Industrial manufacturing operation management method based on hybrid enhanced intelligence

Abstract

The invention relates to the technical field of industrial production management, and discloses an industrial manufacturing operation management method based on hybrid enhanced intelligence, and the method comprises the following steps: S1, constructing a multi-source knowledge base, and building a task matching initial model based on multiple dimensions; s2, performing autonomous task planning, task decomposition, task allocation and task release based on the task matching initial model, and updating a work allocation method knowledge base according to a real-time task execution condition; and S3, according to the updated work distribution method knowledge base and real-time feedback, cooperatively adjusting a management strategy so as to ensure that the knowledge base is synchronously updated with an actual demand, and optimizing a task distribution strategy. By constructing a knowledge-driven intelligent decision closed loop, efficient and collaborative operation, dynamic resource optimization, management cost reduction and continuous self-adaption of the system are realized, and the overall manufacturing intelligence level is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial production management, and specifically to an industrial manufacturing operation management method based on hybrid augmented intelligence. Background Art

[0002] As modern industrial manufacturing advances toward intelligent and customized processes, production operations management is becoming increasingly complex. Enterprises face the daunting challenge of efficiently integrating multi-source information, precisely scheduling heterogeneous resources, rapidly responding to market trends, and continuously optimizing operational efficiency. Traditional management models and technologies are increasingly limiting in addressing these challenges, making them unable to meet the modern manufacturing industry's demands for high efficiency, flexibility, and refined management.

[0003] In existing technologies, industrial manufacturers generally use enterprise resource planning (ERP) systems, manufacturing execution systems (MES), and various types of project management or office automation software to assist with operations management. These systems achieve a certain degree of digitization and standardization of business processes, such as using preset rules for production scheduling, material requirement calculations, and work order issuance. However, these systems often rely on manual configuration and offline analysis and adjustment to deeply understand tasks, make dynamic decisions in complex situations, and integrate cross-domain knowledge, lacking inherent intelligence and adaptability.

[0004] The shortcomings of existing technologies are mainly reflected in the following aspects: First, their ability to intelligently decompose complex tasks, dynamically allocate resources, and optimize resource matching under multi-dimensional constraints is limited. They rely heavily on static rules or manual intervention, making it difficult to adapt to the real-time changes and diverse needs of the production site, resulting in inefficient resource allocation. Secondly, various systems and AI applications often present information silos or functional fragmentation. There is a lack of a unified knowledge hub that can continuously learn and evolve to support global, collaborative intelligent decision-making, causing adjustments to operational strategies to lag behind actual needs and making it difficult to achieve closed-loop optimization based on real-time feedback. Furthermore, the decision-making process of traditional systems often lacks transparency, making it difficult to trace key factors and systematically extract and consolidate best practices from historical experience to guide future operations management.

[0005] Therefore, the present invention proposes an industrial manufacturing operation management method based on hybrid augmented intelligence to address the deficiencies of the existing technology. Summary of the Invention

[0006] In response to the shortcomings of the existing technology, the present invention provides an industrial manufacturing operation management method based on hybrid augmented intelligence, which solves the shortcomings of the traditional industrial manufacturing operation management model in task decomposition, resource allocation, real-time linkage and overall coordination. By introducing explainable artificial intelligence and hybrid augmented intelligence technology, it realizes the collaborative management and optimization of the entire process and all factors, thereby improving production efficiency, reducing management costs and enhancing adaptability.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: an industrial manufacturing operation management method based on hybrid augmented intelligence, comprising the following steps: S1. Build a multi-source knowledge base and establish an initial task matching model based on multiple dimensions; S2. Based on the task matching initial model, autonomous task planning, task decomposition, task allocation and task release are performed, and the work allocation method knowledge base is updated according to the real-time task execution status; S3. Based on the updated work allocation method knowledge base and real-time feedback, collaboratively adjust the management strategy to ensure that the knowledge base is updated synchronously with actual needs and optimize the task allocation strategy.

[0008] Preferably, the step S1 includes: Construct a multi-source knowledge base, which includes a task model, a resource model, and a technical database: The task model uses AI knowledge base tools to vectorize the enterprise's existing workflow; the resource model forms the initial knowledge of resource library data by sorting out the enterprise's resource configuration; the technical database is obtained by vectorizing the enterprise's existing technical data; An initial task matching model is established based on at least the dimensions of organization, qualification, historical quality, and load.

[0009] Preferably, in step S1, the multi-source knowledge base includes a relational database, a NoSQL database and a graph database, which are used to store structured data, unstructured data and knowledge graphs respectively.

[0010] Preferably, the step S2 includes: The demander initiates the task; The project manager AI performs autonomous task planning, task decomposition, task allocation, work knowledge addition, and task release based on the task model. The project manager AI uses a combination of rule-based reasoning and case-based reasoning to allocate tasks. The operations manager reviews the task decomposition and allocation results of the project portfolio manager AI and adjusts resource allocation based on actual conditions; The worker receives the task and provides feedback on the task execution status in real time, including the progress of the task completion and any problems encountered. The work allocation AI agent updates the knowledge base of work allocation methods based on real-time feedback.

[0011] Preferably, the step S3 includes: Operations managers collaborate with AI agents to regularly review process bottlenecks and adjust strategies to keep the knowledge base updated with actual needs, achieving continuous improvement. Through continuous learning and optimization of algorithms, the quality and efficiency of task completion are improved to achieve optimal resource allocation. The continuous learning adopts reinforcement learning algorithms to continuously optimize task allocation strategies based on feedback from workers.

[0012] The present invention also provides an industrial manufacturing operation management system based on hybrid augmented intelligence, comprising: Multi-source knowledge base module, used to store task models, resource models and technical databases; The project manager AI module is used to autonomously plan tasks, decompose tasks, assign tasks, add work knowledge, and publish tasks based on the task model; The operations manager interface module is used for the operations manager to view the task decomposition and allocation results of the project portfolio manager AI and adjust resource allocation according to actual conditions; The worker interface module is used for workers to receive tasks and provide feedback on real-time task execution status; The work allocation AI agent module is used to update the work allocation method knowledge base based on real-time feedback.

[0013] Preferably, it further includes a self-learning module, which is configured to continuously iteratively optimize the task model and resource matching algorithm based on accumulated data and feedback.

[0014] Preferably, it further includes an explainability module, which is configured to trace and analyze key factors of core decisions and the explainability module process.

[0015] Preferably, it also includes an AI workflow and intelligent agent creation module, which is configured to define the logical framework of the AI ​​workflow and create and manage various types of intelligent agents.

[0016] Preferably, it also includes at least one application programming interface, which is used to establish a communication connection between the industrial manufacturing operation management system based on hybrid augmented intelligence and the enterprise's existing enterprise resource planning system, manufacturing execution system or supply chain management system for data exchange.

[0017] The present invention provides an industrial manufacturing operations management method based on hybrid augmented intelligence. It has the following beneficial effects: 1. This invention achieves highly automated and dynamically optimized scheduling of industrial manufacturing operations tasks by building a comprehensive multi-source knowledge base and utilizing a project manager AI module for intelligent task planning, decomposition, and task allocation based on a multi-dimensional matching model. This, combined with real-time worker feedback and continuous optimization mechanisms such as reinforcement learning employed by the self-learning module, enables highly automated and dynamically optimized scheduling of industrial manufacturing operations tasks. This closed-loop management approach, based on hybrid augmented intelligence, enables more accurate and efficient task execution, enabling rapid response to production demands and market changes, significantly improving overall production and operational efficiency and agility.

[0018] 2. This invention establishes a unified and continuously evolving multi-source knowledge base, integrating structured task models, resource models, and vectorized technical data. It then utilizes a work assignment AI agent and a self-learning module to continuously iterate and update this knowledge base based on collaboratively adjusted strategies and real-time feedback from operations managers. This dynamic knowledge management and accumulation mechanism ensures that every task assignment and management decision is based on more comprehensive, accurate, and up-to-date information, significantly improving the quality and scientific nature of industrial manufacturing operations management decisions and enabling management strategies to evolve in sync with actual needs.

[0019] 3. This invention uses the Project Manager AI module to perform preliminary screening of task matching models based on multi-dimensional factors (such as organization, qualifications, historical quality, and workload). Combined with the self-learning module, this system uses advanced algorithms such as reinforcement learning to continuously and deeply optimize resource allocation strategies. This enables more optimal allocation and utilization of various manufacturing resources, including production equipment and manpower. This intelligent resource scheduling minimizes idle resources, conflicts, and improper allocations, optimizes the overall operational flow, effectively reduces operating costs caused by improper resource utilization, and improves the input-output efficiency of assets.

[0020] 4. This invention utilizes an explainability module to effectively trace and clearly display the core decision logic and key influencing factors of the project manager AI and self-learning module when making critical task allocation decisions or optimizing management strategies. This transparent approach to the AI ​​decision-making process significantly enhances operations managers' understanding of and trust in hybrid augmented intelligence systems. This not only facilitates effective human oversight, intervention, and system correction, but also promotes the reliable implementation and widespread application of intelligent technologies in complex industrial manufacturing operations management scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 is a flow chart of the method of the present invention; Figure 2 This is a system architecture diagram of the present invention. DETAILED DESCRIPTION

[0022] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the present specification. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0023] Please see the attached Figure 1 , an embodiment of the present invention provides an industrial manufacturing operation management method based on hybrid augmented intelligence, comprising the following steps: S1. Build a multi-source knowledge base and establish an initial task matching model based on multiple dimensions; S2. Based on the task matching initial model, autonomous task planning, task decomposition, task allocation and task release are carried out, and the work allocation method knowledge base is updated according to the real-time task execution status; S3. Based on the updated work allocation method knowledge base and real-time feedback, collaboratively adjust management strategies to ensure that the knowledge base is updated in sync with actual needs and optimize task allocation strategies.

[0024] Regarding step S1, in this embodiment, in order to realize intelligent management and optimization of the industrial manufacturing operation process, the initialization phase: First, execute step S1 to build a comprehensive multi-source heterogeneous knowledge base, and on this basis establish a preliminary task matching model to lay a solid data and model foundation for subsequent dynamic decision-making and continuous learning.

[0025] The process of building a multi-source knowledge base aims to integrate various information resources related to manufacturing operations to form a unified, machine-understandable knowledge representation. This is the core support for the intelligent management of this invention. The multi-source knowledge base includes a task model, a resource model, and a technical database.

[0026] The task model aims to structure and standardize the representation of the complex workflows and task requirements of an enterprise. Preferably, an AI knowledge base tool (for example, combining natural language processing and process mining technology) is used to analyze the enterprise's existing workflow documents, historical work orders, and operating procedures. The tool extracts task nodes, logical relationships between tasks (such as sequence, parallelism, and dependencies), and task attributes (such as type, priority, required skills, quality standards, etc.), and generates vectorized versions. This vectorization process can, for example, use word embedding or document embedding technology to map tasks or their attributes to a high-dimensional vector space, so that semantically similar tasks are closer in the vector space, providing an effective computational basis for subsequent task understanding, similarity comparison, and intelligent matching. The task model generated in this way can dynamically reflect the actual operation mode of the enterprise and has good scalability.

[0027] The resource model aims to comprehensively organize and digitally describe all resources available for production operations within an enterprise. These resources include, but are not limited to, production equipment, human resources (including their skill levels and qualifications), material resources, and other supporting resources. By systematically organizing the enterprise's resource allocation, it collects each resource's technical parameters, current status (e.g., idle, occupied, under repair), geographic location, cost information, and maintenance records, creating a structured resource library. This data constitutes the initial knowledge of the resource model. This resource model not only records the static properties of resources but also provides a foundational framework for real-time updates of their dynamic status, making it key to achieving optimized resource allocation and efficient scheduling.

[0028] The technical database aims to digitally integrate and intelligently utilize the vast amount of technical documents, process files, operating manuals, quality standards, historical failure cases, and best practices accumulated by the enterprise. After content extraction and preprocessing of the enterprise's existing large amount of multi-format technical data, it is also obtained through vectorization processing. For example, text analysis technology can be used to extract key information, and then combined with knowledge graph construction technology to extract concepts, entities, and their relationships in the document and convert them into vector representations or graph structures. This vectorization processing enables the system to efficiently search, semantically understand, and intelligently answer questions for unstructured or semi-structured technical data, providing immediate and effective knowledge support for solving technical problems encountered in the production process and assisting decision-making.

[0029] To effectively store and manage the aforementioned diverse types and structures of knowledge data, in this embodiment, the multi-source knowledge base preferably utilizes a hybrid storage architecture, specifically comprising relational databases, NoSQL databases, and graph databases. This hybrid architecture leverages the strengths of different database types: relational databases primarily store highly structured data, such as basic task information, resource ledgers, and personnel records; NoSQL databases store semi-structured or unstructured data, such as log data generated during task execution, real-time data collected by equipment sensors, and vectorized technical documentation; and graph databases focus on storing and managing knowledge graph data, which consists of entities (such as tasks, resources, and skills) and the complex relationships between them (such as task dependencies and resource assignments). Graph databases enable efficient complex association queries and path analysis, which are crucial for uncovering the deep connections hidden in data and supporting advanced reasoning. By integrating these three database types, supplemented by a unified data interface and data synchronization mechanism, a multi-source knowledge base system is constructed that comprehensively, efficiently, and flexibly manages manufacturing operations knowledge.

[0030] After building a comprehensive multi-source knowledge base, another key step in step S1 is to establish an initial task matching model based on multiple dimensions. This initial model aims to make preliminary and reasonable matching recommendations based on task requirements and resource attributes, providing high-quality candidate solutions for subsequent manual confirmation or further intelligent optimization.

[0031] The establishment of the initial task matching model fully considers multiple key factors that affect the efficiency and quality of task execution. In this embodiment, the model is constructed based on at least the organizational dimension, qualification dimension, historical quality dimension, and load dimension: The organizational dimension mainly considers the rationality of task allocation at the enterprise organizational structure level, and evaluates the degree of match between the organizational ownership of resources (especially human resources) and task requirements to ensure clear responsibilities, smooth communication and efficient collaboration.

[0032] The qualification dimension focuses on whether a resource possesses the necessary qualifications and authorizations to perform the task. This can include specific skills certifications and job qualifications for human resources, or precision levels and specific process licenses for equipment resources. The model rigorously verifies that candidate resources meet the specific qualification requirements outlined in the task.

[0033] Historical quality is a key empirical consideration. The model considers the quality performance of candidate resources (such as specific employees or equipment) when performing similar tasks in the past, such as product qualification rates and rework rates, prioritizing those with a high historical record of task completion quality.

[0034] The load dimension focuses on the current and expected workload of resources. The model assesses the current busyness of candidate resources (such as personnel or equipment) and the length of their already accepted task queues. The goal is to avoid assigning new tasks to already overloaded resources, ensuring timely task initiation and smooth progress, while also achieving balanced resource utilization.

[0035] It should be noted that the above four dimensions are the basic considerations for establishing the initial model of task matching. It is understandable that according to the complexity and specific needs of the actual application scenario, the matching dimensions can be further expanded or refined. For example, the cost dimension, time urgency dimension, geographic location dimension, resource coordination dimension, etc. can be added.

[0036] In practice, the initial task matching model can be expressed as a multi-factor evaluation function or a set of matching rules. For example, a corresponding evaluation method and weight coefficient can be set for each dimension to comprehensively calculate the matching score between the task and each candidate resource. A schematic matching degree calculation method can be expressed as: ; Where, To assign to a specific task With specific candidate resources The total initial match score, which quantifies the resource's initial suitability for the task; Represents a specific task to be assigned, which includes attribute information such as task type, priority, required skills, etc. Represents a specific candidate resource, which contains attribute information such as resource type, current status, capability parameters, etc. An index representing an evaluation dimension, which includes at least organizational dimension, qualification dimension, historical quality dimension and load dimension; Indicates in Under each evaluation dimension, specific tasks to be assigned With specific candidate resources The matching degree or similarity evaluation value between them reflects the satisfaction of the resource with the task requirements in a specific dimension, which is usually normalized; Indicates the The preset or dynamically adjusted weight coefficients of the evaluation dimensions reflect the relative importance of different evaluation dimensions in the comprehensive decision-making; It means summing up the weighted matching degrees of all evaluation dimensions to obtain a comprehensive weighted score; Represents a comprehensive evaluation function, which can be a simple weighted sum or a more complex nonlinear function. The system selects resources with high scores or those that meet the preset threshold as initial matching suggestions.

[0037] Through the above-mentioned process of constructing a multi-source knowledge base and establishing an initial task matching model based on multiple dimensions, the method of the present invention provides comprehensive, accurate, and structured data input and preliminary, guiding matching solutions for the subsequent hybrid enhanced intelligent decision-making center to perform dynamic task planning, decomposition, allocation, and continuous self-learning and optimization.

[0038] Step S2, in this embodiment, is the core stage of dynamic adjustment and execution. It aims to achieve intelligent planning, decomposition, allocation, and release of tasks based on existing models and knowledge. Combined with real-time execution feedback, it continuously updates knowledge related to work assignments, forming a closed-loop, adaptive manufacturing operations management system.

[0039] The specific process of step S2 begins with the requester initiating a task. When an operational demand, such as a new order, production instruction, or maintenance request, is received, the task request, including a basic description and objectives, is entered into the system as the starting point for subsequent intelligent processing.

[0040] The Program Manager AI then steps in as the core intelligent decision-making unit. It first autonomously plans tasks based on the task model, invoking the vectorized task model built in S1 to deeply analyze requirements and plan the overall execution path and strategy for the tasks. For complex tasks, the Program Manager AI then decomposes them into smaller, more manageable and assignable subtasks or process packages based on the task model, reducing complexity and facilitating refined matching.

[0041] In the critical task allocation process, the Program Manager AI refers to S1's task matching initial model and preferably uses a combination of rule-based reasoning (RBR) and case-based reasoning (CBR) to allocate tasks. The rule-based reasoning module generates allocation recommendations based on predefined business rules (such as resource capabilities and task priority constraints). The case-based reasoning module retrieves historical successful cases similar to the current task from the knowledge base as a reference. To integrate the two reasoning results, the Program Manager AI's reasoning engine uses a dynamic weight allocation mechanism, which can be adjusted based on factors such as historical decision accuracy, task type adaptability, the amount of relevant case data, and task time sensitivity. The final decision plan can be generated through a comprehensive scoring model, for example: Final decision score = (rule reasoning weight Rule reasoning score) + (case reasoning weight Case-Based Reasoning Score); The system selects the solution with the highest overall score.

[0042] Once the allocation plan is finalized, the Project Manager AI performs work knowledge addition, intelligently retrieving and filtering relevant operating procedures, process parameters, quality standards, and other knowledge from the technical database built by S1, appending them to the task instructions to provide guidance to the workers. The Project Manager AI then issues the task, assigning it to the designated execution resource and simultaneously updating the system task status.

[0043] To ensure the rationality of decision-making and address the complexity of the real-world environment, human collaboration is introduced. Operations managers can view the task decomposition and allocation results of the program manager AI through the system interface and have the authority to adjust resource allocation based on actual circumstances. For example, they can intervene when an unexpected situation arises that the AI ​​model does not cover, or when managers, based on their implicit experience, determine that adjustments are necessary. This human-machine collaborative mechanism enhances the robustness of the system, and all manual adjustments are recorded for subsequent learning of the AI ​​model.

[0044] After a task is issued, workers receive instructions containing additional work knowledge and provide real-time feedback on the task's progress during execution. This feedback includes at least the progress of the process (such as completion percentage and key milestones) and any problems encountered (such as equipment failure, material shortages, and quality anomalies). Workers can conveniently input this information.

[0045] To achieve continuous learning and self-optimization, the work assignment AI agent continuously monitors and collects real-time feedback from workers and manual adjustments made by operations managers. Based on this real-time feedback, it updates the work assignment method knowledge base. This update includes, but is not limited to, revising or adding rules to the rule base; converting successful or failed execution instances into new cases to enrich the case base; and adjusting parameters for similarity calculations in case-based reasoning or the dimension weights of the initial task matching model.

[0046] Through a series of operations in step S2, the method of the present invention not only realizes autonomous task management based on AI, but also incorporates artificial intelligence for supervision and optimization, and drives the continuous evolution of the knowledge base through real-time feedback.

[0047] In this embodiment, step S3 represents the system's continuous optimization phase. Its core is to dynamically adjust management strategies based on the updated work allocation method knowledge base and real-time feedback, through human-machine collaboration and autonomous learning. This ensures that the knowledge base is updated synchronously with actual operational needs, and further optimizes task allocation strategies to continuously improve overall operational efficiency.

[0048] Key implementation aspects of step S3 include: First, the collaborative working mechanism between operations managers and AI agents is emphasized. Operations managers (such as production supervisors and process engineers) leverage their domain knowledge and experience to interact and collaborate with AI agents (such as work allocation AI agents or strategy analysis and optimization AI modules) on a regular basis or when key events trigger them.

[0049] A core activity is regularly reviewing process bottlenecks. Using data analysis capabilities, AI agents identify potential bottlenecks from task execution data, resource utilization data, and worker feedback, and present them to operations managers. Both parties work together to identify the root causes of the bottlenecks.

[0050] Based on this, operations managers and AI agents collaboratively adjust management strategies. These adjustments may involve: revising standard operating procedures and feeding them back into S1's task model; adjusting resource allocation priorities or the HR skills matrix; and changing the default processing logic for specific tasks, all of which impact the relevant models and rules in the knowledge base. This collaborative adjustment aims to synchronize the knowledge base with actual needs, achieving continuous improvement and ensuring the vitality and adaptability of the system's knowledge base.

[0051] Secondly, to fundamentally improve the intelligence level and operational performance of task allocation, this step focuses on optimizing core algorithms, especially task allocation strategies, through continuous learning. The goal is to improve the quality and efficiency of task completion and achieve optimal resource allocation.

[0052] Continuous learning preferably uses reinforcement learning (RL) algorithms. A dedicated RL agent continuously optimizes task allocation strategies based on worker feedback (such as process completion progress, encountered problems, and quality evaluations) and objective task execution data recorded by the system (such as actual time and resource consumption).

[0053] The key components of the reinforcement learning algorithm in this scenario include: State space (S): The state of the environment perceived by the RL agent, preferably including the characteristics of the task to be assigned, the real-time status of available resources, the overall state of the production environment, and a summary of historical execution data.

[0054] Action space (A): The decisions that the RL agent can take, primarily including assigning subtasks to specific resource combinations, but can also be extended to deciding to postpone tasks, choose different process paths, etc.

[0055] Reward function (R): quantifies the degree of success after taking an action in a specific state and guides the direction of learning. The reward function aims to reflect the quality of task completion, efficiency, and optimal resource allocation, and incorporates worker feedback. An exemplary reward function can be designed as a multi-objective weighted form: ; Where, Represents the reward component related to task execution efficiency, such as the difference between actual working hours and planned working hours; Represents the reward component related to the quality of task completion, such as product qualification rate or defect-free operation time; Represents the reward component related to the completion of the task on time, such as a timing calculation based on whether the delivery deadline is met; Represents the reward component related to the rationality of resource utilization, such as calculation based on equipment utilization or personnel load balance; Represents the penalty items related to negative consequences during task execution, such as task delays, defective products, or resource conflicts; for The weight coefficient of for The weight coefficient of for The weight coefficient of for The weight coefficient of for The weight coefficient of .

[0056] Reinforcement learning algorithm: Considering the complexity of state and action space, it is preferred to use a deep reinforcement learning algorithm based on value function (such as DQN and its variants) or policy gradient (such as Actor-Critic method). Taking DQN as an example, the system constructs a deep neural network ( Network) approximates the optimal action-value function , the network input is the state, and the output is the action Values. Techniques such as experience replay and target networks are used during training. Learning parameters such as learning rate, discount factor, and exploration strategy are carefully designed and tuned.

[0057] By continuously interacting with the environment, collecting rewards, and updating its internal model using RL algorithms, the RL agent can gradually learn a more optimal task allocation strategy. This strategy can autonomously discover deeper patterns from large-scale interaction experience, achieving dynamic and adaptive optimization of task allocation.

[0058] In summary, step S3 ensures that the method of the present invention not only efficiently handles the current task, but also continuously learns and evolves, adapts to the dynamic manufacturing environment, and continuously improves operational goals, thereby achieving hybrid enhanced intelligence through strategy adjustment of human-machine collaboration and autonomous strategy optimization based on reinforcement learning.

[0059] Please see the attached Figure 2 The present invention also provides an industrial manufacturing operation management system based on hybrid augmented intelligence, comprising: Multi-source knowledge base module, used to store task models, resource models and technical databases; The project manager AI module is used to autonomously plan tasks, decompose tasks, assign tasks, add work knowledge, and release tasks based on the task model; The operations manager interface module is used for operations managers to view the task decomposition and allocation results of the project portfolio manager AI and adjust resource allocation according to actual conditions; The worker interface module is used for workers to receive tasks and provide feedback on real-time task execution status; The work allocation AI agent module is used to update the work allocation method knowledge base based on real-time feedback.

[0060] It also includes a self-learning module, which is configured to continuously iteratively optimize the task model and resource matching algorithm based on accumulated data and feedback.

[0061] It also includes an explainability module that is configured to trace back and analyze the key factors of core decisions and their processes.

[0062] It also includes an AI workflow and intelligent agent creation module, which is configured to define the logical framework of the AI ​​workflow and create and manage various types of intelligent agents.

[0063] It also includes at least one application programming interface, which is used to establish a communication connection between an industrial manufacturing operation management system based on hybrid augmented intelligence and an enterprise's existing enterprise resource planning system, manufacturing execution system or supply chain management system for data exchange.

[0064] For the Multi-Source Knowledge Base module: In this embodiment, the multi-source knowledge base module is the core data and knowledge storage hub of the entire system. This module is configured to build and store a comprehensive, multi-source heterogeneous knowledge base. Specifically, the knowledge base contains at least: Task model: used to store structured task information generated by vectorizing the company's existing workflow through AI knowledge base tools, including task nodes, logical relationships between tasks, task attributes (such as type, priority, required skills, etc.), etc.

[0065] Resource model: used to store resource library data formed by organizing the enterprise resource configuration, including detailed technical parameters, current status, geographical location, cost information, maintenance records, and skill qualifications of production equipment, human resources, material resources, etc.

[0066] Technical Database: This module is used to store knowledge content obtained through vectorized processing of the company's existing technical data (such as operating procedures, process documents, quality standards, historical failure cases, best practices, etc.). Preferably, the multi-source knowledge base module adopts a hybrid storage architecture, for example, combining relational databases (for storing structured data), NoSQL databases (for storing unstructured or semi-structured data such as logs, sensor data, and vectorized documents), and graph databases (for storing knowledge graphs, representing entities and their complex relationships) to efficiently manage different types of data. This module provides data support and a knowledge foundation for the system's other AI modules, corresponding to the knowledge base construction portion of the aforementioned method S1.

[0067] Portfolio Manager AI Module: In this embodiment, the Project Manager AI module is the core intelligent agent for the system to perform autonomous task planning and execution. This module is configured to perform the following operations based on the task model and other information provided by the multi-source knowledge base module: Autonomous task planning: In-depth analysis of received task requirements and overall execution path planning.

[0068] Task decomposition: Decompose complex tasks into a series of smaller, more manageable subtasks or process packages based on the task model.

[0069] Task allocation: Using the built-in reasoning engine (preferably a combination of rule-based reasoning and case-based reasoning, and referring to the results of the task matching initial model), the most appropriate execution resource is selected for each subtask. The decision-making process can consider multiple factors such as organization, qualifications, historical quality, load, etc., and can use methods such as: Final decision score = (rule reasoning weight Rule reasoning score) + (case reasoning weight Case-Based Reasoning Score); comprehensive evaluation in a comprehensive manner.

[0070] Work knowledge addition: Intelligently retrieve and add relevant operating procedures, process parameters and other knowledge to task instructions from the technical database.

[0071] Task release: The completed task is formally assigned to the designated execution resource.

[0072] This module corresponds to the core AI decision-making and execution part in the aforementioned method S2.

[0073] Operations manager interface module: In this embodiment, the operation manager interface module is the key interactive interface for achieving human-machine collaboration and management intervention. This module is configured as follows: Display task decomposition and allocation results: This allows operations managers (such as workshop supervisors and production planners) to clearly view the task decomposition plan, resource allocation suggestions and their basis generated by the Program Manager AI module.

[0074] Support for manual adjustments: Allows operations managers to modify, confirm, or reject the resource allocation plan given by AI based on their actual experience, judgment of emergencies, or more advanced strategic considerations.

[0075] Strategy review and adjustment interface: Preferably, this module also provides an interface for operations managers and AI agents to collaboratively review process bottlenecks and participate in adjusting management strategies, such as modifying business rules, correcting case weights, etc., to ensure that the knowledge base is updated synchronously with actual needs.

[0076] This module corresponds to the part of operations manager participation and collaboration in the aforementioned methods S2 and S3.

[0077] Worker interface module: In this embodiment, the worker interface module is the channel for information exchange between the task execution layer and the system. This module is configured as follows: Task reception and display: It allows frontline workers (such as operators and technicians) to receive detailed task instructions, additional work knowledge, and related process documents on their work terminals (such as tablet computers, industrial PDAs, equipment control panels, etc.).

[0078] Real-time task execution feedback: Allows workers to quickly and easily provide feedback on the real-time status of tasks, including process completion progress (such as completion percentage, key milestones achieved), actual working hours, material consumption, and problems encountered during execution (such as equipment failure, quality anomalies, safety hazards, etc.).

[0079] This module corresponds to the part in the aforementioned method S2 where the worker receives tasks and provides feedback on the execution status.

[0080] Work allocation AI agent module: In this embodiment, the work allocation AI agent module is a key component to achieve continuous learning of the system and dynamic updating of the knowledge base. This module is configured as follows: Collect and analyze feedback: Continuously monitor and collect real-time task execution feedback from the worker interface module and manual adjustment records from the operations manager interface module.

[0081] Update the work allocation method knowledge base: Based on collected feedback and execution results, update and optimize the knowledge related to task allocation in the multi-source knowledge base module (such as the rule base, case base, resource capability assessment model, and task matching model parameters). For example, modify or add inference rules, convert successful or failed execution instances into new cases, and adjust case similarity calculation parameters.

[0082] This module corresponds to the content about knowledge base updating and preliminary learning at the end of S2 and the beginning of S3 of the aforementioned method.

[0083] Self-study modules: In this embodiment, in order to further enhance the intelligence level and adaptive capability of the system, a self-learning module is preferably included. The self-learning module is configured as follows: Deep learning based on accumulated data and feedback: Utilize the extensive historical task data, resource usage data, worker feedback, manual adjustment records, and the final results of task execution (such as quality, efficiency, and cost) accumulated over the long term of the system to conduct deeper machine learning.

[0084] Iteratively optimize the task model and resource matching algorithm: This module preferably uses a reinforcement learning algorithm (e.g., as discussed in Method S3 above, which optimizes policy based on state, action, and reward functions) to iteratively optimize the task allocation strategy, the understanding of task characteristics in the task model, the assessment of resource capabilities in the resource model, and the matching algorithm between tasks and resources. The goal is to enable the system to autonomously discover optimal operating models and resource allocation solutions.

[0085] This module corresponds to the core content of reinforcement learning and continuous algorithm optimization in the aforementioned method S3.

[0086] Interpretability Module: In this embodiment, in order to enhance the user's trust and understanding of the AI ​​decision-making process, an explainability module is preferably also included. The explainability module is configured as follows: Trace the decision path: When the project manager AI module or self-learning module makes a key decision (such as a specific task assignment or strategy adjustment suggestion), the module can record and trace back the key data, activated rules, reference cases or important features in the model on which the decision relied.

[0087] Analyze and present key factors: This approach presents complex AI decision-making processes to operations managers or system analysts in a human-understandable format (e.g., natural language explanations, visual charts, and ranking of influencing factors), explaining why a particular decision was made over other options and the contribution of each factor to the decision. This helps users understand AI behavior, identify potential issues, and implement more effective intervention and tuning.

[0088] AI workflow and agent creation module: In this embodiment, in order to provide flexibility and scalability of the system, it is preferred to further include an AI workflow and agent creation module. The AI ​​workflow and agent creation module is configured as follows: Define the logical framework of AI workflows: This allows system administrators or advanced users to define and configure the interaction logic, data flow paths, and trigger conditions between various AI modules in the system (such as the project manager AI, the work allocation AI agent, the RL agent in the self-learning module, etc.) through a graphical interface or scripting language.

[0089] Creating and managing various AI agents: Tools are provided for creating, configuring, deploying, and monitoring various AI agent instances within the system, including parameter settings, model version management, and permission control. This enables the system to flexibly combine and customize AI capabilities based on different business scenarios and needs.

[0090] At least one application programming interface (API): In this embodiment, to achieve seamless integration and data collaboration between this system and the enterprise's existing information systems, the system also includes at least one application programming interface (API). The API (e.g., based on a standard protocol such as RESTful, SOAP, OPC UA, etc.) is configured as follows: Establishing communication connections: Used to establish secure and reliable communication connections between an industrial manufacturing operation management system based on hybrid augmented intelligence and the company's existing enterprise resource planning (ERP) system, manufacturing execution system (MES), supply chain management (SCM) system, product lifecycle management (PLM) system, etc.

[0091] Data exchange: Supports two-way data exchange, such as obtaining order information and material master data from the ERP system and real-time production progress and equipment status from the MES system, and providing feedback to these systems on task execution results and resource consumption data. This ensures that the system can make decisions based on the latest information across the enterprise and effectively communicate its decision results to other related systems, achieving a closed-loop enterprise information flow.

[0092] Through the collaborative work of the above modules, the system of this embodiment can build a comprehensive knowledge base, realize intelligent task planning, allocation and release, support human-machine collaborative decision-making, and have the ability of continuous learning and self-optimization, thereby significantly improving the efficiency, flexibility and intelligence level of industrial manufacturing operation management.

[0093] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An industrial manufacturing operation management method based on hybrid augmented intelligence, characterized in that: The following steps are involved: S1. Build a multi-source knowledge base and establish an initial task matching model based on multiple dimensions; S2. Based on the task matching initial model, autonomous task planning, task decomposition, task allocation and task release are performed, and the work allocation method knowledge base is updated according to the real-time task execution status; S3. Based on the updated work allocation method knowledge base and real-time feedback, collaboratively adjust the management strategy to ensure that the knowledge base is updated synchronously with actual needs and optimize the task allocation strategy.

2. The industrial manufacturing operation management method based on hybrid augmented intelligence according to claim 1 is characterized in that: The steps of S1 include: Construct a multi-source knowledge base, which includes a task model, a resource model, and a technical database: The task model uses AI knowledge base tools to vectorize the enterprise's existing workflow; the resource model forms the initial knowledge of resource library data by sorting out the enterprise's resource configuration; the technical database is obtained by vectorizing the enterprise's existing technical data; An initial task matching model is established based on at least the dimensions of organization, qualification, historical quality, and load.

3. The industrial manufacturing operation management method based on hybrid augmented intelligence according to claim 1 is characterized in that: In step S1, the multi-source knowledge base includes a relational database, a NoSQL database and a graph database, which are used to store structured data, unstructured data and knowledge graphs respectively.

4. The industrial manufacturing operation management method based on hybrid augmented intelligence according to claim 1 is characterized in that: The steps of S2 include: The demander initiates the task; The project manager AI performs autonomous task planning, task decomposition, task allocation, work knowledge addition, and task release based on the task model. The project manager AI uses a combination of rule-based reasoning and case-based reasoning to allocate tasks. The operations manager reviews the task decomposition and allocation results of the project portfolio manager AI and adjusts resource allocation based on actual conditions; The worker receives the task and provides feedback on the task execution status in real time, including the progress of the task completion and any problems encountered. The work allocation AI agent updates the knowledge base of work allocation methods based on real-time feedback.

5. The industrial manufacturing operation management method based on hybrid augmented intelligence according to claim 1 is characterized in that: The steps of S3 include: Operations managers collaborate with AI agents to regularly review process bottlenecks and adjust strategies to keep the knowledge base updated with actual needs, achieving continuous improvement. Through continuous learning and optimization of algorithms, the quality and efficiency of task completion are improved to achieve optimal resource allocation. The continuous learning adopts reinforcement learning algorithms to continuously optimize task allocation strategies based on feedback from workers.

6. An industrial manufacturing operation management system based on hybrid augmented intelligence, applied to an industrial manufacturing operation management method based on hybrid augmented intelligence as claimed in any one of claims 1 to 5, characterized in that: include: Multi-source knowledge base module, used to store task models, resource models and technical databases; The project manager AI module is used to autonomously plan tasks, decompose tasks, assign tasks, add work knowledge, and publish tasks based on the task model; The operations manager interface module is used for the operations manager to view the task decomposition and allocation results of the project portfolio manager AI and adjust resource allocation according to actual conditions; The worker interface module is used for workers to receive tasks and provide feedback on real-time task execution status; The work allocation AI agent module is used to update the work allocation method knowledge base based on real-time feedback.

7. The industrial manufacturing operation management system based on hybrid augmented intelligence according to claim 6 is characterized in that: It also includes a self-learning module, which is configured to continuously iteratively optimize the task model and resource matching algorithm based on accumulated data and feedback.

8. The industrial manufacturing operation management system based on hybrid augmented intelligence according to claim 6 is characterized in that: Also included is an explainability module configured to trace back and analyze key factors of core decisions and their processes.

9. The industrial manufacturing operation management system based on hybrid augmented intelligence according to claim 6 is characterized in that: It also includes an AI workflow and intelligent agent creation module, which is configured to define the logical framework of the AI ​​workflow and create and manage various types of intelligent agents.

10. The industrial manufacturing operation management system based on hybrid augmented intelligence according to claim 6, characterized in that: It also includes at least one application programming interface, which is used to establish a communication connection between the industrial manufacturing operation management system based on hybrid augmented intelligence and the enterprise's existing enterprise resource planning system, manufacturing execution system or supply chain management system for data exchange.

Images