Atomized AI interface packaging calling method and system

Abstract

The invention relates to the field of industrial automation and data processing, and discloses an atomization AI interface packaging and calling method. In order to solve the problems that an existing industrial control system is high in coupling degree, AI is difficult to directly participate in control and dynamic scheduling and the like, the method comprises the steps that a stateless atomic energy unit is constructed based on AST and CFG analysis; the JSON Schema is adopted to carry out MCP protocol packaging; constructing a shared sensing data pool by utilizing multi-modal feature alignment and a vector database; on-demand activation and dynamic mapping of the capability are realized based on an LLM and ReAct reasoning framework; and finally, constructing an edge cloud collaborative three-layer architecture of perceptual analysis, execution control and cognitive analysis. According to the method, through atomic power semantic packaging, the problems that traditional machine software and CAM software are strongly coupled, and the capacity is difficult to combine according to needs are solved, the AI can understand the industrial control capacity and safely call the industrial control capacity, and the expandability, real-time collaboration and intelligent scheduling capacity of an industrial control system are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of industrial automation control and data processing technology, and in particular to an atomic AI interface encapsulation and calling method and system. Background Technology

[0002] Currently, in precision industrial equipment control and PCB-related electronic design automation applications, a common architecture is the collaboration between machine tool control software and CAM data processing software. The machine tool control software is responsible for executing physical control functions such as motion control, visual positioning, and signal acquisition, while the CAM data processing software is responsible for parsing Gerber files, constructing geometric features, and generating test data files. The two typically interact through intermediate files in a fixed format. While this approach met basic testing and manufacturing needs in early automation control scenarios, its shortcomings have become increasingly apparent as user requirements have become more complex and multimodal information input formats have increased.

[0003] For example, in practical applications, user requirements are often presented in natural language descriptions, engineering screenshot annotations, or non-standardized tables. After-sales or engineering personnel need to manually convert this unstructured information into a standard file format, which is then parsed by the CAM software and finally executed by the machine software. This process heavily relies on manual understanding and translation, which is not only inefficient but also prone to ambiguity and errors, leading to mismatches between the generated test data and actual requirements. Furthermore, the machine software and CAM software are tightly coupled through a specific file format; if test parameters or parsing logic change, both software systems often need to be upgraded simultaneously, resulting in a long development chain and high maintenance costs.

[0004] Furthermore, most existing control software executes tasks based on fixed logic paths and hard-coded parameters, lacking a mechanism for adaptive capability scheduling based on real-time process constraints. When encountering special process requirements or complex constraints, the system cannot flexibly combine different capabilities to handle them like a human engineer; it can only adapt by modifying the underlying code, making it difficult to meet the flexibility and intelligence requirements of modern precision manufacturing scenarios.

[0005] Therefore, in complex industrial scenarios characterized by the coexistence of multi-source heterogeneous data, frequent multimodal demands, and the need for dynamic combination of capabilities, existing technologies cannot fully meet the requirement of "dynamically invoking industrial capabilities according to task semantics and achieving collaborative execution." There is an urgent need for an atomic AI interface encapsulation and invocation method that can still achieve on-demand combination, semantic understanding, and intelligent scheduling of industrial control capabilities without relying on fixed file interactions and manual intermediaries, in order to improve the flexibility, scalability, and intelligence level of industrial control systems. Summary of the Invention

[0006] To address the aforementioned technical shortcomings, the purpose of this invention is to propose an atomic AI interface encapsulation and invocation method. This method aims to solve the technical problem that in the existing technology, machine tool software and CAM software are strongly coupled through a fixed file format, making it difficult to support AI's direct participation in industrial control. This is especially true in industrial control scenarios where different control and parsing capabilities need to be dynamically invoked according to multimodal requirements, making it impossible to achieve on-demand combination, real-time collaboration, and safe scheduling of capabilities.

[0007] To solve the above technical problems, the present invention adopts the following technical solution: The present invention provides an atomic AI interface encapsulation and calling method.

[0008] The atomic AI interface encapsulation and invocation method includes: Step S10: Obtain the motion control logic, vision recognition logic, and signal acquisition logic from the original machine tool control software, as well as the geometric analysis logic and topology construction logic from the original CAM data processing software. Combine the functional semantic independence discrimination mechanism to perform the functional decoupling task and output a set of stateless atomic capability units. Step S20: Based on the set of stateless atomic capability units, perform the atomic capability semantic encapsulation task using a standardized input-output semantic description mechanism, and output a set of atomic capability services that conform to the MCP protocol; Step S30: Based on the atomic capability service set, a multi-source heterogeneous data semantic alignment and mapping mechanism is used to perform the perception data preprocessing task and output a perception data pool under a unified semantic space; Step S40: Based on the perception data pool and the atomic capability service set, the atomic capability dynamic scheduling mapping task is executed using a task context-driven on-demand activation scheduling mechanism, and the atomic capability scheduling mapping table semantically associated with the current task is output. Step S50: Based on the atomic capability scheduling mapping table, perform atomic capability hierarchical classification and system construction, and output a three-layer atomic capability architecture including a perception and analysis layer, an execution control layer, and a cognitive analysis layer.

[0009] Preferably, step S10, which involves acquiring the motion control logic, visual recognition logic, and signal acquisition logic from the original machine tool control software, as well as the geometric analysis logic and topology construction logic from the original CAM data processing software, and combining this with a functional semantic independence discrimination mechanism to perform a functional decoupling task and output a set of stateless atomic capability units, specifically includes: Step S101: Obtain the motion control logic, visual recognition logic, and signal acquisition logic from the original machine tool control software; perform functional semantic decomposition on the motion control logic, visual recognition logic, and signal acquisition logic; and identify the first functional unit with independent physical control meaning. Step S102: Obtain the geometric analysis logic and topology construction logic in the original CAM data processing software, perform functional semantic decomposition on the geometric analysis logic and topology construction logic, and identify the second functional unit with independent data analysis meaning; Step S103: Perform context dependency determination on the split first and second functional units, filter out the functional units that can run only by relying on input parameters, and form a set of stateless atomic capability units.

[0010] Preferably, step S20, which involves performing an atomic capability semantic encapsulation task based on a standardized input-output semantic description mechanism using a set of stateless atomic capability units, and outputting a set of atomic capability services conforming to the MCP protocol, specifically includes: Step S201: Establish a unified input semantic description structure for each stateless atomic capability unit. The input semantic description structure is used to define the calling conditions and input parameter semantics of the capability. Step S202: Establish a unified output semantic description structure for each stateless atomic capability unit. The output semantic description structure is used to define the execution result of the capability and its physical or data semantics. Step S203: Based on the input semantic description structure and the output semantic description structure, construct atomic capability service description objects that conform to the MCP protocol to form an atomic capability service set.

[0011] Preferably, the output semantic description structure includes at least physical coordinate semantics, geometric feature semantics, sensor feedback semantics, and control parameter semantics, which are used to support AI in performing capability scheduling based on semantics.

[0012] Preferably, step S30, which involves performing a perceptual data preprocessing task based on a multi-source heterogeneous data semantic alignment and mapping mechanism using an atomic capability service set, and outputting a perceptual data pool under a unified semantic space, specifically includes: Step S301: Acquire multi-source heterogeneous data from the sensor end, vision end and CAM resolution end. The multi-source heterogeneous data includes additional unified timestamp information. Step S302: Perform coordinate system mapping and time series alignment processing on the multi-source heterogeneous data, and map the processed multi-source heterogeneous data to a unified semantic space, outputting a perception data pool under the unified semantic space.

[0013] Preferably, step S40, which involves executing the dynamic scheduling mapping task of atomic capabilities using a task context-driven on-demand activation scheduling mechanism based on the perception data pool and the atomic capability service set, and outputting an atomic capability scheduling mapping table semantically associated with the current task, specifically includes: Step S401: Analyze the current task requirements based on the perception data pool and extract the task semantic set; Step S402: Filter the atomic capabilities in the atomic capability service set that match the task semantic set; Step S403: Establish a dynamic scheduling mapping relationship between atomic capabilities and preset task contexts, and output an atomic capability scheduling mapping table that is semantically associated with the current task.

[0014] Preferably, step S50, which involves classifying and categorizing atomic capabilities according to the atomic capability scheduling mapping table and constructing the system to output a three-layer atomic capability architecture including a perception and analysis layer, an execution control layer, and a cognitive analysis layer, specifically includes: Step S501: Obtain the input-output semantic attributes, real-time control semantic attributes, and signal processing and decision semantic attributes of the atomic capabilities from the atomic capability scheduling mapping table; Step S502: Based on the input and output semantic attributes of atomic capabilities, classify atomic capabilities into perceptual analysis capabilities; based on the real-time control semantic attributes of atomic capabilities, classify atomic capabilities into execution control capabilities; based on the signal processing and decision-making semantic attributes of atomic capabilities, classify atomic capabilities into cognitive analysis capabilities. Step S503: Construct the three-layer atomic capability architecture based on the results of atomic capability partitioning.

[0015] This invention also provides an atomic AI interface encapsulation and calling system, comprising: The atomic capability decoupling module is used to acquire the motion control logic, vision recognition logic, signal acquisition logic in the original machine tool control software, and the geometric analysis logic and topology construction logic in the original CAM data processing software. It combines the functional semantic independence discrimination mechanism to perform the functional decoupling task and output a set of stateless atomic capability units. The semantic encapsulation module is used to perform atomic capability semantic encapsulation tasks based on a set of stateless atomic capability units using a standardized input-output semantic description mechanism, and outputs a set of atomic capability services that conform to the MCP protocol. The perception data construction module is used to perform perception data preprocessing tasks based on the atomic capability service set and adopt a multi-source heterogeneous data semantic alignment and mapping mechanism to output a perception data pool in a unified semantic space. The dynamic scheduling mapping module is used to execute atomic capability dynamic scheduling mapping tasks based on the perception data pool and the atomic capability service set, using a task context-driven on-demand activation scheduling mechanism, and outputs an atomic capability scheduling mapping table that is semantically associated with the current task. The layered capability building module is used to classify and build the atomic capability hierarchy based on the atomic capability scheduling mapping table, and outputs a three-layer atomic capability architecture including a perception and analysis layer, an execution control layer, and a cognitive analysis layer.

[0016] The present invention also provides an atomic AI interface encapsulation and invocation device, comprising: a memory, a processor, and an atomic AI interface encapsulation and invocation program stored in the memory and executable on the processor, wherein the atomic AI interface encapsulation and invocation program implements the atomic AI interface encapsulation and invocation method when executed by the processor.

[0017] The present invention also provides a computer program product, including an atomic AI interface encapsulation calling program, wherein the atomic AI interface encapsulation calling program implements the atomic AI interface encapsulation calling method when executed by a processor.

[0018] The beneficial effects of this invention are as follows: By deconstructing the original machine control function and CAM data processing function into stateless atomic capability units that conform to the MCP protocol and establishing a unified semantic description system, this invention transforms the originally highly coupled software modules into a set of capabilities that can be independently called by AI according to task semantics, significantly reducing system coupling and version dependency, and improving capability reusability and system scalability between different machines and different factory areas.

[0019] This invention forms a shared perception data pool by unifying the timeline alignment and semantic mapping of sensor data, visual data, geometric analysis data, and historical log data. This enables AI to perform capability scheduling and decision-making based on a real-time and complete industrial semantic environment, avoiding the problems of data fragmentation and information lag in the traditional file interaction mode, and improving system-level collaborative efficiency.

[0020] This invention activates and dynamically schedules atomic capabilities on demand based on task context semantics, avoiding the resource waste and execution inefficiency caused by full loading and fixed calling paths in traditional systems. It enables the system to adaptively select the optimal capability combination according to the current task complexity, equipment load status and data requirements, significantly improving the flexibility, real-time performance and intelligence level of industrial control systems. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart illustrating the first embodiment of an atomic AI interface encapsulation and invocation method according to the present invention.

[0023] Figure 2 This is a schematic diagram of the overall process operation of the first embodiment of the atomic AI interface encapsulation and calling method of the present invention.

[0024] Figure 3 This is a schematic diagram of the abstract syntax tree parsing and control flow graph analysis process of the first embodiment of the atomic AI interface encapsulation and calling method of the present invention.

[0025] Figure 4 This is a schematic diagram of the stateless atomic capability unit mapping process of the first embodiment of the atomic AI interface encapsulation and calling method of the present invention.

[0026] Figure 5 This is a schematic diagram of the process of using a multimodal feature extraction operator and a vector database, which is the first embodiment of the atomic AI interface encapsulation and calling method of the present invention.

[0027] Figure 6 This is a schematic diagram of the directed acyclic graph scheduling sequence flow of the first embodiment of the atomic AI interface encapsulation and calling method of the present invention.

[0028] Figure 7 This is a schematic diagram of the edge-cloud collaborative three-layer atomic capability architecture of the first embodiment of the atomic AI interface encapsulation and calling method of the present invention.

[0029] Figure 8 This is a schematic diagram of a device for an atomic AI interface encapsulation and invocation method according to the present invention. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] Example 1: As Figure 1 The diagram shown is a flowchart of the first embodiment of the atomic AI interface encapsulation and invocation method of the present invention, which presents the first embodiment of the atomic AI interface encapsulation and invocation method of the present invention.

[0032] In the first embodiment, the atomic AI interface encapsulation invocation method includes: Step S10: Obtain the motion control logic, vision recognition logic, and signal acquisition logic from the original machine tool control software, as well as the geometric analysis logic and topology construction logic from the original CAM data processing software. Combine the functional semantic independence discrimination mechanism to perform the functional decoupling task and output a set of stateless atomic capability units. It should be noted that the functional semantic independence discrimination mechanism includes: parsing the original software code using Abstract Syntax Tree (AST) and Control Flow Graph (CFG) to remove implicit global state dependencies and explicitly express variables dependent on hardware state as input parameters; the functional semantic independence discrimination mechanism refers to the semantic independence analysis and judgment of different functional modules in the original machine tool control software and CAM data processing software. In specific implementation, this invention introduces Abstract Syntax Tree (AST) parsing and Control Flow Graph (CFG) analysis techniques. By parsing the AST of the original software source code or compiled interface, functions and methods that do not contain persistent state variables (such as global variables and static variables) are identified and extracted. For logic dependent on hardware context, this invention implements "implicit state explicitization" processing, that is, it forces the hardware state (such as the current probe coordinates and the current machine tool temperature) to be passed as explicit parameters to the functional unit, thereby ensuring that each extracted first and second functional units satisfy the absolute "stateless" characteristic. This mechanism minimizes the dependencies within the module by semantically decomposing the input and output features of each functional unit, allowing each module to exist independently. Specifically, it includes modules such as motion control, visual recognition, signal acquisition, geometric analysis, and topology construction. Each module is defined by standardized input and output semantics, enabling it to operate independently without relying on the state of other modules. In this way, flexible combination and on-demand scheduling of functions can be achieved.

[0033] Understandably, by transforming traditional highly coupled functional modules into independent atomic capability units, this invention achieves flexible scheduling and dynamic combination of the original machine control and CAM data processing functions. This innovative design brings great flexibility, allowing not only the invocation of required functional units for specific tasks but also the adjustment of the execution order and combination of atomic capabilities according to changes in the task. Its technical effects are reflected in: firstly, the ability to adjust individual functional modules without modifying other modules; secondly, this flexible scheduling method greatly improves the adaptability and scalability of industrial control systems, better meeting ever-changing market demands. For example, as... Figure 2 As shown, the overall process flow of the atomic AI interface encapsulation and calling method is demonstrated. By integrating atomic capability decoupling, MCP protocol semantic encapsulation, perception data pool construction, dynamic scheduling mapping, and a three-layer architecture, the technical effect of transforming the traditional highly coupled industrial control logic into an atomic service cluster that AI can understand and schedule on demand is achieved, which significantly reduces the system coupling and improves the intelligent scheduling capability.

[0034] Step S20: Based on the set of stateless atomic capability units, perform the atomic capability semantic encapsulation task using a standardized input-output semantic description mechanism, and output a set of atomic capability services that conform to the MCP protocol; It should be noted that the standardized input-output semantic description mechanism adopts the JSON Schema format, mapping stateless atomic capability units to tools, resources, and prompts under the MCP protocol. In this step, the Model Context Protocol (MCP) is used for semantic encapsulation, enabling the large language model to directly understand and invoke industrial hardware capabilities. The specific mapping rules are as follows: 1. Resource mapping: Real-time sensor data and CAM parsing logs of the machine are encapsulated as MCPResources, allowing AI to read the machine status at any time through a unified URI. 2. Tool mapping: Active execution commands such as motion control and visual positioning of the machine are encapsulated as MCP Tools. Simultaneously, the input and output parameters (such as physical coordinate semantics and control parameter semantics) of each Tool are defined using the native JSON Schema format of the large model, specifying the data type and value range. 3. Prompt Mapping: Specific process specifications in industrial scenarios (such as "high-precision pressure constraint") are encapsulated into MCP Prompt templates, standardizing the context of AI calls. In this way, each atomic capability unit not only possesses independence but can also be called by other modules without depending on the specific implementation, ensuring that it can be extended and interacted through standardized interfaces.

[0035] Understandably, this step achieves high portability and reusability of atomic capability units through standardized input / output semantic descriptions. For example, in industrial automation systems, modules such as machine control, vision recognition, and signal acquisition often have different input / output formats, which increases the difficulty of system integration. This invention, by introducing a standardized description mechanism, unifies the input / output format of each atomic capability unit into a standard format conforming to the MCP protocol, enabling different functional modules to interface through a unified semantic description. In this way, systems across different machines, devices, and factories can exchange data and share capabilities through a standardized protocol. For instance, in traditional industrial control systems, it is necessary to test a new device whose interface and input / output format differ from existing systems. In traditional methods, to make the new device compatible with existing systems, it is often necessary to modify existing control software and data processing flows, or even redefine the interface. However, in this invention, with the help of a standardized semantic description mechanism, we only need to perform corresponding semantic mapping and definition of the input / output parameters of the new device, and then connect it to the existing system through the MCP protocol. At this point, the functional modules of the new equipment can be seamlessly integrated with the existing control system and can be flexibly scheduled and used without modifying the existing system architecture. This approach significantly improves scalability and adaptability.

[0036] For example, such as Figure 3 As shown, by using Abstract Syntax Tree (AST) parsing and Control Flow Graph (CFG) analysis techniques, the functional logic of the original machine control and CAM software is deeply decomposed, and "implicit state explicitization" is implemented. This achieves the technical effect of stripping away global dependencies and outputting a set of stateless atomic capability units, laying the foundation for the independent invocation and flexible combination of subsequent capabilities.

[0037] Step S30: Based on the atomic capability service set, a multi-source heterogeneous data semantic alignment and mapping mechanism is used to perform the perception data preprocessing task and output a perception data pool under a unified semantic space; It should be noted that this step involves semantic alignment and mapping of data from different sources and in different formats. In its implementation, this invention employs a vector database and multimodal feature alignment technology. For heterogeneous data from multiple sources, such as textual requirements, visual images, and one-dimensional waveforms from sensors, a contrastive language-image pre-trained model (such as CLIP) and an industrial-domain fine-tuned embedding model are used to transform unstructured data into high-dimensional feature vectors. Simultaneously, combined with an industrial control ontology, equipment status data is transformed into standardized semantic knowledge graph nodes. By introducing a semantic alignment and mapping mechanism for heterogeneous data from multiple sources, this step can uniformly process these data, mapping them to a standardized semantic space. This ensures that different types of data can collaborate and integrate within a unified semantic framework, thereby generating a complete perception data pool. This perception data pool provides reliable real-time data support for subsequent AI scheduling, decision-making, and control.

[0038] Understandably, multi-source heterogeneous data semantic alignment and mapping mechanisms can transform data from different hardware devices (such as sensors, vision systems, and execution modules from different manufacturers) into the same semantic expression through standardized semantic mapping rules. This unified semantic expression enables heterogeneous data to be processed in a unified format, avoiding data parsing difficulties caused by differences between devices. This mechanism effectively solves problems such as inconsistent data formats and difficulties in data interoperability caused by differences between hardware devices, ensuring that data generated by different devices can interoperate within the same semantic framework. For example, visual data can be combined with force sensor data to form a more complete environmental perception model.

[0039] It should be understood that in traditional automated control systems, sensed data typically comes from multiple sources, and this data cannot always be processed directly within the same framework. Usually, each data type needs to be processed according to its own format before being merged, a process prone to information loss, data mismatch, or high latency. In contrast, this invention, through semantic alignment and mapping mechanisms, not only ensures the uniformity of data sources but also enables various types of data to be effectively accessed and processed simultaneously through a unified semantic space. For example, as... Figure 4 As shown, stateless atomic capability units are mapped to tools, resources and prompt word templates that conform to the Model Context Protocol (MCP) based on the JSON Schema format, achieving the technical effect of interface semantic standardization and cross-platform reuse, and realizing the direct understanding and secure invocation of the underlying capabilities of industrial hardware by the large language model.

[0040] For example, in a certain industrial application, it is necessary to process image data from a vision camera and feedback data from a force sensor simultaneously. In traditional systems, vision data and sensor data typically use different formats and protocols, requiring separate parsing and interfacing of the two data formats before subsequent calculations and analysis. However, in this invention, through a multi-source heterogeneous data semantic alignment and mapping mechanism, vision image and force sensor data can be unified under the same semantic framework, ensuring that this data can be shared and utilized in real time within a collaborative sensing data pool. This allows for a more accurate understanding of the overall scene information during processing, such as the relationship between force and displacement, thereby enabling more precise control or decision-making.

[0041] Step S40: Based on the perception data pool and the atomic capability service set, the atomic capability dynamic scheduling mapping task is executed using a task context-driven on-demand activation scheduling mechanism, and the atomic capability scheduling mapping table semantically associated with the current task is output. It should be noted that the dynamic scheduling and mapping task is executed by a Large Language Model (LLM) agent. Based on the ReAct (Thought-Action-Observation) framework, it combines Retrieval-Enhanced Generation (RAG) technology to match atomic capabilities in the service set and generate a directed acyclic graph (DAG) scheduling sequence. The key to this step is to achieve dynamic scheduling and mapping of atomic capabilities through a task context-driven on-demand activation scheduling mechanism. This invention uses a Large Language Model (LLM)-based agent as the scheduling core. First, the LLM parses natural language or multimodal task context, performs intent understanding and task decomposition, and generates a sub-task DAG containing temporal and logical dependencies. Second, using Retrieval-Enhanced Generation (RAG) technology, it retrieves the most matching MCP capability in the atomic capability service set through semantic similarity. Finally, the LLM uses the ReAct (Thought-Action-Observation) closed-loop mechanism for dynamic scheduling. The large model outputs "Thinking (analyzing current process requirements)," generates "Action (calling the corresponding MCP atomic tool)," and dynamically corrects the next scheduling decision based on the returned sensor feedback semantics (Observation), ultimately generating an atomic capability scheduling mapping table associated with the current task semantics. The task context includes the current task's goals, requirements, input data, and constraints. This information determines which atomic capability units need to be activated in different task scenarios and how to schedule them to complete specific tasks. In traditional automation systems, function calls are often static and pre-defined. This invention, however, dynamically selects and activates appropriate atomic capabilities based on the task context, thereby flexibly adapting to different task requirements and achieving precise scheduling and resource optimization.

[0042] Understandably, driven by the task context, the scheduling of atomic capabilities is no longer blind or preset, but dynamically selected based on real-time task changes and requirements. For example, the scheduling method for the same functional module may differ under different process scenarios. Traditional methods often cannot adjust flexibly, while this invention intelligently selects the most suitable atomic capability service based on the specific requirements and semantics of the current task before each task is executed. This avoids the "repetitive, invalid, or inappropriate" capability scheduling in traditional methods, thereby improving overall efficiency and responsiveness.

[0043] It should be understood that, unlike traditional hard-coded task execution paths, the on-demand activation scheduling mechanism of this invention can automatically determine which atomic capabilities to execute based on real-time task semantics, and dynamically adjust the execution path accordingly. Traditionally, when tasks change, the execution flow and path must be rewritten. However, this invention, through the transmission and semantic parsing of task context, ensures accurate execution of each task while avoiding redundant development and modification. For example, as... Figure 5 As shown, relying on multimodal feature extraction operators and vector databases, sensor signals, visual images, and process texts are uniformly aligned to a high-dimensional vector space, achieving the technical effect of constructing a unified semantic space perception data pool. This solves the problem of data fragmentation in traditional models and provides a real-time, complete, and comprehensive perception foundation for AI decision-making. Figure 6 As shown, by utilizing the Large Language Model (LLM) based on the ReAct (reasoning and action) framework, atomic capabilities are dynamically retrieved and matched according to the real-time task context to generate a directed acyclic graph (DAG) scheduling sequence, achieving the technical effects of on-demand activation and adaptive execution, which significantly improves the system's response flexibility and execution efficiency under complex working conditions.

[0044] For example, in a PCB testing task requiring precise control, suppose the task context includes high-precision probe depth requirements for a specific area. In traditional automated control systems, this might require pre-setting the probe depth and rigidly executing it. However, in this invention, services related to probe depth control, motion trajectory planning, and visual alignment are selected from the atomic capability service set based on the specific requirements of the task context, and activated and scheduled on demand. This dynamic scheduling allows for flexible adjustment of the control strategy according to real-time changes in the task. For instance, the image recognition module is automatically activated during visual alignment, and the probe's depressor depth is dynamically adjusted after confirming accuracy, ensuring smooth task execution without wasting resources. This approach significantly improves efficiency and accuracy in complex tasks or industrial scenarios with fluctuating requirements.

[0045] Step S50: Based on the atomic capability scheduling mapping table, perform atomic capability hierarchical classification and system construction, and output a three-layer atomic capability architecture including a perception and analysis layer, an execution control layer, and a cognitive analysis layer.

[0046] It should be noted that this step, based on the atomic capability scheduling mapping table generated in the previous steps, categorizes atomic capability services according to their functions and roles. These capability units are organized into three main layers: the perception and analysis layer, the execution control layer, and the cognitive analysis layer. Each layer is responsible for different functional tasks. The perception and analysis layer handles data acquisition and preliminary processing, the execution control layer is responsible for actual physical control and task execution, and the cognitive analysis layer performs advanced data analysis and intelligent decision-making. This hierarchical capability architecture design ensures that each task is accurately scheduled and assigned according to its needs, while improving scalability and flexibility. To adapt to the computing power requirements of large AI models and the real-time requirements of industrial control, the three-layer atomic capability architecture of this invention adopts an "edge-cloud collaborative" physical deployment mode. The perception and analysis layer and the execution control layer are deployed on edge computing nodes close to the machine to ensure millisecond / microsecond real-time response to sensor data acquisition and physical motion commands; the cognitive analysis layer is deployed on a local server or cloud with strong computing power to carry out complex reasoning of large-scale language models, natural language intent parsing, and ReAct policy generation. The edge and cloud sides achieve efficient decoupling and semantic interaction through the MCP protocol.

[0047] Understandably, by hierarchically classifying atomic capabilities, this invention clearly divides functional modules into different responsibilities and ensures that each level can efficiently complete its specific task. The perception and analysis layer is mainly responsible for collecting and preprocessing perception data, the execution and control layer is responsible for executing specific control commands to the hardware, and the cognitive analysis layer is responsible for in-depth analysis of perception data and making decisions. This hierarchical structure maximizes the integration of the capabilities of each functional module, avoids functional overlap and resource waste, and helps to achieve independence and collaborative work between modules during the execution of complex tasks. For example, in a PCB inspection task requiring precise control, it is necessary to process sensor data and perform precise physical control simultaneously. In traditional control systems, all functions may be implemented in a unified module, resulting in high coupling between modules during task execution. In this invention, the perception and analysis layer first acquires environmental data through cameras and sensors and performs preliminary analysis; then, the execution and control layer receives this data, performs motion control, and adjusts the probe depth; finally, the cognitive analysis layer performs advanced analysis of the data to determine whether the probe has reached the expected position and adjusts the next operation based on the feedback results. This hierarchical capability architecture not only enables efficient task execution but also allows for flexible responses to changing needs during the task process, achieving precise control and adaptive adjustment, while avoiding the complex interdependencies between functional modules in traditional systems.

[0048] For example, such as Figure 7 As shown, based on the semantic attributes of capabilities, they are classified and merged into the perception and analysis layer, the execution control layer, and the cognitive analysis layer, achieving the technical effect of constructing a three-layer atomic capability architecture system for edge-cloud collaboration, realizing hierarchical management of industrial control capabilities and efficient collaborative scheduling at the system level.

[0049] Example 2: Furthermore, the atomic AI interface encapsulation and invocation system provided by this invention employs an atomic AI interface encapsulation and invocation method from the above embodiments, which can solve the technical problem of atomic AI interface encapsulation and invocation. The beneficial effects of the atomic AI interface encapsulation and invocation system provided by this invention are the same as those of the atomic AI interface encapsulation and invocation method provided by the above embodiments, and other technical features of the atomic AI interface encapsulation and invocation system are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0050] Example 3: This invention provides an atomized AI interface encapsulation and calling device. Please refer to... Figure 8An atomic AI interface encapsulation and invocation device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to execute the atomic AI interface encapsulation and invocation method described in Embodiment 1 above. An atomic AI interface encapsulation and invocation device in this embodiment may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. An atomic AI interface encapsulation and invocation device is merely an example and should not impose any limitations on the functionality and scope of use of this embodiment. An atomic AI interface encapsulation and invocation device may include a processing device 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory 1002 or a program loaded from a storage device 1003 into a random access memory 1004. Random access memory 1004 also stores various programs and data required for the operation of an atomic AI interface encapsulation calling device. Processing device 1001, read-only memory 1002, and random access memory 1004 are interconnected via bus 1005. I / O interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows an atomic AI interface encapsulation calling device to communicate wirelessly or wiredly with other devices to exchange data. Although an atomic AI interface encapsulation calling device with various systems is shown in the figure, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems can be implemented alternatively.

[0051] Example 4: This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the atomic AI interface encapsulation and invocation method described above. The computer program product provided by this invention can solve the technical problem of atomic AI interface encapsulation and invocation. Compared with the prior art, the beneficial effects of the computer program product provided by this invention are the same as the beneficial effects of the atomic AI interface encapsulation and invocation method provided in the above embodiments, and will not be repeated here.

[0052] In particular, according to the embodiments disclosed in this invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this invention.

[0053] It should be understood that the various parts disclosed in this invention can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics may be combined in any suitable manner in one or more embodiments or examples.

[0054] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. An atomic AI interface encapsulation and invocation method, characterized in that, The methods include: Step S10: Obtain the motion control logic, vision recognition logic, and signal acquisition logic from the original machine tool control software, as well as the geometric analysis logic and topology construction logic from the original CAM data processing software. Combine the functional semantic independence discrimination mechanism to perform the functional decoupling task and output a set of stateless atomic capability units. Step S20: Based on the set of stateless atomic capability units, perform the atomic capability semantic encapsulation task using a standardized input-output semantic description mechanism, and output a set of atomic capability services that conform to the MCP protocol; Step S30: Based on the atomic capability service set, a multi-source heterogeneous data semantic alignment and mapping mechanism is used to perform the perception data preprocessing task and output a perception data pool under a unified semantic space; Step S40: Based on the perception data pool and the atomic capability service set, the atomic capability dynamic scheduling mapping task is executed using a task context-driven on-demand activation scheduling mechanism, and the atomic capability scheduling mapping table semantically associated with the current task is output. Step S50: Based on the atomic capability scheduling mapping table, perform atomic capability hierarchical classification and system construction, and output a three-layer atomic capability architecture including a perception and analysis layer, an execution control layer, and a cognitive analysis layer.

2. The atomic AI interface encapsulation and invocation method as described in claim 1, characterized in that, Step S10 involves acquiring the motion control logic, vision recognition logic, and signal acquisition logic from the original machine tool control software, as well as the geometric analysis logic and topology construction logic from the original CAM data processing software. It then combines this with a functional semantic independence discrimination mechanism to perform a functional decoupling task and output a set of stateless atomic capability units. Specifically, this includes: Step S101: Obtain the motion control logic, visual recognition logic, and signal acquisition logic from the original machine tool control software; perform functional semantic decomposition on the motion control logic, visual recognition logic, and signal acquisition logic; and identify the first functional unit with independent physical control meaning. Step S102: Obtain the geometric analysis logic and topology construction logic in the original CAM data processing software, perform functional semantic decomposition on the geometric analysis logic and topology construction logic, and identify the second functional unit with independent data analysis meaning; Step S103: Perform context dependency determination on the split first and second functional units, filter out the functional units that can run only by relying on input parameters, and form a set of stateless atomic capability units.

3. The atomic AI interface encapsulation and invocation method as described in claim 1, characterized in that, Step S20, which involves performing an atomic capability semantic encapsulation task based on a standardized input-output semantic description mechanism using a set of stateless atomic capability units, and outputting a set of atomic capability services conforming to the MCP protocol, specifically includes: Step S201: Establish a unified input semantic description structure for each stateless atomic capability unit. The input semantic description structure is used to define the calling conditions and input parameter semantics of the capability. Step S202: Establish a unified output semantic description structure for each stateless atomic capability unit. The output semantic description structure is used to define the execution result of the capability and its physical or data semantics. Step S203: Based on the input semantic description structure and the output semantic description structure, construct atomic capability service description objects that conform to the MCP protocol to form an atomic capability service set.

4. The atomic AI interface encapsulation and invocation method as described in claim 3, characterized in that, The output semantic description structure includes at least physical coordinate semantics, geometric feature semantics, sensor feedback semantics, and control parameter semantics, which are used to support AI in performing capability scheduling based on semantics.

5. The atomic AI interface encapsulation and invocation method as described in claim 1, characterized in that, Step S30, which involves performing a perceptual data preprocessing task based on a multi-source heterogeneous data semantic alignment and mapping mechanism using an atomic capability service set, and outputting a perceptual data pool in a unified semantic space, specifically includes: Step S301: Acquire multi-source heterogeneous data from the sensor end, vision end and CAM resolution end. The multi-source heterogeneous data includes additional unified timestamp information. Step S302: Perform coordinate system mapping and time series alignment processing on the multi-source heterogeneous data, and map the processed multi-source heterogeneous data to a unified semantic space, outputting a perception data pool under the unified semantic space.

6. The atomic AI interface encapsulation and invocation method as described in claim 1, characterized in that, Step S40, based on the perception data pool and the atomic capability service set, employs a task context-driven on-demand activation scheduling mechanism to execute the dynamic scheduling mapping task of atomic capabilities, and outputs an atomic capability scheduling mapping table semantically associated with the current task. This step specifically includes: Step S401: Analyze the current task requirements based on the perception data pool and extract the task semantic set; Step S402: Filter the atomic capabilities in the atomic capability service set that match the task semantic set; Step S403: Establish a dynamic scheduling mapping relationship between atomic capabilities and preset task contexts, and output an atomic capability scheduling mapping table that is semantically associated with the current task.

7. The atomic AI interface encapsulation and invocation method as described in claim 1, characterized in that, In step S50, the atomic capabilities are hierarchically classified and system-constructed according to the atomic capability scheduling mapping table, outputting a three-layer atomic capability architecture including a perception and analysis layer, an execution control layer, and a cognitive analysis layer. Specifically, this includes: Step S501: Obtain the input-output semantic attributes, real-time control semantic attributes, and signal processing and decision semantic attributes of the atomic capabilities from the atomic capability scheduling mapping table; Step S502: Based on the input and output semantic attributes of atomic capabilities, classify atomic capabilities into perceptual analysis capabilities; based on the real-time control semantic attributes of atomic capabilities, classify atomic capabilities into execution control capabilities; based on the signal processing and decision-making semantic attributes of atomic capabilities, classify atomic capabilities into cognitive analysis capabilities. Step S503: Construct the three-layer atomic capability architecture based on the results of atomic capability partitioning.

8. An atomic AI interface encapsulation and invocation system, applied to the atomic AI interface encapsulation and invocation method according to any one of claims 1 to 7, characterized in that, The atomic AI interface encapsulation and calling system includes: The atomic capability decoupling module is used to acquire the motion control logic, vision recognition logic, signal acquisition logic in the original machine tool control software, and the geometric analysis logic and topology construction logic in the original CAM data processing software. It combines the functional semantic independence discrimination mechanism to perform the functional decoupling task and output a set of stateless atomic capability units. The semantic encapsulation module is used to perform atomic capability semantic encapsulation tasks based on a set of stateless atomic capability units using a standardized input-output semantic description mechanism, and outputs a set of atomic capability services that conform to the MCP protocol. The perception data construction module is used to perform perception data preprocessing tasks based on the atomic capability service set and adopt a multi-source heterogeneous data semantic alignment and mapping mechanism to output a perception data pool in a unified semantic space. The dynamic scheduling mapping module is used to execute atomic capability dynamic scheduling mapping tasks based on the perception data pool and the atomic capability service set, using a task context-driven on-demand activation scheduling mechanism, and outputs an atomic capability scheduling mapping table that is semantically associated with the current task. The layered capability building module is used to classify and build the atomic capability hierarchy based on the atomic capability scheduling mapping table, and outputs a three-layer atomic capability architecture including a perception and analysis layer, an execution control layer, and a cognitive analysis layer.

9. An atomized AI interface encapsulation and calling device, characterized in that, The atomic AI interface encapsulation and invocation device includes: a memory, a processor, and an atomic AI interface encapsulation and invocation program stored in the memory and executable on the processor. When the atomic AI interface encapsulation and invocation program is executed by the processor, it implements an atomic AI interface encapsulation and invocation method according to any one of claims 1 to 7.

10. A computer program product, characterized in that, The computer program product includes an atomic AI interface encapsulation and invocation program, which, when executed by a processor, implements an atomic AI interface encapsulation and invocation method according to any one of claims 1 to 7.

Images