A traceable management method and system for production process control based on AI

By collecting and structuring multi-source data in real time, and using AI models to identify anomalies and build indexes, the problem of low efficiency in product quality traceability in existing technologies has been solved, and efficient backtracking and complete restoration of the production process have been achieved.

CN123117744APending Publication Date: 2026-11-17曾嘉凯
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202610990605.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-03
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In existing technologies, tracing product quality anomalies requires manual data retrieval across multiple systems, resulting in low efficiency and the potential for missing key information, making it impossible to quickly and completely reproduce the production process.

Method used

The system uses AI models to collect multi-source heterogeneous data in real time, converts it into structured data streams, identifies abnormal events, forms abnormal event data packages, and builds multi-dimensional indexes to support rapid retrieval and automatic triggering of production control actions, generating retrospective views and analysis reports.

Benefits of technology

It enables efficient backtracking and complete restoration of the production process of any product, improves data retrieval efficiency, ensures that no key information is missed, and supports rapid response and accurate production control.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123117744A_ABST
    Figure CN123117744A_ABST
Patent Text Reader

Abstract

This invention relates to the field of production management technology, specifically to an AI-based method and system for traceable production process control. The system collects multi-source heterogeneous data and converts it into a structured data stream in a unified format. It uses an AI model to identify anomalies in the data stream and issues trigger signals when key anomalies are detected. In response to the trigger signals, it collects and stores multi-dimensional data within the window before and after the anomaly, using the current product as the main index, forming an anomaly event data package. Based on this data package, it constructs a multi-dimensional hierarchical index, supporting rapid retrieval by different dimensions. It automatically triggers control actions based on the anomaly type and severity level, and records the action information in the data package. In response to a traceability query request, it retrieves the data package using the index and generates a traceability view and anomaly analysis report. This supports efficient traceability and complete reconstruction of the production process for any product.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of production management technology, and in particular to an AI-based production process control and traceable management method and system. Background Technology

[0002] In manufacturing, when product defects occur, it's necessary to trace the product's status throughout the entire production process, including each step, equipment operating parameters, inspection images, environmental data, and operational records. Currently, factories typically rely on Manufacturing Execution Systems (MES) to record production batch information and some process parameters, while using automated optical inspection (AOI) equipment to capture product images at specific stages and provide judgment results. These systems operate independently, accumulating a large amount of production process data.

[0003] However, existing solutions have a significant problem when tracing quality anomalies: when it is necessary to reconstruct the complete production process of a specific product, key information such as equipment operation data, inspection images, and operational records are distributed across multiple independent systems, lacking a spatiotemporal correlation and organizational mechanism centered on a single product. This forces engineers to manually search and align data across multiple systems when investigating quality causes, a cumbersome and inefficient process that is highly susceptible to missing crucial clues due to data gaps, making it impossible to quickly and completely reconstruct the entire production process of a single product.

[0004] Therefore, there is an urgent need for an AI-based production process control and traceability management method to support efficient backtracking and complete restoration of the production process of any product. Summary of the Invention

[0005] The purpose of this invention is to provide an AI-based production process control and traceability management method and system that supports efficient backtracking and complete restoration of the production process of any product.

[0006] To achieve the above objectives, the present invention employs an AI-based traceable management method for production process control, comprising the following steps: Collect multi-source heterogeneous data in real time during the production process and convert it into a structured production data stream in a unified format; The structured production data stream is anomaly identified using a pre-built AI model, and a trigger signal is issued when a key anomaly event is identified. In response to the trigger signal, using the current product as the main index, multi-dimensional on-site data within a preset window before and after the occurrence of the abnormal event is collected and solidified to form an abnormal event data packet. A multi-dimensional hierarchical index is constructed based on the abnormal event data packets to support fast retrieval in different dimensions; Based on the type and severity level of the critical abnormal events, corresponding production control actions are automatically triggered, and the triggering information of the control actions is recorded in the abnormal event data packet; In response to a backtracking query request, the hierarchical index is used to retrieve relevant abnormal event data packets, generating a production process backtracking view and an anomaly analysis report.

[0007] Among these steps, the real-time acquisition of multi-source heterogeneous data during the production process and its conversion into a structured production data stream in a unified format includes: The system connects to production equipment, testing equipment, environmental sensors, and operating terminals of different brands and models through a data gateway, and performs protocol parsing and field mapping on the raw data they output. The parsed data is used to generate structured data records carrying timestamps, device identifiers, and product identifiers according to a unified timestamp benchmark and preset field templates, and then aggregated into the structured production data stream in chronological order.

[0008] In the step of using a pre-built AI model to identify anomalies in the structured production data stream and issuing a trigger signal when a key anomaly event is identified: The structured production data stream is input into the time-series processing branch, visual processing branch, and behavior processing branch of the AI ​​model, respectively. The timing processing branch extracts timing anomaly features from continuous process parameters; the vision processing branch extracts defect semantic features from product inspection images; and the behavior processing branch extracts behavior features from operation logs. The features output from the three branches are fused across modalities to output the category and trigger confidence of key abnormal events. When the confidence exceeds a preset threshold, the trigger signal is issued.

[0009] The training samples for the AI ​​model are obtained in the following way: Extract the time period where quality anomalies have been confirmed from historical production data, and extract the process parameter sequence, inspection images and operation logs within that time period; after manually labeling the anomaly event type, construct a positive and negative sample set for model training.

[0010] In the step of responding to the trigger signal, using the current product as the main index, collecting and storing multi-dimensional on-site data within a preset window before and after the occurrence of the abnormal event, and forming an abnormal event data packet: Based on the generation time of the trigger signal, the length of the backtracking time window for data acquisition is determined; Within the backtracking time window, extract the equipment operating parameters, detection images, environmental data, and operation records associated with the current product; The extracted multidimensional data is aligned to a unified timeline by timestamps, and mapped to the corresponding production process nodes according to the process of the generating equipment of each data, forming an abnormal event data packet organized with a four-dimensional association structure of time, process, parameter, and image. The abnormal event data packets are persistently stored after being correlated with the intermediate inference results and judgment confidence of the AI ​​model.

[0011] Specifically, aligning the extracted multidimensional data to a unified timeline according to timestamps involves: Using the product identification code as the main thread, data records from different devices are corrected to a unified clock reference based on their respective timestamps; and each data record is associated with the corresponding process identifier according to the process position of each device, thereby establishing a time-series data profile of the product at each process node in the entire production process.

[0012] In the step of constructing a multi-dimensional hierarchical index based on the abnormal event data packet to support fast retrieval across different dimensions: For each abnormal event data packet, a product-level index based on product identifier, a batch-level index based on production batch number, a process-level index based on process number, an equipment-level index based on equipment number, and a time-level index based on time are established to support fast retrieval by any dimension of product, batch, process, equipment, and time.

[0013] The step of automatically triggering corresponding production control actions based on the type and severity level of the critical abnormal event, and recording the triggering information of the control actions into the abnormal event data packet: Based on the type and severity level of the critical abnormal events, the corresponding control actions are matched from the preset control strategy library; The matched control actions are sent to the production line for execution. The control actions include at least one of automatic shutdown, defective product rejection, operator alarm, and process parameter limit locking. The trigger time, action type, and execution result of the control action are collected, and the above information is written back to the corresponding abnormal event data packet to form a closed-loop control record.

[0014] In the step of responding to a backtracking query request, retrieving relevant abnormal event data packets using the hierarchical index, and generating a production process backtracking view and anomaly analysis report: Receive user-inputted backtracking query requests and retrieve associated abnormal event data packets in parallel based on the hierarchical index; The multidimensional data in the retrieved abnormal event data package is displayed in a linked manner according to the time axis to generate a production process backtracking view; The pre-trained root cause analysis model is invoked, and the equipment parameter change rate, defect feature change trend, environmental data fluctuation and operation event sequence in the abnormal event data packet are used as input to generate an abnormal analysis report through causal reasoning.

[0015] This invention also provides an AI-based production process control and traceability management system, including a data integration module, an AI anomaly identification module, a snapshot fixing module, an index construction module, a control linkage module, and a traceability analysis module; wherein: The data integration module is used to collect multi-source heterogeneous data in the production process in real time and convert it into a structured production data stream in a unified format. The AI ​​anomaly detection module is used to identify anomalies in the structured production data stream using a pre-built AI model, and to issue a trigger signal when a key anomaly event is detected. The snapshot solidification module is used to respond to the trigger signal, using the current product as the main index, to collect and solidify multi-dimensional on-site data within a preset window before and after the occurrence of the abnormal event, forming an abnormal event data packet; The index building module is used to build a multi-dimensional hierarchical index based on the abnormal event data packet to support fast retrieval in different dimensions; The control and linkage module is used to automatically trigger corresponding production control actions based on the type and severity level of the key abnormal events, and record the triggering information of the control actions into the abnormal event data packet; The backtracking analysis module is used to respond to backtracking query requests, retrieve relevant abnormal event data packets using the hierarchical index, and generate a production process backtracking view and anomaly analysis report.

[0016] This invention discloses an AI-based production process control and traceable management method and system. Through a data integration module, an AI anomaly identification module, a snapshot solidification module, an index construction module, a control linkage module, and a backtracking analysis module, the following steps are implemented: Real-time collection of multi-source heterogeneous data during the production process, converting it into a structured production data stream in a unified format; using a pre-built AI model to identify anomalies in the structured production data stream, and issuing trigger signals when key anomalies are identified; responding to the trigger signals, using the current product as the main index, collecting and solidifying multi-dimensional on-site data within a preset window before and after the anomaly event, forming an anomaly event data package; constructing a multi-dimensional hierarchical index based on the anomaly event data package to support rapid retrieval from different dimensions; automatically triggering corresponding production control actions according to the type and severity level of the key anomaly event, and recording the trigger information of the control actions in the anomaly event data package; responding to a backtracking query request, retrieving relevant anomaly event data packages using the hierarchical index, and generating a production process backtracking view and anomaly analysis report. By employing the above method, efficient backtracking and complete reconstruction of the production process of any product can be achieved. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are 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.

[0018] Figure 1 This is a flowchart of the steps of the AI-based production process control and traceable management method of the present invention.

[0019] Figure 2 This is a flowchart of steps S100 of the present invention.

[0020] Figure 3 This is a flowchart of steps S200 of the present invention.

[0021] Figure 4 This is a flowchart of steps S300 of the present invention.

[0022] Figure 5 This is a flowchart of steps S500 of the present invention.

[0023] Figure 6 This is a flowchart of steps S600 of the present invention.

[0024] Figure 7 This is a schematic diagram of the structural principle of the AI-based production process control and traceability management system of the present invention.

[0025] Figure 8 This is a schematic diagram of the electronic device of the present invention.

[0026] 1-Data integration module, 2-AI anomaly identification module, 3-Snapshot solidification module, 4-Index construction module, 5-Management and control linkage module, 6-Backtracking analysis module. Detailed Implementation

[0027] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application.

[0028] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0029] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0030] Please see Figures 1-6 This invention provides an AI-based traceable management method for production process control, comprising the following steps: S100: Collects multi-source heterogeneous data in real time during the production process and converts it into a structured production data stream in a unified format; In this embodiment, various equipment data, sensor data, and detection data from different sources and in different formats in the workshop are collected in real time and uniformly converted into a data stream in a standard format.

[0031] The specific process of real-time acquisition of multi-source heterogeneous data during the production process and conversion into a structured production data stream in a unified format is as follows: S101: Connects to production equipment, testing equipment, environmental sensors and operating terminals of different brands and models through a data gateway, and performs protocol parsing and field mapping on the raw data output by them; In this embodiment, a data gateway is used to connect production equipment, inspection cameras, temperature and humidity sensors, operating computers, etc. from different manufacturers and models, and translate the diverse raw data they output to extract key information.

[0032] S102: The parsed data is used to generate structured data records carrying timestamps, device identifiers and product identifiers according to a unified timestamp benchmark and preset field templates, and then aggregated into the structured production data stream in chronological order.

[0033] In this embodiment, the translated data is timestamped according to the same clock standard, and a unified table template (which must include time, which device, and which product) is applied to generate a series of neat data records, which are then arranged into a pipeline according to time sequence.

[0034] In the above process: The data gateway is deployed on edge computing nodes in the workshop, supporting the parsing of various industrial communication protocols such as OPC UA, ModbusTCP, MQTT, and HTTP. Protocol adaptation adopts a pluggable adapter architecture; when adding a new device type, only the corresponding adapter plugin needs to be loaded, without modifying the gateway's main logic. Field mapping is completed through a pre-configured mapping table, mapping the register addresses or JSON field paths of different devices to unified data model fields. For time synchronization, the gateway maintains synchronization with a standard time server via the NTP protocol, ensuring that the clock base of all collected data is consistent. The standardized data is published to a distributed message queue using a streaming processing framework for real-time consumption by downstream AI recognition engines.

[0035] S200: Use a pre-built AI model to identify anomalies in the structured production data stream and issue a trigger signal when a key anomaly event is identified; In this implementation, a pre-trained AI model is used to monitor the data stream and automatically determine if there is an anomaly. Once an anomaly is detected, an alarm signal is immediately issued.

[0036] The specific process of using a pre-built AI model to identify anomalies in the structured production data stream and issuing a trigger signal when a key anomaly event is identified is as follows: S201: Input the structured production data stream into the time-series processing branch, visual processing branch, and behavior processing branch of the AI ​​model respectively; In this implementation, a standard data stream is simultaneously fed to three branches of the AI ​​model: one to view continuous parameter curves, one to view quality inspection images, and one to view operation records.

[0037] S202: The timing processing branch extracts timing anomaly features from continuous process parameters; the vision processing branch extracts defect semantic features from product inspection images; the behavior processing branch extracts behavior features from operation logs; In this implementation, the first branch focuses on continuously changing parameters such as temperature, pressure, and speed, identifying abnormal peaks, abrupt changes, or drifts in the curves. The second branch examines product photos taken by quality inspectors to understand the types of defects in the images (e.g., scratches, dirt, cracks). The third branch reads operator action records to determine if there are any non-standard operating behaviors.

[0038] S203: Perform cross-modal fusion on the features output from the above three branches to output the category and trigger confidence of the key abnormal event. When the confidence exceeds the preset threshold, the trigger signal is issued.

[0039] In this implementation, the information seen by the three branches is combined to give a final judgment: whether it is abnormal, what type of abnormality it is, how confident is it, and when the confidence exceeds a set threshold, an alarm signal is issued.

[0040] The training samples for the AI ​​model are obtained in the following way: the time period in which the quality anomaly has been confirmed is extracted from historical production data, and the process parameter sequence, detection images and operation logs within the time period are extracted; after the abnormal event type is manually labeled, a positive and negative sample set is constructed for model training. In this implementation, the time segments where quality problems have been confirmed are identified from past production records. Parameter curves, photos, and operation records from these segments are extracted. Experienced engineers then label these segments (e.g., "caused by equipment failure," "caused by operational error"), creating a training exercise set for the model to learn from repeatedly.

[0041] In the above process, the temporal processing branch adopts a Transformer encoder structure, utilizing a self-attention mechanism to capture long-range dependencies and anomaly patterns in the process parameter sequence; the visual processing branch uses a convolutional neural network pre-trained on an industrial defect detection dataset to output feature vectors of defect type and location; and the behavior processing branch uses a text encoder to semantically encode the operation logs. The output features of the three branches are fused across modalities through an attention weighting mechanism, and the fused features are fed into a fully connected classification layer to output the anomaly event category and confidence level. The confidence level threshold is set according to the balance requirement of false positive rate and false negative rate in actual production scenarios, with a typical value of 0.85. Training samples are extracted from historical data using a sliding time window, with the window centered on the moment of anomaly event confirmation and taking preset time periods before and after to ensure coverage of the entire anomaly evolution process. After the model is trained, it is deployed on edge computing nodes to achieve low-latency real-time inference.

[0042] S300: In response to the trigger signal, using the current product as the main index, collect and solidify multi-dimensional field data within a preset window before and after the occurrence of the abnormal event, forming an abnormal event data packet; In this embodiment, after receiving an alarm signal, the product currently being processed is taken as the main line, and all relevant on-site data from before and after the anomaly is captured, fixed, and saved, forming an evidence package.

[0043] In response to the trigger signal, using the current product as the primary index, the specific process of collecting and storing multi-dimensional on-site data within a preset window before and after the occurrence of the abnormal event, and forming an abnormal event data packet, is as follows: S301: Determine the length of the backtracking time window for data acquisition based on the generation time of the trigger signal; In this implementation, a data collection time range is defined by specifying how long to trace back and how long to observe forward, centered on the time point when the AI ​​issues an alarm.

[0044] S302: Within the backtracking time window, extract the equipment operating parameters, detection images, environmental data, and operation records associated with the current product; In this embodiment, within a defined time frame, all data related to the current product is retrieved, including equipment parameters, quality inspection photos, ambient temperature and humidity, operation records, etc.

[0045] S303: Align the extracted multidimensional data to a unified timeline according to the timestamp, and map it to the corresponding production process node according to the process of the generating equipment of each data, forming an abnormal event data packet organized with a four-dimensional association structure of time, process, parameter and image; In this embodiment, the retrieved data is uniformly calibrated to the same timeline, and each data point is clearly marked to indicate which process it belongs to, ultimately forming a four-dimensional linked archive package with "what time - which process - what parameters - which photo it corresponds to".

[0046] S304: The abnormal event data packet is persistently stored after being associated with the intermediate inference results and judgment confidence of the AI ​​model.

[0047] In this embodiment, the file package is permanently stored together with the intermediate process of AI's reasoning at that time (such as defect feature vectors, heat maps) and the judgment given by AI.

[0048] Specifically, aligning the extracted multidimensional data to a unified timeline according to timestamps involves: Using the product identification code as the main thread, data records from different devices are corrected to a unified clock reference based on their respective timestamps; and each data record is associated with the corresponding process identifier according to the process position of each device, thereby establishing a time-series data profile of the product at each process node in the entire production process.

[0049] In this embodiment, using the product number as a guide, the data recorded by each device is collected, and their respective times are first calibrated to the same standard clock. Then, based on which process each device is in, each data entry is linked to the corresponding process. This allows us to draw a complete cross-sectional diagram of the product's state at each process point in the entire production process.

[0050] In the above process: the backtracking time window length is configured differently according to the type of abnormal event. For process parameter-related anomalies, the window is 3 minutes before and after; for defect detection-related anomalies, it is 1 minute before and after; and for equipment downtime-related anomalies, it is 5 minutes before and after. During multidimensional data extraction, associated data is retrieved in batches from the sliding window cache of the message queue using the product identification code. Simultaneously, corresponding detection images and inference intermediate results (defect heatmaps, feature vectors) are requested from the AOI system and AI quality inspection system. Timestamp alignment uses linear interpolation to unify data from different sampling frequencies to the same time granularity, and the association of process nodes is completed through a static mapping table between equipment and processes. The four-dimensional associated data package adopts a hybrid storage architecture of time-series database and object storage. Metadata and structured parameters are stored in the time-series database, while image files and model inference intermediate results are stored in object storage. The storage path is recorded in the metadata to achieve efficient retrieval.

[0051] S400: Construct a multi-dimensional hierarchical index based on the abnormal event data packet to support fast retrieval in different dimensions; In this embodiment, a hierarchical directory is established for these evidence packages, which can be quickly searched from multiple perspectives (product, batch, equipment, process, time).

[0052] The specific process of constructing a multi-dimensional hierarchical index based on the aforementioned abnormal event data packets to support fast retrieval across different dimensions is as follows: For each abnormal event data packet, a product-level index based on product identifier, a batch-level index based on production batch number, a process-level index based on process number, an equipment-level index based on equipment number, and a time-level index based on time are established to support fast retrieval by any dimension of product, batch, process, equipment, and time.

[0053] In this implementation, an index is created by "Product Identifier" to quickly locate the file when searching for a specific product. An index is created by "Production Batch" to quickly locate the entire batch of products. An index is created by "Process Number" to quickly filter for anomalies that occurred in the third process step. An index is created by "Equipment Number" to quickly retrieve the historical anomaly records of a specific piece of equipment. An index is created by "Time Period" to support quick searches by date, shift, etc. These five dimensions of indexing ensure that searches can be performed quickly from any perspective.

[0054] In the above process: product-level and batch-level indexes are stored in a relational database using a B+ tree structure, supporting exact matching and prefix fuzzy queries; process-level and equipment-level indexes use an inverted index structure, supporting fast aggregation by category; time-level indexes are stored in a partitioned table by date range, combined with a database partitioning and pruning mechanism to achieve efficient time range scanning. All dimension indexes are synchronously written when abnormal event data packets are persisted, ensuring consistency between index data and the data packets themselves. During queries, the optimal index combination is automatically selected based on the user-input query conditions, achieving millisecond-level retrieval response.

[0055] S500: Based on the type and severity level of the critical abnormal event, automatically trigger the corresponding production control action and record the triggering information of the control action in the abnormal event data packet; In this implementation, the system automatically decides whether to shut down, reject defective products, issue an alarm, or lock parameters based on the type and severity of the anomaly. It also records "when and what was done, and whether it was successful" in the evidence package to form a complete handling record.

[0056] The specific process of automatically triggering corresponding production control actions based on the type and severity level of the critical abnormal events, and recording the triggering information of the control actions into the abnormal event data packet, is as follows: S501: Based on the type and severity level of the critical abnormal event, match the corresponding control action from the preset control strategy library; In this embodiment, the most appropriate response is automatically selected from a pre-defined pool of solutions based on the type and severity of the anomaly.

[0057] S502: Send the matched control action to the production line for execution. The control action includes at least one of automatic shutdown, defective product rejection, operator alarm, and process parameter limit locking. In this implementation, the selected instructions are sent to the production line for execution, which may include stopping the machine, kicking out defective products, flashing lights and sounding alarms, or locking parameters to prevent movement.

[0058] S503: Collect the trigger time, action type, and execution result of the control action, and write the above information back to the corresponding abnormal event data packet to form a closed-loop control record.

[0059] In this implementation, the evidence package also records "when the instruction was issued, what the instruction was, and whether it was executed successfully," so that when checking later, even "how the system responded at the time" can be seen in its entirety.

[0060] In the above process, the control strategy library is maintained in the form of rule tables. Each rule defines the mapping relationship between anomaly type, severity level, and control action. Severity levels are divided into three levels: Level 1 (minor) triggers operator alarm prompts; Level 2 (moderate) triggers defective product rejection and process parameter limit locking; Level 3 (severe) triggers automatic shutdown and a full-line alarm. Control actions are sent to the PLC controller for execution via standard industrial communication protocols, and the execution status is confirmed through equipment feedback signals. Action records are appended to the timeline of the anomaly event data packet, using a unified timestamp reference with equipment data, detection images, etc., to ensure the time consistency of control records with other data.

[0061] S600: In response to the backtracking query request, the hierarchical index is used to retrieve the relevant abnormal event data packets, and a production process backtracking view and anomaly analysis report are generated.

[0062] In this implementation, when someone needs to trace back and query, the relevant evidence package can be quickly found through the hierarchical directory, and a report can be generated that allows for a clear replay of the production process and includes an analysis of the causes of the anomalies.

[0063] The specific process of retrieving relevant abnormal event data packets using the hierarchical index in response to a backtracking query request, and generating a production process backtracking view and anomaly analysis report is as follows: S601: Receive a user-inputted backtracking query request and retrieve associated abnormal event data packets in parallel based on the hierarchical index; In this implementation, the system receives user-input query criteria (product number, batch, time period, etc.) and uses a previously created directory to quickly find relevant evidence packages from multiple dimensions.

[0064] S602: Display the multidimensional data in the retrieved abnormal event data package in a linked manner according to the time axis to generate a production process backtracking view; In this embodiment, the data in the found evidence package is displayed in a timeline, with the process on the left, the parameter curve in the middle, the photos on the right, and the abnormal events marked at the bottom, forming a complete playback screen.

[0065] S603: Call the pre-trained root cause analysis model, take the equipment parameter change rate, defect feature change trend, environmental data fluctuation and operation event sequence in the abnormal event data packet as input, and generate an abnormal analysis report through causal reasoning.

[0066] In this implementation, a specialized AI model for finding root causes is called, and the parameter change rate, defect evolution trend, environmental fluctuations, and operation records are fed into it to deduce the causal chain of "A causes B, B causes C" and write it into the analysis report.

[0067] In the above process, backtracking queries support retrieval by single or combined conditions such as product identifier, batch number, time period, process number, and equipment number. The system automatically maps query conditions to the corresponding index dimensions and initiates parallel searches, returning results in a merged and sorted manner by timestamp. The backtracking view adopts a front-end and back-end separated web visualization solution. The front-end uses a time-series chart library and image annotation components to achieve a linked display of parameter curves, quality inspection images, and process status on the same time axis. The root cause analysis model is built on a structural causal model framework, using equipment parameter change rate, defect feature evolution trend, environmental fluctuation sequence, and operation event sequence as observation nodes in the causal graph. The causal inference algorithm estimates the causal effect strength of each node on the occurrence of abnormal events, sorts the root cause candidates and their causal paths according to the effect strength, and generates a structured anomaly analysis report.

[0068] Corresponding to the aforementioned embodiments of the AI-based production process control and traceability management method, this application also provides embodiments of an AI-based production process control and traceability management system.

[0069] Figure 7 This is a block diagram illustrating an AI-based production process control and traceability management system according to an exemplary embodiment. (Refer to...) Figure 7 The system may include a data integration module 1, an AI anomaly identification module 2, a snapshot persistence module 3, an index building module 4, a management and control linkage module 5, and a backtracking analysis module 6; among which: The data integration module 1 is used to collect multi-source heterogeneous data in the production process in real time and convert it into a structured production data stream in a unified format. The AI ​​anomaly identification module 2 is used to identify anomalies in the structured production data stream using a pre-built AI model, and to issue a trigger signal when a key anomaly event is identified. The snapshot solidification module 3 is used to respond to the trigger signal, using the current product as the main index, to collect and solidify multi-dimensional on-site data within a preset window before and after the occurrence of the abnormal event, forming an abnormal event data packet; The index building module 4 is used to build a multi-dimensional hierarchical index based on the abnormal event data packet to support fast retrieval in different dimensions; The control and linkage module 5 is used to automatically trigger corresponding production control actions according to the type and severity level of the key abnormal event, and record the triggering information of the control action into the abnormal event data packet; The backtracking analysis module 6 is used to respond to backtracking query requests, retrieve relevant abnormal event data packets using the hierarchical index, and generate a production process backtracking view and anomaly analysis report.

[0070] In this embodiment, the data integration module 1 collects multi-source heterogeneous data from the production process in real time and converts it into a structured production data stream in a unified format. The AI ​​anomaly identification module 2 uses a pre-built AI model to identify anomalies in the structured production data stream and issues a trigger signal when a key anomaly event is identified. The snapshot solidification module 3, responding to the trigger signal, collects and solidifies multi-dimensional on-site data within a preset window before and after the anomaly event, using the current product as the main index, forming an anomaly event data package. The index construction module 4 constructs a multi-dimensional hierarchical index based on the anomaly event data package to support rapid retrieval from different dimensions. The control and linkage module 5 automatically triggers corresponding production control actions based on the type and severity of the key anomaly event and records the trigger information of the control actions in the anomaly event data package. The backtracking analysis module 6, responding to a backtracking query request, uses the hierarchical index to retrieve relevant anomaly event data packages and generates a production process backtracking view and anomaly analysis report. By employing the above method, efficient backtracking and complete restoration of the production process of any product can be supported.

[0071] Regarding the system in the above embodiments, the specific ways in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0072] For the system embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this application according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0073] Accordingly, this application also provides an electronic device, including: one or more processors; a memory for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the AI-based production process control and traceability management method described above. Figure 8 The diagram shown is a hardware structure diagram of any device with data processing capabilities within an AI-based production process control and traceability management system provided in an embodiment of the present invention. (Except for...) Figure 8 In addition to the processor, memory, and network interface shown, any data processing device in the embodiment may also include other hardware depending on the actual function of the data processing device, which will not be described in detail here.

[0074] Accordingly, this application also provides a computer-readable storage medium storing computer instructions, which, when executed by a processor, implement the aforementioned AI-based production process control and traceability management method. The computer-readable storage medium can be an internal storage unit of any data-processing device as described in any of the foregoing embodiments, such as a hard disk or memory. The computer-readable storage medium can also be an external storage device, such as a plug-in hard disk, smart media card (SMC), SD card, flash card, etc., equipped on the device. Furthermore, the computer-readable storage medium can include both internal storage units of any data-processing device and external storage devices. The computer-readable storage medium is used to store the computer program and other programs and data required by the data-processing device, and can also be used to temporarily store data that has been output or will be output.

[0075] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein.

[0076] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope.

Claims

1. An AI-based production process management traceable management method, characterized by, Includes the following steps: Collect multi-source heterogeneous data in real time during the production process and convert it into a structured production data stream in a unified format; The structured production data stream is anomaly identified using a pre-built AI model, and a trigger signal is issued when a key anomaly event is identified. In response to the trigger signal, using the current product as the main index, multi-dimensional on-site data within a preset window before and after the occurrence of the abnormal event is collected and solidified to form an abnormal event data packet. A multi-dimensional hierarchical index is constructed based on the abnormal event data packets to support fast retrieval in different dimensions; Based on the type and severity level of the critical abnormal events, corresponding production control actions are automatically triggered, and the triggering information of the control actions is recorded in the abnormal event data packet; In response to a backtracking query request, the hierarchical index is used to retrieve relevant abnormal event data packets, generating a production process backtracking view and an anomaly analysis report.

2. The AI-based production process management and control traceable management method of claim 1, wherein, In the step of real-time acquisition of multi-source heterogeneous data during the production process and conversion into a structured production data stream in a unified format: The system connects to production equipment, testing equipment, environmental sensors, and operating terminals of different brands and models through a data gateway, and performs protocol parsing and field mapping on the raw data they output. The parsed data is used to generate structured data records carrying timestamps, device identifiers, and product identifiers according to a unified timestamp benchmark and preset field templates, and then aggregated into the structured production data stream in chronological order. 3.The AI-based production process management and control traceable management method of claim 2, wherein, In the step of using a pre-built AI model to identify anomalies in the structured production data stream and issuing a trigger signal when a key anomaly event is identified: The structured production data stream is input into the time-series processing branch, visual processing branch, and behavior processing branch of the AI ​​model, respectively. The timing processing branch extracts timing anomaly features from continuous process parameters; The visual processing branch extracts defect semantic features from the product inspection images; the behavior processing branch extracts behavior features from the operation logs. The features output from the three branches are fused across modalities to output the category and trigger confidence of key abnormal events. When the confidence exceeds a preset threshold, the trigger signal is issued. 4.The AI-based production process management and control traceable management method of claim 3, wherein, The training samples for the AI ​​model were obtained in the following way: Extract the time period from historical production data where quality anomalies have been confirmed, and extract the process parameter sequence, inspection images, and operation logs within that time period; After manually labeling the types of abnormal events, a set of positive and negative samples is constructed for model training.

5. The AI-based production process management and control traceable management method of claim 4, wherein, In the step of responding to the trigger signal, using the current product as the main index, collecting and storing multi-dimensional on-site data within a preset window before and after the occurrence of the abnormal event, and forming an abnormal event data packet: Based on the generation time of the trigger signal, the length of the backtracking time window for data acquisition is determined; Within the backtracking time window, extract the equipment operating parameters, detection images, environmental data, and operation records associated with the current product; The extracted multidimensional data is aligned to a unified timeline by timestamps, and mapped to the corresponding production process nodes according to the process of the generating equipment of each data, forming an abnormal event data packet organized with a four-dimensional association structure of time, process, parameter, and image. The abnormal event data packets are persistently stored after being correlated with the intermediate inference results and judgment confidence of the AI ​​model.

6. The AI-based production process management and control traceable management method of claim 5, wherein, The step of aligning the extracted multidimensional data to a unified timeline according to timestamps specifically involves: Using the product identification code as the main thread, data records from different devices are corrected to a unified clock reference based on their respective timestamps; and each data record is associated with the corresponding process identifier according to the process position of each device, thereby establishing a time-series data profile of the product at each process node in the entire production process.

7. The AI-based production process management and control traceable management method of claim 6, wherein, In the step of constructing a multi-dimensional hierarchical index based on the abnormal event data packet to support fast retrieval across different dimensions: For each abnormal event data packet, a product-level index based on product identifier, a batch-level index based on production batch number, a process-level index based on process number, an equipment-level index based on equipment number, and a time-level index based on time are established to support fast retrieval by any dimension of product, batch, process, equipment, and time.

8. The AI-based production process management method of claim 7, wherein, The steps include automatically triggering corresponding production control actions based on the type and severity level of the critical abnormal events, and recording the triggering information of the control actions into the abnormal event data packet: Based on the type and severity level of the critical abnormal events, the corresponding control actions are matched from the preset control strategy library; The matched control actions are sent to the production line for execution. The control actions include at least one of automatic shutdown, defective product rejection, operator alarm, and process parameter limit locking. The trigger time, action type, and execution result of the control action are collected, and the above information is written back to the corresponding abnormal event data packet to form a closed-loop control record. 9.The AI-based production process management and control traceable management method of claim 8, wherein, In the step of responding to a backtracking query request, retrieving relevant abnormal event data packets using the hierarchical index, and generating a production process backtracking view and anomaly analysis report: Receive user-inputted backtracking query requests and retrieve associated abnormal event data packets in parallel based on the hierarchical index; The multidimensional data in the retrieved abnormal event data package is displayed in a linked manner according to the time axis to generate a production process backtracking view; The pre-trained root cause analysis model is invoked, and the equipment parameter change rate, defect feature change trend, environmental data fluctuation and operation event sequence in the abnormal event data packet are used as input to generate an abnormal analysis report through causal reasoning.

10. An AI-based production process management and control traceable management system employing the AI-based production process management and control traceable management method according to claim 1, characterized by, It includes a data integration module, an AI anomaly detection module, a snapshot persistence module, an index building module, a management and control linkage module, and a backtracking analysis module; among which: The data integration module is used to collect multi-source heterogeneous data in the production process in real time and convert it into a structured production data stream in a unified format. The AI ​​anomaly detection module is used to identify anomalies in the structured production data stream using a pre-built AI model, and to issue a trigger signal when a key anomaly event is detected. The snapshot solidification module is used to respond to the trigger signal, using the current product as the main index, to collect and solidify multi-dimensional on-site data within a preset window before and after the occurrence of the abnormal event, forming an abnormal event data packet; The index building module is used to build a multi-dimensional hierarchical index based on the abnormal event data packet to support fast retrieval in different dimensions; The control and linkage module is used to automatically trigger corresponding production control actions based on the type and severity level of the key abnormal events, and record the triggering information of the control actions into the abnormal event data packet; The backtracking analysis module is used to respond to backtracking query requests, retrieve relevant abnormal event data packets using the hierarchical index, and generate a production process backtracking view and anomaly analysis report.