A manufacturing exception-oriented cross-level agent collaborative processing method and system

Abstract

The application provides a manufacturing anomaly-oriented cross-level agent collaborative processing method and system, which comprises the following steps: a local agent collects sensing data, process parameters and running states of the level where the local agent is located, and converts detected anomaly events into anomaly semantic descriptors; each local agent of each level reports the anomaly semantic descriptors to a collaborative management agent; the collaborative management agent constructs an anomaly propagation graph based on the anomaly semantic descriptors, takes the anomaly propagation graph as input, combines current resource states and production constraints of each level, generates a cross-level collaborative disposal strategy, and issues the cross-level collaborative disposal strategy to the local agents; after receiving the cross-level collaborative disposal strategy, each local agent performs adaptive analysis and execution; during the execution process, the disposal effect is monitored, effect deviation information is fed back to the collaborative management agent in real time, and the collaborative management agent corrects the collaborative strategy according to the effect deviation information. The application can effectively improve the response speed, collaborative accuracy and strategy adaptability of anomaly disposal in a complex manufacturing scene.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically to a cross-level intelligent agent collaborative processing method and system for manufacturing anomalies. Background Technology

[0002] In modern intelligent manufacturing systems, the stability and reliability of the production process directly affect a company's operational efficiency and product quality. During manufacturing, anomalies can occur, including equipment failures, process parameter drift, material abnormalities, and environmental disturbances. However, traditional anomaly handling methods primarily rely on human experience or rule-based expert systems, resulting in problems such as delayed response and poor adaptability. Summary of the Invention

[0003] Based on the above problems, this invention proposes a cross-level intelligent agent collaborative processing method and system for manufacturing anomalies. Through this invention, the response speed, collaborative accuracy and strategy adaptability of anomaly handling in complex manufacturing scenarios are effectively improved.

[0004] In view of this, one aspect of the present invention proposes a cross-level intelligent agent cooperative processing method for manufacturing anomalies, comprising: Deploy local intelligent agents with autonomous perception capabilities at the equipment layer, production line layer, and factory layer respectively; Each local agent collects sensor data, process parameters and operating status at its level in real time. It performs local reasoning on the collected data through anomaly detection model and transforms the detected abnormal events into anomaly semantic descriptors that include anomaly type, severity, spatiotemporal markers and confidence level. Local agents at each level will report anomaly semantic descriptors to the collaborative management agent; The collaborative management agent dynamically constructs a cross-level anomaly propagation graph based on the causal relationships, temporal correlations, and hierarchical transmission paths among anomaly semantic descriptors. The collaborative management agent takes the anomaly propagation graph as input, combines the current resource status and production constraints at each level, and generates a cross-level collaborative handling strategy that includes handling priority, task allocation scheme and collaborative timing through an intent reasoning mechanism. The strategy is then issued to the local agents at the relevant levels in the form of intent commands. After receiving the intent command, each local agent adaptively parses and executes the intent command according to the actual constraints of its own level. During the execution process, the effectiveness of the measures is continuously monitored, and the deviation information is fed back to the collaborative management intelligent agent in real time. The collaborative management intelligent agent then dynamically adjusts the collaborative strategy based on the deviation information.

[0005] Optionally, the step of deploying local intelligent agents with autonomous perception capabilities at the equipment layer, production line layer, and factory layer respectively includes: Based on the physical topology of the manufacturing system, the sensing objects are divided into three levels: single-machine equipment at the equipment level, multi-machine collaborative units at the production line level, and the global production system at the factory level. The sensing range and data ownership boundaries of the local intelligent agents at each level are determined. Local agents at each level connect to heterogeneous data sources at their respective levels through a unified access adapter, including sensor signals, control system outputs, and production logs, and convert multi-source heterogeneous data into a standardized data stream in a unified format. Based on the computing resources and anomaly complexity at each level, corresponding anomaly detection models are deployed at the equipment layer, production line layer, and factory layer, respectively, and the anomaly detection models at each level share a unified anomaly semantic output specification. Each level of local agent performs anomaly detection reasoning locally, mapping the identified anomalous events into structured semantic descriptors that include anomaly type, severity, spatiotemporal label and confidence level; Each local intelligent agent continuously monitors its own operating status and dynamically registers its current sensing capabilities, resource load, and online status to the collaborative management intelligent agent.

[0006] Optionally, the step of each local intelligent agent collecting sensor data, process parameters, and operating status at its respective level in real time, and performing local inference on the collected data through an anomaly detection model to transform detected abnormal events into anomaly semantic descriptors containing anomaly type, severity, spatiotemporal markers, and confidence levels includes: Each local intelligent agent synchronously collects sensor signals, process parameters and equipment operating status data at its level according to a unified sampling clock, and performs time alignment of multi-source data based on the collection timestamp to form a time-consistent multi-dimensional data frame. The quality of data in each channel of the multidimensional data frame is evaluated to identify data quality problems such as noise pollution, signal drift and missing values. Based on the distribution of historical data at the same level, missing or abnormal channels are adaptively completed, and a standardized data frame is output. Using standardized data frames as input, time-domain statistical features, frequency-domain energy features, and process state deviation features are extracted respectively, and the three types of features are fused into a multi-dimensional feature vector representing the current level of operation. The anomaly detection model infers from multi-dimensional feature vectors to determine whether there are anomalies in the current data frame, and performs type identification and spatial localization on the detected anomaly events, outputting anomaly type labels and their location identifiers in the physical topology of the corresponding layer. The severity of abnormal events is classified based on the degree of deviation between the anomaly type label and feature vector from the normal baseline; Based on the data quality assessment results and the stability of the model inference, the confidence level of this anomaly detection result is estimated; The anomaly type label, severity level, timestamp and spatial location identifier, and confidence level are uniformly encapsulated into a structured anomaly semantic descriptor.

[0007] Optionally, the step of the collaborative management agent dynamically constructing a cross-level anomaly propagation graph based on the causal correlation, temporal correlation, and hierarchical propagation path between anomaly semantic descriptors includes: The collaborative management agent continuously receives anomaly semantic descriptors reported by local agents at each level. Using anomaly type, level source, and timestamp as index keys, the anomaly semantic descriptors are stored in a sliding time window cache to form an anomaly semantic descriptor sequence that supports time-series querying. Perform temporal correlation analysis on the sequence of abnormal semantic descriptors in the cache window, identify abnormal event pairs that have a sequential triggering relationship in the time dimension, and mark abnormal event pairs with temporal correlation exceeding a preset threshold as candidate association edges; For candidate associated edges involving abnormal event pairs at different levels, the causal direction of the candidate associated edges is determined by combining the physical topology and process dependency of the manufacturing system, distinguishing between root cause anomalies and derived anomalies, and outputting a set of cross-level abnormal associated edges with causal direction labels. Based on the set of cross-level anomaly associated edges, along the hierarchical propagation direction of the equipment layer, production line layer, and factory layer, the propagation path of the anomaly event in the cross-level physical topology is extracted, forming a set of directed propagation paths that characterize the anomaly diffusion trajectory, revealing the complete link from local triggering to cross-level diffusion of the anomaly. Using abnormal events as nodes, cross-level abnormal association edges as connections, and hierarchical propagation paths as the structural skeleton, a cross-level abnormal propagation graph is constructed. As new anomaly semantic descriptors are continuously reported, the nodes, edges, and propagation paths in the graph are incrementally updated to maintain the graph's real-time representation of the current manufacturing anomaly state.

[0008] Optionally, the collaborative management agent takes the anomaly propagation graph as input, combines the current resource status and production constraints at each level, generates a cross-level collaborative handling strategy including handling priority, task allocation scheme and collaborative timing through an intent reasoning mechanism, and issues it to the local agents at the relevant levels in the form of intent commands. This step includes: The collaborative management agent analyzes the current anomaly propagation graph, extracts the root cause anomaly node, derived anomaly node and their hierarchical propagation path, and infers the anomaly status type of the current manufacturing system by combining the severity and confidence of each anomaly node, thus forming an overall intention description for this collaborative handling. The collaborative management agent queries the current resource load, available disposal capacity, and online status of the local agents dynamically registered at each level, and obtains the current production plan, process constraints, and safety boundaries, forming a resource-constraint joint state view. Based on the overall intent description, and considering the severity, transmission risk, and root cause status of each anomalous node in the graph, under the constraints of the resource-constraint joint state view, the anomalous nodes to be dealt with are prioritized, and an ordered list of disposal targets reflecting the urgency and scope of impact is output. Based on the orderly list of disposal targets, each disposal target is assigned to a local intelligent agent at the corresponding level with the corresponding disposal capability. The disposal responsibilities and collaborative relationships of each intelligent agent are clarified, and a cross-level task allocation scheme covering the equipment layer, production line layer and factory layer is generated. Based on the dependencies and resource competition constraints among agents at each level in the task allocation scheme, the start sequence and concurrency boundary of each processing task are planned to generate a collaborative timing plan; The processing priority, task allocation scheme and collaborative timing plan are encapsulated into structured intent instructions, which are then distributed to local agents at each level as target nodes. The intent instructions retain the overall intent description as context, supporting adaptive parsing by local agents during the execution phase.

[0009] Optionally, the step of each local agent adaptively parsing and executing the intent instruction based on the actual constraints of its level after receiving the intent instruction includes: Each level of local agent receives structured intent instructions issued by the collaborative management agent, parses the overall intent description, the handling responsibility of this level, the collaborative timing requirements and related level information contained in the instructions, and restores the context position of this handling task in cross-level collaboration; Before execution, the local intelligent agent collects the current real-time constraints at this level, including the available status of equipment, process window limits, and safe operation boundaries. It compares the real-time constraints with the preset resource assumptions in the intention command and identifies the deviation items between the preset conditions of the command and the current actual state. Based on the deviation items, the handling responsibilities in the intent instructions are adapted to this level. While keeping the overall intent semantics unchanged, the specific operation parameters, execution paths or resource call methods are adjusted to generate an executable handling plan that meets the real-time constraints of this level, and the consistency between the executable handling plan and the collaborative timing requirements is verified. The action is carried out step by step according to the feasible action plan. After each step, the status of the equipment, process parameters and abnormal response data at this level are collected and compared with the expected action effect in real time to form a step-by-step status monitoring record during the execution process. Based on the step-by-step status monitoring records, assess the effectiveness of the current response plan and determine whether the progress of anomaly elimination meets expectations; If the deviation exceeds the acceptable range, mark the type and source of the deviation and trigger the feedback reporting process. If the execution results meet expectations, update the state of the agent at this level and mark the task as completed; The deviation information, along with the current real-time constraint status and execution process monitoring records at this level, are encapsulated into a structured feedback message and reported to the collaborative management agent.

[0010] Optionally, the step of continuously monitoring the treatment effect during the execution process, feeding back the effect deviation information to the collaborative management intelligent agent in real time, and dynamically correcting the collaborative strategy based on the effect deviation information includes: After the collaborative handling strategy is issued, the collaborative management agent receives structured feedback messages reported by local agents at each level. Based on the collaborative timing plan, the agent aligns and aggregates the feedback messages at each level according to the timing, forming a panoramic view of the multi-level handling effect covering the equipment layer, production line layer and factory layer. The collaborative management agent evaluates the overall effectiveness of cross-level collaborative handling based on a panoramic view of the multi-level handling effects and compares the expected elimination progress of each abnormal node in the current abnormal propagation map. Identify the abnormal nodes with performance deviations and their levels, and locate the key deviation points affecting the collaborative effect by combining the deviation type and the deviation source; For key deviation points, analyze whether the deviation is caused by resource assumption failure, improper task allocation, collaboration timing conflict or hierarchical constraint change. Determine the required policy correction range based on the type of deviation cause, trigger the corresponding level or global collaboration policy correction process, and clarify the correction target and correction boundary. Based on the latest abnormal response data in the feedback messages at each level, the abnormal nodes and associated edges that have been eliminated, added or changed in status in the abnormal propagation graph are updated synchronously to ensure that the revised collaborative strategy is based on the graph that reflects the latest abnormal situation. Based on the updated anomaly propagation map and deviation cause analysis results, the collaborative management agent locally modifies the affected handling priority, task allocation scheme or collaborative timing while maintaining the continuous execution of undevised tasks. It generates incremental correction intention instructions, which are then directed to the relevant local agents at the relevant level, triggering them to re-execute the adaptive parsing and execution process. After the incremental correction intention command is executed, the collaborative management agent continues to gather a new round of feedback messages from relevant levels to verify the execution effect of the correction strategy; If the elimination of all abnormal nodes in the abnormal propagation graph progresses as expected, the current round of cross-level collaborative handling is considered complete, and the collaborative handling result record is output; otherwise, it returns to continue iterative correction.

[0011] Optionally, in the process of converting detected anomalous events into anomalous semantic descriptors, the overall confidence of each local agent in the anomalous event is calculated as follows:

[0012] in, The overall confidence level of the anomaly semantic descriptor; The score is the inference output of the anomaly detection model based on the current feature vector, reflecting the model's confidence in the anomaly determination. This is the quality assessment score for the current data frame; a higher value indicates better data integrity and signal-to-noise ratio. This is the data quality sensitivity coefficient, which controls the rate at which data quality affects confidence levels. This represents the current computational resource load rate of the local agent, reflecting the inhibitory effect of resource scarcity on inference reliability. This is the load suppression coefficient; To adapt weights to different levels, the settings are based on the differences in the complexity of the detection models for the equipment layer, production line layer, and factory layer. This represents the number of missing or completed channels in the current data frame, reflecting the degree of integrity loss of the original data. This represents the total number of data channels collected by the agent at this level. The missing channel ratio is the percentage of channels missing; the higher the missing channel ratio, the greater the penalty to the confidence level. This is the channel missing penalty coefficient, which controls the degree to which the missing ratio reduces the confidence level; The historical detection consistency coefficient is calculated based on the stability of the local agent's detection results of the same type of anomaly within a recent time window. A higher value is taken if the same type of anomaly is repeatedly confirmed recently, and a lower value is taken if the detection conclusion is frequently reversed. This reflects the historical reliability of the model under the current working conditions.

[0013] Optionally, during the dynamic construction of cross-level anomaly propagation graphs, the causal correlation strength between any two anomaly event nodes is calculated as follows:

[0014] in, , These are two anomalous event nodes in the anomalous propagation graph; The time overlap length between two abnormal events within the time window; The length of the union of the time spans of the two anomalous events is given by the ratio of the two, which reflects the degree of temporal co-occurrence. It represents the difference in the trigger times of two abnormal events, indicating the time interval between them; This is the hierarchical propagation time constant, set based on the typical delay of anomaly propagation between different levels; The topological correlation coefficient is based on... , The connection relationship of the node at the same level in the physical topology of the manufacturing system is assigned a value: direct connection takes a higher value, and indirect connection across levels takes a lower value. , These are abnormal nodes. , The combined confidence level of the corresponding anomaly semantic descriptor, the product of which reflects the joint degree of trustworthiness of the information from the two nodes simultaneously; The joint confidence sensitivity coefficient controls the rate at which the joint confidence level affects the strength of the causal association. When the confidence levels of both nodes are high, this term approaches 1. When the confidence level of either node is low, it significantly suppresses the association strength. The exception type compatibility coefficient is based on... , The causal compatibility of the anomaly type labels in the manufacturing process knowledge base is assigned a value. If there is a known causal relationship between the types, a higher value is taken. If there is no process relationship between the types, a lower value is taken. Domain knowledge is introduced to correct the pure data-driven relationship.

[0015] Another aspect of the present invention provides a cross-level intelligent agent collaborative processing system for manufacturing anomalies, for executing a cross-level intelligent agent collaborative processing method for manufacturing anomalies, comprising: local intelligent agents with autonomous perception capabilities deployed at the equipment layer, production line layer and factory layer respectively; and a collaborative management intelligent agent; The local agent is configured as follows: Real-time acquisition of sensor data, process parameters and operating status at the current level; local inference of the acquired data through an anomaly detection model; conversion of detected abnormal events into anomaly semantic descriptors containing anomaly type, severity, spatiotemporal markers and confidence level. Report the abnormal semantic descriptor to the collaborative management agent; The collaborative management agent is configured as follows: Based on the causal relationships, temporal correlations, and hierarchical propagation paths among anomaly semantic descriptors, a cross-level anomaly propagation graph is dynamically constructed. Using the anomaly propagation graph as input, and combining the current resource status and production constraints at each level, a cross-level collaborative handling strategy containing handling priorities, task allocation schemes and collaborative timing is generated through an intent reasoning mechanism, and then issued to the local intelligent agents at the relevant levels in the form of intent commands. The local agent is also configured to: Upon receiving the intent instruction, the intent instruction is adaptively parsed and executed based on the actual constraints of this level; During the execution process, the effectiveness of the measures is continuously monitored, and the deviation information is fed back to the collaborative management intelligent agent in real time, so that the collaborative management intelligent agent can dynamically adjust the collaborative strategy based on the deviation information.

[0016] The technical solution of this invention provides a cross-level intelligent agent collaborative processing method for manufacturing anomalies, comprising: deploying local intelligent agents with autonomous perception capabilities at the equipment layer, production line layer, and factory layer; each local intelligent agent collects sensor data, process parameters, and operating status at its respective layer in real time, performs local inference on the collected data through anomaly detection models, and transforms detected abnormal events into anomaly semantic descriptors containing anomaly type, severity, spatiotemporal markers, and confidence levels; each level of local intelligent agent reports the anomaly semantic descriptors to the collaborative management intelligent agent; the collaborative management intelligent agent, based on the causal relationships, temporal correlations, and layer-specific relationships between the anomaly semantic descriptors... A hierarchical transmission path is established, dynamically constructing a cross-level anomaly propagation graph. The collaborative management agent, using this graph as input and considering the current resource status and production constraints at each level, generates a cross-level collaborative handling strategy through an intent reasoning mechanism. This strategy includes handling priorities, task allocation schemes, and collaborative timing, and is then issued as intent commands to the relevant local agents. Upon receiving the intent commands, each local agent adaptively parses and executes them based on the actual constraints of its level. During execution, the handling effect is continuously monitored, and deviation information is fed back to the collaborative management agent in real time. The collaborative management agent then dynamically adjusts the collaborative strategy based on this deviation information. This approach achieves lossless transmission of cross-level anomaly information through a unified semantic descriptor, reveals the cross-level causal transmission patterns of manufacturing anomalies through a dynamic anomaly graph, and balances global optimization with hierarchical autonomy through an intent-driven mechanism. This effectively improves the response speed, collaborative accuracy, and strategy adaptability of anomaly handling in complex manufacturing scenarios. Attached Figure Description

[0017] Figure 1 This is a flowchart of a cross-level intelligent agent collaborative processing method for manufacturing anomalies provided in an embodiment of the present invention; Figure 2 This is a schematic block diagram of a cross-level intelligent agent collaborative processing system for manufacturing anomalies provided in one embodiment of the present invention. Detailed Implementation

[0018] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0019] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0020] The terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. 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 apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0021] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0022] The following reference Figures 1 to 2 This invention describes a cross-level intelligent agent collaborative processing method and system for manufacturing anomalies, provided by some embodiments of the present invention.

[0023] like Figure 1 As shown, one embodiment of the present invention provides a cross-level intelligent agent collaborative processing method for manufacturing anomalies, including: Deploy local intelligent agents with autonomous perception capabilities at the equipment layer, production line layer, and factory layer respectively; Each local intelligent agent collects sensor data, process parameters and operating status at its level in real time. It performs local reasoning on the collected data through anomaly detection model and transforms the detected abnormal events into anomaly semantic descriptors containing anomaly type, severity, spatiotemporal markers and confidence level, which serve as a unified information carrier for cross-level collaboration. Local agents at each level will report anomaly semantic descriptors to the collaborative management agent; The collaborative management agent dynamically constructs a cross-level anomaly propagation map based on the causal relationship, temporal correlation, and hierarchical transmission path between anomaly semantic descriptors, thereby realizing a structured expression of the entire chain of manufacturing anomalies from local triggering to cross-level diffusion. The collaborative management agent takes the anomaly propagation graph as input, combines the current resource status and production constraints at each level, and generates a cross-level collaborative handling strategy that includes handling priority, task allocation scheme and collaborative timing through an intent reasoning mechanism. The strategy is then issued to the local agents at the relevant levels in the form of intent commands. After receiving the intent command, each local agent adaptively parses and executes the intent command according to the actual constraints of its own level. During the execution process, the effectiveness of the measures is continuously monitored, and the deviation information is fed back to the collaborative management intelligent agent in real time. The collaborative management intelligent agent dynamically corrects the collaborative strategy based on the deviation information, forming a closed-loop iterative measures mechanism across levels.

[0024] The technical solution adopted in this embodiment achieves lossless transmission of cross-level anomaly information through a unified semantic descriptor, reveals the cross-level causal transmission law of manufacturing anomalies through a dynamic anomaly map, and balances global optimization and hierarchical autonomy through an intent-driven mechanism, effectively improving the response speed, collaborative accuracy and strategy adaptability of anomaly handling in complex manufacturing scenarios.

[0025] In some possible embodiments of the present invention, the step of deploying local intelligent agents with autonomous perception capabilities at the equipment layer, production line layer, and factory layer respectively includes: Based on the physical topology of the manufacturing system, the sensing objects are divided into three levels: single-machine equipment at the equipment level, multi-machine collaborative units at the production line level, and the global production system at the factory level. The sensing range and data ownership boundaries of the local intelligent agents at each level are determined. Local agents at each level connect to heterogeneous data sources at their respective levels through a unified access adapter. These heterogeneous data sources include sensor signals, control system outputs, and production logs. The adapter converts the multi-source heterogeneous data into a standardized data stream in a unified format, which serves as the input for subsequent local inference. Based on the computational resources and anomaly complexity of each level (i.e., equipment level, production line level, and factory level), corresponding anomaly detection models are deployed at the equipment level, production line level, and factory level, respectively. Among them, a lightweight anomaly detection model for single equipment signals is deployed at the equipment level, a medium-complexity model for multi-equipment collaborative states is deployed at the production line level, and a comprehensive detection and analysis model for global process deviations is deployed at the factory level. The anomaly detection models at each level share a unified anomaly semantic output specification. In this step, the available computing resource limits for each level—equipment, production line, and factory—are collected, including processor computing power, memory capacity, and inference latency constraints. Simultaneously, based on the signal dimensions of the monitored objects at each level, the coupling relationships between devices, and the scope of process influence, the input and inference complexity of the anomaly detection tasks at each level are evaluated, forming resource-complexity adaptation benchmarks for each level, which serve as the basis for model selection and tailoring. Based on these benchmarks, a lightweight anomaly detection model is constructed at the equipment level, using a single device signal as input and edge inference as the target, compressing the model parameters to meet the strict computing power and latency constraints of the equipment level. At the production line level, a medium-complexity model is constructed, using multi-device collaborative state sequences as input and balancing detection accuracy and inference efficiency. At the factory level, a comprehensive detection and analysis model is constructed, using global process parameters and cross-production line states as input and supporting multi-dimensional feature fusion. The three models are deployed separately. Deploy the model to the corresponding local agent; establish a unified anomaly semantic output specification, stipulating that the anomaly semantic descriptor output by each level model must include four standard fields: anomaly type label, severity level, spatiotemporal marker, and confidence level. The field format and value range should be consistent across levels; each level model should add a specification adaptation module to the output layer to map its internal inference results to the unified semantic output specification, ensuring that the outputs of models with different complexities can be directly parsed and compared by the collaborative management agent; using the detection results of the production line layer model for the same anomaly event as a benchmark, perform cross-layer calibration on the outputs of the equipment layer lightweight anomaly detection model and the factory layer comprehensive detection and analysis model to correct the anomaly type label deviation and severity level drift caused by the difference in model complexity, ensuring the consistency of the three-layer models at the semantic output level, and providing a reliable semantic alignment basis for cross-layer node association in the subsequent anomaly propagation graph construction. By using resource-complexity adaptation benchmarks to drive the selection of hierarchical models, we ensure that each level of model achieves optimal detection capabilities under computational constraints. By using unified semantic output specifications and cross-level calibration, we eliminate semantic inconsistencies caused by differences in model complexity, ensuring the comparability of information in cross-level collaboration.

[0026] Each level of local agent performs anomaly detection reasoning locally, mapping the identified anomalous events into structured semantic descriptors that include anomaly type, severity, spatiotemporal label and confidence level, thereby achieving a unified expression of anomaly information across levels; Each local agent continuously monitors its own operational status and dynamically registers its current sensing capabilities, resource load, and online status with the collaborative management agent. This ensures that the collaborative management agent has a real-time grasp of the availability of all agents in the network, providing a resource view for the generation of subsequent collaborative strategies.

[0027] In this embodiment, the orderly organization of multi-level perception capabilities is achieved through the delineation of hierarchical perception boundaries and the normalization of heterogeneous data; the consistency and synergy of cross-level anomaly information are ensured through hierarchical adaptation models and unified semantic output.

[0028] In some possible embodiments of the present invention, the steps of each local intelligent agent collecting sensor data, process parameters, and operating status at its respective level in real time, performing local inference on the collected data through an anomaly detection model, and converting detected abnormal events into anomaly semantic descriptors containing anomaly type, severity, spatiotemporal markers, and confidence levels include: Each local intelligent agent synchronously collects sensor signals, process parameters and equipment operating status data at its level according to a unified sampling clock, and performs time alignment of multi-source data based on the collection timestamp to form a time-consistent multi-dimensional data frame. The quality of data in each channel of the multidimensional data frame is evaluated to identify data quality problems such as noise pollution, signal drift and missing values. Based on the distribution of historical data at the same level, missing or abnormal channels are adaptively completed, and a standardized data frame is output. Using standardized data frames as input, time-domain statistical features, frequency-domain energy features, and process state deviation features are extracted respectively. The three types of features (time-domain statistical features, frequency-domain energy features, and process state deviation features) are then fused into a multi-dimensional feature vector representing the current level of operation. The anomaly detection model infers from multi-dimensional feature vectors to determine whether there are anomalies in the current data frame, and performs type identification and spatial localization on the detected anomaly events, outputting anomaly type labels and their location identifiers in the physical topology of the corresponding layer. In this step, the anomaly detection model deployed on each local agent takes the current multidimensional feature vector as input and compares it with the normal baseline feature distribution constructed based on the historical normal operation data of this level. It calculates the comprehensive deviation of the current feature vector from the normal baseline. When the comprehensive deviation exceeds the judgment threshold set according to the hierarchical anomaly complexity, the current data frame is determined to be abnormal, triggering subsequent type identification and spatial localization processes. Otherwise, a normal label is output and the process returns to the next data frame. For data frames determined to be abnormal, the anomaly detection model analyzes the deviation patterns of each feature dimension in the multidimensional feature vector, matches the current deviation pattern with the feature templates of each type in the predefined anomaly type feature library, identifies the anomaly type that best matches the current deviation pattern, and outputs an anomaly type label. When the deviation pattern matches multiple anomaly types simultaneously... The primary and secondary anomaly type labels are determined based on the dominant contribution of deviations in each dimension. Based on the mapping relationship between each data channel in the multi-dimensional feature vector and the physical topology nodes at this level, the set of data channels that contributes most to the anomaly type determination is identified. This set of channels is then mapped to the corresponding equipment nodes or production line units in the physical topology at this level to determine the spatial source range of the anomaly event and output the location identifier of the anomaly event in the physical topology at this level. The anomaly type label and location identifier are jointly verified, and the physical node corresponding to the location identifier is checked against the manufacturing process knowledge base to see if it possesses the process conditions to generate this type of anomaly. If the joint verification passes, the output is confirmed; if it fails, the anomaly type label is downgraded to a candidate label and a location doubt marker is added. This label, along with the location identifier, is included in the anomaly semantic descriptor for reference in subsequent confidence estimation. This process achieves an orderly connection between anomaly determination and type identification through normal baseline comparison; precise spatial positioning is achieved through the mapping of data channels to physical topology; and joint verification using process knowledge suppresses type misjudgment and location misalignment, improving the accuracy and reliability of the anomaly semantic descriptor.

[0029] The severity of abnormal events is classified based on the degree of deviation between the anomaly type label and feature vector from the normal baseline; Based on the data quality assessment results and the stability of the model inference, the confidence level of this anomaly detection result is estimated; Anomaly type labels, severity levels, timestamps, spatial location identifiers, and confidence levels are uniformly encapsulated into structured anomaly semantic descriptors, which serve as the standard information carrier for local agents to report to collaborative management agents, supporting the subsequent construction of dynamic anomaly maps and the generation of collaborative strategies.

[0030] In this embodiment, the reliability of inference input is ensured through time alignment and data quality control; the accuracy and interpretability of anomaly identification are improved through multi-dimensional feature fusion and anomaly localization; and a quantitative and reliable basis for anomaly information is provided for cross-level collaborative decision-making through severity rating and confidence estimation, thereby enhancing the targeted nature of collaborative handling.

[0031] In some possible embodiments of the present invention, the step of the collaborative management agent dynamically constructing a cross-level anomaly propagation graph based on the causal correlation, temporal correlation, and hierarchical propagation path between anomaly semantic descriptors includes: The collaborative management agent continuously receives anomaly semantic descriptors reported by local agents at each level. Using anomaly type, level source, and timestamp as index keys, the anomaly semantic descriptors are stored in a sliding time window cache to form an anomaly semantic descriptor sequence that supports time-series querying. Perform temporal correlation analysis on the sequence of abnormal semantic descriptors in the cache window to identify abnormal event pairs that have a sequential triggering relationship in the time dimension. Mark abnormal event pairs with temporal correlation exceeding a preset threshold as candidate association edges, which serve as the input candidate set for causal inference. For candidate associated edges involving abnormal event pairs at different levels, the causal direction of the candidate associated edges is determined by combining the physical topology and process dependency of the manufacturing system, distinguishing between root cause anomalies and derived anomalies, and outputting a set of cross-level abnormal associated edges with causal direction labels. Based on the set of cross-level anomaly associated edges, along the hierarchical propagation direction of the equipment layer, production line layer, and factory layer, the propagation path of the anomaly event in the cross-level physical topology is extracted, forming a set of directed propagation paths that characterize the anomaly diffusion trajectory, revealing the complete link from local triggering to cross-level diffusion of the anomaly. Using abnormal events as nodes, cross-level abnormal association edges as connections, and hierarchical propagation paths as the structural skeleton, a cross-level abnormal propagation graph is constructed. As new anomaly semantic descriptors are continuously reported, the nodes, edges, and propagation paths in the graph are incrementally updated to maintain the graph's real-time representation of the current manufacturing anomaly state.

[0032] In this embodiment, the root cause anomaly and the derived anomaly are effectively distinguished through temporal correlation analysis and causal direction determination; through hierarchical transmission path extraction and dynamic map updating, a real-time structured expression of the cross-layer anomaly diffusion process is formed, providing a traceable causal basis for the accurate generation of collaborative treatment strategies.

[0033] In some possible embodiments of the present invention, the collaborative management agent takes the anomaly propagation graph as input, combines the current resource status and production constraints at each level, generates a cross-level collaborative handling strategy including handling priority, task allocation scheme and collaborative timing through an intent reasoning mechanism, and issues it to the local agents at the relevant levels in the form of intent commands. The steps include: The collaborative management agent analyzes the current anomaly propagation graph, extracts the root cause anomaly node, derived anomaly node and their hierarchical propagation path, and infers the anomaly status type of the current manufacturing system by combining the severity and confidence of each anomaly node, thus forming an overall intention description for this collaborative handling. The collaborative management agent queries the current resource load, available disposal capacity, and online status of the local agents dynamically registered at each level, and obtains the current production plan, process constraints, and safety boundaries, forming a resource-constraint joint state view. Based on the overall intent description, and considering the severity, transmission risk, and root cause status of each anomalous node in the graph, under the constraints of the resource-constraint joint state view, the anomalous nodes to be dealt with are prioritized, and an ordered list of disposal targets reflecting the urgency and scope of impact is output. Based on the orderly list of disposal targets, each disposal target is assigned to a local intelligent agent at the corresponding level with the corresponding disposal capability. The disposal responsibilities and collaborative relationships of each intelligent agent are clarified, and a cross-level task allocation scheme covering the equipment layer, production line layer and factory layer is generated. Based on the dependencies and resource competition constraints among agents at each level in the task allocation scheme, the start sequence and concurrency boundary of each disposal task are planned, and a collaborative timing plan is generated to ensure that cross-level disposal actions are connected in an orderly manner in the time dimension and to avoid resource conflicts. The processing priority, task allocation scheme and collaborative timing plan are encapsulated into structured intent instructions, which are then distributed to local agents at each level as target nodes. The intent instructions retain the overall intent description as context, supporting adaptive parsing by local agents during the execution phase.

[0034] In this embodiment, intent recognition is driven by graph parsing to ensure semantic consistency between policy generation and the current abnormal situation; task allocation and timing planning are constrained by the resource-constraint joint view to improve the feasibility and execution efficiency of collaborative strategies; and the accuracy of local intelligent agent adaptive execution is enhanced by carrying context through intent instructions.

[0035] In some possible embodiments of the present invention, the step of each local intelligent agent adaptively parsing and executing the intent instruction based on the actual constraints of its own level after receiving the intent instruction includes: Each level of local agent receives structured intent instructions issued by the collaborative management agent, parses the overall intent description, the handling responsibility of this level, the collaborative timing requirements and related level information contained in the instructions, and restores the context position of this handling task in cross-level collaboration; Before execution, the local intelligent agent collects the current real-time constraints at this level, including the available status of equipment, process window limits, and safe operation boundaries. It compares the real-time constraints with the preset resource assumptions in the intention command and identifies the deviation items between the preset conditions of the command and the current actual state. Based on the deviation items, the handling responsibilities in the intent instructions are adapted to this level. While keeping the overall intent semantics unchanged, the specific operation parameters, execution paths or resource call methods are adjusted to generate an executable handling plan that meets the real-time constraints of this level, and the consistency between the executable handling plan and the collaborative timing requirements is verified. The action is carried out step by step according to the feasible action plan. After each step, the status of the equipment, process parameters and abnormal response data at this level are collected and compared with the expected action effect in real time to form a step-by-step status monitoring record during the execution process. Based on the step-by-step status monitoring records, assess the effectiveness of the current response plan and determine whether the progress of anomaly elimination meets expectations; If the deviation exceeds the acceptable range, mark the type and source of the deviation and trigger the feedback reporting process. If the execution results meet expectations, update the state of the agent at this level and mark the task as completed; The effect deviation information, along with the current real-time constraint status and execution process monitoring records of this level, are encapsulated into a structured feedback message and reported to the collaborative management agent, providing a complete execution-side basis for the collaborative management agent to dynamically correct cross-level collaborative strategies.

[0036] In this embodiment, context restoration and real-time constraint awareness ensure semantic fidelity and constraint adaptation of intent commands at the hierarchical execution side; step-by-step execution and effect evaluation enable fine-grained monitoring of the handling process; and structured deviation feedback provides reliable execution-side information support for the closed-loop iterative correction of collaborative strategies.

[0037] In some possible embodiments of the present invention, the step of continuously monitoring the treatment effect during the execution process, feeding back the effect deviation information to the collaborative management intelligent agent in real time, and the collaborative management intelligent agent dynamically correcting the collaborative strategy based on the effect deviation information includes: After the collaborative handling strategy is issued, the collaborative management agent receives structured feedback messages reported by local agents at each level. Based on the collaborative timing plan, the agent aligns and aggregates the feedback messages at each level according to the timing, forming a panoramic view of the multi-level handling effect covering the equipment layer, production line layer and factory layer. The collaborative management agent evaluates the overall effectiveness of cross-level collaborative handling based on a panoramic view of the multi-level handling effects and compares the expected elimination progress of each abnormal node in the current abnormal propagation map. Identify the abnormal nodes with performance deviations and their levels, and locate the key deviation points affecting the collaborative effect by combining the deviation type and the deviation source; For key deviation points, analyze whether the deviation is caused by resource assumption failure, improper task allocation, collaboration timing conflict or hierarchical constraint change. Determine the required policy correction range based on the type of deviation cause, trigger the corresponding level or global collaboration policy correction process, and clarify the correction target and correction boundary. Based on the latest abnormal response data in the feedback messages at each level, the abnormal nodes and associated edges that have been eliminated, added or changed in status in the abnormal propagation graph are updated synchronously to ensure that the revised collaborative strategy is based on the graph that reflects the latest abnormal situation. Based on the updated anomaly propagation map and deviation cause analysis results, the collaborative management agent locally modifies the affected handling priority, task allocation scheme or collaborative timing while maintaining the continuous execution of undevised tasks. It generates incremental correction intention instructions, which are then directed to the relevant local agents at the relevant level, triggering them to re-execute the adaptive parsing and execution process. After the incremental correction intention command is executed, the collaborative management agent continues to gather a new round of feedback messages from relevant levels to verify the execution effect of the correction strategy; If the elimination of all abnormal nodes in the abnormal propagation graph progresses as expected, the current round of cross-level collaborative handling is considered complete, and the collaborative handling result record is output; otherwise, it returns to continue iterative correction.

[0038] In this embodiment, multi-level feedback timing alignment and panoramic view construction ensure that strategy correction is based on complete and consistent execution-side information; deviation cause classification drives accurate correction range determination to avoid over-adjustment interfering with normal task execution; and map synchronization update and incremental instruction issuance enable lightweight and efficient closed-loop iteration of collaborative strategies.

[0039] In some possible embodiments of the present invention, during the process of converting detected anomalous events into anomalous semantic descriptors, the overall confidence level of each local agent for the anomalous event is calculated as follows:

[0040] in, The overall confidence level of the anomaly semantic descriptor; The score is the inference output of the anomaly detection model based on the current feature vector, reflecting the model's confidence in the anomaly determination. This is the quality assessment score for the current data frame; a higher value indicates better data integrity and signal-to-noise ratio. This is the data quality sensitivity coefficient, which controls the rate at which data quality affects confidence levels. This represents the current computational resource load rate of the local agent, reflecting the inhibitory effect of resource scarcity on inference reliability. This is the load suppression coefficient; To adapt weights to different levels, the settings are based on the differences in the complexity of the detection models for the equipment layer, production line layer, and factory layer. This represents the number of missing or completed channels in the current data frame, reflecting the degree of integrity loss of the original data. This represents the total number of data channels collected by the agent at this level. The missing channel ratio is the percentage of channels missing; the higher the missing channel ratio, the greater the penalty to the confidence level. This is the channel missing penalty coefficient, which controls the degree to which the missing ratio reduces the confidence level; The historical detection consistency coefficient is calculated based on the stability of the local agent's detection results of the same type of anomaly within a recent time window. A higher value is taken if the same type of anomaly is repeatedly confirmed recently, and a lower value is taken if the detection conclusion is frequently reversed. This reflects the historical reliability of the model under the current working conditions.

[0041] In some possible embodiments of the present invention, during the process of dynamically constructing a cross-level anomaly propagation graph, the causal correlation strength between any two anomaly event nodes is calculated as follows:

[0042] in, , These are two anomalous event nodes in the anomalous propagation graph; The time overlap length between two abnormal events within the time window; The length of the union of the time spans of the two anomalous events is given by the ratio of the two, which reflects the degree of temporal co-occurrence. It represents the difference in the trigger times of two abnormal events, indicating the time interval between them; This is the hierarchical propagation time constant, set based on the typical delay of anomaly propagation between different levels; The topological correlation coefficient is based on... , The connection relationship of the node at the same level in the physical topology of the manufacturing system is assigned a value: direct connection takes a higher value, and indirect connection across levels takes a lower value. , These are abnormal nodes. , The combined confidence level of the corresponding anomaly semantic descriptor, the product of which reflects the joint degree of trustworthiness of the information from the two nodes simultaneously; The joint confidence sensitivity coefficient controls the rate at which the joint confidence level affects the strength of the causal association. When the confidence levels of both nodes are high, this term approaches 1. When the confidence level of either node is low, it significantly suppresses the association strength. The exception type compatibility coefficient is based on... , The causal compatibility of the anomaly type labels in the manufacturing process knowledge base is assigned a value. If there is a known causal relationship between the types, a higher value is taken. If there is no process relationship between the types, a lower value is taken. Domain knowledge is introduced to correct the pure data-driven relationship.

[0043] In some possible embodiments of the present invention, during the process of generating a cross-level collaborative handling strategy that includes handling priorities, the handling priority score of each abnormal node is calculated as follows:

[0044] in, These are abnormal nodes to be evaluated. This is the normalized value for the severity level of this node; Indicates As the root cause For the derived directed causal edge, This corresponds to the strength of the causal relationship; for The weighted transmission risk of all downstream derived anomalies reflects the potential for the spread of root cause anomalies. for Level The current available resource ratio; the more scarce the resources, the higher the priority. These are the weighting coefficients for severity, transmission risk, and resource scarcity, respectively. , These are the root cause abnormal nodes. With derived abnormal nodes The overall confidence level of the corresponding anomaly semantic descriptor is used to weight the severity score with confidence, so as to avoid low-confidence anomaly nodes receiving unreasonable processing priority due to inflated severity labels; Abnormal node The duration of persistence in the anomaly propagation map reflects the accumulated unaddressed time of the anomaly; The threshold for the longest tolerable period of non-action, set based on the type and severity of the anomaly; The ratio of abnormal sustained pressure is such that the closer the ratio is to or the greater than 1, the more the abnormality has not been dealt with for a long time and the higher the risk of timeliness. The abnormal continuous pressure weighting coefficient.

[0045] In some possible embodiments of the present invention, when each local agent performs adaptive parsing of the intent instruction, the semantic conformity between the execution scheme and the intent instruction is evaluated as follows:

[0046] in, The semantic conformity of the execution plan to the intent instruction is such that the closer the value is to 1, the higher the conformity. The number of targets to be processed contained in the intent instructions at this level; and These represent the values ​​of the operation parameters corresponding to the obj-th disposal target in the execution plan and intent instruction, respectively. The penalty weight for parameter deviation of the obj-th target; This is a flag indicating the deviation between the execution timing of the obj-th disposal target and the coordinated timing plan. It is set to 1 if there is a timing conflict, and 0 otherwise. Weights for time-series deviation penalties; The flag for the safety boundary violation of the objth target is set to 1 if the corresponding operation parameter in the execution plan exceeds the safety operation boundary of this level, otherwise it is set to 0, reflecting whether the adaptive adjustment touches the safety constraint. The penalty weight for the security violation of the j-th target is the penalty weight for the deviation from the parameter. Setting it to a larger value ensures that violations of security boundaries impose a stronger penalty on semantic compliance. The priority score for handling the abnormal node corresponding to the obj-th handling target; This represents the maximum priority score of all disposal targets in the current collaborative disposal strategy; For normalized priority ratio, The two factors together constitute the priority adaptive tolerance factor, which appropriately relaxes the semantic compliance requirements for low-priority targets and tightens the compliance standards for high-priority targets.

[0047] In some possible embodiments of the present invention, during the process of dynamically correcting the collaborative strategy by the collaborative management agent, the urgency of correcting feedback deviations at each level is evaluated as follows:

[0048] in, For levels where there is a performance bias; The overall confidence level of the reported anomaly semantic descriptors at this level reflects the credibility of the deviation information itself. This represents the semantic compliance of the current execution plan at this level; the lower the compliance, the higher the urgency of correction. The duration of the task at this level has been executed; The ratio of the processing deadline set for this level in the coordinated timeline plan to the timeline plan reflects the degree of time pressure. This is the timing compression amplification factor; The cascade impact coefficient of the abnormal node at this level in the abnormal propagation map is calculated based on the number of downstream derived nodes of the abnormal node at this level and the strength of causal relationship, reflecting the risk of cross-level diffusion that may be caused if the deviation is not corrected in time. hierarchical The number of strategy corrections triggered during this round of collaborative handling reflects the historical frequency of repeated deviations at this level; To repeatedly correct the amplification factor, the more times the correction is performed, the more persistent the execution deviation at that level should be, and the urgency of correction should be further increased to encourage the collaborative management agent to focus on that level. hierarchical Current available resource ratio; This represents the maximum available resource ratio across all tiers with deviations. To normalize the resource sufficiency ratio, The resource abundance inhibition coefficient and the two together constitute the resource abundance inhibition term. This appropriately reduces the urgency of correction for levels with relatively abundant resources, guiding the collaborative management agent to prioritize correction of levels with scarce resources and serious deviations.

[0049] Please refer to Figure 2 Another embodiment of the present invention provides a cross-level intelligent agent collaborative processing system for manufacturing anomalies, for executing a cross-level intelligent agent collaborative processing method for manufacturing anomalies, including: local intelligent agents with autonomous perception capabilities deployed at the equipment layer, production line layer and factory layer respectively; and a collaborative management intelligent agent; The local agent is configured as follows: Real-time acquisition of sensor data, process parameters and operating status at the current level; local inference of the acquired data through an anomaly detection model; conversion of detected abnormal events into anomaly semantic descriptors containing anomaly type, severity, spatiotemporal markers and confidence level. Report the abnormal semantic descriptor to the collaborative management agent; The collaborative management agent is configured as follows: Based on the causal relationships, temporal correlations, and hierarchical propagation paths among anomaly semantic descriptors, a cross-level anomaly propagation graph is dynamically constructed. Using the anomaly propagation graph as input, and combining the current resource status and production constraints at each level, a cross-level collaborative handling strategy containing handling priorities, task allocation schemes and collaborative timing is generated through an intent reasoning mechanism, and then issued to the local intelligent agents at the relevant levels in the form of intent commands. The local agent is also configured to: Upon receiving the intent instruction, the intent instruction is adaptively parsed and executed based on the actual constraints of this level; During the execution process, the effectiveness of the measures is continuously monitored, and the deviation information is fed back to the collaborative management intelligent agent in real time, so that the collaborative management intelligent agent can dynamically adjust the collaborative strategy based on the deviation information.

[0050] It should be known that, Figure 2 The block diagram of the cross-level intelligent agent collaborative processing system for manufacturing anomalies shown is for illustrative purposes only, and the number of modules shown does not limit the scope of protection of this invention. The cross-level intelligent agent collaborative processing system for manufacturing anomalies provided in this embodiment can be used to execute various embodiments of the corresponding cross-level intelligent agent collaborative processing method for manufacturing anomalies. For specific implementation details, please refer to the descriptions of the respective method embodiments, which will not be repeated here.

[0051] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0052] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0053] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical or other forms.

[0054] The units described above as separate components may or may not be physically separate. 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 units can be selected to achieve the purpose of this embodiment according to actual needs.

[0055] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0056] If the integrated units described above are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0057] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0058] The embodiments of this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

[0059] While the present invention has been disclosed above, it is not limited thereto. Any person skilled in the art can easily conceive of variations or substitutions without departing from the spirit and scope of the present invention, and various modifications and alterations can be made, including combinations of the different functions and implementation steps described above, as well as software and hardware implementation methods, all of which are within the protection scope of the present invention.

Claims

1. A cross-level intelligent agent collaborative processing method for manufacturing anomalies, characterized in that, include: Deploy local intelligent agents with autonomous perception capabilities at the equipment layer, production line layer, and factory layer respectively; Each local agent collects sensor data, process parameters and operating status at its level in real time. It performs local reasoning on the collected data through anomaly detection model and transforms the detected abnormal events into anomaly semantic descriptors that include anomaly type, severity, spatiotemporal markers and confidence level. Local agents at each level will report anomaly semantic descriptors to the collaborative management agent; The collaborative management agent dynamically constructs a cross-level anomaly propagation graph based on the causal relationships, temporal correlations, and hierarchical transmission paths among anomaly semantic descriptors. The collaborative management agent takes the anomaly propagation graph as input, combines the current resource status and production constraints at each level, and generates a cross-level collaborative handling strategy that includes handling priority, task allocation scheme and collaborative timing through an intent reasoning mechanism. The strategy is then issued to the local agents at the relevant levels in the form of intent commands. After receiving the intent command, each local agent adaptively parses and executes the intent command according to the actual constraints of its own level. During the execution process, the effectiveness of the measures is continuously monitored, and the deviation information is fed back to the collaborative management intelligent agent in real time. The collaborative management intelligent agent then dynamically adjusts the collaborative strategy based on the deviation information.

2. The cross-level intelligent agent collaborative processing method for manufacturing anomalies according to claim 1, characterized in that, The step of deploying local intelligent agents with autonomous perception capabilities at the equipment layer, production line layer, and factory layer respectively includes: Based on the physical topology of the manufacturing system, the sensing objects are divided into three levels: single-machine equipment at the equipment level, multi-machine collaborative units at the production line level, and the global production system at the factory level. The sensing range and data ownership boundaries of the local intelligent agents at each level are determined. Local agents at each level connect to heterogeneous data sources at their respective levels through a unified access adapter, including sensor signals, control system outputs, and production logs, and convert multi-source heterogeneous data into a standardized data stream in a unified format. Based on the computing resources and anomaly complexity at each level, corresponding anomaly detection models are deployed at the equipment layer, production line layer, and factory layer, respectively, and the anomaly detection models at each level share a unified anomaly semantic output specification. Each level of local agent performs anomaly detection reasoning locally, mapping the identified anomalous events into structured semantic descriptors that include anomaly type, severity, spatiotemporal label and confidence level; Each local intelligent agent continuously monitors its own operating status and dynamically registers its current sensing capabilities, resource load, and online status to the collaborative management intelligent agent.

3. The cross-level intelligent agent collaborative processing method for manufacturing anomalies according to claim 2, characterized in that, Each local agent collects sensor data, process parameters, and operating status at its respective level in real time. It then performs local inference on the collected data using an anomaly detection model, transforming detected abnormal events into anomaly semantic descriptors containing anomaly type, severity, spatiotemporal markers, and confidence levels. This process includes: Each local intelligent agent synchronously collects sensor signals, process parameters and equipment operating status data at its level according to a unified sampling clock, and performs time alignment of multi-source data based on the collection timestamp to form a time-consistent multi-dimensional data frame. The quality of data in each channel of the multidimensional data frame is evaluated to identify data quality problems such as noise pollution, signal drift and missing values. Based on the distribution of historical data at the same level, missing or abnormal channels are adaptively completed, and a standardized data frame is output. Using standardized data frames as input, time-domain statistical features, frequency-domain energy features, and process state deviation features are extracted respectively, and the three types of features are fused into a multi-dimensional feature vector representing the current level of operation. The anomaly detection model infers from multi-dimensional feature vectors to determine whether there are anomalies in the current data frame, and performs type identification and spatial localization on the detected anomaly events, outputting anomaly type labels and their location identifiers in the physical topology of the corresponding layer. The severity of abnormal events is classified based on the degree of deviation between the anomaly type label and feature vector from the normal baseline; Based on the data quality assessment results and the stability of the model inference, the confidence level of this anomaly detection result is estimated; The anomaly type label, severity level, timestamp and spatial location identifier, and confidence level are uniformly encapsulated into a structured anomaly semantic descriptor.

4. The cross-level intelligent agent collaborative processing method for manufacturing anomalies according to claim 3, characterized in that, The collaborative management agent dynamically constructs a cross-level anomaly propagation graph based on the causal relationships, temporal correlations, and hierarchical propagation paths among anomaly semantic descriptors, including the following steps: The collaborative management agent continuously receives anomaly semantic descriptors reported by local agents at each level. Using anomaly type, level source, and timestamp as index keys, the anomaly semantic descriptors are stored in a sliding time window cache to form an anomaly semantic descriptor sequence that supports time-series querying. Perform temporal correlation analysis on the sequence of abnormal semantic descriptors in the cache window, identify abnormal event pairs that have a sequential triggering relationship in the time dimension, and mark abnormal event pairs with temporal correlation exceeding a preset threshold as candidate association edges; For candidate associated edges involving abnormal event pairs at different levels, the causal direction of the candidate associated edges is determined by combining the physical topology and process dependency of the manufacturing system, distinguishing between root cause anomalies and derived anomalies, and outputting a set of cross-level abnormal associated edges with causal direction labels. Based on the set of cross-level anomaly associated edges, along the hierarchical propagation direction of the equipment layer, production line layer, and factory layer, the propagation path of the anomaly event in the cross-level physical topology is extracted, forming a set of directed propagation paths that characterize the anomaly diffusion trajectory, revealing the complete link from local triggering to cross-level diffusion of the anomaly. Using abnormal events as nodes, cross-level abnormal association edges as connections, and hierarchical propagation paths as the structural skeleton, a cross-level abnormal propagation graph is constructed. As new anomaly semantic descriptors are continuously reported, the nodes, edges, and propagation paths in the graph are incrementally updated to maintain the graph's real-time representation of the current manufacturing anomaly state.

5. The cross-level intelligent agent collaborative processing method for manufacturing anomalies according to claim 4, characterized in that, The collaborative management agent takes the anomaly propagation graph as input, combines the current resource status and production constraints at each level, and generates a cross-level collaborative handling strategy including handling priority, task allocation scheme and collaborative timing through an intent reasoning mechanism. This strategy is then issued to the local agents at the relevant levels in the form of intent commands. The steps include: The collaborative management agent analyzes the current anomaly propagation graph, extracts the root cause anomaly node, derived anomaly node and their hierarchical propagation path, and infers the anomaly status type of the current manufacturing system by combining the severity and confidence of each anomaly node, thus forming an overall intention description for this collaborative handling. The collaborative management agent queries the current resource load, available disposal capacity, and online status of the local agents dynamically registered at each level, and obtains the current production plan, process constraints, and safety boundaries, forming a resource-constraint joint state view. Based on the overall intent description, and considering the severity, transmission risk, and root cause status of each anomalous node in the graph, under the constraints of the resource-constraint joint state view, the anomalous nodes to be dealt with are prioritized, and an ordered list of disposal targets reflecting the urgency and scope of impact is output. Based on the orderly list of disposal targets, each disposal target is assigned to a local intelligent agent at the corresponding level with the corresponding disposal capability. The disposal responsibilities and collaborative relationships of each intelligent agent are clarified, and a cross-level task allocation scheme covering the equipment layer, production line layer and factory layer is generated. Based on the dependencies and resource competition constraints among agents at each level in the task allocation scheme, the start sequence and concurrency boundary of each processing task are planned to generate a collaborative timing plan; The processing priority, task allocation scheme and collaborative timing plan are encapsulated into structured intent instructions, which are then distributed to local agents at each level as target nodes. The intent instructions retain the overall intent description as context, supporting adaptive parsing by local agents during the execution phase.

6. The cross-level intelligent agent collaborative processing method for manufacturing anomalies according to claim 5, characterized in that, After receiving the intent command, each local agent adaptively parses and executes the intent command based on the actual constraints of its level, including: Each level of local agent receives structured intent instructions issued by the collaborative management agent, parses the overall intent description, the handling responsibility of this level, the collaborative timing requirements and related level information contained in the instructions, and restores the context position of this handling task in cross-level collaboration; Before execution, the local intelligent agent collects the current real-time constraints at this level, including the available status of equipment, process window limits, and safe operation boundaries. It compares the real-time constraints with the preset resource assumptions in the intention command and identifies the deviation items between the preset conditions of the command and the current actual state. Based on the deviation items, the handling responsibilities in the intent instructions are adapted to this level. While keeping the overall intent semantics unchanged, the specific operation parameters, execution paths or resource call methods are adjusted to generate an executable handling plan that meets the real-time constraints of this level, and the consistency between the executable handling plan and the collaborative timing requirements is verified. The action is carried out step by step according to the feasible action plan. After each step, the status of the equipment, process parameters and abnormal response data at this level are collected and compared with the expected action effect in real time to form a step-by-step status monitoring record during the execution process. Based on the step-by-step status monitoring records, assess the effectiveness of the current response plan and determine whether the progress of anomaly elimination meets expectations; If the deviation exceeds the acceptable range, mark the type and source of the deviation and trigger the feedback reporting process. If the execution results meet expectations, update the state of the agent at this level and mark the task as completed; The deviation information, along with the current real-time constraint status and execution process monitoring records at this level, are encapsulated into a structured feedback message and reported to the collaborative management agent.

7. The cross-level intelligent agent collaborative processing method for manufacturing anomalies according to claim 6, characterized in that, The steps of continuously monitoring the treatment effect during the execution process, feeding back the effect deviation information to the collaborative management intelligent agent in real time, and dynamically correcting the collaborative strategy based on the effect deviation information include: After the collaborative handling strategy is issued, the collaborative management agent receives structured feedback messages reported by local agents at each level. Based on the collaborative timing plan, the agent aligns and aggregates the feedback messages at each level according to the timing, forming a panoramic view of the multi-level handling effect covering the equipment layer, production line layer and factory layer. The collaborative management agent evaluates the overall effectiveness of cross-level collaborative handling based on a panoramic view of the multi-level handling effects and compares the expected elimination progress of each abnormal node in the current abnormal propagation map. Identify the abnormal nodes with performance deviations and their levels, and locate the key deviation points affecting the collaborative effect by combining the deviation type and the deviation source; For key deviation points, analyze whether the deviation is caused by resource assumption failure, improper task allocation, collaboration timing conflict or hierarchical constraint change. Determine the required policy correction range based on the type of deviation cause, trigger the corresponding level or global collaboration policy correction process, and clarify the correction target and correction boundary. Based on the latest abnormal response data in the feedback messages at each level, the abnormal nodes and associated edges that have been eliminated, added or changed in status in the abnormal propagation graph are updated synchronously to ensure that the revised collaborative strategy is based on the graph that reflects the latest abnormal situation. Based on the updated anomaly propagation map and deviation cause analysis results, the collaborative management agent locally modifies the affected handling priority, task allocation scheme or collaborative timing while maintaining the continuous execution of undevised tasks. It generates incremental correction intention instructions, which are then directed to the relevant local agents at the relevant level, triggering them to re-execute the adaptive parsing and execution process. After the incremental correction intention command is executed, the collaborative management agent continues to gather a new round of feedback messages from relevant levels to verify the execution effect of the correction strategy; If the elimination of all abnormal nodes in the abnormal propagation graph progresses as expected, the current round of cross-level collaborative handling is considered complete, and the collaborative handling result record is output; otherwise, it returns to continue iterative correction.

8. The cross-level intelligent agent collaborative processing method for manufacturing anomalies according to claim 7, characterized in that, In the process of converting detected anomalous events into anomalous semantic descriptors, the overall confidence of each local agent in the anomalous event is calculated as follows: in, The overall confidence level of the anomaly semantic descriptor; The score is the inference output of the anomaly detection model based on the current feature vector, reflecting the model's confidence in the anomaly determination. This is the quality assessment score for the current data frame; a higher value indicates better data integrity and signal-to-noise ratio. This is the data quality sensitivity coefficient, which controls the rate at which data quality affects confidence levels. This represents the current computational resource load rate of the local agent, reflecting the inhibitory effect of resource scarcity on inference reliability. This is the load suppression coefficient; To adapt weights to different levels, the settings are based on the differences in the complexity of the detection models for the equipment layer, production line layer, and factory layer. This represents the number of missing or completed channels in the current data frame, reflecting the degree of integrity loss of the original data. This represents the total number of data channels collected by the agent at this level. The missing channel ratio is the percentage of channels missing; the higher the missing channel ratio, the greater the penalty to the confidence level. This is the channel missing penalty coefficient, which controls the degree to which the missing ratio reduces the confidence level; The historical detection consistency coefficient is calculated based on the stability of the local agent's detection results of the same type of anomaly within a recent time window. A higher value is taken if the same type of anomaly is repeatedly confirmed recently, and a lower value is taken if the detection conclusion is frequently reversed. This reflects the historical reliability of the model under the current working conditions.

9. The cross-level intelligent agent collaborative processing method for manufacturing anomalies according to claim 8, characterized in that, During the dynamic construction of cross-level anomaly propagation graphs, the causal correlation strength between any two anomaly event nodes is calculated as follows: in, , These are two anomalous event nodes in the anomalous propagation graph; The time overlap length between two abnormal events within the time window; The length of the union of the time spans of the two anomalous events is given by the ratio of the two, which reflects the degree of temporal co-occurrence. It represents the difference in the trigger times of two abnormal events, indicating the time interval between them; This is the hierarchical propagation time constant, set based on the typical delay of anomaly propagation between different levels; The topological correlation coefficient is based on... , The connection relationship of the node at the same level in the physical topology of the manufacturing system is assigned a value: a higher value is taken for direct connection and a lower value is taken for indirect connection across levels. , These are abnormal nodes. , The combined confidence level of the corresponding anomaly semantic descriptor, the product of which reflects the joint degree of trustworthiness of the information from the two nodes simultaneously; The joint confidence sensitivity coefficient controls the rate at which the joint confidence level affects the strength of the causal association. When the confidence levels of both nodes are high, this term approaches 1. When the confidence level of either node is low, it significantly suppresses the association strength. The exception type compatibility coefficient is based on... , The causal compatibility of the anomaly type labels in the manufacturing process knowledge base is assigned a value. If there is a known causal relationship between the types, a higher value is taken. If there is no process relationship between the types, a lower value is taken. Domain knowledge is introduced to correct the pure data-driven relationship.

10. A cross-level intelligent agent cooperative processing system for manufacturing anomalies, used to execute the cross-level intelligent agent cooperative processing method for manufacturing anomalies as described in any one of claims 1 to 9, characterized in that, include: Local intelligent agents with autonomous perception capabilities are deployed at the equipment layer, production line layer, and factory layer, respectively. and collaborative management intelligent agents; The local agent is configured as follows: Real-time acquisition of sensor data, process parameters and operating status at the current level; local inference of the acquired data through an anomaly detection model; conversion of detected abnormal events into anomaly semantic descriptors containing anomaly type, severity, spatiotemporal markers and confidence level. Report the abnormal semantic descriptor to the collaborative management agent; The collaborative management agent is configured as follows: Based on the causal relationships, temporal correlations, and hierarchical propagation paths among anomaly semantic descriptors, a cross-level anomaly propagation graph is dynamically constructed. Using the anomaly propagation graph as input, and combining the current resource status and production constraints at each level, a cross-level collaborative handling strategy containing handling priorities, task allocation schemes and collaborative timing is generated through an intent reasoning mechanism, and then issued to the local intelligent agents at the relevant levels in the form of intent commands. The local agent is also configured as follows: Upon receiving the intent instruction, the intent instruction is adaptively parsed and executed based on the actual constraints of this level. During the execution process, the effectiveness of the measures is continuously monitored, and the deviation information is fed back to the collaborative management intelligent agent in real time, so that the collaborative management intelligent agent can dynamically adjust the collaborative strategy based on the deviation information.

Images