Project quality control method and system

Abstract

The invention discloses a project quality management and control method and system, and belongs to the technical field of quality management and control, and the method comprises the steps: building a multi-level coding system of a production process route, and representing a processing constraint relation through topological sorting; automatically binding the QCP inspection items with the coding nodes by adopting an RETE algorithm of a rule engine; when a process change triggering event occurs, the rule engine calls the difference analysis module, compares the historical process parameter library and dynamically adds and deletes inspection items; the component state is defined, and the QCP inspection result triggers state conversion in real time and is synchronized to a production planning module through an API; constructing a Gantt chart-state linkage model based on the process route dependency relationship; based on a knowledge graph technology, aggregating historical quality data, and establishing a'defect mode-root cause-repair scheme 'association rule base; aPI linkage equipment self-adjustment is carried out, and when the QCP inspection is not passed, a parameter adjustment instruction is automatically issued; and rechecking is automatically triggered after correction. The method has the effect of guaranteeing the stable quality of the whole production process.

Description

Technical Field

[0001] The present application relates to the technical field of quality control, and in particular to a project quality control method and system. Background Art

[0002] In the equipment manufacturing industry, especially in high-end equipment manufacturing scenarios like quay cranes and ships, quality control is directly related to product safety, reliability, and market competitiveness. As a core component guiding quality inspections during the production process, the Quality Control Plan (QCP) plays a key role in ensuring product quality meets standards.

[0003] Currently, traditional QCP has significant limitations in practical application. First, the development of its inspection standards relies heavily on manual experience. In a high-variety, small-batch production model, it is difficult to quickly and accurately adjust inspection requirements in the face of frequent dynamic process changes, resulting in insufficient timeliness and accuracy in quality control. Second, inspection data suffers from isolated storage issues. The data from the Manufacturing Execution System (MES) and quality inspection equipment is separated, making cross-process quality traceability and root cause analysis difficult to achieve, and unable to effectively support the rapid identification and resolution of quality issues. Summary of the Invention

[0004] In order to automatically generate a dynamic QCP based on real-time process parameters and historical quality data, this application provides a project quality control method and system.

[0005] This application provides a project quality control method and system that adopts the following technical solutions:

[0006] In a first aspect, the present application provides a project quality control method, comprising the following steps:

[0007] Establish a multi-level coding system for production process routes and represent processing constraints through topological sorting;

[0008] Adopt the RETE algorithm of the rule engine to automatically bind QCP inspection items to coding nodes;

[0009] When a process change triggers an event, the rule engine calls the difference analysis module to compare the historical process parameter library and dynamically add or delete inspection items;

[0010] Define component status, including pending inspection, qualified, repaired, and scrapped. The QCP inspection results trigger status transitions in real time and are synchronized to the production planning module via API;

[0011] Based on the process route dependency, a Gantt chart-state linkage model is constructed;

[0012] Based on knowledge graph technology, historical quality data is aggregated to establish a "defect mode-root cause-repair solution" association rule base;

[0013] Perform API-linked device self-adjustment. When a QCP inspection fails, the root cause database matches the top three remediation suggestions. Call the CNC device interface via the RESTful API to automatically issue parameter adjustment instructions.

[0014] Re-inspection is automatically triggered after correction.

[0015] Furthermore, the steps of using the RETE algorithm of the rule engine also include: associating the coding rules of the process route with the conditional judgment logic in the rule engine to achieve dynamic control and optimization of the process flow;

[0016] Design a rule engine based on a data binding framework. By using the data binding framework, dynamic evaluation of conditions and execution of triggers are achieved, and the rule engine is bound to the coding rules of the process route.

[0017] Integrate the rule engine with business rules. The business rules are stored, versioned, and configured through the rule management platform. The rule engine reads and executes the rule set to dynamically control the process route.

[0018] Plan process routes based on cases and rules, build a hierarchical process route case library, and determine the weights of case feature attributes by combining the Delfi method and the hierarchical analysis method;

[0019] Through the backbone constraint matching method, based on group technology and expert system technology, the commonalities of different parts process routes are summarized to form a process route backbone. Combined with constraints such as processing methods and production batches, the process backbone is automatically matched and corrected to launch a process route for new parts.

[0020] Optimize the process route based on the rule engine, introduce processing elements and topological sorting to represent the part processing constraints, establish a machining process route with high efficiency as the goal, and use the interactive platform for optimization;

[0021] Rapidly generate process tasks based on the rule engine. By establishing a mapping description model and an association matrix, combined with enterprise business rules, rapid generation of process task sets is achieved.

[0022] Based on the rule engine mapping process route code, by binding the process route code with the conditional judgment logic in the rule engine, dynamic control and optimization of the process flow can be achieved.

[0023] Furthermore, before matching the root cause database with the top 3 remediation suggestions, the following steps are also included:

[0024] Collect operating data from CNC equipment through API interface and transmit the data to the history recorder server through the network;

[0025] Storing the collected data in a historian server, the historian server comprising a core / master node and distributed / mirror nodes;

[0026] In root cause analysis, the knowledge graph is used to represent the service call relationship. By building the knowledge graph, the efficiency of fault location is improved.

[0027] Through the historical data ECG interface, the historical data of CNC equipment can be filtered, scaled, marked, differentiated, and spectrum calculated;

[0028] By introducing artificial intelligence, the Internet of Things (IoT), and big data technologies, we can collect equipment operation data in real time, achieve self-adjustment and optimization through data analysis, and improve production efficiency and equipment reliability.

[0029] Furthermore, it also includes:

[0030] Establish a unified data center based on the Industrial Internet platform, integrate the MES system, sensor network, and quality inspection equipment data sources through standardized interfaces, and achieve millisecond-level synchronization of process parameters, equipment status, and test results;

[0031] Use edge computing nodes to pre-process real-time data and use time series databases to store highly concurrent data streams, solving the problem of cross-system data fragmentation.

[0032] Deploy blockchain nodes in key processes, store process parameter hash values ​​on the chain in real time, and establish a tamper-proof process quality supervision system;

[0033] Automatically trigger quality inspection tasks through smart contracts to ensure mandatory correlation between data collection and inspection activities;

[0034] Use deep reinforcement learning (DRL) to build a dynamic mapping model between process parameters and quality indicators;

[0035] In the offline training phase, historical process data and corresponding quality inspection results are input, and the thresholds of key control parameters are mined through the Actor-Critic algorithm;

[0036] During the online decision-making phase, the inspection frequency and tolerance are dynamically adjusted based on real-time process flow data;

[0037] A Bayesian network quality transmission model was constructed and nodes were defined, with the quality deviation of the preceding process as the parent node and the defect rate of the subsequent process as the child node. Dynamic reasoning was performed to automatically increase the UT inspection frequency of the closure weld when the pretreatment cleanliness exceeded the standard, and to provide root cause analysis suggestions.

[0038] Perform anomaly detection and use the KNN-PCA algorithm to identify process parameter deviations and automatically trigger a level 3 alarm;

[0039] Conduct cause-and-effect analysis, quantify the impact intensity between processes based on Transfer Entropy, and locate key fault propagation paths;

[0040] Establish a self-evolutionary improvement mechanism and transform the root cause analysis results into a FMEA rule base by establishing a quality improvement knowledge base;

[0041] Through reinforcement learning, the system dynamically optimizes the control strategy. When a new failure mode is identified, it automatically generates a process parameter adjustment plan and verifies its effectiveness, thus achieving continuous iteration of the QCP.

[0042] Blockchain-AI collaborative verification, deploying smart contracts at key QCP execution nodes:

[0043] Automatically verify the conformity of test data with process standards;

[0044] Freeze production instructions for substandard processes and trigger corrective measures;

[0045] Preview process plans in a virtual environment, predict quality risks through Monte Carlo simulation, and optimize the layout of QCP inspection points in advance.

[0046] Furthermore, it also includes:

[0047] Integrate multi-dimensional data including production process routes, process parameters, inspection standards, and historical defect cases;

[0048] Associate each process node with inspection items and disposal measures to form a dynamically scalable triplet network;

[0049] Cross-modal alignment technology is used to integrate structured process parameters with unstructured expert experience to resolve semantic ambiguity in complex inspection items.

[0050] The introduction of an uncertainty reasoning module automatically triggers a Bayesian network-based confidence assessment when new data conflicts with existing knowledge, ensuring knowledge reliability.

[0051] When process parameters change, a multi-layer graph convolutional network (GCN) traverses the knowledge graph to identify affected nodes;

[0052] Introducing the Transformer attention mechanism to quantify the influence strength between nodes and accurately locate the new inspection items that need to be added;

[0053] Design a dual-threshold decision-making mechanism to automatically update the QCP rule base when the inference confidence level is ≥90%; trigger the expert review process when the confidence level is 70-90%; and initiate supplementary training of the reinforcement learning model when the confidence level is below 70%;

[0054] Breaking through the limitations of traditional fixed thresholds, we introduced a sliding window adaptive algorithm to dynamically adjust the confidence threshold based on historical accuracy.

[0055] Establish a real-time weight feedback mechanism, and the test results directly drive the update of graph node weights;

[0056] Build a reinforcement learning optimization strategy and a dual-agent reinforcement learning model;

[0057] Deeply integrated with the MES system, when the weight of the graph node changes, the production plan priority is automatically adjusted and quality inspection resources are dynamically allocated.

[0058] Furthermore, the step of traversing the knowledge graph using the multi-layer graph convolutional network (GCN) further includes:

[0059] Collect historical data and store them in the process database to build the initial diagram based on process fault diagnosis (PFD);

[0060] Divide the dataset into training and test sets, and use the training set to train the model;

[0061] The GCN model extracts and aggregates features through graph convolution operations and activation functions;

[0062] During the training process, graph exploration and knowledge filtering are performed to improve prediction accuracy. If the prediction accuracy drops below a threshold, the model is retrained; otherwise, the model is tested.

[0063] Load the trained GCN model and adjacency matrix, collect and preprocess process data in real time, and feed the real-time input data into the GCN model to obtain the prediction results of key quality variables.

[0064] Furthermore, the step of directly driving the graph node weight update based on the test result also includes:

[0065] Dynamic adjustment of risk weights can be achieved through dynamic capture of spatiotemporal characteristics and quantification of network relationship graphs;

[0066] Adopting federated learning weight pools and quantum reinforcement learning mechanisms to achieve cross-institutional joint modeling and strategic space exploration in complex scenarios;

[0067] Dynamic adjustment of production status is achieved through edge intelligent decision nodes and real-time feedback mechanisms;

[0068] By collecting real-time production status data, the dynamic weight adjustment of each scheduling rule is calculated, and the production scheduling strategy is executed according to the rule priority;

[0069] In a dynamic production environment, the machine vision CNN process can autonomously update feature weights based on disturbances such as lighting changes and equipment vibration to maintain decision stability.

[0070] Through the dynamic control balance mechanism, dynamic real-time processing information is conveyed according to the rolling scheduling of the fault point, and the planning information and scheduling are reorganized;

[0071] By monitoring message backlogs, average latency, and throughput in real time, queue weights are dynamically adjusted to ensure maximum processing efficiency for high-priority queues.

[0072] In the adaptive consensus decision model, dynamic adjustment is achieved by recalculating the attribute-level consensus and updating the attribute weights based on the consensus;

[0073] By shortening the deviation rate between time and energy consumption index values ​​and average aggregation distance, the index weight coefficient is dynamically adjusted to achieve hierarchical optimization of equipment and processes;

[0074] In the digital workshop cloud platform, an improved Nginx dynamic weight load balancing algorithm is used to achieve dynamic distribution of user requests and improve system performance.

[0075] In a second aspect, the present application provides a project quality control system, comprising:

[0076] The coding system establishment module is used to establish a multi-level coding system for production process routes and represent processing constraint relationships through topological sorting;

[0077] The rule engine module is used to automatically bind QCP inspection items to coding nodes using the RETE algorithm of the rule engine;

[0078] Dynamic addition and deletion module: When a process change triggers an event, the rule engine calls the difference analysis module, compares the historical process parameter library, and dynamically adds and deletes inspection items;

[0079] The component definition module is used to define component status, including pending inspection, qualified, repaired, and scrapped. The QCP inspection results trigger status transitions in real time and are synchronized to the production planning module through the API;

[0080] Model building module, used to build Gantt chart-state linkage model based on process route dependencies;

[0081] The rule base association module is used to aggregate historical quality data based on knowledge graph technology and establish a "defect mode-root cause-repair solution" association rule base;

[0082] The parameter adjustment module is used for API-linked equipment self-adjustment. When the QCP inspection fails, the root cause library matches the top three remediation suggestions. The CNC equipment interface is called through the RESTful API to automatically issue parameter adjustment instructions.

[0083] Re-inspection module, used to automatically trigger re-inspection after correction.

[0084] In a third aspect, the present application provides an intelligent terminal comprising a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and execute the above-mentioned project quality control method.

[0085] In a fourth aspect, the present application provides a computer-readable storage medium storing a computer program that can be loaded by a processor and execute the above-mentioned project quality control method.

[0086] In summary, compared with the prior art, the above technical solution has the following beneficial effects:

[0087] The project quality control method described in this application comprehensively improves the quality control efficiency through the collaborative operation of multiple modules. The coding system establishment and rule engine module realize the intelligent binding of inspection items, and the dynamic addition and deletion module flexibly responds to process changes; the component definition and model construction module ensures real-time status monitoring and visual presentation; the rule library association, parameter adjustment and re-inspection module form an "analysis-repair-verification" closed loop, which not only realizes the rapid location and processing of quality problems, but also reduces human intervention with the help of equipment self-adjustment, greatly improving the automation, intelligence level and production efficiency of quality control, and effectively ensuring the quality stability of the entire process of complex equipment production. BRIEF DESCRIPTION OF THE DRAWINGS

[0088] Figure 1 It is a flow chart of a project quality control method according to an embodiment of the present application. DETAILED DESCRIPTION

[0089] The present application is further described in detail below in conjunction with all the accompanying drawings.

[0090] The present application discloses a project quality control method and system, referring to Figure 1 , a project quality control method includes:

[0091] S101. Establish a multi-level coding system for production process routes and represent processing constraint relationships through topological sorting.

[0092] Specifically, at the initial project launch, the quality control system organizes key elements such as processes, equipment, and materials according to the production process route. Each process is assigned a unique process ID, each production equipment is given a unique equipment ID, and each batch of materials is labeled with a material batch ID. This information is integrated into a multi-level code (such as a "process ID-equipment ID-material batch" combination). After coding is completed, all processes are processed using a topological sorting algorithm, arranging the processes into an ordered sequence based on the processing sequence and dependencies. This clearly presents the processing constraints of the entire production process and provides a basic framework for subsequent operations.

[0093] S102. Use the RETE algorithm of the rule engine to automatically bind the QCP inspection items to the coding nodes.

[0094] Specifically, the quality control system introduces a rule engine and uses the RETE algorithm to pre-enter the inspection requirements, standards, and other information for various QCP inspection items (such as welding strength testing and dimensional tolerance measurement). The system automatically scans each node in the established multi-level coding system and matches the corresponding inspection items with the coding nodes based on pre-set binding rules. For example, for coding nodes involving welding processes, the system automatically binds inspection items related to welding strength, achieving a precise association between inspection items and production links, ensuring that each key node in the production process has a corresponding inspection standard.

[0095] S103. When a process change triggers an event, the rule engine calls the difference analysis module, compares the historical process parameter library, and dynamically adds or deletes inspection items.

[0096] Specifically, the quality control system monitors the production process in real time, and once a process change event (such as material replacement, equipment parameter adjustment, etc.) is captured, the rule engine is immediately triggered. The rule engine calls the difference analysis module to perform a detailed comparison of the changed process parameters with the data in the historical process parameter library. After analyzing the specific differences, the corresponding inspection items are dynamically added or deleted according to the pre-set inspection item increase and decrease rules. For example, when a material replacement occurs, the system automatically adds a "material composition spectral analysis" inspection item and incorporates it into the inspection process of the corresponding coding node to ensure that the inspection requirements are updated in a timely manner as the process changes.

[0097] S104: Define component status. The QCP inspection result triggers status transition in real time and is synchronized to the production planning module through the API.

[0098] Specifically, the quality control system clearly defines four states for components: pending inspection, qualified (marked in green), repair (marked in red), and scrapped (marked in black). After the QCP inspection is completed, the system automatically triggers the component status transition based on the inspection results. For example, the status of components that pass the inspection is updated to "qualified" and displayed in green, while components that fail the inspection are converted to "repair" status and marked in red. At the same time, the component status information is synchronized to the production planning module in real time through the API interface, so that the production plan can be adjusted in time according to the actual quality situation.

[0099] S105. Based on the process route dependency, a Gantt chart-state linkage model is constructed.

[0100] Specifically, the quality control system constructs a Gantt chart-status linkage model based on the established process route dependency and coding system. The planned start time, end time and progress of each process are displayed in the Gantt chart, and the component status is associated with the Gantt chart. When the status of a component in a process changes to "rework", the system automatically marks the process and its subsequent dependent processes in red in the Gantt chart, and triggers a work abnormality reminder, notifies relevant personnel via SMS, and highlights the abnormal information on the production dashboard. In addition, the system starts a resource rescheduling algorithm to reasonably reallocate resources based on the current usage of production resources (equipment, working hours, etc.) to ensure the smooth progress of the production plan.

[0101] S106. Based on knowledge graph technology, aggregate historical quality data and establish a "defect mode-root cause-repair solution" association rule library.

[0102] Specifically, the quality control system uses knowledge graph technology to conduct in-depth mining and integration of historical quality data. It collects various quality-related data such as welding current fluctuation frequency and tool wear curves, and analyzes the inherent connections between the data. Through data mining and machine learning algorithms, it summarizes the relationship between "defect pattern-root cause-repair solution" and constructs an association rule base. For example, when it is found that the welding current fluctuates frequently and welding defects occur, the possible root cause is analyzed to be unstable welding machine voltage, and the corresponding repair solution (calibrating the welding machine voltage) is entered into the rule base to provide a reference basis for subsequent quality problem handling.

[0103] S107: Perform API linkage device self-adjustment.

[0104] Specifically, when the quality control system indicates a failure in the QCP inspection, the system quickly matches the top three repair suggestions from the association rule library. For example, if the inspection reveals that the welding quality does not meet the standard, the system recommends a repair solution such as "calibrate the welding machine voltage" based on historical data and the rule library. Then, by calling the CNC device interface through a RESTful API, the repair suggestions are converted into specific parameter adjustment instructions, which are automatically issued to the corresponding equipment to achieve self-adjustment. After the equipment completes the parameter adjustment and performs the correction operation, the system automatically triggers the re-inspection process and re-inspects the component until it passes the inspection, forming a complete "inspection-analysis-correction" quality control closed loop.

[0105] In another embodiment, the steps of using the RETE algorithm of the rule engine further include:

[0106] During the system initialization phase, elements such as process IDs, equipment IDs, and material batches within the multi-level coding system are mapped to the conditional judgment logic within the rule engine. For example, a corresponding inspection standard rule is configured for a specific process ID. When the rule engine detects production activity for that process ID, the relevant conditional judgment is automatically triggered. This association allows the system to dynamically adjust the process flow based on the coding rules. For example, when a material batch changes, the relevant material inspection process is automatically triggered, achieving dynamic control and optimization of the process flow.

[0107] The quality control system's rule engine is designed based on a data binding framework, employing the dependency property mechanism and MVVM pattern within WPF to build its architecture. Process parameters, inspection standards, and other data are defined as dependency properties, enabling two-way data binding. The conditional evaluation logic for the rules is implemented in the ViewModel layer, automatically triggering the re-evaluation of related rules when data changes. This tightly binds the rule engine to the coded rules of the process routing, ensuring that dynamic rule evaluation and trigger execution can respond to changes in production data in real time.

[0108] The quality control system integrates a rules engine with business rules to create a rules management platform for storing, versioning, and configuring business rules related to production scheduling. Rules are stored in a standardized format (such as XML or JSON) and contain information such as conditional expressions, execution actions, and priorities. At runtime, the rules engine reads the currently effective rule set from the platform and dynamically executes it based on production status. For example, when the production plan changes, the rules engine automatically reads the corresponding scheduling rules and adjusts the process routing and resource allocation, enabling dynamic control of the process routing.

[0109] Based on case and rule-based process routes, a hierarchical process route case library is constructed, categorizing and storing historical process routes by dimensions such as part type and processing characteristics. Domain experts are invited to assess the importance of case feature attributes using the Delphi method, and the weights of each attribute are determined using the Analytic Hierarchy Process (AHP). When planning a new part process route, the system searches the case library based on part characteristics, matches the most similar cases, and adjusts and applies the case rules based on the weights, improving the standardization and efficiency of process route planning.

[0110] Process routes are generated using a backbone constraint matching method. Parts are categorized using grouping technology, and common characteristics across different part process routes are summarized to form a backbone. Expert system technology is used to convert constraints such as processing methods, production batches, and equipment capacity into a set of rules. When a new part process route is generated, the system first matches the closest backbone. It then modifies and optimizes the backbone based on specific constraints to automatically generate a process route that meets the requirements.

[0111] A rule-based engine optimizes process routes, introducing the concept of processing elements to decompose complex machining processes into basic processing units. Topological sorting is used to characterize part processing constraints. A high-efficiency machining process route model is established, converting optimization objectives such as machining time and equipment utilization into rules. Using an interactive platform, process engineers input optimization requirements, and the rule engine adjusts and optimizes the process route based on pre-set rules, ultimately generating a process plan that meets these requirements.

[0112] A mapping model is established between process tasks, product structure, and process routing. The dependencies and constraints between tasks are clearly defined through an association matrix. Integrating enterprise business rules (such as resource allocation principles and production priorities), the rule engine automatically analyzes the product structure and process routing to generate a corresponding set of process tasks. During this generation process, the system dynamically adjusts task parameters based on real-time resource status and production plans to ensure task feasibility and efficiency.

[0113] XML rule entries define the mapping between process route codes and the rule engine's conditional logic. Create a trigger class responsible for converting XML rules into executable conditional objects. When the state of a coded node changes during production, the trigger class automatically captures the event, converts the coded information into a conditional object, and performs logical operations. Based on the results, the trigger class executes corresponding actions (such as adjusting inspection items or triggering equipment adjustments), enabling dynamic control and optimization of the process flow.

[0114] In another embodiment, before the step of matching the root cause database with the top 3 remediation suggestions, the step further includes building a historical root cause database:

[0115] Deploy an API interface module to support various data source connection methods, such as OPC DA Collector and OPC UA Collector, and establish communication connections with CNC equipment via network protocols. Configure the corresponding API interface (such as cnc_api) for each type of CNC equipment and call its data acquisition method (such as get_data()) to collect real-time equipment operating parameters, including current, voltage, temperature, speed, and vibration frequency. Also, set the data collection frequency and range to ensure sufficient information about the equipment's operating status.

[0116] Build a history record server cluster consisting of core / master nodes and distributed / mirror nodes. The core nodes are responsible for receiving data collected from CNC equipment and performing preliminary data cleaning and format conversion to ensure data standardization. Distributed nodes store processed data and also perform diagnostic management and alarm archiving functions. Introducing edge computing technology, deploying edge computing nodes close to data sources allows them to pre-process real-time data, such as fault alarm identification and tool life prediction. This reduces data transmission volume and latency, improving system response speed.

[0117] Adhering to industry standards such as API TR 7CR and API Std 20G, we designed interface specifications between CNC equipment and external systems, ensuring consistency in data formats and transmission protocols. We also developed interfaces such as the Device Shared Data Service API and the Device Health Data Service API to enable cross-system data sharing. For example, through the Device Health Data Service API, the quality control system can obtain real-time device status information, providing data support for root cause analysis. Furthermore, we established API documentation and testing mechanisms to ensure interface stability and scalability.

[0118] Model the relationship between knowledge graph construction and services, collect multi-source data such as equipment operation data, process parameters, and quality inspection results, and build the basic data set for the knowledge graph. Define entity types (such as equipment, process, quality defects, etc.) and relationship types (such as "cause," "association," "call," etc.) in the knowledge graph. Utilize graph database technology to model service call relationships as nodes and edges in the knowledge graph. For example, service A calling service B is represented as a directed edge from node A to node B. Through the construction of the knowledge graph, a network of associations between equipment operating status and quality defects is formed, providing structured data support for root cause analysis.

[0119] Develop an ECG interface for historical data, integrating functions such as data filtering, scaling, tagging, differencing, and spectrum calculation. This interface allows analysts to intuitively view trends in historical device data and identify unusual fluctuations. Use the tagging function to annotate key events, such as the time of equipment failure or quality defect occurrence. Use differencing and spectrum calculation functions to deeply analyze data variation characteristics and identify potential fault signs. Furthermore, integrate a history record service (such as Activiti7's HistoryService) to record and query historical system operation information, providing support for root cause analysis.

[0120] Introducing artificial intelligence, the Internet of Things (IoT), and big data technologies to build an intelligent CNC system. IoT sensors are deployed on equipment to collect real-time operating data and transmit it to cloud servers via the network. Big data technologies are used to store and manage massive amounts of historical data, and artificial intelligence algorithms are employed to conduct in-depth analysis of this data, uncovering underlying patterns. Based on these analysis results, the system can self-regulate and optimize, such as automatically adjusting equipment parameters, predicting tool life, and providing early warning of potential failures. This improves production efficiency and equipment reliability, providing more accurate and comprehensive data support for the construction of a root cause library.

[0121] In another embodiment, a project quality control method further includes:

[0122] We built a data center architecture centered around the Industrial Internet of Things (IIoT) platform, deploying standardized communication protocol modules at the system's bottom layer, supporting industry-standard interfaces such as Modbus and OPC UA. We developed data access plug-ins tailored to different data sources, including MES systems, sensor networks, and quality inspection equipment, converting process parameters, equipment operating status, and quality inspection results into a unified format. Using message queues and distributed caching technologies, we achieved millisecond-level synchronous transmission of multi-source data, ensuring data timeliness and consistency.

[0123] Edge computing nodes are deployed at production sites and configured with data preprocessing algorithm modules to perform operations such as denoising and normalization on real-time sensor data. In this embodiment, a sliding average filter algorithm is used to eliminate high-frequency noise, and linear normalization is used to normalize data of different dimensions to a standard range. The processed data is directly written to a time series database. Leveraging its high concurrency write and time series query capabilities, it efficiently stores real-time data streams from device operations, breaking down the data silos between traditional systems.

[0124] A data fusion engine was designed to establish a correlation model for the three data sources of "process flow, quality inspection flow, and operation and maintenance flow." Through the unified data view of the data center, process execution records, quality inspection results, and equipment operation and maintenance logs were aligned and associated with production batches by timestamp. For example, the welding process current parameters (process flow), weld flaw detection results (quality inspection flow), and welding machine maintenance records (operation and maintenance flow) were integrated into a single dataset, providing comprehensive data input for the dynamic quality control plan (QCP).

[0125] Blockchain nodes are deployed at key process stations, such as welding and heat treatment. These nodes have built-in hash calculation modules that perform real-time hashing of process parameters (such as welding current and heat treatment temperature curves) and package the hash values ​​into blocks. Using consortium chain technology, each process node is connected into a network to ensure data immutability. Smart contracts are also written to set triggering conditions for quality inspection tasks. For example, once a welding process is completed and the parameter hash values ​​are uploaded to the blockchain, a weld flaw detection task is automatically assigned to quality inspectors, forcing the linkage of data collection and inspection activities.

[0126] Build a full-link trusted traceability system based on blockchain-stored evidence and multi-source information from the data center. Each production batch is assigned a unique traceability code. By scanning the product QR code or RFID tag, data from the entire process, from raw material procurement and production to finished product quality inspection, can be retrieved. The system establishes data indexes and association maps, enabling "forward tracing" (from raw materials to finished product) and "reverse tracing" (from finished product to raw materials), significantly improving product recall efficiency.

[0127] A deep reinforcement learning model was constructed. During the offline training phase, historical process data (such as temperature, pressure, vibration, and other parameters) and corresponding quality inspection results were collected to construct a training dataset. A deep reinforcement learning (DRL) model was built using the Actor-Critic algorithm. Through continuous trial and error and reward feedback mechanisms, the correlation between process parameters and quality indicators was explored, and the thresholds for key control parameters were determined. During the online decision-making phase, the model received real-time production site data and dynamically adjusted the inspection frequency and quality tolerance based on environmental changes. For example, the inspection standard for welding deformation of the quay crane main beam was automatically adjusted based on the ambient temperature and humidity.

[0128] The node-edge relationships of the Bayesian network are defined, with the quality parameters of the previous process (such as steel pretreatment cleanliness) as the parent node and the defect rate of the subsequent process (such as the pass rate of the segmented closure weld) as the child node. A conditional probability table is then established to describe the dependencies between the nodes. When the system detects an abnormality in the parameters of the parent node (such as pretreatment cleanliness exceeding the standard), the Bayesian inference algorithm is triggered, automatically increasing the inspection frequency of the corresponding process of the child node (such as increasing the number of ultrasonic inspections of welds). Root cause analysis suggestions are then pushed based on probability calculations, enabling a rapid response to quality issues.

[0129] A multimodal root cause location system was established. The anomaly detection layer uses the KNN-PCA algorithm to analyze real-time process parameters. After dimensionality reduction through principal component analysis, the K-nearest neighbor algorithm is used to identify parameter offsets. When an anomaly is detected (such as a sudden change in temperature during ship shafting processing), three levels of alarm (yellow, orange, and red) are triggered based on the degree of offset. The causal analysis layer, based on the transfer entropy algorithm, quantifies the impact of parameter changes between different processes. By constructing an influence network diagram, it locates the fault propagation path. For example, it discovered the potential impact of paint shop humidity on electrical equipment processes, solving the problem of multi-factor fault diagnosis in complex equipment manufacturing.

[0130] A quality improvement knowledge base was established to store root cause analysis results and improvement measures. For example, the analysis conclusion that "a welding current fluctuation of ±5A causes a 30% increase in porosity" was converted into a Failure Mode and Effects Analysis (FMEA) rule and updated to the rule base. A reinforcement learning algorithm was used to continuously optimize the control strategy. When the system identified a new failure mode, it automatically generated a process parameter adjustment plan. Its effectiveness was verified through small-batch trial production, enabling dynamic, iterative upgrades of the QCP.

[0131] A blockchain-AI collaborative verification mechanism is deployed, with smart contracts embedded in the AI ​​verification module at key QCP execution nodes (such as stress testing before quay crane installation). The smart contracts capture test data (such as bolt preload values) in real time and compare them with process standards (such as a ±2% tolerance band). If the data falls short of the standard, production instructions are automatically frozen and corrective measures, such as notifying technicians to recalibrate equipment, are triggered. By leveraging the immutable nature of blockchain and automated AI verification, the entire inspection, judgment, and disposal process is automated.

[0132] Build a digital twin of equipment manufacturing by integrating CAD design models, physical simulation rules, and real-time IoT data. Import production process plans into a virtual environment and use Monte Carlo simulation algorithms to predict quality risks through large-scale random sampling. Based on simulation results, optimize the layout of inspection points and frequency in the QCP to prevent potential quality issues before physical production begins.

[0133] In another embodiment, a project quality control method further includes:

[0134] A data integration platform was established to clean and standardize heterogeneous data such as production process routes, process parameters, inspection standards, and historical defect cases. Data parsing modules were designed for structured data (such as process parameters) and unstructured data (such as expert experience documents) to extract key information. A hypergraph data structure was constructed, defining "process parameters," "inspection items," and "defect types" as node types. The relationships between nodes (such as "welding current > 30A" associated with the "crack risk" inspection item) were represented as hyperedges, forming a dynamically scalable triple network.

[0135] The Apache Flink real-time data streaming engine was deployed, establishing a data channel with the MES system through standardized interfaces, ingesting real-time production data at a rate of thousands of records per second. An incremental update algorithm was designed to automatically parse the data semantics and map it into a hypergraph structure as new data enters the system. For example, when a new process parameter (such as the use of a new welding material) is detected, the system automatically creates a corresponding node and, based on pre-set rules, generates associations with existing inspection items and defect types, enabling the dynamic growth of the knowledge graph.

[0136] Graph embedding alignment technology is used to semantically fuse structured process parameters with unstructured expert experience. For structured data, feature extraction algorithms are used to convert process parameters into low-dimensional vector representations. For unstructured text (such as expert inspection reports), natural language processing techniques are used to extract keywords and generate semantic vectors. By designing a cross-modal loss function to minimize the distance between data from different modalities in the embedding space, semantic ambiguity in complex inspection items such as "weld metallographic structure inspection" is resolved, ensuring the consistency of knowledge representation.

[0137] An uncertainty reasoning module is introduced into the knowledge graph, associating a confidence score with each piece of knowledge. When new data conflicts with existing knowledge (e.g., new process parameters conflict with historical defect cases), a Bayesian network is triggered to perform a confidence assessment. The system collects relevant evidence (such as historical performance of similar processes and expert opinions) and calculates the posterior probability of the conflicting knowledge. Based on the evaluation results, the knowledge is graded: high-confidence knowledge is directly updated in the graph, while low-confidence knowledge is marked as pending verification, ensuring the reliability of the knowledge base.

[0138] A multi-layer graph convolutional network (GCN) is constructed. When process parameters change, it automatically traverses the knowledge graph and identifies affected nodes and their relationships. For example, when an increase in welding current is detected, GCN uses propagation calculations to identify related nodes such as "expansion of the heat-affected zone" and "increased crack risk." The Transformer attention mechanism is introduced to quantify the influence intensity between nodes. By training the model, an attention weight is assigned to each relationship (for example, the weight of the current parameter on the crack risk is 0.93). Based on the weight, the inspection items that need to be added (such as "weld metallographic structure inspection") are accurately located to achieve intelligent decision-making.

[0139] A dual-threshold decision-making mechanism was designed, with two confidence thresholds set at 90% and 70%. When the inference confidence level is ≥90%, the system automatically updates the QCP rule base. When the confidence level is between 70% and 90%, the expert review process is triggered, and the decision request is pushed to the quality control expert terminal. When the confidence level is below 70%, the reinforcement learning model is activated, using historical data and current conflict cases for supplementary training. A sliding window adaptive algorithm was introduced to analyze historical decision accuracy. During periods of stable quality, the confidence threshold was lowered to 85% to improve decision efficiency, and raised to 95% during periods of quality fluctuation to ensure decision reliability.

[0140] A real-time weighted feedback system is established to directly link inspection results with knowledge graph node weights. When a process fails inspection, a negative feedback mechanism is triggered, exponentially reducing the weight of the relevant nodes (e.g., multiplying by 0.7). The production task is automatically downgraded to "high risk," and the inspection frequency is increased. When 10 consecutive batches pass inspection, a positive feedback mechanism is activated, reducing the weight of redundant inspection items (e.g., UT flaw detection frequency -1), and reallocating quality inspection resources to high-risk nodes. This dynamic adjustment allows for optimal allocation of quality inspection resources.

[0141] A dual-agent reinforcement learning model (Dual-Agent RL) is deployed, with local agents deployed at edge computing nodes to process real-time data and make rapid decisions. The local agents fine-tune inspection frequency based on pre-set policies, keeping response latency below 800ms and ensuring timely responses to production anomalies. A global agent is deployed in the cloud, utilizing quantum reinforcement learning technology (512-QPU) to simulate 500,000 weight combinations per second. By optimizing the long-term quality cost function, the globally optimal quality control strategy is developed. These two levels of agents work collaboratively to achieve a balance between real-time performance and global optimization.

[0142] Develop a two-way integration interface with the MES system to realize the linkage between the knowledge graph and the production plan. When the weight of the knowledge graph node changes (such as the risk level of a process increases), the system automatically sends a priority adjustment request to the MES system to queue up high-risk tasks. At the same time, quality inspection resources are dynamically allocated according to the node weight, such as prioritizing the allocation of high-end inspection equipment such as infrared thermal imagers to high-risk processes. Through this integration, full-process automated collaboration from quality control to production execution is achieved.

[0143] In another embodiment, the step of traversing the knowledge graph using a multi-layer graph convolutional network (GCN) further includes:

[0144] Deploy a data acquisition system to collect real-time process data from production equipment through standardized interfaces (such as OPC UA), including controller outputs, control valve status, loop parameters, and transmitter readings, and store them in a process database. Based on the process fault diagnosis (PFD) framework, analyze the causal relationships between each link in the production process, map entities such as equipment, parameters, and failure modes into graph nodes, and define the physical connections, data flows, and causal influences between entities as graph edges to construct an initial knowledge graph. For example, in a debutanizer system, devices such as temperature sensors, pressure controllers, and reflux valves are used as nodes, and the material flow and control relationships between devices are used as edges.

[0145] Develop a data preprocessing module to clean the collected raw data and remove noise and outliers. Design feature extraction algorithms for different parameter types (such as temperature, pressure, and flow) to extract time-domain features (such as mean and variance) and frequency-domain features (such as spectral energy). Use normalization techniques (such as Min-Max scaling) to unify multi-dimensional and multi-dimensional features into a standard range, ensuring that the data format is suitable for GCN processing. For example, normalize the controller output current signal (4-20mA) to the range [0,1].

[0146] The preprocessed dataset is divided into training and test sets in proportion. The GCN model architecture is constructed, typically consisting of multiple layers of graph convolutional layers, with each layer sequentially performing graph convolution operations and nonlinear activation function operations. During training, the model continuously adjusts parameters through iterative optimization to learn the mapping relationship between node features and graph structure. At the same time, a graph exploration mechanism is implemented to dynamically discover hidden association patterns in the graph; a knowledge filtering strategy is introduced to eliminate noise associations and improve prediction accuracy. A threshold for accuracy drop (e.g., 5%) is set, and when the prediction accuracy drops below the threshold, the model retraining process is triggered.

[0147] Design a multi-layer GCN model structure, adjusting the number of layers and neurons based on the specific application scenario. For example, in control loop fault diagnosis, a two-layer GCN structure is used. The first layer extracts local features, the second layer aggregates global information, and finally, a pooling layer performs dimensionality reduction to obtain a seven-category classification result (such as normal, sensor failure, actuator failure, etc.). To improve model performance, try different activation functions (such as ReLU, GELU) and optimizers (such as Adam, SGD), and adjust hyperparameters such as the learning rate. Use cross-validation to evaluate model performance under different parameter configurations and select the optimal solution.

[0148] Use the test set to comprehensively evaluate the trained GCN model, calculating key metrics such as classification accuracy, loss, and recall. If model performance falls short of expectations, analyze the source of the error and make targeted adjustments to the model structure (such as increasing the number of GCN layers) or optimize the training strategy (such as introducing regularization methods). Use hyperparameter search algorithms (such as grid search and random search) to automatically find the optimal parameter combination. Furthermore, introduce model fusion techniques to combine GCN with other models (such as the Transformer) to leverage their respective strengths and enhance overall predictive capabilities.

[0149] An online prediction system is deployed at the production site, loaded with a trained GCN model and pre-calculated adjacency matrix. Process data is collected in real time, pre-processed, and fed into the model to obtain predictions for key quality variables. For example, during the operation of a debutanizer, the system collects parameters such as temperature and pressure in real time and uses the KBCE-GCN model to predict the purity of the overhead product. A visualization interface for prediction results is developed to compare predicted values ​​with actual values. Anomaly warning thresholds are also set to automatically trigger alarms when predicted results deviate from the normal range.

[0150] In order to improve the efficiency and scalability of the model, D 2 The GCN framework uses a dynamic skip connection strategy based on multi-granularity data dependencies to reduce redundant computation and lower resource consumption. New production data is regularly collected, and incremental learning methods are used to update the model to adapt to process changes. We continue to explore model fusion technologies, such as combining the graph structure modeling capabilities of GCN with the sequence modeling advantages of Transformer to build a more powerful hybrid model architecture, further improving the accuracy and robustness of process parameter reasoning.

[0151] In another embodiment, the step of directly driving the graph node weight update based on the inspection result further includes:

[0152] An electronic fence monitoring module has been installed in relevant systems. For manufacturing plants, a 500-meter radius electronic fence is set. Real-time electronic fence status data is collected. When deviations exceed the limit are detected, a weight adjustment mechanism is immediately triggered, increasing the weight of the corresponding risk node by 30%. For cross-border e-commerce scenarios, GPS positioning technology is used to obtain the real-time coordinates of overseas warehouses and compare them with the coordinates of the customs declaration location. If the distance exceeds 2,000 kilometers, the system automatically issues a manual review instruction.

[0153] Build a federated learning platform and establish data exchange channels with commercial banks, logistics platforms, power grid companies, and other institutions. Each participant contributes their data to the federated learning weight pool. The platform analyzes and processes this data through quantum reinforcement learning mechanisms, dynamically adjusting weighting factors. For example, based on capital flow data provided by commercial banks, cargo transportation trajectory data from logistics platforms, and electricity consumption data from power grid companies, it comprehensively assesses and adjusts risk weights related to project quality.

[0154] Deploy edge intelligent decision-making nodes in high-risk areas such as digital asset trading. These nodes collect real-time on-site data, such as digital asset transaction records. Based on pre-set rules and algorithms, they analyze and make decisions based on this data, enabling dynamic adjustments to production status. Decision results are also fed back to relevant systems in real time, enabling timely further adjustments.

[0155] A real-time production status data collection system was established to comprehensively collect various data from the production process, including equipment operating status, order progress, and personnel efficiency. Based on this data, dynamic weight adjustments for each scheduling rule were calculated. For example, for orders with tight delivery deadlines, their production scheduling priority was increased accordingly. Production scheduling strategies were then executed based on the rule priorities to optimize production processes and improve production efficiency.

[0156] In dynamic production environments, machine vision technology is used to monitor the production process. A CNN (convolutional neural network) process processes and analyzes image data to identify product quality characteristics. When encountering disturbances such as lighting changes and equipment vibration, the CNN process automatically adjusts feature weights to maintain decision stability. For example, when light intensity fluctuates, the recognition weight of product edge features is automatically increased to ensure accurate judgment of product quality.

[0157] A dynamic control and balance mechanism has been established within the workshop production management system. When a fault occurs, a rolling scheduling process is initiated, re-adjusting production plans and schedules based on the fault's scope and severity. By communicating dynamic processing information in real time, each process can make timely adjustments to achieve efficient production. For example, if a key piece of equipment fails, the system quickly adjusts the production tasks of other equipment and rearranges the production sequence to minimize the impact on the overall production schedule.

[0158] A message queue monitoring module is set up in the system to monitor metrics such as message backlog, average latency, and throughput in real time. Based on these metrics, queue weights are dynamically adjusted to ensure maximum processing efficiency for high-priority queues. For example, if a large message backlog is detected in a production instruction queue, the weight of that queue is automatically increased, prioritizing production instructions and ensuring smooth production.

[0159] An adaptive consensus decision-making model is introduced into the decision-making system. By recalculating attribute-level consensus, the consistency and reliability of each decision factor are evaluated. Then, based on the consensus level, the attribute weights based on consensus are updated, enabling dynamic adjustments. For example, when evaluating product quality, the consensus level of multiple quality indicators is comprehensively considered and the weights are adjusted to make decisions more accurate and reasonable.

[0160] Real-time data collection of equipment metrics such as operating time and energy consumption is performed, and the deviation rate from the average cluster distance is calculated. Dynamically adjust the indicator weight coefficient based on the deviation rate to achieve hierarchical optimization of equipment and processes. For example, for equipment with high energy consumption, the weight of the energy consumption indicator is appropriately increased to encourage the implementation of energy-saving measures; for processes with low production efficiency, the weight is increased to focus on optimization and improvement.

[0161] The digital workshop cloud platform uses an improved Nginx dynamic weighted load balancing algorithm. This algorithm monitors the load of each server node in real time and dynamically adjusts the node weights based on the load conditions, enabling dynamic distribution of user requests. For example, if a server node is overloaded, its weight is automatically reduced, and user requests are distributed to other, less loaded nodes, thereby improving overall system performance.

[0162] Based on the above method, the embodiment of the present application also discloses a project quality control system. A project quality control system includes:

[0163] The coding system establishment module is used to establish a multi-level coding system for production process routes and represent processing constraint relationships through topological sorting;

[0164] The rule engine module is used to automatically bind QCP inspection items to coding nodes using the RETE algorithm of the rule engine;

[0165] Dynamic addition and deletion module: When a process change triggers an event, the rule engine calls the difference analysis module, compares the historical process parameter library, and dynamically adds and deletes inspection items;

[0166] The component definition module is used to define component status, including pending inspection, qualified, repaired, and scrapped. The QCP inspection results trigger status transitions in real time and are synchronized to the production planning module through the API;

[0167] Model building module, used to build Gantt chart-state linkage model based on process route dependencies;

[0168] The rule base association module is used to aggregate historical quality data based on knowledge graph technology and establish a "defect mode-root cause-repair solution" association rule base;

[0169] The parameter adjustment module is used for API-linked equipment self-adjustment. When the QCP inspection fails, the root cause library matches the top three remediation suggestions. The CNC equipment interface is called through the RESTful API to automatically issue parameter adjustment instructions.

[0170] Re-inspection module, used to automatically trigger re-inspection after correction.

[0171] An embodiment of the present application further discloses an intelligent terminal, which includes a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and execute a project quality control method such as the above-mentioned one.

[0172] The present application also discloses a computer-readable storage medium. The computer-readable storage medium stores a computer program that can be loaded by a processor and execute the project quality control method described above. The computer-readable storage medium includes, for example, a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, among other media capable of storing program code.

[0173] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the scope of protection of the invention. Obviously, the embodiments described are only some embodiments of the present invention, rather than all embodiments. Based on these embodiments, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in this field can still combine, add, delete or make other adjustments to the features in the various embodiments of the present invention according to the circumstances without conflict, without making creative work, so as to obtain different other technical solutions that do not deviate from the concept of the present invention in essence, and these technical solutions also fall within the scope of protection of the present invention.

Claims

1. A project quality control method, characterized in that: The following steps are involved: Establish a multi-level coding system for production process routes and represent processing constraints through topological sorting; Adopt the RETE algorithm of the rule engine to automatically bind QCP inspection items to coding nodes; When a process change triggers an event, the rule engine calls the difference analysis module to compare the historical process parameter library and dynamically add or delete inspection items; Define component status, including pending inspection, qualified, repaired, and scrapped. The QCP inspection results trigger status transitions in real time and are synchronized to the production planning module via API; Based on the process route dependency, a Gantt chart-state linkage model is constructed; Based on knowledge graph technology, historical quality data is aggregated to establish a "defect mode-root cause-repair solution" association rule base; Perform API-linked device self-adjustment. When a QCP inspection fails, the root cause database matches the top three remediation suggestions. Call the CNC device interface via the RESTful API to automatically issue parameter adjustment instructions. Re-inspection is automatically triggered after correction.

2. A project quality control method according to claim 1, characterized in that: The steps of using the RETE algorithm of the rule engine also include: associating the coding rules of the process route with the conditional judgment logic in the rule engine to achieve dynamic control and optimization of the process flow; Design a rule engine based on a data binding framework. By using the data binding framework, dynamic evaluation of conditions and execution of triggers are achieved, and the rule engine is bound to the coding rules of the process route. Integrate the rule engine with business rules. The business rules are stored, versioned, and configured through the rule management platform. The rule engine reads and executes the rule set to dynamically control the process route. Plan process routes based on cases and rules, build a hierarchical process route case library, and determine the weights of case feature attributes by combining the Delfi method and the hierarchical analysis method; Through the backbone constraint matching method, based on group technology and expert system technology, the commonalities of different parts process routes are summarized to form a process route backbone. Combined with constraints such as processing methods and production batches, the process backbone is automatically matched and corrected to launch a process route for new parts. Optimize the process route based on the rule engine, introduce processing elements and topological sorting to represent the part processing constraints, establish a machining process route with high efficiency as the goal, and use the interactive platform for optimization; Rapidly generate process tasks based on the rule engine. By establishing a mapping description model and an association matrix, combined with enterprise business rules, rapid generation of process task sets is achieved. Based on the rule engine mapping process route code, by binding the process route code with the conditional judgment logic in the rule engine, dynamic control and optimization of the process flow can be achieved.

3. A project quality control method according to claim 1, characterized in that: Before matching the root cause database with the top 3 remediation suggestions, the following steps are also included: Collect operating data from CNC equipment through API interface and transmit the data to the history recorder server through the network; Storing the collected data in a historian server, the historian server comprising a core / master node and distributed / mirror nodes; In root cause analysis, the knowledge graph is used to represent the service call relationship. By building the knowledge graph, the efficiency of fault location is improved. Through the historical data ECG interface, the historical data of CNC equipment can be filtered, scaled, marked, differentiated, and spectrum calculated; By introducing artificial intelligence, the Internet of Things (IoT), and big data technologies, we can collect equipment operation data in real time, achieve self-adjustment and optimization through data analysis, and improve production efficiency and equipment reliability.

4. A project quality control method according to claim 1, characterized in that: Also includes: Establish a unified data center based on the Industrial Internet platform, integrate the MES system, sensor network, and quality inspection equipment data sources through standardized interfaces, and achieve millisecond-level synchronization of process parameters, equipment status, and test results; Use edge computing nodes to pre-process real-time data and use time series databases to store highly concurrent data streams, solving the problem of cross-system data fragmentation. Deploy blockchain nodes in key processes, store process parameter hash values ​​on the chain in real time, and establish a tamper-proof process quality supervision system; Automatically trigger quality inspection tasks through smart contracts to ensure mandatory correlation between data collection and inspection activities; Use deep reinforcement learning (DRL) to build a dynamic mapping model between process parameters and quality indicators; In the offline training phase, historical process data and corresponding quality inspection results are input, and the thresholds of key control parameters are mined through the Actor-Critic algorithm; During the online decision-making phase, the inspection frequency and tolerance are dynamically adjusted based on real-time process flow data; A Bayesian network quality transmission model was constructed and nodes were defined, with the quality deviation of the preceding process as the parent node and the defect rate of the subsequent process as the child node. Dynamic reasoning was performed to automatically increase the UT inspection frequency of the closure weld when the pretreatment cleanliness exceeded the standard, and to provide root cause analysis suggestions. Perform anomaly detection and use the KNN-PCA algorithm to identify process parameter deviations and automatically trigger a level 3 alarm; Conduct cause-and-effect analysis, quantify the impact intensity between processes based on Transfer Entropy, and locate key fault propagation paths; Establish a self-evolutionary improvement mechanism and transform the root cause analysis results into a FMEA rule base by establishing a quality improvement knowledge base; Through reinforcement learning, the system dynamically optimizes the control strategy. When a new failure mode is identified, it automatically generates a process parameter adjustment plan and verifies its effectiveness, thus achieving continuous iteration of the QCP. Blockchain-AI collaborative verification, deploying smart contracts at key QCP execution nodes: Automatically verify the conformity of test data with process standards; Freeze production instructions for substandard processes and trigger corrective measures; Preview process plans in a virtual environment, predict quality risks through Monte Carlo simulation, and optimize the layout of QCP inspection points in advance.

5. A project quality control method according to claim 1, characterized in that: Also includes: Integrate multi-dimensional data including production process routes, process parameters, inspection standards, and historical defect cases; Associate each process node with inspection items and disposal measures to form a dynamically scalable triplet network; Cross-modal alignment technology is used to integrate structured process parameters with unstructured expert experience to resolve semantic ambiguity in complex inspection items. The introduction of an uncertainty reasoning module automatically triggers a Bayesian network-based confidence assessment when new data conflicts with existing knowledge, ensuring knowledge reliability. When process parameters change, a multi-layer graph convolutional network (GCN) traverses the knowledge graph to identify affected nodes; Introducing the Transformer attention mechanism to quantify the influence strength between nodes and accurately locate the new inspection items that need to be added; Design a dual-threshold decision-making mechanism to automatically update the QCP rule base when the inference confidence level is ≥90%; trigger the expert review process when the confidence level is 70-90%; If the rate is lower than 70%, the reinforcement learning model will be started for supplementary training; Breaking through the limitations of traditional fixed thresholds, we introduced a sliding window adaptive algorithm to dynamically adjust the confidence threshold based on historical accuracy. Establish a real-time weight feedback mechanism, and the test results directly drive the update of graph node weights; Build a reinforcement learning optimization strategy and a dual-agent reinforcement learning model; Deeply integrated with the MES system, when the weight of the graph node changes, the production plan priority is automatically adjusted and quality inspection resources are dynamically allocated.

6. A project quality control method according to claim 1, characterized in that: The step of traversing the knowledge graph using the multi-layer graph convolutional network (GCN) further includes: Collect historical data and store them in the process database to build the initial diagram based on process fault diagnosis (PFD); Divide the dataset into training and test sets, and use the training set to train the model; The GCN model extracts and aggregates features through graph convolution operations and activation functions; During the training process, graph exploration and knowledge filtering are performed to improve prediction accuracy. If the prediction accuracy drops below a threshold, the model is retrained; otherwise, the model is tested. Load the trained GCN model and adjacency matrix, collect and preprocess process data in real time, and feed the real-time input data into the GCN model to obtain the prediction results of key quality variables.

7. A project quality control method according to claim 5, characterized in that: The step of directly driving the graph node weight update based on the test result also includes: Dynamic adjustment of risk weights can be achieved through dynamic capture of spatiotemporal characteristics and quantification of network relationship graphs; Adopting federated learning weight pools and quantum reinforcement learning mechanisms to achieve cross-institutional joint modeling and strategic space exploration in complex scenarios; Dynamic adjustment of production status is achieved through edge intelligent decision nodes and real-time feedback mechanisms; By collecting real-time production status data, the dynamic weight adjustment of each scheduling rule is calculated, and the production scheduling strategy is executed according to the rule priority; In a dynamic production environment, the machine vision CNN process can autonomously update feature weights based on disturbances such as lighting changes and equipment vibration to maintain decision stability. Through the dynamic control balance mechanism, dynamic real-time processing information is conveyed according to the rolling scheduling of the fault point, and the planning information and scheduling are reorganized; By monitoring message backlogs, average latency, and throughput in real time, queue weights are dynamically adjusted to ensure maximum processing efficiency for high-priority queues. In the adaptive consensus decision model, dynamic adjustment is achieved by recalculating the attribute-level consensus and updating the attribute weights based on the consensus; By shortening the deviation rate between time and energy consumption index values ​​and average aggregation distance, the index weight coefficient is dynamically adjusted to achieve hierarchical optimization of equipment and processes; In the digital workshop cloud platform, an improved Nginx dynamic weight load balancing algorithm is used to achieve dynamic distribution of user requests and improve system performance.

8. A project quality control system, characterized in that: include: The coding system establishment module is used to establish a multi-level coding system for production process routes and represent processing constraint relationships through topological sorting; The rule engine module is used to automatically bind QCP inspection items to coding nodes using the RETE algorithm of the rule engine; Dynamic addition and deletion module: When a process change triggers an event, the rule engine calls the difference analysis module, compares the historical process parameter library, and dynamically adds and deletes inspection items; The component definition module is used to define component status, including pending inspection, qualified, repaired, and scrapped. The QCP inspection results trigger status transitions in real time and are synchronized to the production planning module through the API; Model building module, used to build Gantt chart-state linkage model based on process route dependencies; The rule base association module is used to aggregate historical quality data based on knowledge graph technology and establish a "defect mode-root cause-repair solution" association rule base; The parameter adjustment module is used for API-linked equipment self-adjustment. When the QCP inspection fails, the root cause library matches the top three remediation suggestions. The CNC equipment interface is called through the RESTful API to automatically issue parameter adjustment instructions. Re-inspection module, used to automatically trigger re-inspection after correction.

9. An intelligent terminal, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that A computer program is stored which can be loaded by a processor and execute the method according to any one of claims 1 to 7.

Images