Industrial edge intelligent task collaboration method and system based on AI

Abstract

The invention belongs to the technical field of industrial artificial intelligence, and discloses an AI-based industrial edge intelligent task collaboration method and system, and the method comprises the steps: actively generating a main execution DAG and a plurality of probabilistic DAGs based on the probability distribution prediction of a future resource state, achieving the instant switching when a preset failure mode is detected, and greatly improving the system robustness; deep diagnosis is carried out by using a causal inference engine, an intelligent agent is guided to carry out skill specialization and skill pruning, and intelligent ecology capable of self-purification and evolution is constructed; and a cross-topology element learning framework is adopted, so that the system masters the ability of carrying out few-sample rapid adaptation on brand new and unseen tasks. According to the invention, through three innovations of multipath or natural planning, causal-driven skill life cycle management and cross-topology element learning, the effects of nearly zero delay of task recovery, collective intelligent efficient evolution and rapid generalization of new tasks are brought, and the application value of the system in flexible manufacturing and other complex and changeable industrial scenes is enhanced.

Description

Technical Field

[0001] This invention belongs to the field of industrial artificial intelligence technology, specifically relating to an AI-based industrial edge intelligent task collaboration method and system. Background Technology

[0002] Industrial edge intelligence aims to bring AI capabilities down to the production floor to meet the demands for low latency and high reliability. Existing technologies often decompose tasks into directed acyclic graphs (DAGs) and utilize multi-agent systems (MAS) and reinforcement learning (RL) for collaborative scheduling. However, in modern industrial environments characterized by flexible manufacturing, the diverse task types, rapid production pace, and dynamic working environments (such as network jitter and equipment load fluctuations) pose unprecedented challenges to intelligent collaborative systems, highlighting the increasing limitations of existing technologies. In realizing this invention, the inventors discovered at least the following problems in the existing technologies:

[0003] 1. Existing proactive or predictive scheduling methods typically only prevent the most likely type of failure. When other unforeseen or less probable failures occur, the system still needs to spend a lot of time replanning and scheduling, resulting in task interruption and reduced production efficiency, which cannot meet the stringent requirements of business continuity in industrial scenarios.

[0004] 2. In multi-agent collaboration, although knowledge-sharing mechanisms such as federated learning exist, there is a lack of verification of the effectiveness of knowledge. A skill effective in one scenario may result in negative transfer in another, thereby degrading system performance. Existing mechanisms cannot proactively identify and manage such toxic knowledge, leading to inefficient or even misguided evolution of collective intelligence.

[0005] 3. Traditional scheduling strategies are typically optimized for a specific type of task (i.e., a fixed DAG topology). When a production line switches to a completely new product or process, the corresponding task DAG topology changes fundamentally, and the original scheduling strategy often becomes completely ineffective. The system needs to start from scratch and undergo a lengthy learning process to regain a usable level, making it unable to adapt to the rapid changeover requirements of flexible production. Summary of the Invention

[0006] The present invention aims to at least partially solve the aforementioned technical problems. Therefore, the objective of the present invention is to provide an AI-based method and system for collaborative intelligent tasks at the industrial edge.

[0007] The technical solution adopted in this invention is as follows:

[0008] An AI-based method for intelligent task collaboration in industrial edge computing includes the following steps:

[0009] S1. Receive the industrial collaborative task and decompose it into an initial directed acyclic graph (DAG) consisting of multiple sub-task nodes.

[0010] S2. Based on probabilistic predictions of edge environment states, multiple potential failure modes are identified;

[0011] S3. For the multiple potential failure modes, generate a main execution DAG and at least one corresponding probabilistic DAG, wherein the probabilistic DAG is a backup task graph optimized to deal with a specific failure mode;

[0012] S4. A collaborative scheduler is used to allocate and instruct execution agents on edge devices to collaboratively execute sub-tasks according to the main execution DAG;

[0013] S5. During execution, the edge environment is monitored in real time. When an actual failure mode matching one of the potential failure modes is detected, the execution of the main execution DAG is immediately stopped, and the corresponding probabilistic DAG is switched to continue executing the task.

[0014] S6. After the task is executed, the execution data is analyzed using a diagnostic engine based on causal inference. When the root cause of failure is found to be the lack of ability of a certain execution agent in a specific situation, the agent is triggered to distill the general model into an expert skill model.

[0015] S7. When the diagnostic engine for causal inference discovers that a certain shared skill has a negative transfer to some agents, it triggers skill pruning of the skill library of these agents;

[0016] S8. A meta-learning framework is adopted to train on DAG tasks with various different topologies in order to learn a scheduling meta-policy that can quickly adapt to new DAG topologies with few samples.

[0017] Preferably, in S5, switching to the corresponding probabilistic DAG includes: starting execution from a node in the probabilistic DAG that corresponds to the execution progress when the main execution DAG is aborted, and inheriting the results of the subtasks already completed by the main execution DAG.

[0018] Preferably, the step of distilling the general model into an expert skill model in S6 includes: collecting contextual data in the specific context as supervision information, and using the data to distill the knowledge in the general model related to the specific context into a lightweight expert skill model.

[0019] Preferably, in S7, the skill pruning includes: prohibiting a specific agent from loading or calling a shared skill model that has been diagnosed as generating negative transfer.

[0020] Preferably, in S8, the meta-learning framework optimizes the scheduling meta-policy on a large number of historical DAG tasks, so that when faced with a new, unseen DAG topology, it can converge to an efficient scheduling policy with only a small number of scheduling instances.

[0021] Preferably, the method further includes: when the diagnostic engine of causal inference attributes the failure to inaccurate prediction in S2, the execution data that caused the failure is used as a negative sample to optimize the model on which the probabilistic prediction depends.

[0022] Preferably, in S2, the probabilistic prediction of the edge environment state is achieved through a temporal graph neural network.

[0023] Preferably, the industrial collaborative task is multi-process quality inspection or predictive maintenance of equipment groups on a flexible production line.

[0024] This invention also provides an AI-based industrial edge intelligent task collaboration system, comprising:

[0025] The task decomposition module is used to execute step S1 in the above method;

[0026] The probability planning module is used to execute steps S2 and S3 in the above method;

[0027] The collaborative execution and switching module is used to execute steps S4 and S5 in the above method;

[0028] The causal-driven skill management module is used to execute steps S6 and S7 in the above method;

[0029] The cross-topology learning module is used to perform step S8 in the above method.

[0030] Preferably, the causal-driven skill management module is further configured to maintain a skill profile that records the source of each agent's skills, applicable contexts, performance, and negative migration records, in order to support the decision-making process for skill pruning.

[0031] The beneficial effects of this invention are as follows:

[0032] This invention achieves zero-latency contingency plan switching in the event of a failure by pre-generating and caching a probabilistic Directed Acyclic Graph (DAG) for various possible faults. This enables the system to possess unprecedented fault recovery speed and business continuity assurance capabilities, allowing it to maintain an extremely high mission success rate in highly unstable real-world industrial environments.

[0033] Furthermore, the system can not only learn but also proactively differentiate into experts and eliminate harmful knowledge. This adaptive intelligent ecological evolution mechanism greatly enhances the overall performance and stability of collective intelligence, especially when dealing with complex and heterogeneous problems. Attached Figure Description

[0034] Figure 1 This is a flowchart of the AI-based industrial edge intelligent task collaboration method of the present invention.

[0035] Figure 2 This is a functional block diagram of the AI-based industrial edge intelligent task collaboration system of the present invention. Detailed Implementation

[0036] The technical solutions of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0037] It should be understood that, and also noted, in the embodiments, the functions / actions may appear in a different order than those shown in the figures. For example, depending on the functions / actions involved, they may actually be performed substantially concurrently, or sometimes the two figures shown consecutively may be performed in reverse order.

[0038] like Figure 1 As shown, this embodiment uses the visual inspection task of various mobile phone mid-frames on a flexible production line as an example to illustrate the implementation process of the method of the present invention, specifically including:

[0039] S1. Receiving and Decomposing Tasks:

[0040] The system receives the task "Detect the mid-frame of a Type A mobile phone", and the task decomposition module decomposes it into an initial DAG.

[0041] S2-S3. Active multi-path contingency planning:

[0042] Core Implementation: The probabilistic planning module is activated. Its built-in Temporal Graphical Neural Network (T-GNN) analyzes the historical state of each node in the edge network and outputs a probability distribution of the resource state within the next minute. The analysis shows: there is a 70% probability that everything is normal; there is a 20% probability that the GPU of Node-05 (responsible for high-precision size measurement) will be preempted, resulting in insufficient computing power; and there is a 10% probability that the network link from Node-02 to Node-05 will experience severe congestion.

[0043] Generate a multi-path DAG: Based on this prediction, the system generates:

[0044] Main execution DAG: Optimal scheduling is performed based on 70% of normal conditions.

[0045] Contingency DAG-1 (to address insufficient computing power): Replace the high-precision measurement task on Node-05 with a slightly less accurate but faster algorithm, or migrate it to the backup node Node-09.

[0046] Contingency DAG-2 (to address network congestion): Add a data compression subtask to Node-02 to reduce the amount of data transmitted.

[0047] All three DAGs are cached.

[0048] S4. Collaborative Execution:

[0049] Based on the main execution DAG, the collaborative execution and switching module uses its internal collaborative scheduler to allocate each subtask node in the DAG to the corresponding edge agents on the production line (such as edge servers with cameras, robotic arm controllers, etc.) and issue execution instructions. The task then begins to be executed collaboratively along the path of the main DAG.

[0050] S5. Monitoring and Real-time Switching:

[0051] During execution, the system's status monitoring unit continuously receives heartbeat packets and key performance indicators from each edge agent, and monitors the edge environment in real time.

[0052] When the task was halfway through, the monitoring unit detected that the GPU utilization of Node-05 surged to 99% in a short period of time, which matched the preset failure mode due to insufficient computing power.

[0053] The collaborative execution and switching module immediately triggers the switching mechanism: it issues a stop command to the relevant agents executing the main DAG and seamlessly switches to the cached probabilistic DAG-1. The system inherits the results of all completed subtasks and resumes execution directly from the node in the probabilistic DAG-1 that matches the interrupted state (e.g., the backup measurement task on Node-09). The entire switching process takes milliseconds, avoiding the huge overhead of replanning from scratch.

[0054] S6-S7. Self-optimizing closed loop (cause-and-effect guided skill management and system self-learning):

[0055] Core implementation: After the task is completed, regardless of success or failure, the cause-and-effect driven skill management module will be activated to perform in-depth analysis of the execution data.

[0056] Skill Specialization: Suppose a task fails, and the quality inspection report's accuracy is substandard. Causal engine analysis reveals the root cause is that the agent on Node-03 performs poorly with its general defect detection model when handling a specific scenario of high-glare surface reflection. The system automatically triggers skill specialization: collecting all failed samples (high-glare surface images), forcing the Node-03 agent to distill the feature extraction layer from the general model into a specialized expert skill model that suppresses highlights and enhances edges, labeling it with the context: high glare surface.

[0057] Skill pruning: Assuming the aforementioned highlight suppression expert skill is shared, when the Node-08 agent (responsible for detecting planar scratches) invokes it, it actually leads to a decrease in its ability to recognize fine scratches. After several iterations, the causal engine discovers that this skill has a significant negative causal effect on Node-08's task. The system then triggers skill pruning, updating Node-08's skill profile and prohibiting it from invoking this skill again, thereby purifying the collective knowledge base.

[0058] Predictive Model Self-Optimization: Suppose a task interruption is due to an unpredictable network congestion event. Afterwards, the causal engine attributes the failure to this. Further analysis reveals that the root cause of the interruption is not poor execution, but rather that the T-GNN model in S2 failed to predict this congestion pattern. At this point, the system treats this failed prediction case (including the pre-event environmental state sequence and the final congestion result) as a new negative sample, feeding it back to the T-GNN model for incremental training or fine-tuning, thereby improving the accuracy and coverage of its future predictions, forming a complete self-optimization closed loop.

[0059] S8. Cross-topological meta-learning:

[0060] Core implementation: The factory's flexible production line may need to switch between producing three different mobile phone models, A, B, and C, within a single day. The testing task for each model corresponds to a completely different DAG topology.

[0061] Meta-policy learning: The cross-topology meta-learning module works continuously in the background. It collects data pairs of all past A, B, and C type tasks (DAG topology, successful scheduling trajectories). Utilizing the Model-Independent Meta-Learning (MAML) algorithm, it does not learn a specific scheduler for A, B, or C, but rather a scheduling meta-policy. This meta-policy contains meta-knowledge on how to adjust scheduling parameters based on topological features such as the number of nodes, edge density, and critical path length of the graph.

[0062] Rapid generalization with few samples: When the production line receives a brand-new D-type mobile phone inspection task the next day, the system faces an unprecedented DAG topology. At this point, the scheduler, loaded with a meta-policy, no longer needs to start from scratch with random exploration. Utilizing the good initial parameters provided by the meta-policy, it can quickly adapt to the topological characteristics of the D-type task and converge to a near-optimal scheduling policy after only a few (e.g., less than 10) online scheduling attempts. This represents an efficiency improvement of several orders of magnitude compared to traditional RL, which requires thousands of attempts to learn.

[0063] To gain a clearer understanding of the system implementation of this invention, please refer to... Figure 2 This system serves as the physical or logical platform for implementing the aforementioned methods. It can be deployed on cloud servers, edge servers, or a combination thereof, and communicates with on-site edge devices (executing agents) via a network. The system primarily includes: a task decomposition module, a probabilistic planning module, a collaborative execution and switching module, a causal-driven skill management module, and a cross-topology meta-learning module.

[0064] like Figure 2 As shown, the various modules of the system of the present invention will be described in detail below:

[0065] The task decomposition module is the system's entry point. It receives macro-level industrial tasks from the upper-level MES (Manufacturing Execution System) or user input, such as "complete a comprehensive quality inspection of workpieces in batch P008." Internally, this module can be a rule-based parsing engine or a trained large-scale planning model. It parses and decomposes the macro-level task into a series of subtasks with dependencies, outputting them in the format of an initial directed acyclic graph (DAG). For example, comprehensive quality inspection is decomposed into nodes such as loading, surface cleanliness inspection, dimensional measurement, crack detection, and unloading, defining the execution order between them.

[0066] The probabilistic planning module is the core of the system's proactive robustness. It receives the initial DAG output from the task decomposition module. Its core function is to generate a complete and immediately executable multipath planning scheme based on probabilistic predictions of the future state of the environment.

[0067] This module embeds a time-series prediction model and a DAG generator. It continuously interacts with the environmental knowledge graph to acquire historical and real-time status data of edge nodes (such as cameras, robots, and edge servers), including CPU / GPU load, network bandwidth, and device temperature. Based on this data, the time-series prediction model outputs multiple failure modes and their probabilities that may occur within a future period (e.g., the next 5 minutes), for example: 20% probability of insufficient computing power for node A, and 10% probability of congestion for link B.

[0068] For each high-probability failure mode, the DAG generator optimizes the initial DAG to generate a corresponding probabilistic DAG. Simultaneously, it also generates a main execution DAG based on the most likely scenario. Finally, this module packages this set of DAGs (one main DAG and multiple probabilistic DAGs) and sends it to the coordinated execution and switching module.

[0069] The collaborative execution and switching module is responsible for assigning subtasks in the DAG to specific edge agents for execution, monitoring the execution process in real time, and performing pre-plan switching when necessary. The module includes a collaborative scheduler and a state monitoring and switching unit. The scheduler efficiently assigns subtask nodes of the main execution DAG to edge agents based on the meta-policy provided by the cross-topology meta-learning module. The state monitoring and switching unit continuously receives real-time status reports from edge agents and matches them against failure modes preset by the probabilistic planning module.

[0070] When a failure mode is detected (such as insufficient computing power of node A), the unit immediately issues a stop command to all relevant agents and seamlessly switches to the corresponding probabilistic DAG, instructing the agents to continue executing tasks in the new graph from the point of interruption. All data during execution, including success and failure logs, is sent to the causal-driven skill management module.

[0071] The causal-driven skill management module performs in-depth, attributable analysis of task execution results and manages the knowledge (i.e., the skill model of the agent) within the system throughout its entire lifecycle.

[0072] The internal structure and process of a cause-and-effect driven skills management module:

[0073] Causal Inference Engine: Receives execution data from the collaborative execution and switching module. It uses causal discovery algorithms (such as PC algorithm and FCI algorithm) and counterfactual inference to locate the root cause of task failure or inefficiency from complex execution logs, rather than just correlations.

[0074] Skill specialization and pruning: When the root cause is identified as a skill deficiency of an agent in a specific context, the module triggers skill specialization, instructing the agent to distill an expert model using failure samples. When the root cause is identified as a shared skill causing negative transfer to a specific agent, skill pruning is triggered.

[0075] Skill Profile Database: This is a structured database used for fine-grained management of each skill model in the system. The database maintains a record for each skill, including: skill ID, source (which agent generated it in which task), version number, applicable context (e.g., high-gloss reflective surface detection), historical performance (success rate, average time, etc.), and a critical negative migration list. When skill pruning is triggered, the system adds the agent ID that generated the negative migration to the skill's negative migration list, thus preventing that agent from loading this skill in future task scheduling. This profile provides a solid data foundation for the accurate invocation, evaluation, and cleanup of skills.

[0076] Self-optimizing feedback: When the causal inference engine attributes the failure to prediction error, it generates an instruction and sends it to the probabilistic planning module to use this failure case as a new sample to optimize its internal time series prediction model.

[0077] The cross-topology meta-learning module is used to learn a meta-capability of "learning how to schedule," rather than a scheduling strategy specific to a particular task. It operates in the background or offline, collecting vast amounts of historical task data compiled by causal-driven skill management modules, forming data pairs of (DAG topology, efficient scheduling strategy). Using meta-learning algorithms such as MAML, the module extracts a general scheduling meta-strategy from this data spanning different topologies.

[0078] The trained meta-policy is loaded into the scheduler of the cooperative execution and switching module. This enables the scheduler to quickly find an efficient scheduling scheme with minimal trial and error (few-shot learning) when faced with a previously unseen DAG topology, starting from the meta-policy, thereby achieving rapid generalization to new tasks.

[0079] In summary, the system of the present invention, through the close collaboration of the above five modules, forms a complete closed loop from task reception, multi-path planning, robust execution to deep analysis and self-evolution.

[0080] This invention is not limited to the above-described optional embodiments. Anyone can derive other various forms of products under the guidance of this invention. However, regardless of any changes made in their shape or structure, any technical solution that falls within the scope of the claims of this invention shall be protected by this invention.

Claims

1. An AI-based industrial edge intelligent task collaboration method, characterized in that, Includes the following steps: S1. Receive the industrial collaborative task and decompose it into an initial directed acyclic graph consisting of multiple sub-task nodes; S2. Based on probabilistic predictions of edge environment states, multiple potential failure modes are identified; S3. For the multiple potential failure modes, generate a main execution DAG and at least one corresponding probabilistic DAG, wherein the probabilistic DAG is a backup task graph optimized to deal with a specific failure mode; S4. A collaborative scheduler is used to allocate and instruct execution agents on edge devices to collaboratively execute sub-tasks according to the main execution DAG; S5. During execution, the edge environment is monitored in real time. When an actual failure mode matching one of the potential failure modes is detected, the execution of the main execution DAG is immediately stopped, and the corresponding probabilistic DAG is switched to continue executing the task. S6. After the task is executed, the execution data is analyzed using a diagnostic engine based on causal inference. When the root cause of failure is found to be the lack of ability of a certain execution agent in a specific situation, the agent is triggered to distill the general model into an expert skill model. S7. When the diagnostic engine for causal inference discovers that a certain shared skill has a negative transfer to some agents, it triggers skill pruning of the skill library of these agents; S8. A meta-learning framework is adopted to train on DAG tasks with various different topologies in order to learn a scheduling meta-policy that can quickly adapt to new DAG topologies with few samples.

2. The method according to claim 1, characterized in that, In S5, switching to the corresponding probabilistic DAG includes: starting execution from a node in the probabilistic DAG that corresponds to the execution progress when the main execution DAG is aborted, and inheriting the results of the subtasks that have been completed in the main execution DAG.

3. The method according to claim 1, characterized in that, The step of distilling the general model into an expert skill model in S6 includes: collecting contextual data in the specific context as supervision information, and using the data to distill the knowledge in the general model related to the specific context into a lightweight expert skill model.

4. The method according to claim 1, characterized in that, In S7, the skill pruning includes: prohibiting a specific agent from loading or calling a shared skill model that has been diagnosed as generating negative transfer.

5. The method according to claim 1, characterized in that, In S8, the meta-learning framework optimizes the scheduling meta-policy on a large number of historical DAG tasks, so that when faced with a new, unseen DAG topology, it can converge to an efficient scheduling policy with only a small number of scheduling instances.

6. The method according to claim 1, characterized in that, The method further includes: when the diagnostic engine of the causal inference attributes the failure to the inaccuracy of the prediction in S2, the execution data that caused the failure is used as a negative sample to optimize the model on which the probabilistic prediction depends.

7. The method according to claim 1, characterized in that, In S2, the probabilistic prediction of the edge environment state is achieved through a temporal graph neural network.

8. The method according to claim 1, characterized in that, The industrial collaborative task refers to multi-process quality inspection or predictive maintenance of equipment groups on flexible production lines.

9. An AI-based industrial edge intelligent task collaboration system, characterized in that, include: The task decomposition module is used to execute step S1 as described in claim 1; The probability planning module is used to perform steps S2 and S3 as described in claim 1; A collaborative execution and switching module is used to execute steps S4 and S5 as described in claim 1; A causal-driven skill management module is used to perform steps S6 and S7 as described in claim 1; A cross-topology learning module is used to perform step S8 as described in claim 1.

10. The system according to claim 9, characterized in that, The causal-driven skill management module is further configured to maintain a skill profile that records the source of each agent's skills, applicable contexts, performance, and negative transfer records, in order to support the decision-making process for skill pruning.

Images