Method for quality early warning and hetero-cause diagnosis of aviation product manufacturing process based on digital twinning

Abstract

The application discloses an aviation product manufacturing process quality early warning and hetero-cause diagnosis method based on digital twinning. First, the digital detection report is automatically collected and analyzed by RPA and two-network interconnection technology to obtain a structured quality characteristic data set. Then, based on the fusion of light-weight models and measured point cloud data, an aviation product part digital twinning body geometric model and a process capability model are constructed. At the data layer, process quality indexes, key quality characteristics and process quality BOM are generated through recursive decomposition, RPA collection and information association. At the model layer, a high-precision digital model set is constructed, which includes risk calculation, SPC discrimination, control chart pattern recognition, fluctuation source reasoning and self-learning expert knowledge base. Finally, at the simulation layer, virtual and real synchronization and scenario deduction are carried out, and at the service layer, visual interaction and auxiliary decision-making are realized, so as to promote the transformation of product quality from after-the-fact inspection to pre-emptive prevention.

Description

Technical Field

[0001] This invention relates to the field of manufacturing process quality early warning and anomaly diagnosis technology, specifically to a method for quality early warning and anomaly diagnosis of aerospace product manufacturing processes based on digital twins. Background Technology

[0002] As global manufacturing accelerates its transformation towards high-end and intelligent production, aerospace product manufacturing has become a core area for measuring a nation's industrial competitiveness. These products typically consist of tens of thousands of parts, requiring high assembly precision and involving complex processes. Their quality is easily affected by a combination of factors, including manufacturing errors, assembly stress, and tool wear, posing a significant challenge to traditional quality control methods.

[0003] I. Industry Pain Points: Limitations of Traditional Quality Control Currently, quality control in the manufacturing of complex products such as aviation still relies on a "human experience + offline sampling inspection" model. The core problem lies in the mismatch between this static control model and dynamic, complex systems, manifesting as data fragmentation, poor real-time performance, over-reliance on human intervention, and a lack of system-level risk control. The three main bottlenecks are as follows: 1. Data silos and delayed analysis hinder real-time decision-making. Because quality data is scattered across independent systems in the supply chain, production, and quality inspection, information cannot be shared and integrated in real time, leading to delays in anomaly detection and lagging decision-making. For example, an aero-engine manufacturer discovered an issue with excessive coating thickness in a batch of turbine blades only after assembly was completed due to a mismatch between supplier material data and production line inspection results. This resulted in a 25-day delay and rework costs exceeding 8 million yuan.

[0004] 2. Relying primarily on post-production inspection leads to high costs and risks. Quality control is concentrated at the end of production (such as pre-shipment testing), neglecting process control, resulting in high rework costs and delays. For example, in the manufacturing of a certain ship, welding defects were discovered later, requiring cutting and reconstruction, which increased costs several times over.

[0005] 3. Inefficient root cause tracing of anomalies. Complex product quality defects are often caused by multiple factors transmitted across processes. For example, the assembly error of a satellite bracket may be caused by the coupling effect of deformation in the previous heat treatment and loosening of the current tooling fixture. However, traditional FMEA (Failure Mode and Effects Analysis) relies on manual experience databases, and the average diagnosis time exceeds 48 hours, which seriously restricts production efficiency.

[0006] II. Current Technological Status: Existing Technological Bottlenecks and Challenges To overcome the aforementioned limitations, digital twin technology, due to its "physical-virtual real-time interaction" characteristics, has been widely applied to manufacturing process optimization in recent years. According to a 2023 review in IEEE Transactions on Industrial Informatics, 62% of the world's leading manufacturing companies have deployed digital twin prototyping systems; however, significant technological gaps still exist in quality early warning and anomaly diagnosis scenarios. 1. Information space is separated from physical space.

[0007] In recent years, guided and driven by national strategies, aerospace product manufacturing units have actively responded to national strategic requirements, promoted the integration of informatization and industrialization, and achieved intelligent manufacturing. However, currently, within enterprises, there is still a physical isolation between the enterprise intranet and the industrial control system, resulting in information gaps in the "dynamic perception" stage of intelligent manufacturing, making subsequent "real-time analysis, precise execution, and autonomous decision-making" impossible. Digital twin technology emphasizes real-time interaction between the information space and physical space; the existence of this virtual-physical gap hinders the research and application of this technology.

[0008] 2. Insufficient model building and real-time performance.

[0009] Existing digital twin models are mostly focused on offline simulations of single equipment or partial processes (typically 4-8 hours per cycle), and do not incorporate multi-dimensional data, resulting in insufficient model fidelity and difficulty in supporting real-time decision-making. For example, the literature (X. Li et al., "Digital Twin-Driven Machining Process Monitoring," Journal of Manufacturing Systems, vol. 65, pp. 25-37, 2022.) constructs a twin model of CNC machine tool machining based on finite element simulation. Due to the lack of integration of real-time cutting force feedback data, its thermal deformation prediction error reaches 15%. The literature (Y. Wang et al., "LSTM-Based ToolWear Prediction in Milling Operations," IEEE IoT Journal, vol. 10, no. 8, pp. 7123-7134, 2023.) uses LSTM networks to predict tool wear, but does not introduce constraints on tool material wear mechanisms, resulting in a sharp drop in generalization under small sample conditions.

[0010] On the other hand, the application scenarios of the above modeling methods are limited, and they can only be applied to a single piece of equipment or a partial process, which cannot meet the general quality control requirements of enterprises for the entire process of complex product manufacturing in a timely manner.

[0011] 3. Limitations of quality early warning methods.

[0012] In multi-variety, small-batch manufacturing environments, traditional threshold alarms are often based on fixed rules (such as 3σ), which cannot adapt to the dynamically changing working conditions during the manufacturing process of complex products. For example, changes in parameters such as equipment wear and material property fluctuations can lead to frequent false alarms of static thresholds (false alarm rate as high as 35%). In addition, knowledge base systems suffer from the problem of rule "explosion," with reasoning efficiency dropping by 60% after the number of rule entries exceeds 5,000, making it difficult to adapt to dynamic working conditions.

[0013] 4. Difficulty in acquiring diagnostic knowledge and poor interpretability of diagnoses.

[0014] In the field of process quality diagnosis, scholars both domestically and internationally often introduce technologies such as neural networks and expert systems to analyze the sources of process quality fluctuations, thereby improving the capability and accuracy of process quality diagnosis. Expert systems, as one of the core technologies for process anomaly diagnosis, have advantages such as interactivity and transparency in fault diagnosis, but they still face a bottleneck in knowledge acquisition. Furthermore, existing anomaly diagnosis methods rely excessively on black-box models. The literature (G. Liu et al., "Deep Reinforcement Learning for Fault Diagnosis," Mechanical Systems and Signal Processing, vol.185, 2023.) uses deep reinforcement learning to locate anomaly sources, but its decision-making logic is uninterpretable, making it difficult for engineers to trust and adopt. Summary of the Invention

[0015] The purpose of this invention is to provide a method for quality early warning and anomaly diagnosis in the manufacturing process of aerospace products based on digital twins. This method can realize dynamic quality early warning, anomaly early warning and anomaly diagnosis in the product manufacturing process. It has the characteristics of high modeling efficiency, strong early warning timeliness, high reliability of anomaly location, and support for decision support, thus promoting the transformation from post-event inspection to pre-event prevention.

[0016] To achieve the above objectives, the present invention employs the following technical solution: Methods for quality early warning and anomaly diagnosis in aerospace product manufacturing processes based on digital twins include: The RPA (Robotic Process Automation) robot is used to obtain digital inspection reports of aerospace products from digital measurement equipment, and then transmit the digital inspection reports from the industrial control network to the enterprise intranet for parsing and processing. After the parsed data is processed by data denoising and point cloud filtering in the data layer, a digital twin of the aerospace product parts is created based on the lightweight model of the aerospace product parts using reverse modeling and dynamic lightweight modeling technology. At the same time, a digital twin that can reflect the manufacturing process capabilities of the aerospace product parts is also created. At the data layer, a recursive algorithm is used to quickly decompose the process quality indicators of the product manufacturing process and generate process-level quality indicator data; RPA process automation robot technology is used to automatically collect key quality characteristic information of aerospace product supporting parts from the CAPP / 3D process design system; and information on the status changes of the process "man, machine, material, method, and environment" is collected to dynamically generate the quality BOM of aerospace product supporting processes. Establish a data model for aerospace product parts: collect operational information on the "man, machine, material, method, and environment" processes, and use statistical analysis methods such as regression analysis, analytic hierarchy process, and fuzzy evaluation to establish a process quality risk calculation model; use BOM structure transformation and data mapping algorithms to establish a process quality traceability model based on the bill of materials for aerospace product processes; Establish a fault behavior model and use preset limit control chart technology to establish an SPC anomaly detection and early warning model; A machine learning-based method for identifying sources of abnormal fluctuations in process operations is adopted, and a mechanism model is established at the model layer. This mechanism model combines process anomalies with information on quality risks and changes in quality factors to infer and identify sources of process quality fluctuations and provide corresponding improvement measures. A data-driven model is established, and a machine learning method based on expert knowledge for process anomaly diagnosis is used to enable the self-learning and memorization of anomaly diagnosis knowledge, thereby optimizing and improving the accuracy of predicting the causes of process anomalies.

[0017] Furthermore, the process of building digital twins for aerospace product parts is as follows: Obtain 3D point cloud data for measuring aerospace product parts from the analyzed digital inspection report; Based on the CAITA file for measuring points of aerospace product parts in the PDM system, the geometric information set and theoretical measurement point information set of aerospace product parts are obtained; the geometric information set of parts includes the three-dimensional model of the parts, the spatial coordinate system, and the positioning datum of the parts; the theoretical measurement point information set includes measurement point information. For the 3D model of the part in the theoretical model information set, the associated theoretical measurement point information set is retained, and the geometric description and load-bearing information of the 3D model of the part are simplified; the lightweighting process is as follows: On the model processing server, a scheduled task is created to call the script file and synchronize the EBOM data in the 3D process design system. The scheduled task is then triggered to call the 3D process design service and retrieve the model. It checks if a CATIA model exists in the download path and synchronizes it. Once the model synchronization in the EBOM is complete, the model lightweighting process is triggered. First, the affected VCIs are analyzed and traversed, collecting all 3D models into the conversion workspace. An impact analysis is performed upwards from the VCIs to obtain the corresponding ICI model. If multiple ICI models exist, only one is used. If some VCI models have incorrect assembly positions, the ICI model is used for model conversion. The ICI models obtained from the impact analysis are then lightweighted. Automatic processing includes merging facet nodes, preserving high-curvature contours, voxel-based simplification, deleting useless nodes, and deleting reference model nodes to obtain the lightweight model. The collected measured point cloud data of aerospace product parts is transformed into spatial coordinates, and the transformed point cloud data is fused with the lightweight model. The measured point cloud dataset is mapped to the corresponding theoretical measurement point locations on the lightweight model, thus visualizing the measured point cloud dataset on the lightweight model and achieving the replacement of theoretical measurement points with 3D point clouds. Based on the 3D point cloud, new 3D measured points are generated to encrypt the point cloud data. Subsequently, digital twins of aerospace product parts are created using the encrypted point cloud data. For geometrically shaped surfaces, surface reconstruction is performed through fitting or sectional drawing. For complex surfaces, a feature curve mesh is first fitted, followed by surface fitting, and finally surface reconstruction.

[0018] Furthermore, the process of creating a digital twin that reflects the manufacturing process capabilities of aerospace product parts is as follows: Collect 3D point cloud data of the quality characteristics of a certain process in aerospace parts production; If the process only processes one product and the quality characteristic is measured only once, the measured value of the spatial dimension of the quality characteristic is calculated based on the three-dimensional point cloud data. If the process involves batch processing of multiple products, the arithmetic mean and Kalman filter method are used to filter the measured values. Obtain lightweight models of aerospace product parts and annotate the models with mass characteristics and theoretical measurement points; The filtered measured values ​​are fused with the lightweight model, and the measured values ​​and the theoretical values ​​corresponding to the theoretical measurement points are distinguished by shape and color. The state can be traced back according to any measurement time sequence to obtain a digital twin. Using the filtered measured values, the consistency between the actual average value of the process and the center value of the quality characteristic specification is calculated, and the accuracy of the process is evaluated. The mean deviation between the process average value and the center value of the specification is calculated, and process adjustment suggestions are proposed based on the magnitude of the deviation.

[0019] Furthermore, the RPA robot is controlled to log into the CAPP process design system, download key and process specification data for each model according to the rules, and merge the data tables downloaded for each model.

[0020] Control the RPA robot to filter data by model and extract process content, and replace the value of process content according to the engineering symbol table; Control the RPA robot to log in to the MES manufacturing execution system, and download the quality control records of key processes for each model according to the unit, model, batch, key process number, and process specification number, including sample size, measured value, unit, model, batch, key process number, and process specification number; The RPA robot is controlled to use the unit, model, batch, key process number, quality characteristic identifier, and process specification number as unique matching conditions to match the CAPP process query results with the MES key process quality control record table and fill in the values ​​to form a key process measured value dataset. Control the RPA robot to log in to the Integrated Quality System (IQS), import the dataset of measured values ​​of key processes according to the predefined fields in the data collection form, and generate a key quality characteristic measurement data collection form. By combining the key quality characteristic measurement data collection form and the status change information of "man, machine, material, method, environment", the quality BOM of supporting processes for aerospace products is dynamically generated.

[0021] Furthermore, a process quality risk calculation model is established, including: Starting from five dimensions—"people, machines, materials, methods, and environment"—the manufacturing process of key product processes is analyzed, quality risk factors under each dimension are identified, and a quality risk system framework diagram for key product processes is established. The framework diagram consists of a criterion layer and an indicator layer. Among them, "people, machines, materials, methods, and environment" are the five risk factors in the criterion layer; the corresponding factors under each factor in the criterion layer constitute the indicator layer. The weight vectors of each factor in the criterion layer are constructed using the analytic hierarchy process. We construct the weight vectors of the corresponding indicator layer factors under each factor of the criterion layer; and combine the comparison and quantification value table between each indicator with the calculated weight vectors of each factor of the criterion layer, we finally calculate the weight value of each factor. Introducing factor quality risk coefficients, and combining the characteristics of each factor in the indicator layer, mathematical methods such as regression analysis and fuzzy evaluation are used to establish a quantitative evaluation model for the quality risk of each factor, and the quality risk coefficient of the factor under the current state is calculated. Based on the quality risk coefficients of each factor, the overall quality risk coefficient of key processes is calculated by weighting. When the coefficient is less than the set threshold, it indicates that the product quality risk is low and at an acceptable level under the current manufacturing conditions. When it exceeds the set threshold, the factors with higher quality risk need to be improved before product processing to reduce the quality risk.

[0022] Furthermore, a process quality traceability model based on the bill of materials for aerospace product processes is established, including: Acquire and synchronize the structural information of the model layer, station layer, assembly process layer, supporting layer, and parts process specification layer in the bill of materials for aviation product processes; By using the unique identifier of the assembly process number of aviation products, the actual manufacturing quality information of "man, machine, material, method, environment, and measurement" in the assembly product manufacturing record of the assembly MES system is read. The "man, machine, material, method, environment, and measurement" information of each assembly process / step in the assembly product manufacturing record is associated with the assembly process and step in the aviation product process material list and stored in the assembly process quality layer. By matching the material codes of the part MES system with the material codes of the process bill of materials, the "man, machine, material, method, environment, and measurement" quality information of the part product manufacturing record in the part MES system is read and stored in the part process quality layer. Identify the addition, reduction, and modification of assembly / parts processes for aerospace products, as well as information on changes in process quality. Employ data derivation, filtering, and replacement mapping rules to dynamically update the process quality traceability BOM. The process quality traceability BOM structure should at least include the model layer, station layer, assembly process layer, assembly process quality layer, matching layer, parts process specification layer, and parts process quality layer.

[0023] Create a forward traceability path for part information. Based on the unique material code, trace the BOM from the bottom layer to the top layer of the process quality traceability BOM, execute and retrieve the matching part process specifications, part process quality, assembly process, station position and assembly frame information in sequence. Create a reverse traceability path for process failures / abnormalities. Based on the unique identifier of the assembly process or the unique identifier of the faulty part, trace the process quality from the top level of the BOM downwards, and execute and sequentially search the assembly process layer, assembly process quality information layer, matching layer, part process specification layer, and part process quality sub-layer to obtain the assembly process quality, matching, part process specification / finished product information, and part process quality information of the faulty / abnormal process.

[0024] Furthermore, using preset limit control chart technology, an SPC anomaly detection and early warning model is established, including: For the production process of aerospace product parts manufacturing, the set values ​​of its process quality indicators are obtained, and it is assumed that the production process is under statistical control and the quality characteristic values ​​of the production process follow a normal distribution. If the mean value of the machining dimension does not coincide with the distribution center of the tolerance zone, calculate the corresponding process capability index setting value according to the preset offset and process quality index setting value, and finally obtain the sample standard deviation setting value. Calculate the mean derived value based on the actual offset direction of the sample. Based on the "3σ" control limit determination principle, the center line and upper and lower control limits of the mean preset limit control chart are calculated, thereby establishing the mean preset limit control chart; Therefore, for the sample size range of 5 ≤ n < 100, the F-test is used to determine the actual deviation of the sample. Compared with the standard deviation setting value The variances are homogeneous; if there is no significant difference between the two variances, a T-test is required to determine the difference between the actual sample mean u and the derived mean. Consistency: If there is no significant difference between the two means, determine whether the process is under steady-state control based on the preset limit control chart of the means; if there is a difference between the two means, calculate the actual offset of the sample means. ,like >1.5 Then the process conditions need to be adjusted to make u as close as possible to the tolerance center value; if ≤1.5 This indicates that the process is in a stable state. Furthermore, the process of constructing the mechanistic model is as follows: Obtain measurement data of a certain quality characteristic of a process or part, and use a control chart pattern recognition method based on machine learning to identify the control chart pattern of the quality characteristic. If it is identified as an abnormal pattern, establish an abnormal pattern matrix M. For trend and step-type patterns, the trend analysis method is used to distinguish the factors that cause the abnormal pattern, whether they are accidental or systematic. If it is determined to be an accidental factor, no fault diagnosis is required. If it is determined to be a systematic factor, the cause of the process abnormality is further analyzed. If a certain process or part is determined to be affected by systematic factors, it is necessary to further identify the sources of process quality fluctuations. Fuzzy statistical methods or expert experience-based methods are used to determine the degree of membership between the abnormal patterns and causes of the control charts, and finally, a fuzzy relation matrix Q is established. Integrating matrices Q and M yields the failure probability matrix T. The elements in matrix T are then merged, dividing [0,1] into different intervals, with each interval assigned a confidence value. The elements T in the matrix are then... i Based on the interval it falls into, the corresponding confidence value is used to diagnose the cause of the anomaly. The potential quality risks in the "man, machine, material, method, and environment" aspects of the production process or parts are defined and quantified. Potential quality risk coefficients for various factors in the process are set, and dynamic risk thresholds for the process are set. The mapping relationship between the distribution of abnormal causes and process risk indicators is established. Combined with the predicted causes of process failures and potential quality risk coefficients, the causes of process abnormalities are analyzed.

[0025] Furthermore, the process of building a data-driven model is as follows: According to different aviation product processing specialties and processes, the database of non-conforming aviation products is classified. The non-conforming product drawing number is used as a unique identifier. The product manufacturing record barcode of the MES system is matched to obtain the total sample size and sample measured value information of the processing / manufacturing batch in which the non-conforming quality characteristics are located. The Monte Carlo simulation method was used to obtain sample data with the same quality characteristics as the actual production. After preprocessing, the data was input into the BP neural network for control chart pattern recognition to obtain model M1. The measured values ​​of quality characteristics are preprocessed and used as input to a BP neural network. The trained model M1 is then used to identify control chart patterns, yielding the actual results of control chart pattern recognition. The database of non-conforming products is grouped according to different manufacturing processes of aviation products and processing specialties. A BP neural network is constructed and trained in each group to obtain the nonlinear mapping relationship from the abnormal mode of the control chart to the cause of failure. The parameters of the BP neural network are written into the database.

[0026] Furthermore, the method also includes: In the simulation layer of the digital twin platform, when aerospace product parts undergo in-flight digital inspection, a real-time simulation engine is used to achieve virtual-real synchronization. Based on the digital twin, process capability prediction, process quality risk prediction, process anomaly prediction, process anomaly fluctuation source prediction, and scenario simulation are performed.

[0027] A terminal device includes a processor, a memory, and a computer program stored in the memory; when the processor executes the computer program, it implements the method for quality early warning and anomaly diagnosis of aviation product manufacturing processes based on digital twins.

[0028] A computer-readable storage medium storing a computer program; when executed by a processor, the computer program implements the method for quality early warning and anomaly diagnosis of aerospace product manufacturing processes based on digital twins.

[0029] Compared with the prior art, the present invention has the following technical features: 1. Optimize manufacturing costs.

[0030] This reduces scrap rates caused by process control failures, lowers rework costs and material waste, and reduces labor and time costs associated with tracing product quality issues. For example, a CNC machining plant processing aerospace parts saw its average quarterly number of product quality problems decrease from 79 to 42, a 47% reduction, and quarterly quality losses decrease from 1.96 million to 890,000, a 55% reduction.

[0031] 2. Improved production efficiency.

[0032] By implementing real-time closed-loop quality control, production line downtime for troubleshooting is avoided, improving manufacturing continuity. The mean time to recovery (MTTR) of a certain machine body CNC machining production line was reduced from 4.2 hours to 15 minutes, and the overall efficiency (OEE) of the production line improved by 27%.

[0033] 3. Quality traceability capability.

[0034] Based on the Bill of Materials (BBOM) for the actual manufacturing processes of aerospace products, structured storage of process quality information is achieved, generating a process quality traceability BOM. Utilizing BOM structure transformation and data mapping algorithms, the system automatically responds to information changes during production, ensuring dynamic traceability of aerospace product process quality information. A bidirectional traceability path is established from aerospace product parts to the assembly process, improving the efficiency of tracing the causes of process failures / abnormalities by 95%. Attached Figure Description

[0035] Figure 1 A schematic diagram illustrating the measurement report setting format of an embodiment of the method of the present invention; Figure 2 A schematic diagram of the interconnection and integration relationship between the two networks in the method of this invention; Figure 3 A measurement point planning diagram for aerospace product parts according to an embodiment of the method of the present invention; Figure 4 A fusion diagram of measured data and a lightweight part model from one embodiment of the method of the present invention; Figure 5 A schematic diagram of a NURBS curve fitting method according to an embodiment of the present invention; Figure 6 This is a data mapping diagram of the assembly process layer of the BOM for process quality traceability according to an embodiment of the method of the present invention; Figure 7 This is a schematic diagram of the BOM structure for process quality traceability in the method of the present invention; Figure 8 This is a schematic diagram of the dynamic process quality traceability BOM and system implementation process of the method of the present invention; Figure 9 This is a schematic diagram of the actual mean shift of a sample in one embodiment of the method of the present invention; Figure 10This is a schematic diagram of the abnormal control chart mode of the method of the present invention; Figure 11 Trend analysis of the quality characteristic measurement values ​​of the method of the present invention; Figure 12 This is a mapping diagram of process abnormality causes and risk indicators in one embodiment of the method of the present invention; Figure 13 This is a flowchart of the method of the present invention; Figure 14 The following is a flowchart illustrating the logic for updating the digital twin model in the method of this invention; Figure 15 This is a technical architecture diagram of the method of the present invention. Detailed Implementation

[0036] This invention provides a method for quality early warning and anomaly diagnosis in the manufacturing process of aerospace products based on digital twins. It constructs a technical system of "perception-modeling-prediction-optimization" with data-driven approaches and virtual-real interaction as its core. The architecture is divided into the following layers: 1. Physical layer.

[0037] The system collects real-time operational data (covering the "human, machine, material, method, and environment" aspects of the manufacturing process of aerospace product parts) through digital measurement equipment, RPA robots, and system integration interfaces. It adopts a digital detection data transmission and analysis method based on RPA and interconnected networks, and transmits the relevant data to the enterprise intranet through the interconnection channel between the industrial control network and the enterprise intranet.

[0038] 2. Data layer.

[0039] It integrates inherent data from physical space (aviation product parts), real-time digital measurement data, and manufacturing process data ("man, machine, material, method, environment"). It also integrates technologies such as data noise reduction, point cloud filtering preprocessing, data standardization coding preprocessing, and process quality BOM model traceability to manage multi-source heterogeneous data (structured, unstructured, and time-series data) in a unified manner, providing basic support for upper-level models.

[0040] 3. Model Layer: Constructs high-precision digital models based on the physical characteristics of entities, covering: Geometric model (creating digital twins of parts, digital twins that reflect process capabilities); data model (establishing process quality risk assessment model, establishing quality traceability model based on dynamic process quality BOM); fault behavior model (establishing SPC anomaly detection and early warning model under multi-variety small batch mode, establishing control chart anomaly pattern recognition model based on machine learning methods); mechanism model (establishing process anomaly "reasoning-association-decision" rule model); data-driven model (adopting a process anomaly diagnosis expert knowledge machine learning model based on neural networks).

[0041] 4. Simulation Layer: When aerospace product parts are digitally inspected in-flight, a real-time simulation engine is used to achieve virtual-real synchronization, supporting process capability prediction, process quality risk prediction, process anomaly prediction and diagnosis, process anomaly fluctuation source prediction, and scenario simulation.

[0042] 5. Service Layer: This layer contains semi-autonomous functional modules formed by model combination, which support flexible expansion and combination, such as process capability analysis, process indicator achievement prediction, process anomaly early warning, process anomaly fluctuation source identification, improvement measure feedback and other functions, and provides visualization interaction, AI process anomaly analysis and decision optimization capabilities.

[0043] The specific steps of the method of the present invention are as follows: Step 1: Using an RPA (Robotic Process Automation) robot, the digital inspection report of the aviation product is automatically collected from the digital measurement equipment. The measurement report is then transmitted from the industrial control network to the enterprise intranet, and after parsing and processing, it is automatically transmitted to the application system on the enterprise intranet.

[0044] Step 101: Standardize the format of the measurement report content, automatically create a digital test report through the digital measurement equipment terminal, and store it in the specified path on the computer terminal.

[0045] Step 102: Deploy an RPA robot in the computer terminal connected to the digital inspection equipment, and control the RPA robot to access the digital inspection report in the specified path in the computer terminal.

[0046] Step 103: Control the RPA robot to log in to the designated website and enter the "Industrial Control Network Measurement Report Upload" page. Control the RPA robot to click the "Measurement Report Upload" button, then click "Browse" to select the digital inspection report file and complete the upload of the measurement report.

[0047] Step 104: Based on the interconnection of the two networks, establish a file data transmission channel between the enterprise intranet and the industrial control network. At the same time, establish a data monitoring service on the industrial control network to periodically monitor the measurement report files uploaded by the RPA robot. When a new measurement report is received, the measurement report data is transmitted to the enterprise intranet application server.

[0048] Step 105: Based on the enterprise intranet MES system application, establish a measurement report data parsing algorithm model and monitoring program to monitor the measurement report files transmitted from the industrial control network in real time. When a new measurement report is received, the parsing algorithm is used to parse the digital test report line by line.

[0049] Step 106: When parsing the digital inspection report, use the "key feature name" in the "feature description (N)" section as the anchor point n, split the string by the ":" character in the anchor point n, and extract the character at the subscript "1" to obtain the key feature name data.

[0050] Step 107: Define the worksheet in the Excel file of the digital inspection report with the y-axis vertically and the x-axis horizontally. Using anchor point n as the positioning reference, extract other relevant data from the "Characteristic Description (N)" section one by one downwards along the y-axis.

[0051] Step 108: Parse the data in the "Title Block (M)" and "Measurement Result (Q)" sections of the digital inspection report and store them in the target database. Based on the key identification information retrieved after parsing, associate it with the product manufacturing records in the MES system.

[0052] For example, during the machining of a support component for an aerospace product, a digital measuring machine (DMM) was used to inspect the overall dimensions of the component. The acquired measurement data and information on the "man, machine, material, method, and environment" aspects of the machining process were recorded in the measurement report. Figure 1 The measurement report was analyzed to obtain the following data: The measurement report consists of three parts, with the "Title Block (M)" section containing the following information: M = {Product Manufacturing Record Number (CP-3755-01), Part Number Under Test (01), Measurement Process Number (50), Measurement Equipment Name (DELTA 30.80.20), Measurement Equipment Number (DESL000006), Part Drawing Number (**), Part Name Under Test (Support), Part Number Template (A), Quantity of Parts Under Test (1), Measurement Unit (MM), Measurement Date (2024-8-26), Measurement Personnel (Zhang San), Measurement Ambient Temperature (22℃), Measurement Ambient Humidity 48%}; Information N in the “Characteristic Description (N)” section: N = {hole spacing, G102, 50, location of key features, feature description (location 3, size location (circle 11), diameter, theoretical value: 8, measured value: 7.88627, upper deviation: 0.05, lower deviation: -0.05, deviation data in all directions and overall: -0.1137, out-of-tolerance data in all directions and overall: 0.06373)}.

[0053] Information in the "Measurement Results (Q)" section: Q: Q = {Name of the element being measured (POINT79), Measured content of the element (size and location (point 1)), Direction description (X), Theoretical data (463.46814), Measured value (463.466), Lower deviation (0.01), Upper deviation (-0.01), Deviation data in each direction and overall (-0.00249), Out-of-tolerance data in each direction and overall (0)}.

[0054] Step 2: After the parsed data is processed by the data layer of the digital twin platform for data denoising and point cloud filtering, a digital twin of the aerospace product parts is created based on the lightweight model of the aerospace product parts using reverse modeling and dynamic lightweight modeling technology. At the same time, a digital twin that can reflect the manufacturing process capabilities of the aerospace product parts is also created.

[0055] The process of building digital twins for aerospace product parts is as follows: Step 201: Obtain the three-dimensional point cloud data of the aerospace product parts from the parsed digital inspection report.

[0056] Step 202: Based on the CAITA file for measuring points of aerospace product parts in the PDM system, obtain the geometric information set and theoretical measurement point information set of the aerospace product parts; the geometric information set includes the three-dimensional model of the part, the spatial coordinate system, and the positioning datum of the part; the theoretical measurement point information set includes measurement point information.

[0057] For example: the CAITA file for measuring point planning of aerospace product parts read from the PDM system, see... Figure 3 All webs with a thickness of 3mm are on the same horizontal plane, and the web surface is Z=0 as shown in the figure.

[0058] Step 203: For the 3D model of the part in the theoretical model information set, retain the associated theoretical measurement point information set, and simplify the geometric description and load-bearing information of the 3D model of the part to obtain a lightweight model; the lightweighting process is as follows: Step 20301: On the model processing server, create a scheduled task to call the script file and synchronize the EBOM data in the 3D process design system.

[0059] Step 20302: Trigger a scheduled task to call the 3D process design service and retrieve the model; Step 20303: Determine if a CATIA model exists in the download path. Based on the existence status of the CATIA model and the cgr file, set the corresponding catiastatus attribute to complete the model synchronization from the 3D process design system.

[0060] Step 20304: When the model synchronization in the EBOM is completed, the model lightweighting process will be triggered. First, the affected VCIs will be analyzed and traversed, and all 3D models will be collected into the conversion workspace as a model conversion preparation environment.

[0061] Step 20305: Perform an impact analysis upwards using VCI to obtain the corresponding ICI model. If multiple ICI models exist, only one is selected. If some VCI models have incorrect assembly positions, the model is converted using the ICI model.

[0062] Step 20306 involves performing a lightweight transformation on the ICI model obtained from the traversal influence analysis. Automatic processing includes merging facet nodes, preserving high curvature contours, voxel-based simplification, deleting useless nodes, deleting reference model nodes, and renaming nodes. For example, for this aerospace product part, a geometric feature recognition algorithm was used to remove 5 invalid reference nodes, reducing the file size from 340MB to 8.5MB. Table 1 shows a comparison of the lightweighting effects of a three-dimensional model of an aerospace product support part according to an embodiment.

[0063] Table 1 Comparison of lightweighting effects of 3D models of aircraft product support parts in one embodiment

[0064] Step 20307: Determine whether a lightweight file already exists in the system for the corresponding version. If a lightweight file exists, delete the relationship between the original lightweight file and the corresponding version. If not, create a lightweight file object and associate it with the corresponding version, indicating that the model lightweighting was successful.

[0065] Step 204: Perform spatial coordinate transformation on the collected measured point cloud data of aerospace product parts, and store the transformed point cloud.

[0066] Step 205: Fuse the transformed point cloud data with the lightweight model. Step 206: Map the 3D point cloud in the measured point cloud dataset to the corresponding theoretical measurement point locations on the lightweight model, completing the visualization of the measured point cloud dataset on the lightweight model, that is, realizing the position replacement of the theoretical measurement points by the 3D point cloud; on the basis of the 3D point cloud, perform interpolation generation, fitting sampling and other methods to generate new 3D measured points, realizing the encryption of the point cloud data; subsequently, use the encrypted point cloud data to create a digital twin of the aerospace product parts.

[0067] For example: The measured data after coordinate transformation is fused with the lightweight model of the part to generate a new lightweight model (see...). Figure 4 After generating a large amount of new point cloud data, digital twins of aerospace product parts are created.

[0068] Step 207: For regular geometric surfaces such as planes, cylinders, and spheres on aerospace product parts, surface reconstruction is performed by fitting or sectional drawing. Step 208: For complex parts surfaces such as strip-shaped, rotary, and gently sloping surfaces, first fit a feature curve mesh, then perform surface fitting and reconstruction, and finally visualize the digital twin and 3D model of the aerospace product part according to color and shape, such as... Figure 5 As shown.

[0069] The process of creating a digital twin that reflects the manufacturing capabilities of aerospace product parts is as follows: Step 209: Collect three-dimensional point cloud data of the quality characteristics of a certain process of aerospace parts products.

[0070] Step 210: If the process only processes one product and only measures the quality characteristic once, calculate the measured value of the spatial dimension of the quality characteristic based on the three-dimensional point cloud data, and establish a digital twin according to the method described in steps 212-214.

[0071] Step 211: If the process involves batch processing of multiple products, the measured values ​​are filtered using arithmetic mean and Kalman filtering based on the spatiotemporal characteristics of the quality characteristic measurement process.

[0072] Step 212: Obtain a lightweight model of the aerospace product parts and mark the mass characteristics and theoretical measurement points on the model.

[0073] Step 213: The filtered 3D point cloud data is fused with the lightweight model, and the measured values ​​and theoretical values ​​corresponding to the theoretical measurement points are distinguished by shape and color. The state can be backtracked according to any measurement time sequence.

[0074] Step 214: Using the filtered measured values, calculate the consistency between the actual average value of the process and the center value of the quality characteristic specification, and evaluate the process accuracy Ca; calculate the mean deviation between the process average value and the center value of the specification, and propose process adjustment suggestions based on the magnitude of the deviation.

[0075] Ca=( -M) / (T / 2); in, M is the actual average value of the process; T is the center value of the quality characteristic specification; and T is the specification tolerance.

[0076] The process capability digital twin model is evaluated according to the following process accuracy levels: Level A ~|Ca|≤12.5%; Level B ~12.5%<|Ca|≤25%; Level C ~25%<|Ca|≤50%; Level D ~|Ca|>50%.

[0077] For example, |Ca|=26.7%, according to the digital twin model evaluation mode of process capability, the process accuracy level of this process is C.

[0078] Step 3: In the data layer of the digital twin platform, a recursive algorithm is used to quickly decompose the process quality indicators of the product manufacturing process and generate process-level quality indicator data; RPA process automation robot technology is used to automatically collect key quality characteristic information of aerospace product supporting parts from the CAPP / 3D process design system; collect the status change information of the process (physical entity) of "man, machine, material, method, and environment" and dynamically generate the quality BOM of aerospace product supporting processes.

[0079] Step 301: In the data layer of the digital twin platform, a recursive algorithm is used to quickly decompose the process quality indicators of complex product manufacturing processes and generate process-level quality indicator data.

[0080] Step 302 involves automatically collecting key quality characteristic information of aerospace product components from a CAPP / 3D process design system using RPA (Robotic Process Automation) technology; specifically including: Step 30201: Control the RPA robot to log in to the CAPP process design system, download key and process specification data for each model according to the rules, and merge the data tables downloaded for each model.

[0081] Step 30202: Control the RPA robot to filter data by model and extract process content, replace the value of process content according to the engineering symbol table, and save.

[0082] Step 30203: Control the RPA robot to log in to the MES manufacturing execution system. Based on the screening conditions such as unit, model, batch, key process number, and process specification number, download the quality control records of key processes for each model. These records mainly include sample size, measured value, unit, model, batch, key process number, and process specification number.

[0083] Step 30204: Control the RPA robot to match the CAPP process query results with the MES key process quality control record table using the unit, model, batch, key process number, quality characteristic identifier, process specification number, etc. as unique matching conditions, and fill in the values ​​to form a key process measured value dataset, and save the data.

[0084] Step 30205: Control the RPA robot to log in to the Integrated Quality System (IQS), import the dataset of measured values ​​of key processes according to the predefined fields in the data collection form, and automatically generate a data collection form for key quality characteristics measurement.

[0085] Step 303: Collect information on the status changes of the "man, machine, material, method, and environment" of the process (physical entity), and dynamically generate the quality BOM of the supporting process for aerospace products by combining the key quality characteristic measurement data collection form.

[0086] Step 4: Establish a data model for aerospace product parts: Collect operational information on the "man, machine, material, method, and environment" of the process (physical entity), and use statistical analysis methods such as regression analysis, analytic hierarchy process (AHP), and fuzzy evaluation method (FCE) to establish a process quality risk calculation model; use BOM structure transformation and data mapping algorithms to establish a process quality traceability model based on the bill of materials for aerospace product processes.

[0087] Step 401, Establish a data model: Collect operational information on the "man, machine, material, method, and environment" of the process (physical entity), and use statistical analysis methods such as regression analysis, analytic hierarchy process (AHP), and fuzzy evaluation method (FCE) to establish a process quality risk calculation model, as detailed below: Step 40101: Starting from the five dimensions of "people, machine, material, method, and environment", the manufacturing process of key product processes is sorted out, the quality risk factors under each dimension are fully identified, and a quality risk system framework diagram of key product processes is established. The framework diagram consists of a criterion layer and an indicator layer. Among them, "people, machine, material, method, and environment" are the five risk factors of the criterion layer. The corresponding factors under each factor of the criterion layer constitute the indicator layer.

[0088] Step 40102: Construct the weight vectors of each factor in the criterion layer using the Analytic Hierarchy Process (AHP).

[0089] Step 40103: Construct the weight vectors of the corresponding indicator layer factors under each factor of the criterion layer; and combine the quantitative value table of comparison between each indicator and the calculated weight vectors of each factor of the criterion layer to finally calculate the weight value of each factor.

[0090] Step 40104: Introduce factor quality risk coefficients, and combine the characteristics of each factor in the indicator layer with mathematical methods such as regression analysis and fuzzy evaluation to establish a quantitative evaluation model for the quality risk of each factor, and calculate the quality risk coefficient of the factor under the current state.

[0091] Step 40105: Calculate the overall quality risk coefficient of the key processes of the product (the weighted sum of the quality risk coefficients) that takes into account the weight of each factor. When the coefficient is less than the set threshold, it means that the product quality risk is low under the current manufacturing conditions and is at an acceptable level. When it exceeds the set threshold, the factors with greater quality risk need to be improved before the product is processed in order to reduce the quality risk.

[0092] Step 40106: Develop a key process quality risk visualization assessment module in the Quality Management Information System (QMS). Obtain current product processing status data from information systems such as MES, Human Resources System (HR), ERP, and CAPP, calculate the key process quality risk coefficient, and realize quantitative assessment and accurate prediction of key process quality risk based on dynamic information perception.

[0093] Step 402: Using BOM structure transformation and data mapping algorithms, establish a process quality traceability model based on the bill of materials for aerospace product processes; specifically as follows: Step 40201: Obtain and synchronize the structural information of the model layer, station layer, assembly process layer, accessory layer, and part process specification layer in the bill of materials for aerospace product processes. The model layer is the top-level node of the process bill of materials, representing the model identifier of the aerospace product, used to distinguish process quality traceability BOM instances for different product models. The station layer is a direct child node of the model layer, representing the physical workstation or assembly station in the product assembly process, used to group assembly processes according to the geographical location or scope of responsibility of the production organization. The assembly process layer is a child node of the station layer, containing several assembly processes. The types of assembly processes in the assembly process layer include standard assembly processes, rework / repair processes, and temporary modification processes. The accessory layer is a child node of the assembly process layer, containing several accessory material nodes. Accessory types include assemblies, self-made parts, and finished products. The data types in the part process specification layer include part process specifications, rework / repair process specifications, and temporary modification process specifications.

[0094] Step 40202: Using the unique identifier of the assembly process number of the aviation product, read the actual manufacturing quality information of "man, machine, material, method, environment, and measurement" of the assembly product manufacturing record in the assembly MES system, associate the "man, machine, material, method, environment, and measurement" information of each assembly process / step in the assembly product manufacturing record with the assembly process and step in the aviation product process material list, and store it in the assembly process quality layer. Step 40303: By matching the material code of the part MES system with the material code of the process bill of materials (BOM) matching layer, the "Man, Machine, Material, Method, Environment, and Measurement" quality information of the part product manufacturing record in the part MES system is read and stored in the process step of the part process specification layer of the process BOM; if the matching type is a finished product, the basic information such as the finished product name and number is read and stored in the part process quality layer of the process BOM; if the matching type is a part, the "Man, Machine, Material, Method, Environment, and Measurement" quality information is stored in the part process quality layer of the process BOM by matching the unique identifier of the part process specification.

[0095] Step 40404: Identify the addition, reduction, modification, and process quality change information of the assembly / parts process of aviation products. Use data derivation, filtering, and replacement mapping rules to dynamically update the process quality traceability BOM. The process quality traceability BOM structure should include at least the model layer, station layer, assembly process layer, assembly process quality layer, matching layer, part process specification layer, and part process quality layer.

[0096] Step 40505: Create a forward traceability path for part information. Based on the unique material code, trace the BOM from the bottom layer of the process quality upwards, execute and sequentially search the matching layer, assembly process layer, station layer and model layer to obtain information such as the process specification, part process quality, assembly process, station and assembly frame of the matching part. The matching material can be divided into assembly, finished product and part.

[0097] A process quality traceability BOM assembly process layer data mapping diagram according to an embodiment of the method of the present invention is shown below. Figure 6 The schematic diagram of the BOM structure for process quality traceability in this invention is shown below. Figure 7 The schematic diagram of the dynamic process quality traceability BOM and system implementation process of the present invention is shown below. Figure 8 .

[0098] Step 40506: Create a reverse traceability path for process faults / abnormalities. Based on the unique identifier of the assembly process or the unique identifier of the faulty part, start from the top level of the process quality traceability BOM and proceed downwards, sequentially retrieving the assembly process layer, assembly process quality information layer, matching layer, part process specification layer, and part process quality sub-layer to obtain the assembly process quality, matching, part process specification / finished product information, and part process quality information of the faulty / abnormal process.

[0099] Step 5: Establish a fault behavior model and use SPC techniques such as preset limit control charts to establish an SPC anomaly detection and early warning model under the multi-variety small-batch mode.

[0100] Step 501: Establish a fault behavior model at the model layer, and use SPC techniques such as preset limit control charts to establish an SPC anomaly detection and early warning model under the multi-variety small-batch mode, as detailed below: Step 50101: For the production process of aerospace product parts manufacturing, obtain the set value of its process quality index, and assume that the production process is under statistical control and the production process quality characteristic value follows a normal distribution; wherein, the quality characteristic value refers to the key quantitative parameter for measuring the quality level of the production process, which is the processing dimension in this scheme.

[0101] Step 50102: If the mean value of the machining dimension does not coincide with the distribution center of the tolerance zone, based on domestic and international production practices, under normal circumstances, the offset is usually about 1.5σ. Calculate the corresponding process capability index setting value based on the offset of 1.5σ and the process quality index setting value, and finally obtain the sample standard deviation setting value. Calculate the mean derived value based on the actual offset direction of the sample; where the sample is the measured value of the machining dimension.

[0102] Step 50103: Based on the “3σ” control limit determination principle, calculate the center line and upper and lower control limits of the mean preset limit control chart, thereby establishing the mean preset limit control chart.

[0103] Step 50104: Since the minimum sample size requirement for calculating the process capability index is not met when the sample size n < 100, the F-test is used to determine the actual deviation of the sample for the sample size interval 5 ≤ n < 100. Compared with the standard deviation setting value The variances are homogeneous; if there is no significant difference between the two variances, a T-test is required to determine the difference between the actual sample mean u and the derived mean. Consistency: If there is no significant difference between the two means, determine whether the process is under steady-state control based on the preset limit control chart of the means; if there is a difference between the two means, calculate the actual offset of the sample means. ,like >1.5 Then the process conditions need to be adjusted to make u as close as possible to the tolerance center value; if ≤1.5 This indicates that the process is in a stable state. Step 50105: If the actual deviation is verified by the F-test. Compared with the standard deviation setting value The existence of significant differences indicates that the two data points are not identically distributed, so further judgment is needed. and Logical size: If > This indicates that the process does not meet the process control requirements; if < Then it is necessary to further calculate the actual offset of the sample mean. ,when >1.5 Then the process conditions need to be adjusted to make u as close as possible to the tolerance center value; when ≤1.5 This indicates that the process meets and exceeds the process indicator control requirements.

[0104] Step 50106: When the sample size n ≥ 100, it is defined as falling into the category of mass production mode, a conventional control chart is drawn, the process capability index is calculated, and it is judged whether the process is under steady-state control according to the abnormality judgment criteria specified in GB / T 4091-2001 "Conventional Control Charts".

[0105] Specifically, the control chart that can be integrated with the digital twin model is not limited to the preset-limit control chart, but also includes the individual value control chart, cumulative sum control chart, mean control chart and the like.

[0106] For example: after calculation, the measured mean value u of the sample of the aviation product support part shifts upward, and the shift amount =0.19 The process quality characteristic fluctuation analysis is carried out according to the abnormality judgment rule described in step 60106, as shown in Figure 9 .

[0107] Step 6: Adopt a machine learning-based process abnormal fluctuation source identification method to establish a mechanism model at the model layer; the mechanism model combines process special causes with quality risks and quality factor change information, infers and identifies process quality fluctuation sources, and provides corresponding improvement measures. The construction process of the mechanism model is as follows: Step 601: Obtain measurement data of a certain quality characteristic of a process or part, use a machine learning-based control chart pattern recognition method to identify the control chart pattern of the quality characteristic, and if an abnormal pattern is identified, establish an abnormal pattern matrix M. There are 5 types of abnormal control chart patterns: upward trend pattern (I), downward trend pattern (II), upward step pattern (III), downward step pattern (IV), and periodic pattern (V). The schematic diagram of the abnormal control chart patterns is shown in Figure 10 .

[0108] Step 602: For trend and step-type patterns, adopt a trend analysis method to distinguish the factors leading to the abnormal pattern, i.e., accidental or systematic factors. If it is determined to be an accidental factor, no fault diagnosis is required; if it is determined to be a systematic factor, further analyze the cause of the process abnormality.

[0109] For abnormal patterns of upward (downward) trend and upward (downward) step type, the trend analysis method is used to distinguish the factors leading to the abnormal pattern, i.e., accidental or systematic factors. The measured value of a certain quality characteristic changes with time, as shown in Figure 11 , calculate the slope k of the curve n ; when the process is in a steady state, set the maximum standard slope of the quality characteristic observation curve as k0, if k n <k0, it is determined that the trend change of the measured value of the quality characteristic is small, and it is only affected by unavoidable accidental factors; if k nIf the value is greater than k0, it indicates that the trend of the quality characteristic measurement value is changing significantly, suggesting that the personnel, equipment / tools, materials, processes, environment, etc. in the production process are in an abnormal state and are affected by system factors. Further analysis and diagnosis of the cause of the abnormality should be carried out.

[0110] Step 603: For a certain process or part, if it is determined that the production process is affected by systematic factors, it is necessary to further identify the source of process quality fluctuation. Using fuzzy statistical method or expert experience-based method, the degree of membership (weight) of the abnormal pattern and cause of the control chart is determined, and finally a fuzzy relation matrix Q is established.

[0111] Step 604: Integrate matrices Q and M to obtain the failure probability matrix T. Merge the elements in matrix T, dividing [0,1] into different intervals, and assigning a confidence value to each interval. Then, sort the elements T in the matrix... i Based on the interval it falls into, the corresponding confidence value is used to diagnose the cause of the anomaly.

[0112] Table 2. Data and results of the quality risk assessment process in the examples.

[0113] Table 3. Common Causes of Failures in the Production Process of the Examples of Processes or Parts

[0114] Table 4. Correspondence between Abnormal Patterns and Potential Causes of Abnormalities

[0115] For example, after merging matrix T, we get T = [0, 0.6, ..., 0]. Then, we can... i After comparing the values ​​with Table 3, it was concluded that the cause of this abnormal pattern "may be" the second abnormal cause, namely "operational error / mistake (R2)" in the human factors.

[0116] Step 605: Define and quantify the potential quality risks in terms of "manpower, machinery, materials, methods, and environment" during the production of processes or parts, set the potential quality risk coefficient υ for each aspect of the process, set the dynamic risk threshold for the process, establish the mapping relationship between the distribution of abnormal causes and the process risk indicators, and conduct process abnormal cause analysis by combining the predicted causes of process failures and the potential quality risk coefficient υ.

[0117] For example, according to the embodiment, the dynamic risk coefficients of the four aspects of "machine, material, method and environment" are set to 0. The dynamic risk coefficient υ1 of the personnel aspect of the process is calculated only from the three process risk indicators of personnel factors. The three factors are the operator's working status, the operator's working time and the operator's skill level, and υ1=0.55 is obtained.

[0118] υ(P1)=2*10 -6 P1 3 +3*10 -4 P1 2 -0.0272P1+0.5567;0≤P1≤75; In the formula, P1 is the total working time (hours) of the first 7 days, and υ(P1) is the product process quality risk coefficient for the operator to continue working under this time. υ(P2) = -1*10 -5 P2 4 +8*10 -4 P2 3 -0.0175P2 2 +0.065P2+0.8407; 0≤P2≤36; In the formula, P2 represents the length of service (in years), and υ(P2) represents the product process quality risk coefficient of the operator during that length of service. υ(P3) = 1 / P3; In the formula, P3 is the operator's qualification examination score, and υ(P3) is the quality risk coefficient of the corresponding process processed by the operator who obtained the score.

[0119] Calculate the total risk coefficient σ of the process. [0,1], σ=O υ T =O1* υ(P1)+ O2* υ(P2)+ O3* υ(P3)=υ1=0.55.

[0120] In the formula, O is the weight value vector of each factor in the process (described as 18 factors in the embodiment), O1 is the weight value vector of the risk index of the operator's working status, O2 is the weight value vector of the risk index of the operator's working time, O3 is the weight value vector of the risk index of the operator's skill level, and υ is the quality risk coefficient vector under the current processing status. According to the value of the total risk coefficient σ of the process, the process risk is divided into four levels: low risk (0 < σ < 0.25), moderate risk (0.25 ≤ σ < 0.5), relatively high risk (0.5 ≤ σ < 0.75), and high risk (0.75 ≤ σ ≤ 1). When σ ≥ 0.5, process adjustment measures need to be taken when necessary.

[0121] See the mapping relationship of one embodiment. Figure 12 By combining the predicted causes of process failures with the potential quality risk coefficient υ, the causes of process abnormalities are analyzed.

[0122] Step 606: Based on the unique identifier of the process number / part number, read the process quality factor (man, machine, material, method, environment) change information in the process quality BOM. Through data integration, output and display the process quality fluctuation source identification result W. The result set includes abnormal cause diagnosis results and confidence value, potential quality risk coefficient, process risk level, process quality factor changes, preventive measures, etc. The preventive measures are formulated according to the actual situation.

[0123] For example: After identifying process quality factor change information using the method described in step 606, it is determined that the operator for this process has changed within the past 10 days, from Zhang San to Li Si. Other process quality factors have not changed. Operator Li Si has a one-year tenure, which is relatively short. Output the process quality fluctuation source identification result W: W = {Operational error / mistake R2 (predicted cause of process failure), 0.6 (confidence level), 0.55 (potential risk coefficient υ), higher risk (risk level), operator change for this process (change in process quality factors), operator replacement recommended (preventive measures)}.

[0124] Step 7: Establish a data-driven model. Through the machine learning method of expert knowledge for process anomaly diagnosis, the model enables the self-learning and memorization of anomaly diagnosis knowledge, thereby optimizing and improving the accuracy of process anomaly cause prediction.

[0125] Step 701: Classify the non-conforming product database data according to different aviation product processing specialties and processes (parts), use the non-conforming product drawing number as a unique identifier, match it with the product manufacturing record barcode in the MES system, and obtain production information such as the total sample size and actual measured value of the processing / manufacturing batch in which the non-conforming quality characteristics are located.

[0126] The aforementioned non-conforming quality characteristics refer to the collective term for a specific quality indicator (such as dimensions, geometric tolerances, surface quality, etc.) that exceeds the tolerance range specified in the design or process during the manufacturing process of aerospace products, resulting in the product or part being judged as a non-conforming product.

[0127] Step 702: Using the Monte Carlo data simulation method, a large amount of sample data with the same quality characteristics as actual production is obtained. After preprocessing (standardization and encoding), the data is input into the BP neural network for control chart pattern recognition to obtain model M1 and output the theoretical results of control chart pattern recognition.

[0128] Step 703: The measured values ​​of the quality characteristics are preprocessed and used as input to the BP neural network. The trained model M1 is then used to identify the control chart pattern to obtain the actual results of the control chart pattern recognition.

[0129] For example, using 25 quality characteristic measurement data parameters of a panel part as 25 input layer nodes of the M1 model, and 5 control chart anomaly pattern parameters as 5 output layer nodes of the M1 model, the machine learning algorithm identifies the quality characteristic control chart pattern of the non-conforming part as an "upward trend pattern". The result set W is denoted as: W={ Manufacturing record barcode-ZZJL001, dispatch number-N001, model-Jxx, batch / shelf number-0101, operation number-50, quality characteristic name-panel thickness, quality characteristic number-TX001, part / assembly drawing number-Jxx.a.100, control chart mode category-upward trend mode}.

[0130] Step 704: Record and store information such as the control chart anomaly mode, the actual product failure mode, the cause of failure, the failure cause code, and the corrective measures for the non-conforming process, forming a basic database for process quality diagnosis.

[0131] WQ=W+Q={Manufacturing record barcode w1, dispatch number w2, model w3, batch number w4, process number w5, quality characteristic name w6, quality characteristic number w7, part / assembly drawing number w8, control chart mode category w9, fault cause q1, fault cause code q2, corrective action q3}; Where Q = {fault cause q1, fault cause code q2, corrective action q3}.

[0132] Step 705: Group the non-conforming product database according to different manufacturing processes (parts) of the same processing specialty for aviation products. Construct a BP neural network in each group. Use the abnormal control chart pattern and the cause of failure as the input and output samples of the neural network for training. Obtain the nonlinear mapping relationship from the abnormal control chart pattern to the cause of failure. Write the neural network parameters (weights, biases, hyperparameters, etc.) into the database. For example, according to the embodiment, the sub-neural network is set to have 12 input layer nodes i, 20 hidden layer nodes s, 8 output layer nodes n, a maximum number of iterations of 12000, and a learning step size of... With a target error E of 0.01 and a value of 0.1, the fault cause output is "mold deformation" obtained through sample training. The non-conforming product database is retrieved; the cause code is G6, and the corrective action is "replace the mold". The machine learning result of the process anomaly cause diagnosis expert knowledge is recorded and denoted as JG. JG={Panel thickness out of tolerance, step 50, feature number TX001, upward trend mode, fault cause - mold deformation, cause code G6, corrective action - replace mold}.

[0133] Step 8: In the simulation layer of the digital twin platform, when the aerospace product parts are digitally inspected in the air, the real-time simulation engine is used to achieve virtual-real synchronization. Based on the digital twin, process capability prediction, process quality risk prediction, process anomaly prediction, process anomaly fluctuation source prediction, and scenario simulation are performed.

[0134] Step 9: In the service layer of the digital twin platform, construct a semi-autonomous functional module with multiple model combinations. This module supports flexible expansion and combination. Through this module, perform process capability status analysis, process indicator achievement prediction, process risk estimation, process anomaly warning, and improvement measure feedback. It also provides visualization interaction, AI process anomaly analysis, and decision optimization capabilities.

[0135] Compared with existing technologies, the present invention has the following features: 1. Breakthrough in technological capabilities.

[0136] 1.1 Data acquisition and transmission capabilities.

[0137] 1.1.1 Inspection Data Acquisition: For the part inspection data generated by imported digital inspection equipment, the RPA process automation robot realizes automatic capture and uploading from the industrial control network terminal, which solves the problem that imported equipment cannot obtain production and inspection data due to technical barriers and breaks the manufacturer's technical monopoly.

[0138] 1.1.2 Product Inspection Efficiency: By using the interconnection and parsing method between the industrial control network and the enterprise intranet, data monitoring services are built on both the industrial control network and the enterprise intranet. This enables measurement reports to be automatically transmitted from the industrial control network to the enterprise intranet application server, significantly reducing manual recording and data entry time. At the same time, it avoids the physical storage of traditional paper measurement reports, ensuring data accuracy and timeliness, and greatly improving the efficiency of inspection work.

[0139] 1.1.3 Key Product Characteristics Data Acquisition: The use of RPA robots to automatically collect key quality characteristic data of aerospace products in the process platform solves the problems of difficult unstructured data acquisition, low efficiency of manual statistical analysis, and poor data accuracy.

[0140] For parts inspection data generated by imported digital inspection equipment, the system has transformed from being unable to obtain data due to technical barriers to achieving automated and efficient data collection and acquisition. For aerospace product parts inspection, the system has transformed from archiving printed paper inspection records and manually recording inspection results to automatically collecting, recording, and storing inspection information. The time spent on product inspection has been reduced from 30 minutes to 2-3 minutes, improving inspection efficiency by approximately 92% and data accuracy by 30%. The collection of key characteristics of aerospace product components has been reduced from 10 days / person to 10 minutes / person, and the efficiency of statistical analysis has been improved by 98%.

[0141] 1.2 Geometric model construction capability.

[0142] 1.2.1 Geometric Model Construction Efficiency: A novel model construction method is proposed. Point cloud data is collected through digital measurement, and the measured point cloud is fused with a lightweight model to generate the overall point cloud of the part. Then, a non-uniform rational B-spline (NURBS) surface reconstruction algorithm is used to achieve efficient geometric modeling of complex aerospace parts. This solves the problem of difficulty in reverse modeling due to incomplete scanning of the internal structure of complex parts, and also solves the problem of high economic and time costs of traditional aerospace product supporting parts modeling methods.

[0143] 1.2.2 Lightweight Model: Lightweight modeling methods are employed to improve model running speed, thereby enhancing the timeliness of decision-making in online simulation of aircraft parts based on digital twin models. For critical areas of aerospace product parts, a vertex curvature distribution algorithm is used to preserve high-curvature regions, ensuring that key geometric features are not compromised and meeting the high-precision requirements of key aerospace parts characteristics. Mesh simplification algorithms, such as quadratic error measurement and edge folding operations, are used to reduce the number of vertices and faces, significantly reducing model complexity. For non-critical areas of aerospace product parts, a voxel mesh algorithm is used to further reduce model size.

[0144] 1.2.3 Geometric accuracy control: Based on the spatiotemporal characteristics of the process measurement, the arithmetic mean method and Kalman filtering method are used to fuse the measurement data to improve the accuracy of the process capability digital twin modeling.

[0145] 1.2.4 Dynamic Adaptability: On the model processing server, the engine is called through script tasks to synchronize the 3D model of the supporting parts of aerospace products, and then the lightweight processing process is automatically triggered. This solves the problem that the lightweight model cannot be updated by itself due to frequent changes in the design model of aerospace product parts, and improves the dynamic adaptability of the digital twin geometric model. The modeling time for a certain aerospace component was reduced from 66 hours in traditional manual reverse engineering to 2 hours, improving modeling efficiency by 33 times; the number of model faces was optimized from 180 million to 20,000 (retaining key features), achieving an optimization rate of 99.99%; the file size of a single aerospace product component model was optimized from 580MB to 8.5MB, achieving an optimization rate of 99.0%; the file size of the model of supporting parts for an entire aerospace product was optimized from 80GB to 2.5GB, achieving an optimization rate of 96.9%; in the inspection of the compressor blade disk of a certain aerospace engine, the maximum deviation between the measured value and the model prediction value of key dimensions (such as blade installation angle) was 4.3µm, meeting the requirements of the AS9100 standard.

[0146] 1.3 Multi-source heterogeneous data fusion capability.

[0147] 1.3.1 Interoperability: Enables bidirectional mapping and dynamic interaction of single aerospace product quality characteristic data, CNC machining measurement data, process "man, machine, material, method, environment, and measurement" data, historical quality data, and process quality index data.

[0148] 1.3.2 Real-time performance: The core algorithm utilizes RPA (Robotic Process Automation) robots to collect key quality characteristic data of a single aviation product, enabling dynamic identification and reading of key characteristic data. Through data transmission and integrated interaction via two interconnected networks at the data layer, the network import / export devices on which the core algorithm relies support a maximum of 5 million concurrent connections, can handle 4 gigabits per second of data traffic, support a data transmission rate of 1 Gbps, and the digital twin model has a response latency in the millisecond range, meeting the needs of large-scale data process processing and high-concurrency connection application scenarios.

[0149] 1.3.3 Scalability: Based on a layered model architecture (data layer - model layer - simulation layer - service layer), it can dynamically integrate new sensors / digital inspection equipment, quality characteristics or process parameters to adapt to the multi-scale needs of aerospace manufacturing.

[0150] Multi-source data is correlated based on identifiers such as process number and part drawing number, with an accuracy of approximately 99.7%; cross-system collaborative analysis is supported, reducing the time required by the traditional method from 10.5 hours to 0.5 hours by the method of this invention, an efficiency improvement of approximately 21 times.

[0151] 1.4 High-fidelity real-time quality early warning accuracy.

[0152] 1.4.1 Fidelity: The digital virtual body and the physical entity achieve millimeter-level simulation mapping in terms of geometric structure, state (process quality risk, process quality factor change, design change), and temporality (processing or inspection sequence).

[0153] 1.4.2 Dynamic Multidimensional Early Warning Algorithm: Utilizing a preset limit control chart anomaly detection algorithm suitable for multi-variety, small-batch production, early warnings are achieved for process indicator achievement and process anomalies. A mathematical model of control chart anomaly patterns is established, and a large amount of model training data is generated using the Monte Carlo method. Machine learning algorithms are used to identify control chart anomaly patterns, and this is combined with the SPC anomaly detection method to break through the limitations of traditional single, static early warning rules. Based on the process dynamic risk assessment algorithm, the five-dimensional quality risk of the process ("man, machine, material, method, environment") is calculated, and early warnings are issued through risk thresholds.

[0154] 1.4.3 Integration of Early Warning Mechanism and Digital Twin: Digital twins are created for both part and process capabilities. Control chart early warning mechanisms are integrated with geometric model colors or contours, allowing for visualization of process warnings and capabilities. Quality tools that can be integrated with the digital twins include: rainbow charts, individual value control charts, mean control charts, cumulative sum control charts, and preset limit control charts. By combining traditional SPC technology with digital twins, the visualization effect and application potential of early warnings are enhanced.

[0155] The abnormality detection rate of key processes is 99.5%, and the false alarm rate is <0.3%, which is 22 percentage points higher than the traditional manual sampling inspection (detection rate of 78%).

[0156] 1.5 Intelligent diagnosis of abnormal causes.

[0157] 1.5.1 Closed-loop nature: A quality control closed loop is formed through the service layer (diagnosis-prediction-decision function), for example, after the source of fluctuation is located, improvement measures or process optimization instructions are automatically triggered.

[0158] 1.5.2 Causal Reasoning Engine: It employs a control chart pattern recognition and heterogeneous nonlinear mapping causal reasoning engine based on artificial neural networks, combined with process risk assessment modeling algorithms and dynamic traceability methods for process quality factors. Based on the comprehensive analysis results, it locates the sources of process quality fluctuations (such as tool wear, fixture offset, and material batch defects) and generates a diagnostic report.

[0159] 1.5.3 Self-optimization of heterogeneous cause prediction accuracy By using a machine learning algorithm based on expert knowledge for process anomaly diagnosis, the system can learn and memorize anomaly diagnosis knowledge, thereby optimizing and improving the accuracy of predicting the causes of process anomalies.

[0160] The diagnostic time has been reduced from an average of 8 hours to 3 minutes, and the accuracy of locating quality fluctuation sources is >95% (40% improvement compared to the traditional FMEA method). It also supports automatic recommendation of process improvement or optimization measures.

[0161] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for quality early warning and cause diagnosis of an aviation product manufacturing process based on digital twinning, characterized in that, include: The RPA (Robotic Process Automation) robot is used to obtain digital inspection reports of aerospace products from digital measurement equipment, and then transmit the digital inspection reports from the industrial control network to the enterprise intranet for parsing and processing. After the parsed data is processed by data denoising and point cloud filtering in the data layer, a digital twin of the aerospace product parts is created based on the lightweight model of the aerospace product parts using reverse modeling and dynamic lightweight modeling technology. At the same time, a digital twin that can reflect the manufacturing process capabilities of the aerospace product parts is also created. At the data layer, a recursive algorithm is used to quickly decompose the process quality indicators of the product manufacturing process and generate process-level quality indicator data; RPA process automation robot technology is used to automatically collect key quality characteristic information of aerospace product supporting parts from the CAPP / 3D process design system; and information on the status changes of "man, machine, material, method, and environment" of the process is collected to dynamically generate the quality BOM of aerospace product supporting processes. Establish a data model for aerospace product parts: collect operational information on the "man, machine, material, method, and environment" processes, and use statistical analysis methods such as regression analysis, analytic hierarchy process, and fuzzy evaluation to establish a process quality risk calculation model; use BOM structure transformation and data mapping algorithms to establish a process quality traceability model based on the bill of materials for aerospace product processes; Establish a fault behavior model and use preset limit control chart technology to establish an SPC anomaly detection and early warning model; A machine learning-based method for identifying sources of abnormal fluctuations in process operations is adopted, and a mechanism model is established at the model layer. This mechanism model combines process anomalies with information on quality risks and changes in quality factors to infer and identify sources of process quality fluctuations and provide corresponding improvement measures. A data-driven model is established, and a machine learning method based on expert knowledge for process anomaly diagnosis is used to enable the self-learning and memorization of anomaly diagnosis knowledge, thereby optimizing and improving the accuracy of predicting the causes of process anomalies. 2.The method according to claim 1, characterized in that, The process of building digital twins of aerospace product parts is as follows: Obtain 3D point cloud data for measuring aerospace product parts from the analyzed digital inspection report; Based on the CAITA file for measuring points of aerospace product parts in the PDM system, the geometric information set and theoretical measurement point information set of aerospace product parts are obtained; the geometric information set of parts includes the three-dimensional model of the parts, the spatial coordinate system, and the positioning datum of the parts; the theoretical measurement point information set includes measurement point information. For the 3D model of the part in the theoretical model information set, the associated theoretical measurement point information set is retained, and the geometric description and load-bearing information of the 3D model of the part are simplified; the lightweighting process is as follows: On the model processing server, a scheduled task is created to call the script file and synchronize the EBOM data in the 3D process design system; the scheduled task is triggered to call the 3D process design service to retrieve the model; and it is checked whether the CATIA model exists in the download path and synchronized accordingly. Once the model synchronization in the EBOM is complete, the model lightweighting process will be triggered. First, the affected VCIs will be analyzed and traversed, and all 3D models will be collected into the conversion workspace. The influence analysis is performed upwards from the VCI to obtain the corresponding ICI model. If multiple ICI models exist, only one is selected. If some VCI models are incorrectly assembled, the model is converted to the ICI model. The ICI model obtained from the traversal influence analysis is then subjected to lightweight conversion. Automatic processing includes merging facet nodes, preserving high curvature contours, voxel-based simplification, deleting useless nodes, and deleting reference model nodes to obtain a lightweight model. The collected measured point cloud data of aerospace product parts is transformed into spatial coordinates, and the transformed point cloud data is fused with the lightweight model. The measured point cloud dataset is mapped to the corresponding theoretical measurement point locations on the lightweight model, thus visualizing the measured point cloud dataset on the lightweight model and achieving the substitution of theoretical measurement points by the 3D point cloud. Based on the 3D point cloud, new 3D measured points are generated to encrypt the point cloud data. Subsequently, digital twins of aerospace product parts are created using the encrypted point cloud data. For geometrically shaped surfaces, surface reconstruction is performed through fitting or sectional drawing. For complex surfaces, a feature curve mesh is first fitted, followed by surface fitting, and finally surface reconstruction. 3.The method according to claim 1, characterized in that, The process of creating a digital twin that reflects the manufacturing capabilities of aerospace product parts is as follows: Collect 3D point cloud data of the quality characteristics of a certain process in aerospace parts production; If the process only processes one product and the quality characteristic is measured only once, the measured value of the spatial dimension of the quality characteristic is calculated based on the three-dimensional point cloud data. If this process involves batch processing of multiple products, then the arithmetic mean and Kalman filter method are used to filter the measured values. Obtain lightweight models of aerospace product parts and annotate the mass characteristics and theoretical measurement points on the models; The filtered measured values ​​are fused with the lightweight model, and the measured values ​​and the theoretical values ​​corresponding to the theoretical measurement points are distinguished by shape and color. The state can be traced back according to any measurement time sequence to obtain a digital twin. Using the filtered measured values, the consistency between the actual average value of the process and the center value of the quality characteristic specification is calculated, and the accuracy of the process is evaluated. The mean deviation between the process average value and the center value of the specification is calculated, and process adjustment suggestions are proposed based on the magnitude of the deviation. 4.The method according to claim 1, characterized in that, The RPA robot is controlled to log into the CAPP process design system, download key and process specification data for each model according to the rules, and merge the data tables downloaded for each model.

5. Control the RPA robot to filter data by model and extract process content, and replace the values ​​of process content according to the engineering symbol table; Control the RPA robot to log in to the MES manufacturing execution system, and download the quality control records of key processes for each model according to the unit, model, batch, key process number, and process specification number, including sample size, measured value, unit, model, batch, key process number, and process specification number; The RPA robot is controlled to use the unit, model, batch, key process number, quality characteristic identifier, and process specification number as unique matching conditions to match the CAPP process query results with the MES key process quality control record table and fill in the values ​​to form a key process measured value dataset. Control the RPA robot to log in to the Integrated Quality System (IQS), import the dataset of measured values ​​of key processes according to the predefined fields in the data collection form, and generate a data collection form for key quality characteristics measurement. By combining key quality characteristic measurement data collection forms and information on the status changes of "man, machine, material, method, and environment", the quality BOM of supporting processes for aerospace products is dynamically generated. 6.The method of claim 1, wherein the method further comprises: determining a cause of the abnormality based on the digital twin of the aircraft product manufacturing process. Establish a process quality risk calculation model, including: Starting from five dimensions—"people, machines, materials, methods, and environment"—the manufacturing process of key product processes is analyzed, quality risk factors under each dimension are identified, and a quality risk system framework diagram for key product processes is established. The framework diagram consists of a criterion layer and an indicator layer. Among them, "people, machines, materials, methods, and environment" are the five risk factors in the criterion layer; the corresponding factors under each factor in the criterion layer constitute the indicator layer. The weight vectors of each factor in the criterion layer are constructed using the analytic hierarchy process. We construct the weight vectors of the corresponding indicator layer factors under each factor in the criterion layer; and combine the comparison and quantification value table between each indicator with the calculated weight vectors of each factor in the criterion layer to finally calculate the weight value of each factor. Introducing factor quality risk coefficients, and combining the characteristics of each factor in the indicator layer, mathematical methods such as regression analysis and fuzzy evaluation are used to establish a quantitative evaluation model for the quality risk of each factor, and the quality risk coefficient of the factor under the current state is calculated. Based on the quality risk coefficients of each factor, the overall quality risk coefficient of key processes is calculated by weighting. When the coefficient is less than the set threshold, it indicates that the product quality risk is low and at an acceptable level under the current manufacturing conditions. When it exceeds the set threshold, the factors with higher quality risk need to be improved before product processing to reduce the quality risk.

7. The method for quality early warning and anomaly diagnosis of aerospace product manufacturing processes based on digital twins as described in claim 1, characterized in that, Establish a process quality traceability model based on the bill of materials for aerospace product processes, including: Acquire and synchronize the structural information of the model layer, station layer, assembly process layer, supporting layer, and parts process specification layer in the bill of materials for aviation product processes; By using the unique identifier of the assembly process number of aviation products, the actual manufacturing quality information of "man, machine, material, method, environment, and measurement" in the assembly product manufacturing record of the assembly MES system is read. The "man, machine, material, method, environment, and measurement" information of each assembly process / step in the assembly product manufacturing record is associated with the assembly process and step in the aviation product process material list and stored in the assembly process quality layer. By matching the material codes of the part MES system with the material codes of the process bill of materials, the "man, machine, material, method, environment, and measurement" quality information of the part product manufacturing record in the part MES system is read and stored in the part process quality layer. Identify the addition, reduction, and modification of assembly / parts processes for aerospace products, as well as information on changes in process quality. Employ data derivation, filtering, and replacement mapping rules to dynamically update the process quality traceability BOM. The process quality traceability BOM structure should at least include the model layer, station layer, assembly process layer, assembly process quality layer, matching layer, parts process specification layer, and parts process quality layer.

8. Create a forward traceability path for part information. Based on the unique material code, trace the BOM from the bottom layer to the top layer of the process quality traceability BOM, execute and retrieve the matching part process specifications, part process quality, assembly process, station position and assembly frame information in sequence. Create a reverse traceability path for process failures / abnormalities. Based on the unique identifier of the assembly process or the unique identifier of the faulty part, trace the process quality from the top level of the BOM downwards, and execute and sequentially search the assembly process layer, assembly process quality information layer, matching layer, part process specification layer, and part process quality sub-layer to obtain the assembly process quality, matching, part process specification / finished product information, and part process quality information of the faulty / abnormal process.

9. The method for quality early warning and anomaly diagnosis of aerospace product manufacturing processes based on digital twins as described in claim 1, characterized in that, A pre-defined limit control chart technique is used to establish an SPC anomaly detection and early warning model, including: For the production process of aerospace product parts manufacturing, the set values ​​of its process quality indicators are obtained, and it is assumed that the production process is under statistical control and the quality characteristic values ​​of the production process follow a normal distribution. If the mean value of the machining dimension does not coincide with the distribution center of the tolerance zone, calculate the corresponding process capability index setting value according to the preset offset and process quality index setting value, and finally obtain the sample standard deviation setting value. Calculate the mean derived value based on the actual offset direction of the sample. Based on the "3σ" control limit determination principle, the center line and upper and lower control limits of the mean preset limit control chart are calculated, thereby establishing the mean preset limit control chart; Therefore, for the sample size interval 5≤n<100, the F test method is used to judge the actual deviation of the sample and the standard deviation set value are homogeneous; if there is no significant difference between the two variances, T test is needed to judge the consistency of the actual mean value u of the sample and the mean value derived value : if there is no significant difference between the two mean values, the mean value preset limit control chart is used to judge whether it belongs to the process steady state control; if there is a difference between the two mean values, the actual offset of the sample mean value is calculated >1.5 , the process conditions need to be adjusted to make u as close as possible to the tolerance center value; if ≤1.5 , it indicates that the process is in a stable state.

10. The method for quality early warning and anomaly diagnosis of aerospace product manufacturing processes based on digital twins as described in claim 1, characterized in that, The process of constructing the mechanistic model is as follows: Obtain measurement data of a certain quality characteristic of a process or part, and use a control chart pattern recognition method based on machine learning to identify the control chart pattern of the quality characteristic. If it is identified as an abnormal pattern, establish an abnormal pattern matrix M. For trend and step-type patterns, the trend analysis method is used to distinguish the factors that cause the abnormal pattern, whether they are accidental or systematic. If it is determined to be an accidental factor, no fault diagnosis is required. If it is determined to be a systematic factor, the cause of the process abnormality is further analyzed. If a certain process or part is determined to be affected by systematic factors, it is necessary to further identify the sources of process quality fluctuations. Fuzzy statistical methods or expert experience-based methods are used to determine the degree of membership between the abnormal patterns and causes of the control charts, and finally, a fuzzy relation matrix Q is established. The matrix Q and M are integrated to obtain a fault occurrence possibility matrix T. The element values in the matrix T are merged to divide different intervals of [0, 1], and a credibility value is taken in each interval. Each element T i According to the corresponding credibility value of the interval into which the element falls, the abnormal reason is diagnosed according to the size of the credibility value. The potential quality risks in the "man, machine, material, method, and environment" aspects of the production process or parts are defined and quantified. Potential quality risk coefficients for various factors in the process are set, and dynamic risk thresholds for the process are set. A mapping relationship between the distribution of abnormal causes and process risk indicators is established. Combined with the predicted causes of process failures and potential quality risk coefficients, the causes of process abnormalities are analyzed.

11. The method for quality early warning and anomaly diagnosis of aerospace product manufacturing processes based on digital twins as described in claim 1, characterized in that, The process of building a data-driven model is as follows: According to different aviation product processing specialties and processes, the database of non-conforming aviation products is classified. The non-conforming product drawing number is used as a unique identifier. The product manufacturing record barcode of the MES system is matched to obtain the total sample size and sample measured value information of the processing / manufacturing batch in which the non-conforming quality characteristics are located. The Monte Carlo simulation method was used to obtain sample data with the same quality characteristics as the actual production. After preprocessing, the data was input into the BP neural network for control chart pattern recognition to obtain model M1. The measured values ​​of quality characteristics are preprocessed and used as input to a BP neural network. The trained model M1 is then used to identify control chart patterns, yielding the actual results of control chart pattern recognition. The database of non-conforming products is grouped according to different manufacturing processes of aviation products and processing specialties. A BP neural network is constructed and trained in each group to obtain the nonlinear mapping relationship from the abnormal mode of the control chart to the cause of failure. The parameters of the BP neural network are written into the database.

12. In the simulation layer of the digital twin platform, when aerospace product parts are digitally inspected in-flight, a real-time simulation engine is used to achieve virtual-real synchronization. Based on the digital twin, process capability prediction, process quality risk prediction, process anomaly prediction, process anomaly fluctuation source prediction, and scenario simulation are performed.

13. A terminal device comprising a processor, a memory, and a computer program stored in the memory; characterized in that, When the processor executes the computer program, it implements the method for quality early warning and anomaly diagnosis of aviation product manufacturing process based on digital twin as described in any one of claims 1-9.

Images