SPC quality monitoring method and system based on AI intelligent identification

Abstract

The application discloses an SPC quality monitoring method and system based on AI intelligent identification, relates to the technical field of industrial AI application, and comprises the following steps: reading real-time data flow in a production control process, performing double-end process control analysis according to an SPC module to obtain an SPC analysis result; an AI intelligent agent performs anti-convolution traceability deconstruction and interference simulation positioning on an abnormal state vector through data interaction with the SPC module, and determines production control data; and according to the SPC analysis result and the production control data, alarm management and production control driving in the production control process are performed. The application solves the technical problems of single analysis dimension, difficult abnormal root cause tracing and low control efficiency of traditional SPC quality monitoring in industrial production, achieves comprehensive real-time monitoring, early warning, accurate root cause positioning and efficient control of industrial production quality, and improves the technical effects of quality control accuracy and efficiency.

Description

Technical Field

[0001] This invention relates to the field of industrial AI application technology, and in particular to an SPC quality monitoring method and system based on AI intelligent recognition. Background Technology

[0002] In the field of industrial automation control, intelligent monitoring and quality control of the production process are crucial for product qualification rate and equipment operational stability, and SPC (Statistical Process Control) is one of the core technologies. Existing technologies mostly employ traditional SPC systems combined with conventional sensing equipment to analyze production parameters and monitor quality, playing a role in stable production scenarios. However, with the increasing demand for industrial intelligence, traditional technologies reveal limitations when applied in complex industrial environments: they struggle to simultaneously handle both operational condition control and disturbance analysis, cannot accurately trace the root causes of anomalies, resulting in incomplete data and delayed early warnings, failing to meet the needs of accurate quality assessment and efficient management under industrial automation control. Summary of the Invention

[0003] This application provides an SPC quality monitoring method and system based on AI intelligent recognition, which solves the technical problems of traditional SPC quality monitoring in industrial production, such as single analysis dimensions, difficulty in tracing the root causes of anomalies, and inefficient control.

[0004] The first aspect of this application provides an AI-based intelligent identification-based SPC quality monitoring method, the method comprising: reading real-time data streams during the production control process; performing dual-end process control analysis based on the SPC module to obtain SPC analysis results, wherein the dual-end process control analysis includes process state vector runaway prediction of the operating condition control part and mass field force line propagation of the operating condition disturbance part; the AI ​​agent, through data interaction with the SPC module, performing deconvolution source tracing deconstruction and interference simulation localization on the abnormal state vectors to determine production control data; and executing alarm management and production control driving during the production control process based on the SPC analysis results and the production control data.

[0005] The second aspect of this application provides an AI-based intelligent recognition-based SPC quality monitoring system, comprising: an SPC analysis result acquisition module, used to read real-time data streams during the production control process, perform dual-end process control analysis based on the SPC module, and obtain SPC analysis results, wherein the dual-end process control analysis includes process state vector runaway prediction of the operating condition control part and mass field force line propagation of the operating condition disturbance part; a production control data acquisition module, used by the AI ​​agent to perform deconvolution source tracing deconstruction and interference simulation localization on abnormal state vectors through data interaction with the SPC module, and determine production control data; and a production control execution module, used to execute alarm management and production control driving during the production control process based on the SPC analysis results and production control data.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0007] This application conducts dual-end SPC analysis on the operating condition control data and operating condition disturbance data of the industrial production process. After processing such as AI intelligent agent deconvolution tracing and interference simulation, production control data is obtained. By combining the SPC analysis results and control data, multi-threaded instruction execution is performed, thereby realizing real-time and comprehensive monitoring of production quality, early warning of anomalies, and precise root cause control. This makes industrial production quality control more precise and efficient, achieving the technical effect of comprehensive real-time monitoring, early warning, precise root cause location, and efficient control of industrial production quality, and improving the accuracy and efficiency of quality control. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 This is a flowchart illustrating the SPC quality monitoring method based on AI intelligent recognition provided in this application embodiment.

[0010] Figure 2 This is a schematic diagram of the structure of the SPC quality monitoring system based on AI intelligent recognition provided in the embodiments of this application.

[0011] Figure labeling: SPC analysis result acquisition module 1, production control data acquisition module 2, production control execution module 3. Detailed Implementation

[0012] This application provides an SPC quality monitoring method and system based on AI intelligent recognition, which solves the technical problems of traditional SPC quality monitoring in industrial production, such as single analysis dimensions, difficulty in tracing the root causes of anomalies, and inefficient control.

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0014] It should be noted that the terms "first," "second," etc., in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules not explicitly listed or inherent to such processes, methods, products, or devices.

[0015] Example 1, as Figure 1 As shown, an SPC quality monitoring method based on AI intelligent recognition is described, wherein the method includes:

[0016] The system reads real-time data streams from the production control process and performs two-way process control analysis based on the SPC module to obtain SPC analysis results. The two-way process control analysis includes process state vector runaway prediction in the operating condition control part and mass field force line propagation in the operating condition disturbance part.

[0017] In this embodiment, the SPC module is a hardware and software integrated functional unit in industrial production scenarios, based on the principles of statistical process control, integrating data acquisition and analysis functions to monitor the stability of production process parameters and identify abnormal fluctuations. The operating condition control section is an active control mechanism in industrial production that actively adjusts control variables such as equipment operating parameters to stabilize controlled parameters such as product quality and production status within a preset range. The operating condition disturbance section is a collection of various factors, besides actively controlled variables, that can cause controlled parameters to deviate from the preset range during industrial production, such as raw material fluctuations and changes in environmental temperature and humidity.

[0018] Specifically, firstly, corresponding sensors and programmable logic controllers (PLCs) are deployed at key equipment and process nodes in the production control process. For example, in the production control scenario of a machining workshop, speed sensors and torque sensors are deployed at the spindle of the CNC machine tool, weight sensors are deployed at the weighing nodes of the raw material conveyor belt, and temperature and humidity sensors and dust concentration sensors are deployed at workshop ventilation openings and around equipment. At the same time, PLCs are integrated in the control cabinets of each equipment. Through the hardware connection between the sensors and the PLCs, a hardware network covering the acquisition of operating condition control and operating condition disturbance parameters is built.

[0019] Next, the OPC UA industrial communication protocol is used to establish transmission links between each programmable logic controller (PLC) and the industrial server in the workshop, completing the data interaction configuration between the hardware and the terminal to ensure that the collected data can be stably transmitted to the server. Subsequently, the acquisition frequency is set according to the parameter monitoring requirements, and each sensor and PLC is activated to acquire the raw parameter values ​​of each monitoring point in real time. Then, the high-frequency fluctuating speed and torque data are processed using the moving average filtering method on the raw values ​​received by the server. Duplicate values ​​of parameters such as weight, temperature, and humidity are eliminated by comparing duplicate data within 1 second at the same monitoring point, finally obtaining a continuous and stable real-time data stream for the production control process.

[0020] Finally, a two-way process control analysis was performed based on the SPC module, clarifying that the SPC module employs parallel deployment of process control and process disturbance branches. Specifically, the process control branch analyzes the first operating condition control data in the real-time data stream to determine the first SPC result, while the process disturbance branch analyzes the second operating condition disturbance data in the real-time data stream to determine the second SPC result. Finally, by fusing the first and second SPC results, a complete SPC analysis result is formed. This step will be explained in detail later.

[0021] The AI ​​agent interacts with the SPC module to perform deconvolution source tracing and interference simulation to locate abnormal vectors and determine production control data.

[0022] In this embodiment of the application, the AI ​​agent is an intelligent system based on artificial intelligence technology that can perceive environmental information, make autonomous decisions, and execute actions to achieve specific goals.

[0023] Optionally, the AI ​​agent is constructed first: First, deconvolutional source tracing is performed based on the anomalous state vector, and the contribution to the state shift is used as the core guide for source tracing, and the first deconstruction layer is deployed accordingly; Second, based on the root cause parameters obtained from source tracing, interference simulation operation is performed, and the location of the minimum root cause set is completed, and the second control layer is deployed accordingly; Third, the first deconstruction layer and the second control layer are cascaded, and the AI ​​agent is formed through this cascade operation. Finally, a data interaction channel between the AI ​​agent and the SPC module is established. This step will be explained in detail in the following content.

[0024] Next, the SPC analysis results are identified. If an anomalous vector is found in the results, the constructed AI agent is activated. Then, the anomalous vector is deconvolved through the AI ​​agent's first deconstruction layer to determine the first root cause parameter, which falls within the first operating condition control data category. Finally, the first root cause parameter is imported into the AI ​​agent's second control layer, which performs parameter interference simulation. By determining whether the process state vector returns to the center of the healthy state cluster after interference, the production control data is determined. The determination of the production control data is based on minimizing the adjustment cost during the iterative interference simulation process; this step will be explained in detail later.

[0025] Based on the SPC analysis results and production control data, alarm management and production control are implemented during the production control process.

[0026] In one embodiment of this application, the SPC analysis results and the determined production control data are first mapped to specific control terminals of the production system, completing the mapping between the control terminals and forming multiple mapping groups. Next, for each mapping group, multiple instruction groups containing early warning instructions and production control instructions for that unit are generated, based on the smallest control unit of the production system. Finally, these instruction groups are sent to the corresponding control terminals using a multi-threaded approach to execute alarm management and production control driving operations during the production process. This step will be described in detail later.

[0027] Furthermore, the method provided in this application embodiment includes:

[0028] The SPC module includes a process control branch and a process disturbance branch deployed in parallel. Based on the process control branch, the first operating condition control data in the real-time data stream is analyzed to determine a first SPC result. Based on the process disturbance branch, the second operating condition disturbance data in the real-time data stream is analyzed to determine a second SPC result. The first SPC result and the second SPC result are fused together to obtain the SPC analysis result.

[0029] Specifically, the process control branch is first constructed, using an industrial server as the data processing carrier, which is then connected to a programmable logic controller (PLC) connected to sensors monitoring the equipment's operating parameters. Data processing rules are configured in the PLC using ladder logic programming to calculate averages and filter fluctuations in data such as equipment speed and processing pressure. An independent industrial Ethernet port is configured on the server to connect to the SPC module, allowing the processed operating condition control data to be transmitted separately, forming a branch that analyzes only the first operating condition control data.

[0030] Next, a process disturbance branch is constructed, using an edge computing gateway as the data processing carrier, which is connected to the acquisition terminal connected to sensors monitoring environmental and raw material parameters. Data processing logic is configured using industrial configuration software to determine outliers and preliminarily identify trends in data such as temperature, humidity, and raw material weight. An additional independent industrial Ethernet port is configured for the edge computing gateway to connect to the SPC module, ensuring that the processed disturbance data is transmitted synchronously and without interference with the control branch data, thus forming a branch that analyzes only the disturbance data of the second operating condition.

[0031] Then, the first operating condition control data is selected from the real-time data stream. Next, multi-source heterogeneous fusion processing is performed on the first operating condition control data to generate a process state vector, which represents the comprehensive production control state at each moment. Then, a vector space is constructed, and a health state cluster center is initialized in real-time within the space. This center dynamically moves according to the production operating conditions at different times. Finally, by comparing the correlation between the process state vector and the health state cluster center, the first SPC result is determined. This step will be explained in detail later.

[0032] Next, the second-condition disturbance data is selected from the real-time data stream. Then, a mass field force line is constructed, and the second-condition disturbance data is located within this mass field force line. Using the propagation and amplification of the disturbance along the mass field force line as the analytical guide, disturbance propagation analysis is performed on the second-condition disturbance data. Finally, the disturbance damping capacity is determined through the above analysis, and this disturbance damping capacity is used as the second SPC result. This step will be explained in detail later.

[0033] Then, the core information of the first and second SPC results is clarified: The first SPC result is the runaway attractor, which needs to be quantified into a runaway risk value. Based on the historical occurrence frequency of the defect mode corresponding to the runaway attractor, the higher the frequency, the higher the runaway risk value. For example, high-frequency defect modes correspond to 5 points, medium-frequency modes correspond to 3 points, and low-frequency modes correspond to 1 point. The second SPC result is the disturbance damping capability, which is also quantified into a damping capability value. Based on the historical normal range of attenuation coefficient and propagation speed, strong damping capability (large attenuation coefficient, slow propagation speed) corresponds to 5 points, medium damping capability corresponds to 3 points, and weak damping capability corresponds to 1 point. This completes the unified quantification of the two types of results, providing basic data for fusion.

[0034] The results were then fused using a weighted summation method. The weights of the two types of results were determined using historical quality data: the proportion of quality problems caused by runaway attractors in the past 12 months was used as the weight for the first SPC result; the proportion of quality problems caused by insufficient damping capacity was used as the weight for the second SPC result, with the sum of the two weights being 1. For example, if runaway attractors accounted for 60% of the problems and insufficient damping capacity accounted for 40%, the weights for the first SPC result would be set to 0.6 and the second to 0.4. Next, the quantified runaway risk value was multiplied by its corresponding weight, and the damping capacity value was multiplied by its corresponding weight. The sum of these two results yielded a comprehensive score, which is the preliminary fusion result.

[0035] Finally, the SPC analysis results are graded based on the overall score: a preset score of 5 indicates a quality and safety level with no quality risk; 3-4 indicates a level of concern requiring enhanced monitoring; and 1-2 indicates a level of warning requiring immediate process adjustment. The grade is verified in conjunction with real-time production conditions. If the overall score is at the level of concern and the real-time collected equipment operating parameters show no abnormal fluctuations, the level of concern is maintained. If the score is at the level of warning and real-time disturbance data shows a continuous decline in damping capacity, the level of warning is confirmed. Ultimately, this grade and the corresponding risk causes, such as the type of runaway attractor or weak points in damping capacity, are integrated into the SPC analysis results.

[0036] Furthermore, the method provided in this application embodiment includes:

[0037] Based on the real-time data stream, first operating condition control data is determined; by performing multi-source heterogeneous fusion on the operating condition control data, a process state vector is determined, wherein the process state vector represents the comprehensive state of production control at each moment; a vector space is constructed and a health state cluster center is initialized in real time, wherein the health state cluster center moves with the operating conditions at different times; based on the process state vector and the health state cluster center, a first SPC result is determined.

[0038] In this embodiment of the application, the process state vector is a multi-dimensional data carrier that represents the comprehensive state of production control at a certain moment after fusing multi-source operating condition control data such as spindle speed, machining pressure, and feed rate during the production process.

[0039] Optionally, firstly, identify the core parameter types corresponding to the first operating condition control data during the production process. These typically include active control parameters such as spindle speed, machining pressure, and feed rate. Use rule-based filtering based on parameter tags to process the real-time data stream. First, preset attribute tags for operating condition control parameters such as spindle speed, machining pressure, and feed rate in the data processing system. Then, iterate through each data point in the real-time data stream, comparing its built-in attribute tags with the preset tags. Extract data with matching tags, while removing data tagged with non-operating condition control parameters such as ambient temperature and humidity or raw material weight. Through this filtering process, the first operating condition control data containing only operating condition control parameters is separated from the real-time data stream.

[0040] Next, since the first operating condition control data contains multi-source heterogeneous data with different units such as rotational speed (r / min), pressure (MPa), and feed rate (mm / s), the min-max standardization method is first used to process each parameter. The maximum and minimum values ​​of each parameter during the historical normal operation of the production system are collected. By substituting these values ​​into the formula: (current parameter value - historical minimum value) / (historical maximum value - historical minimum value), all parameter values ​​are uniformly mapped to the [0,1] interval to obtain standardized parameter values. Then, based on industry production experience, the weights of each parameter are assigned, and the standardized parameter values ​​are weighted to obtain the fused value at each moment. Finally, the fused value at each moment is combined with the standardized parameter values ​​at that moment to form a multi-dimensional array. This multi-dimensional array is the process state vector, which can completely represent the comprehensive state of production control at each moment.

[0041] Next, a vector space was constructed using a Euclidean space model. A corresponding multidimensional Euclidean space was built based on the dimensions of the process state vectors; for example, a 4-dimensional space containing 3 standardized parameters and 1 fused value ensures that each process state vector has a unique coordinate point in the space. The health state cluster centers were initialized using the K-means clustering algorithm. Process state vectors from 100 consecutive time points during the initial normal operation of the production system were selected as the initial sample set. Since the health state is a single category, the number of clusters, K=1, was set. The mean of each dimension of all vectors in the sample set was calculated, and the resulting mean vector is the initial health state cluster center. To enable the health state cluster centers to move with different operating conditions at different times, a sliding window update strategy was adopted. The window size was set to 50 time points. Whenever a new process state vector was added, the vector from the oldest time point within the window was removed, and the mean of each dimension of all vectors within the current window was recalculated. The new mean vector replaced the original health state cluster center, ensuring that the health state cluster centers could adjust in real time according to changes in operating conditions.

[0042] Finally, the geodesic distance between the process state vector and the center of the healthy state cluster is calculated, and this distance is used as the state offset. Next, within the constructed vector space, the short-time evolution path is determined by analyzing the momentum of the process state vector's trajectory. Combining the obtained short-time evolution path and state offset, the runaway attractor is predicted. The runaway attractor is defined by the vector region and defect mode in the vector space, and this runaway attractor is used as the first SPC result. This step will be explained in detail later.

[0043] By filtering and separating operating condition control data according to rules, standardizing and weighting to generate process state vectors, and building Euclidean space and updating cluster centers with sliding windows, a dynamic and accurate representation of the overall state of production control is achieved, providing a real-time state benchmark that fits the actual operating conditions for determining the first SPC result.

[0044] Furthermore, the method provided in this application embodiment includes:

[0045] The geodesic distance between the process state vector and the center of the healthy state cluster is calculated as the state offset; in the vector space, the short-time evolution path is determined by analyzing the trajectory momentum of the process state vector; based on the short-time evolution path and the state offset, the runaway attractor is predicted as the first SPC result, wherein the runaway attractor is defined by the vector region and defect mode located in the vector space.

[0046] Specifically, the process state vector at the current moment and the health state cluster center updated in real time are first obtained; both are multi-dimensional data in vector space. The geodesic distance is calculated using the Euclidean distance method. First, the coordinate values ​​of the corresponding dimensions of the process state vector and the health state cluster center are extracted. The difference between the two vector data in each dimension is calculated, and the differences in all dimensions are squared. All squared results are summed, and the square root of this sum is the geodesic distance. This geodesic distance is defined as the state offset, used to quantify the degree of deviation between the current production control state and the health baseline state.

[0047] Next, process-state vector data from the established vector space are extracted from the most recent consecutive timeframes. The momentum of the motion trajectory is analyzed by examining the differences between process-state vectors at adjacent timeframes; specifically, the difference in the corresponding dimensional data of process-state vectors between every two adjacent timeframes is calculated to obtain momentum data for each adjacent timeframe. The average direction and magnitude of these momentum data changes are statistically analyzed to determine the overall trend of the process-state vector's motion. Based on this average trend, the possible positions of the process-state vectors at a few future timeframes, such as 2-3 timeframes, are predicted. The current timeframe, the positions of the process-state vectors at the most recent consecutive timeframes, and the predicted positions are then connected sequentially to form a short-term evolution path reflecting the short-term motion direction of the process-state vectors.

[0048] Finally, runaway attractors are defined based on historical production data. All process state vector data from past production processes that resulted in quality defects are collected and imported into a vector space. K-means clustering is then used to cluster these data, forming multiple clustered vector regions. The specific defect pattern corresponding to each vector region is recorded, such as dimensional deviations or surface defects. Each vector region associated with a defect pattern is a runaway attractor. Subsequently, runaway attractors are predicted by combining short-time evolution paths and state shifts: first, a normal range threshold for state shifts is determined based on state shift data from historical normal production periods; if the current state shift exceeds this normal threshold, and the short-time evolution path extends towards a defined runaway attractor region, then the runaway attractor corresponding to that region is taken as the first SPC result.

[0049] Furthermore, the method provided in this application embodiment includes:

[0050] Based on the real-time data stream, the disturbance data for the second operating condition is determined; a mass field force line is constructed, and the disturbance data for the second operating condition is located within the mass field force line. The disturbance damping capacity is determined by performing disturbance propagation analysis, which serves as the second SPC result; wherein the analysis is guided by the propagation and amplification of the disturbance along the mass field force line.

[0051] In this embodiment, the mass field force line is a concrete disturbance propagation path within the production system that connects disturbance sources such as machine tool foundations and servo motors with machining accuracy detection nodes, and carries the weight of the disturbance's impact on quality.

[0052] Specifically, the second operating condition disturbance data is first determined from the real-time data stream. Similarly, specific labels for disturbance parameters such as machine tool foundation settlement and servo motor current harmonics are preset in the data processing system. Basic identification thresholds are also set for each disturbance parameter, such as foundation settlement above 0.01 micrometers and specific frequency bands for motor current harmonics. Then, the real-time data stream is traversed, comparing the attribute labels of each data point with the preset disturbance labels. For data with matching labels, their values ​​are further checked to see if they exceed the basic identification thresholds. Regular data with mismatched labels or values ​​below the threshold—that is, irrelevant noise as determined by the SPC—are discarded. Data that meets the criteria, such as machine tool foundation settlement and servo motor current harmonics, are retained, ultimately yielding the second operating condition disturbance data.

[0053] Next, a mass field force line is constructed. For each entity in the production system, a mass potential energy is defined based on the entity's contribution to product quality and the risk of damage. Each entity is assigned a corresponding mass potential energy value, which dynamically changes with different production conditions at different times. Then, using the assigned mass potential energy values ​​of each entity as a benchmark, a transmission path based on quality influence is built between entities. The construction of this path ultimately forms the mass field force line. This step will be explained in detail later.

[0054] Next, the disturbance data of the second working condition is located in the mass field force line, and the disturbance propagation analysis is carried out with the propagation and amplification of the disturbance along the mass field force line as the core analysis guide to determine the disturbance damping capacity: First, the selected disturbance data of the second working condition is mapped to the specific disturbance source node of the mass field force line to ensure that the subsequent analysis is carried out around the force line and does not deviate from the preset propagation path, thus completing the data location. The disturbance source node is a specific individual in the mass-related entities covered by the mass field force line that has experienced the working condition disturbance.

[0055] Subsequently, a time-series data tracing method was used to analyze the disturbance propagation. This involved recording the occurrence time of the disturbance at each node along the mass field force lines using a data acquisition system, calculating the time difference between the source node and subsequent nodes, and combining this with the path length between nodes to determine the disturbance propagation speed. Simultaneously, the intensity values ​​of the disturbance at each node were compared. If the disturbance intensity at a later node was less than that at a previous node, the attenuation coefficient was calculated using the ratio of the disturbance intensity at the later node to that at the previous node. If the disturbance intensity at a later node was greater than that at a previous node, it was identified as an amplification phenomenon along the mass field force lines, and the amplification factor was recorded.

[0056] Finally, the system's damping capability is evaluated by combining propagation speed, attenuation coefficient, and the presence of amplification: Under normal operating conditions, a healthy system should be able to absorb and attenuate small disturbances to maintain operational stability under these disturbances. If the disturbance propagates slowly along the force line, has a large attenuation coefficient, and exhibits no amplification, it indicates that the system can effectively absorb and attenuate the disturbance, demonstrating high immunity. If the disturbance propagates quickly along the force line, has a small attenuation coefficient, or exhibits amplification, it indicates that the system cannot suppress the propagation of the disturbance or even allows it to amplify, resulting in weak damping capability and low immunity. Ultimately, the damping capability of this disturbance is used as the second SPC result.

[0057] By employing steps such as label matching to filter disturbance data, correlation analysis to construct mass field force lines, and time-series tracking and intensity comparison analysis to analyze disturbance propagation, the disturbance damping capability of the production control system is accurately assessed and used as a second SPC result. This overcomes the shortcomings of traditional SPC, which treats such disturbance data as irrelevant noise and ignores the assessment of the system's intrinsic health status.

[0058] Furthermore, the method provided in this application embodiment includes:

[0059] For each entity in the production system, a mass potential energy value is assigned. The mass potential energy is defined by the entity's contribution to product quality and the risk of damage. The mass potential energy value changes with different operating conditions at different times. Based on the mass potential energy value, a transmission path based on quality influence between entities is constructed to form a mass field force line.

[0060] Specifically, the scope of all entities involved in quality formation within the production system is first clearly defined, encompassing units that directly or indirectly affect product quality, such as lasers, robot joints, and raw materials. A method combining expert scoring and historical quality data is used to define the quality potential value of each entity: industry technical experts are first organized to score each entity's contribution to product quality and risk of quality failure on a scale of 1 to 10, based on the entity's functions, such as the energy stability of the laser and the purity of the raw materials. The higher the contribution score and the lower the risk of failure score, the stronger the entity's positive impact on quality.

[0061] Next, product quality data under normal and abnormal operating conditions for the past 12 months were collected. The correlation coefficient between the entity's condition and the quality pass rate was calculated. This coefficient was used to weight and correct the expert scores, with higher correlation coefficients resulting in greater weights, thus obtaining the basic mass potential energy value for each entity. Subsequently, real-time operating parameters of the entity, such as laser energy fluctuations and raw material moisture content, were collected hourly. The deviation rate between real-time parameters and historical normal parameters was compared, and the basic mass potential energy value was dynamically adjusted according to the magnitude of the deviation rate. The larger the deviation rate, the greater the reduction in potential energy value, allowing the mass potential energy value to change with different operating conditions at different times.

[0062] Finally, using the real-time updated mass potential energy values ​​of each entity as a benchmark, correlation analysis is employed to construct a transmission path based on mass influence between entities: First, the mass potential energy value changes of any two entities within the same time period are extracted. The Pearson correlation coefficient is calculated to determine the strength of the mass correlation between the two entities; the closer the absolute value of the correlation coefficient is to 1, the stronger the correlation. Entities with higher mass potential energy values ​​are identified as source entities, and those with lower potential energy values ​​are identified as affected entities, and the direction of influence transmission is determined. The correlation strength between all entities is ranked, and strong correlations with an absolute correlation coefficient greater than 0.6 are retained. A single transmission path is constructed according to the direction from source entity to affected entity. Finally, all single transmission paths are integrated to form a mass field force line covering the main entities of the production system and clearly defining the direction of influence transmission.

[0063] Furthermore, the method provided in this application embodiment includes:

[0064] A first deconvolutional deconstruction layer is deployed based on anomaly vector-based deconvolution, wherein the contribution to the state shift is used as the deconstruction guide; a second control layer is deployed based on interference simulation and minimum root cause set localization based on root cause parameters; the AI ​​agent is constructed by cascading the first deconstruction layer and the second control layer, and an interaction channel between the AI ​​agent and the SPC module is established.

[0065] In one embodiment, historical abnormal state vector data from the production process is first collected. This data originates from process state vector samples whose state offsets exceed normal thresholds, recorded by the SPC module. Simultaneously, the state offset values ​​at corresponding times are compiled to form a dataset linking abnormal state vectors and state offsets. A 1D deconvolution algorithm is used to deconstruct the abnormal state vectors. The abnormal state vectors are initially treated as mixed signals. The filter length for deconvolution is set based on the dimensions of the process state vectors; for example, a 4-dimensional vector corresponds to a filter of length 4. Through deconvolution, the abnormal state vectors are decomposed into multiple independent dimensional components. Each component corresponds to an original parameter dimension in the process state vector, such as spindle speed or machining pressure.

[0066] Next, the contribution of each decomposed component to the state shift is calculated. Using the ratio formula of component change / total state shift, the proportion of the change magnitude of each dimension component to the total value of the state shift is calculated. The components of each dimension are sorted from high to low according to their contribution. The top 3 dimension components with the highest contribution are retained as candidate source directions. The above deconvolution operation logic, contribution calculation rules and candidate direction selection process are solidified into an algorithm module to complete the deployment of the first deconstruction layer. This deconstruction layer can receive abnormal state vectors and output the candidate root cause parameter direction that contributes the most to the state shift.

[0067] Then, specific process parameters, such as spindle speed, machining pressure, and feed rate, are extracted from the candidate root cause parameters output from the first deconstruction layer and used as the initial set of root cause parameters. A controlled variable method is used to perform parameter interference simulation. First, other parameters are kept constant, and only the value of one root cause parameter is adjusted, for example, the spindle speed is gradually adjusted within ±5% of the normal range. The SPC module is used to simulate the change trend of the process state vector under this adjustment, and the change in distance between the process state vector and the center of the healthy state cluster is recorded. The above simulation operation is repeated for each root cause parameter in the initial set, and effective root cause parameters that, after adjustment, reduce the distance between the process state vector and the center of the healthy cluster by more than 20% are selected.

[0068] Next, a step-by-step elimination method is used to locate the minimum root cause set. After eliminating one parameter at a time from the valid root cause parameters, the combined adjustment effect of the remaining parameters is re-simulated. If the combined adjustment effect after elimination, i.e., the decrease in the magnitude of the process state vector approaching the center of the healthy cluster, is less than 10%, then that parameter does not belong to the minimum root cause set. If the decrease exceeds 30%, the parameter is retained. Finally, 2-3 core parameters are selected to form the minimum root cause set. The above control variable simulation logic, valid parameter selection criteria, and step-by-step elimination process are integrated into an algorithm module to complete the deployment of the second control layer. This layer can receive candidate root cause parameters and output the minimum root cause set and the corresponding interference adjustment direction.

[0069] Finally, the output of the first deconstruction layer and the input of the second control layer are directly connected via a data link, enabling the candidate root cause parameters output by the first deconstruction layer to be automatically transmitted to the second control layer. This forms a complete data flow loop: abnormal state vector input, source tracing by the first deconstruction layer, simulation localization by the second control layer, and minimum root cause set output. These two cascaded layers constitute the core functional modules of the AI ​​agent. An interaction channel between the AI ​​agent and the SPC module is built using the commonly used industrial OPC UA data interface. Interface parameters are configured so that the SPC module can transmit real-time detected abnormal state vectors and state offset values ​​to the AI ​​agent. Simultaneously, the AI ​​agent can transmit the minimum root cause set and parameter adjustment suggestions output by the second control layer to the SPC module, achieving bidirectional data interaction between the two. This ensures that the AI ​​agent can perform root cause analysis based on real-time data from the SPC module, and that the analysis results can be fed back to the SPC module to support subsequent control.

[0070] By constructing the first deconvolution layer through deconvolution source tracing, constructing the second control layer through control variable simulation and stepwise elimination, cascading layers, and building a standardized data interface, an AI intelligent agent with the ability to trace the root cause of anomalies and simulate control was constructed. Effective data interaction between the agent and the SPC module was established, providing intelligent support for alarm management and control drive in subsequent production control.

[0071] Furthermore, the method provided in this application embodiment includes:

[0072] The SPC analysis results are identified. If an abnormal state vector exists, the AI ​​agent is activated. The abnormal state vector is deconvolved through a first deconstruction layer to determine a first root cause parameter, which belongs to the first operating condition control data. The first root cause parameter is imported into a second control layer, and parameter interference simulation is performed. The production control data is determined by judging whether the process state vector after interference is pulled back to the center of the healthy state cluster, wherein the minimum adjustment cost under iterative interference simulation is used as the basis for determination.

[0073] Optionally, the state vector data in the SPC analysis results is first identified using a preset threshold comparison method: based on historical normal production data, a normal distance threshold between the state vector and the center of the healthy state cluster is set. The state vector in the SPC analysis results is then compared with this threshold. If the state vector exceeds the normal distance threshold, an abnormal state vector is determined to exist. At this point, the activation command of the AI ​​agent is triggered, and the core functional modules of the AI ​​agent are started through a pre-configured program interface, putting the AI ​​agent into working state, ready to receive subsequent data and conduct analysis.

[0074] Next, the identified anomalous state vectors are input into the first deconstruction layer of the AI ​​agent. A 1D deconvolution algorithm is used to process the anomalous state vectors: first, the anomalous state vectors are decomposed into multiple underlying data sources in the production system. These data sources all belong to the first operating condition control data, such as parameter data corresponding to spindle speed, machining pressure, and feed rate. Then, the contribution of each underlying data source to the state offset is calculated. Based on the formula of parameter change of a single data source ÷ total state offset change, the contribution value of each data source is obtained. The contributions of all data sources are sorted, and the 1-2 underlying data sources with the highest contributions are selected. Their corresponding parameters are determined as the first root cause parameters.

[0075] Subsequently, the first root cause parameter is imported into the second control layer of the AI ​​agent, and parameter interference simulation is performed using the controlled variable method: other operating parameters are kept constant, and only the first root cause parameter is virtually adjusted. The adjustment range is within the allowable range of the production process, such as ±1% or ±2% increments. The SPC module is called to simulate the change trend of the process state vector after each adjustment, and it is observed whether the process state vector moves closer to the center of the healthy state cluster. If a single simulation does not bring the process state vector closer to the center of the healthy state cluster, the parameter range is adjusted again and the simulation is repeated until an adjustment scheme that can pull the process state vector back to the center of the healthy state cluster is found. In this process, the adjustment cost of each effective adjustment scheme is calculated, and the adjustment range of the parameter is used as the cost measurement standard. The smaller the adjustment range, the lower the cost. Through multiple iterative interference simulations, the adjustment scheme with the minimum adjustment cost that can bring the process state vector back to the healthy state is selected, and the parameter adjustment value in this scheme is determined as the production control data.

[0076] Furthermore, the method provided in this application embodiment includes:

[0077] The SPC analysis results and production control data are mapped to the control terminal to determine multiple mapping groups; for the multiple mapping groups, multiple instruction groups are generated, wherein each instruction group contains early warning instructions and production control instructions based on the minimum control unit; the multiple instruction groups are issued in a multi-threaded manner to execute alarm management and production control driving.

[0078] In one embodiment, the core information of the SPC analysis results and production control data is first clarified: the SPC analysis results include quality risk levels, such as warning levels, attention levels, and quality and safety levels, as well as the corresponding abnormal associated control terminal directions; the production control data includes specific parameter adjustment values ​​and adjustment object identifiers. A control terminal identifier matching method is used for mapping to control terminals: first, the unique identifiers of all control terminals in the production system are identified, and a correspondence rule between control terminal identifiers and associated parameter types is established. Then, the identifiers of abnormal associated control terminals in the SPC analysis results and the identifiers of adjustment objects in the production control data are extracted respectively. SPC analysis result fragments with consistent identifiers are bound to production control data fragments to form a single mapping group of control terminal identifier + risk level + parameter adjustment value. This operation is repeated until all associated data is matched, ultimately determining multiple mapping groups.

[0079] Next, for each mapping group, the corresponding minimum control unit is extracted. The minimum control unit is an independent control module that cannot be further divided within the production system, ensuring that each mapping group corresponds to 1-2 minimum control units. Standardized instruction generation templates are used to generate instruction groups: the early warning instruction template includes: control unit ID + anomaly type + early warning message; the production control instruction template includes: control unit ID + target parameter value + execution time limit. The control terminal identifier in the mapping group is converted into the minimum control unit ID, the risk level is converted into anomaly type and message, and the parameter adjustment value is filled into the target parameter value field. These are combined according to the templates to form a single instruction group containing one early warning instruction and one production control instruction. Multiple instruction groups are generated for all mapping groups.

[0080] Finally, the commonly used industrial multi-threaded communication protocol OPC UA is employed to issue multiple command groups. First, in the command issuance system of the control terminal, an independent communication thread is assigned to each command group. Thread priorities are set according to the risk level of the command group; for example, warning-level command groups have higher priority than attention-level groups, and emergency control command groups have higher priority than routine control groups. All threads are simultaneously started, enabling each command group to send data to the communication port of the target minimum control unit through its corresponding thread. During the issuance process, the communication status of each thread is monitored in real time. If a thread indicates that the command has not been received, a retransmission mechanism is triggered, and retransmissions are made up to three times until all minimum control units report successful command reception. Each control unit activates alarm devices (such as audible and visual alarms) based on the received warning command and automatically adjusts parameters according to production control commands, ultimately completing the execution of alarm management and production control.

[0081] In summary, the SPC quality monitoring method based on AI intelligent recognition provided in this application has the following technical effects:

[0082] This application collects real-time production data streams, separates the first operating condition control data from the second operating condition disturbance data, constructs a mass field force line and a cascaded deconstruction layer-control layer AI intelligent agent, analyzes state offset and disturbance damping to obtain SPC analysis results, and after the AI ​​intelligent agent is activated, traces the root cause parameters and performs interference simulation of execution parameters. Through multi-threaded instruction groups, it realizes AI intelligent identification SPC quality monitoring, improves the accuracy of production quality control, and achieves the technical effects of comprehensive real-time monitoring, early warning, accurate root cause location and efficient control of industrial production quality, thereby improving the accuracy and efficiency of quality control.

[0083] Example 2, as Figure 2 As shown, based on the same inventive concept as in Embodiment 1 above, this application provides an SPC quality monitoring system based on AI intelligent recognition, the system comprising:

[0084] SPC analysis result acquisition module 1 is used to read the real-time data stream in the production control process, perform two-end process control analysis based on the SPC module, and obtain SPC analysis results. The two-end process control analysis includes process state vector runaway prediction of the working condition control part and mass field force line propagation of the working condition disturbance part.

[0085] The production control data acquisition module 2 is used by the AI ​​agent to perform deconvolution source tracing and interference simulation positioning on the abnormal state vector through data interaction with the SPC module, and to determine the production control data.

[0086] The production control execution module 3 is used to execute alarm management and production control drive in the production control process based on the SPC analysis results and production control data.

[0087] Furthermore, the SPC analysis result acquisition module 1 is used to perform the following steps:

[0088] The SPC module includes a process control branch and a process disturbance branch deployed in parallel. Based on the process control branch, the first operating condition control data in the real-time data stream is analyzed to determine a first SPC result. Based on the process disturbance branch, the second operating condition disturbance data in the real-time data stream is analyzed to determine a second SPC result. The first SPC result and the second SPC result are fused together to obtain the SPC analysis result.

[0089] Furthermore, the SPC analysis result acquisition module 1 is used to perform the following steps:

[0090] Based on the real-time data stream, first operating condition control data is determined; by performing multi-source heterogeneous fusion on the operating condition control data, a process state vector is determined, wherein the process state vector represents the comprehensive state of production control at each moment; a vector space is constructed and a health state cluster center is initialized in real time, wherein the health state cluster center moves with the operating conditions at different times; based on the process state vector and the health state cluster center, a first SPC result is determined.

[0091] Furthermore, the SPC analysis result acquisition module 1 is used to perform the following steps:

[0092] The geodesic distance between the process state vector and the center of the healthy state cluster is calculated as the state offset; in the vector space, the short-time evolution path is determined by analyzing the trajectory momentum of the process state vector; based on the short-time evolution path and the state offset, the runaway attractor is predicted as the first SPC result, wherein the runaway attractor is defined by the vector region and defect mode located in the vector space.

[0093] Furthermore, the SPC analysis result acquisition module 1 is used to perform the following steps:

[0094] Based on the real-time data stream, the disturbance data for the second operating condition is determined; a mass field force line is constructed, and the disturbance data for the second operating condition is located within the mass field force line. The disturbance damping capacity is determined by performing disturbance propagation analysis, which serves as the second SPC result; wherein the analysis is guided by the propagation and amplification of the disturbance along the mass field force line.

[0095] Furthermore, the SPC analysis result acquisition module 1 is used to perform the following steps:

[0096] For each entity in the production system, a mass potential energy value is assigned. The mass potential energy is defined by the entity's contribution to product quality and the risk of damage. The mass potential energy value changes with different operating conditions at different times. Based on the mass potential energy value, a transmission path based on quality influence between entities is constructed to form a mass field force line.

[0097] Furthermore, the production control data acquisition module 2 is used to perform the following steps:

[0098] A first deconvolutional deconstruction layer is deployed based on anomaly vector-based deconvolution, wherein the contribution to the state shift is used as the deconstruction guide; a second control layer is deployed based on interference simulation and minimum root cause set localization based on root cause parameters; the AI ​​agent is constructed by cascading the first deconstruction layer and the second control layer, and an interaction channel between the AI ​​agent and the SPC module is established.

[0099] Furthermore, the production control data acquisition module 2 is used to perform the following steps:

[0100] The SPC analysis results are identified. If an abnormal state vector exists, the AI ​​agent is activated. The abnormal state vector is deconvolved through a first deconstruction layer to determine a first root cause parameter, which belongs to the first operating condition control data. The first root cause parameter is imported into a second control layer, and parameter interference simulation is performed. The production control data is determined by judging whether the process state vector after interference is pulled back to the center of the healthy state cluster, wherein the minimum adjustment cost under iterative interference simulation is used as the basis for determination.

[0101] Furthermore, the production control execution module 3 is used to perform the following steps:

[0102] The SPC analysis results and production control data are mapped to the control terminal to determine multiple mapping groups; for the multiple mapping groups, multiple instruction groups are generated, wherein each instruction group contains early warning instructions and production control instructions based on the minimum control unit; the multiple instruction groups are issued in a multi-threaded manner to execute alarm management and production control driving.

[0103] The AI-based intelligent recognition SPC quality monitoring system provided in this embodiment of the invention can execute the AI-based intelligent recognition SPC quality monitoring method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0104] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this invention.

[0105] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application. In some cases, the actions or steps described in this application can be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. An SPC quality monitoring method based on AI intelligent recognition, characterized in that, The method includes: Read the real-time data stream of the production control process, perform two-end process control analysis based on the SPC module, and obtain the SPC analysis results. The two-end process control analysis includes the process state vector runaway prediction of the working condition control part and the mass field force line propagation of the working condition disturbance part. The AI ​​agent interacts with the SPC module to perform deconvolution source tracing and interferometric simulation to locate abnormal vectors and determine production control data. Based on the SPC analysis results and production control data, alarm management and production control drive are executed during the production control process; The SPC module includes process control branches and process disturbance branches deployed in parallel. Based on the process control branch, the first operating condition control data in the real-time data stream is analyzed to determine the first SPC result; Based on the process disturbance branch, the second operating condition disturbance data in the real-time data stream is analyzed to determine the second SPC result; The SPC analysis result is obtained by fusing the first SPC result with the second SPC result. Determine the first SPC result, including: Based on the real-time data stream, determine the control data for the first operating condition; By performing multi-source heterogeneous fusion on the aforementioned operating condition control data, the process state vector is determined, wherein the process state vector represents the overall production control state at each moment; A vector space is constructed and a health status cluster center is initialized in real time, wherein the health status cluster center moves with different working conditions at different times; The first SPC result is determined based on the process state vector and the health state cluster center; Based on the process state vector and the health state cluster center, the first SPC result is determined, including: The state offset is calculated by measuring the geodesic distance between the process state vector and the center of the healthy state cluster. In the vector space, the short-time evolution path is determined by analyzing the trajectory momentum of the process state vector. Based on the short-time evolution path and the state shift, a runaway attractor is predicted as the first SPC result, wherein the runaway attractor is defined by the vector region and defect mode located in the vector space.

2. The SPC quality monitoring method based on AI intelligent recognition as described in claim 1, characterized in that, Determine the second SPC result, including: Based on the real-time data stream, determine the disturbance data for the second operating condition; Construct mass field force lines, locate the second working condition disturbance data in the mass field force lines, and determine the disturbance damping capacity by performing disturbance propagation analysis, which serves as the second SPC result; The analysis is guided by the propagation and amplification of disturbances along the mass field force lines.

3. The SPC quality monitoring method based on AI intelligent recognition as described in claim 1, characterized in that, Constructing mass field force lines includes: For each entity in the production system, a mass potential energy value is assigned. The mass potential energy is defined by the entity's ability to contribute to product quality and the risk of damage. The mass potential energy value changes with different operating conditions at different times. Based on the mass potential energy value, a transmission path based on mass influence is constructed between entities to form mass field force lines.

4. The SPC quality monitoring method based on AI intelligent recognition as described in claim 1, characterized in that, The construction of the AI ​​agent before it interacts with the SPC module includes: A first deconstruction layer is deployed using deconvolutional source tracing deconstruction based on anomaly vectors, wherein the contribution to the state shift is used as the source tracing deconstruction guide; A second regulatory layer is deployed based on interference simulation and minimum root cause set localization using source root cause parameters; By cascading the first deconstruction layer and the second control layer, the AI ​​agent is formed, and an interaction channel between the AI ​​agent and the SPC module is established.

5. The SPC quality monitoring method based on AI intelligent recognition as described in claim 4, characterized in that, Before implementing alarm management and production control in the production control process, the following should be included: Identify the SPC analysis results; if an anomalous state vector is found, activate the AI ​​agent. By performing deconvolution operation on the abnormal state vector through the first deconstruction layer, the first source root cause parameter is determined, wherein the first source root cause parameter belongs to the first operating condition control data; The first root cause parameter is imported into the second control layer, and parameter interference simulation is performed. By determining whether the process state vector after interference is pulled back to the center of the healthy state cluster, the production control data is determined, with the minimum adjustment cost under iterative interference simulation as the basis for determination.

6. The SPC quality monitoring method based on AI intelligent recognition as described in claim 5, characterized in that, The process of implementing alarm management and production control includes: The SPC analysis results and production control data are mapped to the same control terminal to determine multiple mapping groups; For the multiple mapping groups, multiple instruction groups are generated, wherein each instruction group includes early warning instructions and production control instructions based on the minimum control unit; The multiple instruction groups are issued in a multi-threaded manner to execute alarm management and production control.

7. An SPC quality monitoring system based on AI intelligent recognition, characterized in that: The system is used to implement the SPC quality monitoring method based on AI intelligent recognition as described in any one of claims 1-6, the system comprising: The SPC analysis result acquisition module is used to read the real-time data stream in the production control process, perform two-way process control analysis based on the SPC module, and obtain the SPC analysis results. The two-way process control analysis includes the process state vector runaway prediction of the working condition control part and the mass field force line propagation of the working condition disturbance part. The production control data acquisition module is used by the AI ​​agent to perform deconvolution source tracing and interference simulation positioning on the abnormal state vector through data interaction with the SPC module, and to determine the production control data. The production control execution module is used to execute alarm management and production control drive in the production control process based on the SPC analysis results and production control data.

Images