is a flowchart illustrating the real-time optimization and quality traceability method for production processes based on multi-source equipment data fusion provided in Embodiment 1 of this application. Figure 2 This is a flowchart illustrating step S2 in the real-time optimization and quality traceability method for production processes based on multi-source device data fusion provided in Embodiment 1 of this application. Figure 3This is a flowchart illustrating step S3 in the real-time optimization and quality traceability method for production processes based on multi-source device data fusion provided in Embodiment 1 of this application. Figure 4 This is a schematic diagram of the structure of the real-time optimization and quality traceability system for the production process based on multi-source equipment data fusion provided in Embodiment 2 of this application. Detailed Implementation

[0018] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, exemplary embodiments will be described in detail below, examples of which are illustrated in the accompanying drawings. In the following description, when referring to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application.

[0019] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used herein are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0020] The following detailed description of the specific implementation methods, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided in detail.

[0021] Example 1

[0022] Please see Figures 1-3 This embodiment provides a method for real-time optimization and quality traceability of production processes based on multi-source device data fusion, including the following steps: S1. Deploy edge computing nodes on the production site, collect heterogeneous data from programmable logic controllers, sensor devices, electronic scales and mobile terminals through industrial protocol converters, and perform timestamp synchronization and semantic annotation on the data based on the mapping table between equipment codes and process positions to generate a structured data stream containing material batch numbers, process codes, timestamps and parameter values. Further, step S1 specifically includes: Edge computing nodes synchronize with the factory clock server via the IEEE 1588 protocol to generate a unified time base; the arrival time of each data packet is recorded by dedicated hardware circuits, and combined with the processing pipeline offset and dynamic network latency compensation based on historical statistics, a globally consistent and accurate time stamp is generated. Extract the device identifier from the data packet, use it as the key to query the pre-set device code and process position mapping table in memory, and obtain the process number, parameter semantic definition and numerical conversion rules corresponding to the device; According to the rules, the original values ​​are converted into standardized parameter values, and three-level semantic tags of equipment function, business meaning and process attribute are added to the data points. The standardized parameter values, time tags, semantic tags and current material batch information are encapsulated into a structured data record in a unified format according to a predetermined structure.

[0023] Furthermore, the edge computing node in S1 has a pre-set dynamic mapping table of equipment codes and process positions. When the equipment layout of the production line changes, the mapping table is updated by the equipment management module of the manufacturing execution system. The edge computing node automatically adjusts the semantic annotation rules of the data acquisition channel according to the updated mapping table.

[0024] Furthermore, after automatically adjusting the semantic annotation rules of the data acquisition channels according to the updated mapping table, it also includes continuous monitoring of the communication status. The edge computing node monitors the communication status of each data acquisition channel in real time. When an abnormality is detected in a channel, it predicts the missing data based on historical data patterns and adds data quality identifiers to the structured data stream.

[0025] Specifically, by using edge computing nodes to achieve real-time acquisition, synchronization, and semantic fusion of multi-source heterogeneous data, a unified structured data stream is generated, which solves the problems of data isolation, lag, and inconsistent standards in the background technology. This enables real-time transparency of all elements of the production process and provides a high-quality data foundation for subsequent optimization and traceability.

[0026] S2. Input the structured data stream into the process-level digital twin model. The model calculates the cumulative deviation of process parameters between processes based on the preset threshold and parameter fluctuation range of the process specifications, evaluates the quality status of intermediate products based on the process quality correlation model, and outputs the quality risk assessment result. Furthermore, the calculation of the cumulative deviation of process parameters between processes in step S2 specifically includes: S21. Extract the real-time process parameters of the current process from the structured data stream, calculate the absolute deviation between the measured value of each parameter and the preset target value in the process specification, and then divide the absolute deviation by the allowable fluctuation range defined in the process specification to generate a normalized instantaneous deviation coefficient. S22. Based on a predefined inter-process deviation transmission model, the normalized instantaneous deviation coefficient vector output by the upstream process is multiplied by its corresponding deviation transmission coefficient matrix to obtain the transmission impact of the upstream process deviation on the current process. The deviation transmission coefficient matrix is ​​determined by analyzing the statistical correlation between the fluctuations of upstream and downstream process parameters in historical production data. S23. Based on the process sequence defined in the production process route, recursively calculate starting from the first process: the total cumulative deviation vector of each process is equal to the sum of the influence of all upstream processes transmitted to this process, and then superimposed with the normalized instantaneous deviation coefficient vector newly generated in this process. S24. For the total cumulative deviation vector output by any process, calculate its Euclidean norm or extract preset key dimension components to generate a scalar deviation index that characterizes the overall deviation of the process. The scalar index will be compared with preset multi-level thresholds to determine the process status and serve as a direct basis for triggering quality risk warnings.

[0027] Furthermore, the output of the quality risk assessment result in step S2 specifically includes: inputting the scalar deviation index and the cumulative deviation vector of related process parameters into a trained process quality correlation model. The process quality correlation model is a multi-layer graph neural network, where network nodes correspond to production processes and edges correspond to the quality influence relationships between processes. Based on the input deviation data, the process quality correlation model performs information propagation and feature aggregation between nodes through the graph neural network layers, deduces the probability distribution of the quality status of the current intermediate product and the predicted final product, calculates the probability of the occurrence of defect categories based on the probability distribution of quality status, maps the probability values ​​to a predefined risk level range, and outputs the corresponding quality risk level or numerical score as a quantitative quality risk assessment result.

[0028] The process-level digital twin model includes: firstly, building high-precision 3D geometric models of the corresponding equipment for each process in Unity3D or a similar 3D engine, based on the physical layout and equipment list of the production line; secondly, configuring corresponding behavior scripts and control logic for each 3D model according to equipment control logic and process principles, enabling it to receive real-time control commands from the PLC and simulate the action response of real equipment; thirdly, associating the process specification database from the manufacturing execution system with the model, setting preset target values, allowable fluctuation ranges, and equipment operation constraints for each process's key process parameters; and fourthly, connecting structured data streams to the twin model through a real-time data interface, driving its status to update synchronously with the physical production line, and enabling it to receive real-time process parameters, perform deviation calculations, call quality correlation models for simulation prediction, and output evaluation results.

[0029] The training and deployment details of the multi-layer graph neural network in the process quality correlation model are as follows: the number of nodes in the input layer of the network is equal to the number of key process parameters, and the features of each node are the time-series statistics of the corresponding parameter (such as the mean and variance of the most recent 10 sampling points); the graph structure is predefined according to the bill of materials (BOM) and process route, and directed edges are established if there is material transfer between processes; the training adopts a supervised learning method, using the final quality inspection results (pass / defect type) of historical batches as labels; the model is implemented using the PyTorch Geometric framework, and after training, it is packaged in ONNX format and deployed on an inference server equipped with a GPU accelerator.

[0030] Specifically, by constructing a process-level digital twin model and a quality correlation model based on graph neural networks, the problems of optimization lag and lack of predictive ability were solved. Real-time transmission and calculation of process parameter deviations and probabilistic prediction of intermediate product quality risks were realized, enabling the production process to shift from post-production recording to pre-production early warning and in-process evaluation, providing accurate risk perception basis for dynamic optimization.

[0031] S3. When the quality risk assessment result exceeds the preset threshold, the process-level digital twin model is used as a simulator, and material balance, equipment capacity, and process specifications are used as constraints. A multi-objective optimization algorithm is used to solve the process parameter adjustment value or process scheduling scheme, and the optimization result is sent to the production execution unit. Furthermore, step S3 employs a multi-objective optimization algorithm to solve for process parameter adjustment values ​​or process scheduling schemes, specifically including: S31. Based on the process that needs adjustment indicated by the quality risk assessment results, delineate the system boundary to be optimized in the digital twin model and establish a mathematical model for a multi-objective optimization problem. S32. Transform the three major categories of constraints—material balance, equipment capacity, and process specifications—into mathematical expressions. Then, based on the equipment health score calculated from real-time monitoring data, dynamically adjust the constraint boundaries of the corresponding process parameters or operating speeds. Initialize the constrained particle swarm optimization algorithm, set the number of particles, the number of iterations, and the learning factor, and randomly generate initial solutions for the particle swarm within the feasible region. S33. In each iteration cycle, the position vector of each particle represents a set of candidate process parameter adjustment values ​​or process scheduling time series. The candidate solutions are input into the digital twin model to drive the model to perform rapid simulation in order to calculate the corresponding objective function value and verify whether it meets all constraints. The fitness value of a particle is determined by the objective function value and the penalty term for the degree of constraint violation. When a particle updates its position and velocity, the search process is guided to a feasible region that satisfies all constraints, especially the dynamic device health constraint, through a feasibility rule or dynamic penalty function mechanism. S34. When the algorithm reaches the termination condition, all non-dominated solutions are selected from the final generation of particle swarm to form the Pareto optimal front of the problem. According to the predefined decision strategy, the final compromise solution is selected from the Pareto front through the preset weight preference or the distance minimization method based on the ideal point. S35. If the final compromise is a process parameter adjustment scheme, then the specific setting value correction instruction to be sent to the equipment controller will be output. If it is a process scheduling scheme, then the adjusted process start time, equipment allocation or work sequence plan will be output and sent to the production scheduling system for execution.

[0032] The specific calculation method for the equipment health score is as follows: based on the historical baseline of equipment operation data (such as current, vibration, and temperature), the isolated forest algorithm is used to calculate the anomaly score of real-time data, and combined with the recent maintenance records of the equipment (such as the last maintenance time and the number of failures) for weighted fusion, and finally a standardized health score between 0 and 1 is generated. The lower the score, the worse the equipment condition, and the greater the shrinkage ratio of its parameter feasible region in optimization.

[0033] Furthermore, the mathematical model of the multi-objective optimization problem includes an objective function with at least two objective components. The first component is used to minimize the deviation of the predicted quality index, and the second component is used to maximize production efficiency or minimize production energy consumption. Each objective component is comprehensively processed by a preset scalarization method (such as the weighted sum method or the ε-constraint method) to form a single scalar fitness function.

[0034] Material balance is reflected in the linear equality constraint between the input and output quantities of the process; equipment capacity is reflected in the linear or nonlinear inequality constraint of the maximum power, speed or processing capacity of the equipment; process specifications are reflected in the boundary inequality constraint of the process parameters.

[0035] Furthermore, after using multi-objective optimization algorithms to solve for process parameter adjustment values ​​or process scheduling schemes, cross-batch collaborative optimization is also included. Specifically, this includes: incorporating the current batch and several subsequent batches of the same type to be produced into a unified extended optimization time domain, establishing a multi-batch joint optimization model, using a long sequence composed of multiple batch process parameter sequences as decision variables, and minimizing the overall quality variance of multiple batches, total production time, or total energy consumption as global objectives. Under the constraints of cross-batch factors such as material supply and equipment capacity, the globally optimal process parameter configuration series that can improve the overall steady-state performance of the production line is solved.

[0036] Specifically, this step uses a digital twin model as a simulator, combined with dynamic constraints on equipment health and multi-objective optimization algorithms, to solve the problems of static, lagging, and lack of adaptive capabilities in the background technology. It realizes real-time closed-loop optimization and cross-batch collaboration of process parameters or production scheduling under multiple constraints, enabling the production process to shift to dynamic optimization and global collaboration, significantly improving quality stability, production efficiency, and equipment reliability.

[0037] S4. Record the complete process data generated by the optimization result, including the original data fragment that triggered the optimization, the constraint settings, the algorithm parameter configuration, the solution process data and the content of the issued instructions, and associate the process data with the structured data stream to form a chain of decision evidence; Furthermore, the decision evidence chain construction in step S4 adopts a causal graph model, with the quality risk assessment results, the final compromise solution, and the actual process parameters and states as nodes of the graph model, and the logical causality derived from the process quality correlation model and the process specifications and equipment capability constraints as directed edges connecting the above nodes. By automatically associating the triggering, decision and execution events in the optimization process, a complete directed causal evidence network is constructed, forming a traceable decision evidence chain and a directed causal evidence network.

[0038] Furthermore, the causal graph model learns online based on historical production data. When a new process event association pattern is detected, it automatically updates the weights and directions of the edges and adjusts the confidence parameters of the causal relationship.

[0039] The online learning of the causal graph model employs an incremental Bayesian structured learning algorithm. The system periodically adds newly generated decision evidence chains from the past 24 hours as an incremental dataset daily, combining it with the existing graph model. The BIC scoring function is used to evaluate the improvement in model fit caused by adding or deleting edges. Furthermore, the Markov chain Monte Carlo method is used to sample the model space, probabilistically updating the graph structure (presence and direction of edges) and parameters (edge ​​weights), thereby achieving a gradual evolution of causal knowledge.

[0040] Specifically, by constructing and continuously optimizing the decision-making evidence chain through a causal graph model, the problems of static, fragmented records and lack of logical connection in the background technology of quality traceability are solved. It realizes interpretable, dynamic correlation and automated recording of the entire process of optimization decision-making, and upgrades quality traceability into an intelligent evidence network with traceable causes and logical consistency, which greatly improves the accuracy and efficiency of root cause localization.

[0041] S5. For each production batch, aggregate the structured data stream, quality risk assessment results, optimization results, and decision evidence chain to construct a batch data cube. Extract the structured features of the data cube, calculate the feature hash value, and store it in the blockchain node to generate a batch traceability identifier.

[0042] Further, step S5 specifically includes: Using the unique identifier of each production batch as an index, the structured data stream, quality risk assessment results, optimization results, and decision evidence chain data of each production batch are aggregated. Based on a unified timestamp and process code, the data is aligned and correlated in the spatiotemporal dimension. In an in-memory database or columnar storage system, a batch data cube that supports multi-dimensional query and analysis is constructed with process time and data category as dimensions, and process parameter values, risk levels, optimization instruction content, and causal event identifiers as measures. According to predefined quality traceability and analysis rules, statistical features of process parameters, risk event distribution features, optimization action sequence features, and causal evidence chain structure features are automatically extracted from the data cube. After normalization of the extracted features, they are organized into a fixed-dimensional structured feature vector in a predetermined order to characterize the core characteristics and quality context of the batch production. The structured feature vector is calculated using a cryptographic hash function to generate a unique feature hash value. The hash value, batch unique identifier, production time range, and feature vector version metadata are encapsulated into a data storage package. The data package is submitted as a transaction by calling the smart contract interface deployed on the permissioned blockchain network. After being verified by network consensus, it is permanently recorded in the blockchain. The generated blockchain transaction hash and the block height together constitute the immutable traceability identifier of the batch.

[0043] Among them, the predefined quality traceability and analysis rules are a set of logics pre-set based on production process specifications, quality indicator systems, and historical anomaly pattern analysis. They define: the categories of key process parameters to be extracted (such as temperature, pressure, and speed) and their statistical characteristic calculation methods (such as mean, standard deviation, extreme values, and process capability index); the classification criteria for quality risk events (such as risk level and alarm type) and their distribution statistics in time and process dimensions (such as frequency, duration, and number of related processes); the classification of optimization decision-making actions (such as parameter adjustment and scheduling change) and their trigger sequence pattern descriptions; and the identification criteria for key nodes in the causal evidence chain (such as root cause events and key propagation paths) and the quantitative indicators of path characteristics (such as impact intensity and path length). These rules together constitute the complete basis for automatically extracting structured feature vectors from batch data cubes.

[0044] Furthermore, the blockchain network adopts a consortium blockchain architecture, jointly maintained by manufacturers, key suppliers, and regulatory agencies as authorized nodes. The notarization smart contract receives the hash value of the structured feature vector and related metadata. While completing the blockchain transaction record, it triggers an off-chain storage operation to store the complete batch data cube index information in the connected InterPlanetary File System (IPS). The content identifier returned from the IPS is then associated with the blockchain transaction record, thereby achieving verifiable, tamper-proof, and efficiently accessible distributed storage of massive amounts of process data.

[0045] Specifically, by aggregating batch full-element data to construct a multi-dimensional data cube and extracting structured features, and combining blockchain and InterPlanetary File System distributed evidence storage technology, the problems of incomplete traceability information, easy tampering, and low credibility in the background technology are solved. This achieves high-fidelity traceability of the product's full life cycle quality archives, which is tamper-proof, verifiable, and efficiently accessible, upgrading quality traceability from information silos to an authoritative digital certificate system that is data-complete, anti-counterfeiting, and reliable.

[0046] Example 2

[0047] Please see Figure 4 This embodiment provides a real-time optimization and quality traceability system for production processes based on multi-source equipment data fusion, applied to a method for real-time optimization and quality traceability of production processes based on multi-source equipment data fusion, including: The data fusion and preprocessing module, which includes an industrial protocol converter, a clock synchronization unit, and a semantic annotation engine, is responsible for collecting multi-source heterogeneous raw data, performing millisecond-level timestamp synchronization and three-level semantic annotation based on a pre-set equipment code and process position mapping table, and outputting a structured real-time data stream. The digital twin and quality assessment module includes a process-level digital twin simulation engine and a quality correlation model based on graph neural networks. It is used to receive structured data streams, drive the twin model to run synchronously with the production line, calculate the cumulative deviation of process parameter transmission, and deduce the probability distribution of the quality status of intermediate products and finished products in real time, and output quantitative quality risk assessment results. The online optimization decision module includes a multi-objective optimization solver and a constraint management unit. When the quality risk exceeds the threshold, it uses a digital twin model as a simulator and, under the condition of satisfying dynamic constraints, employs algorithms such as constrained particle swarm optimization to solve for the optimal process parameter adjustment or production scheduling scheme, and supports cross-batch collaborative optimization. The causal evidence chain construction module, based on the causal graph model, automatically uses optimization trigger events, decision events, and execution events as nodes, establishes directed edges according to the influence relationship between processes and production logic, constructs a traceable decision evidence chain, and has the ability to learn online based on historical data to update causal relationships. The trusted traceability archive generation module includes a data cube construction engine, a feature extractor, and a blockchain interface. It constructs a multi-dimensional data cube by aggregating batch full lifecycle data, extracts structured feature vectors and calculates hash values, and submits the evidence data package to the permissioned blockchain through a smart contract to generate an immutable batch traceability identifier.

[0048] This embodiment transforms multi-source heterogeneous equipment data into a unified structured information flow through spatiotemporal synchronization and semantic fusion at the perception layer, breaking down data silos. At the cognition layer, it constructs a dynamic model of the production system that maps the virtual and real worlds using process-level digital twins and graph neural networks, enabling simulation of process deviation transmission and probabilistic prediction of quality status. Then, at the decision-making layer, using the twin model as a simulation sandbox, it solves for optimal process parameters or scheduling schemes under dynamic constraints through multi-objective optimization algorithms, forming an adaptive control closed loop. Finally, at the traceability layer, it reconstructs the decision logic chain through a causal graph model and uses blockchain technology to permanently store process feature fingerprints, forming an immutable quality archive. This embodiment realizes a complete value chain from raw data to optimization decisions to reliable traceability, enabling the production process to form a transparent, traceable, and dynamically optimized intelligent system.

[0049] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for real-time optimization and quality traceability of production processes based on multi-source equipment data fusion, characterized by: Includes the following steps: S1. Deploy edge computing nodes on the production site to collect heterogeneous data from programmable logic controllers, sensor devices, electronic scales and mobile terminals. Based on the mapping table between equipment codes and process positions, perform timestamp synchronization and semantic annotation on the data to generate a structured data stream containing material batch numbers, process codes, timestamps and parameter values. S2. Input the structured data stream into the process-level digital twin model. The model calculates the cumulative deviation of process parameters between processes according to the process specifications, evaluates the quality status of intermediate products based on the process quality correlation model, and outputs the quality risk assessment results. S3. When the quality risk assessment result exceeds the preset threshold, the process-level digital twin model is used as a simulator, and material balance, equipment capacity, and process specifications are used as constraints. A multi-objective optimization algorithm is used to solve the process parameter adjustment value or process scheduling scheme, and the optimization result is sent to the production execution unit. S4. Record the complete process data generated by the optimization results, and associate the process data with the structured data stream to form a chain of decision evidence; S5. For each production batch, aggregate the structured data stream, quality risk assessment results, optimization results, and decision evidence chain to construct a batch data cube. Extract the structured features of the data cube, calculate the feature hash value, and store it in the blockchain node to generate a batch traceability identifier.

2. The method for real-time optimization and quality traceability of production processes based on multi-source equipment data fusion according to claim 1, characterized in that: Step S1 specifically includes: Edge computing nodes synchronize with the factory clock server via the IEEE 1588 protocol to generate a unified time base; the arrival time of each data packet is recorded by dedicated hardware circuitry, and combined with pipeline offset and dynamic network latency compensation, to generate accurate time stamps. Extract the device identifier from the data packet, and use the device identifier as the key to query the pre-set device code and process position mapping table in memory to obtain the process number, parameter semantic definition and numerical conversion rules corresponding to the device. According to the rules, the original values ​​are converted into standardized parameter values, and three-level semantic tags of equipment function, business meaning and process attribute are added to the data points. The standardized parameter values, time tags, semantic tags and current material batch information are encapsulated into a structured data record in a unified format according to a predetermined structure.

3. The method for real-time optimization and quality traceability of production processes based on multi-source equipment data fusion according to claim 2, characterized in that: Step S1 further includes: a dynamic mapping table of pre-set equipment codes and process positions within the edge computing node; when the equipment layout of the production line changes, the mapping table is updated through the equipment management module of the manufacturing execution system; and the edge computing node automatically adjusts the semantic labeling rules of the data acquisition channel according to the updated mapping table.

4. The method for real-time optimization and quality traceability of production processes based on multi-source equipment data fusion according to claim 1, characterized in that: Step S2 calculates the cumulative deviation of process parameters between processes, specifically including: S21. Extract the real-time process parameters of the current process from the structured data stream, calculate the absolute deviation between the measured value of each parameter and the preset target value in the process specification, and then divide the absolute deviation by the allowable fluctuation range defined in the process specification to generate a normalized instantaneous deviation coefficient. S22. Based on the predefined inter-process deviation transmission model, the normalized instantaneous deviation coefficient vector output by the upstream process is multiplied by its corresponding deviation transmission coefficient matrix to obtain the transmission influence of the upstream process deviation on the current process. The deviation transmission coefficient matrix is ​​determined by analyzing the statistical correlation between fluctuations in upstream and downstream process parameters in historical production data. S23. Based on the process sequence defined in the production process route, recursively calculate starting from the first process: the total cumulative deviation vector of each process is equal to the sum of the influence of all upstream processes transmitted to this process, and then superimposed with the normalized instantaneous deviation coefficient vector newly generated in this process. S24. For the total cumulative deviation vector output by any process, calculate its Euclidean norm or extract the preset key dimension components to generate a scalar deviation index that characterizes the overall deviation of the process.

5. The method for real-time optimization and quality traceability of production processes based on multi-source equipment data fusion according to claim 1, characterized in that: Step S2 outputs the quality risk assessment results, specifically including: inputting the scalar deviation index and the cumulative deviation vector of related process parameters into a trained process quality correlation model. The process quality correlation model is a multi-layer graph neural network, where network nodes correspond to production processes and edges correspond to the quality influence relationships between processes. Based on the input deviation data, the process quality correlation model performs information propagation and feature aggregation between nodes through the graph neural network layers, deduces the probability distribution of the quality status of the current intermediate product and the predicted final product, calculates the probability of the occurrence of defect categories based on the probability distribution of quality status, maps the probability values ​​to a predefined risk level range, and outputs the corresponding quality risk level or numerical score.

6. The method for real-time optimization and quality traceability of production processes based on multi-source equipment data fusion according to claim 1, characterized in that: Step S3 employs a multi-objective optimization algorithm to solve for process parameter adjustment values ​​or process scheduling schemes, specifically including: S31. Based on the process that needs adjustment indicated by the quality risk assessment results, delineate the system boundary to be optimized in the digital twin model and establish a mathematical model for a multi-objective optimization problem. S32. Transform the three major categories of constraints—material balance, equipment capacity, and process specifications—into mathematical expressions, then dynamically adjust the constraint boundaries of the corresponding process parameters or operating speeds based on the equipment health score calculated from real-time monitoring data, and initialize the constrained particle swarm optimization algorithm. S33. In each iteration cycle, the particle position vector representing the candidate process parameter adjustment value or process scheduling scheme is input into the digital twin model for rapid simulation, which is used to calculate the corresponding objective function value and verify the constraint satisfaction. S34. When the algorithm reaches the termination condition, select all non-dominated solutions from the final generation of particle swarm to form the Pareto optimal frontier, and select the final compromise solution according to the predefined decision strategy. S35. If the final compromise is a process parameter adjustment scheme, then the specific setting value correction instruction to be sent to the equipment controller will be output. If it is a process scheduling scheme, then the adjusted process start time, equipment allocation or work sequence plan will be output and sent to the production scheduling system for execution.

7. The method for real-time optimization and quality traceability of production processes based on multi-source equipment data fusion according to claim 6, characterized in that: After using multi-objective optimization algorithms to solve for process parameter adjustment values ​​or process scheduling schemes, cross-batch collaborative optimization is also included. Specifically, this includes: incorporating the current batch and several subsequent batches of the same type to be produced into a unified extended optimization time domain, establishing a multi-batch joint optimization model, using a long sequence formed by connecting multiple batch process parameter sequences as decision variables, and minimizing the overall quality variance, total production time, or total energy consumption of multiple batches as global objectives, and solving for the globally optimal process parameter configuration series under cross-batch constraints.

8. The method for real-time optimization and quality traceability of production processes based on multi-source equipment data fusion according to claim 1, characterized in that: The decision evidence chain construction in step S4 adopts a causal graph model. The quality risk assessment results, the final compromise solution, and the actual process parameters and status are used as nodes of the graph model. The inter-process influence relationship implied by the process quality correlation model and the logical causality derived from the process specifications and equipment capability constraints are used as directed edges to connect the above nodes. By automatically associating the triggering, decision and execution events in the optimization process, a complete directed causal evidence network is constructed. The causal graph model learns online based on historical production data. When a new process event association pattern is detected, it automatically updates the weights and directions of the edges and adjusts the confidence parameters of the causal relationship.

9. The method for real-time optimization and quality traceability of production processes based on multi-source equipment data fusion according to claim 1, characterized in that: Step S5 specifically includes: Using the unique identifier of the production batch as an index, the structured data stream, quality risk assessment results, optimization results and decision evidence chain data of the batch are aggregated. Based on the unified timestamp and process code, the data are aligned and associated in the spatiotemporal dimension to construct a batch data cube with process time and data category as dimensions. According to the predefined quality traceability and analysis rules, the statistical characteristics of process parameters, risk event distribution characteristics, optimization action sequence characteristics and causal evidence chain structure characteristics are automatically extracted from the data cube, and the extracted features are normalized and organized into a fixed-dimensional structured feature vector. The structured feature vector is calculated using an encrypted hash function to generate a unique feature hash value. This hash value, the batch unique identifier, the production time range, and the feature vector version metadata are then encapsulated into a data storage package. By calling the smart contract interface deployed on the permissioned blockchain network, the evidence data packet is submitted as a transaction. After being verified by network consensus, it is permanently recorded in the blockchain. The generated blockchain transaction hash and the block height together constitute the traceability identifier for this batch.

10. A real-time optimization and quality traceability system for production processes based on multi-source equipment data fusion, applied to the real-time optimization and quality traceability method for production processes based on multi-source equipment data fusion as described in any one of claims 1-9, characterized in that: include: The data fusion and preprocessing module is used to collect multi-source heterogeneous raw data, perform millisecond-level timestamp synchronization and three-level semantic annotation according to the preset equipment code and process position mapping table, and output a structured real-time data stream. The digital twin and quality assessment module is used to receive structured data streams, drive the twin model to run synchronously with the production line, calculate the cumulative deviation of process parameter transmission, and deduce the probability distribution of the quality status of intermediate products and finished products in real time, and output quantitative quality risk assessment results. The online optimization decision module uses a digital twin model as a simulator and a constrained particle swarm optimization algorithm to solve for the optimal process parameter adjustment or production scheduling scheme when the quality risk exceeds the threshold. It also supports cross-batch collaborative optimization. The causal evidence chain construction module, based on the causal graph model, automatically uses optimization trigger events, decision events, and execution events as nodes, establishes directed edges according to the influence relationship between processes and production logic, constructs a traceable decision evidence chain, and has the ability to learn online based on historical data to update causal relationships. The trusted traceability archive generation module constructs a multi-dimensional data cube by aggregating batch full lifecycle data, extracts structured feature vectors and calculates hash values, and submits the evidence data package to the permissioned blockchain through a smart contract to generate an immutable batch traceability identifier.

Images