An Impedance Control Method and System Based on AI Agent and MCP Architecture

Abstract

The application relates to the technical field of industrial automation control, and discloses an impedance control method and system based on an AI Agent and an MCP architecture, wherein the method comprises the following steps: acquiring user-side multi-modal test demand data and circuit board engineering data; constructing a standardized demand constraint set; constructing a three-level atomic capability service architecture; generating a standardized test context; generating a candidate action sequence of multiple test heads in parallel; performing safety checking through a multi-sub-proxy cross-attestation mechanism; executing closed-loop adaptive parameter adjustment; and outputting an impedance test result. Compared with the prior art, the software modules are strongly coupled, the manual data conversion efficiency is low, and especially under the condition of multi-head cooperative operation, dynamic safety impedance control cannot be realized. Due to the layered atomic capability architecture constructed by the application, the separation control and adaptive optimization of demand understanding, process scheduling and physical execution are realized, and the impedance test efficiency and industrial safety reliability are improved.

Description

Technical Field

[0001] This invention relates to the field of industrial automation control technology, and in particular to an impedance control method and system based on an AI Agent and MCP architecture. Background Technology

[0002] Currently, in the field of printed circuit board (PCB) impedance testing, automated time-domain reflectometry (TDR) testing systems typically employ a dual-software architecture consisting of "data generation software + machine control software." The data generation software parses Gerber or ODB++ layout files and generates test data files, while the machine control software drives a multi-axis motion system and the TDR testing module to perform test operations based on the test data files.

[0003] However, due to the increasing complexity of PCB design and the extremely high impedance accuracy requirements of high-frequency signals, traditional PID control based on fixed step size or fixed path planning cannot cope with material deformation and dynamic interference from multiple concurrent sources. Therefore, it is necessary to rely on AI agents with reasoning and adaptive capabilities. For example, data generation and machine control are usually coupled through fixed-format files. Once test parameters, process rules, or hardware structure change, multiple software modules often need to be upgraded simultaneously, resulting in high system maintenance costs and complex version compatibility. In the test requirement conversion stage, user-provided natural language descriptions, screenshot annotations, or non-standard table information need to be manually converted into standard format files. The conversion process relies on engineering experience, which can easily lead to misunderstandings and input errors, reducing test preparation efficiency. Furthermore, at the machine control level, existing automatic test systems mostly use hard-coded control logic. When encountering special process constraints (such as pin mark depth limitations, special board materials, or high-density differential line areas), the system lacks adaptive parameter tuning capabilities and cannot dynamically optimize motion and sampling parameters based on real-time feedback, often requiring manual intervention or modification of the underlying program code. Meanwhile, in scenarios where multiple test heads operate concurrently, traditional fixed-area protection or simple limit control is difficult to adapt to the needs of dynamic path planning, which can easily lead to path conflict risks and affect equipment safety and operating efficiency.

[0004] Therefore, under complex process constraints, multimodal input requirements, and the collaborative operation of multiple test heads, there is an urgent need for an impedance control method that can automatically understand requirements, decouple control processes in a hierarchical manner, perform dynamic safety simulations, and enable closed-loop adaptive parameter tuning, so as to improve testing efficiency, system stability, and safety and reliability in industrial settings. Summary of the Invention

[0005] To address the aforementioned technical shortcomings, the purpose of this invention is to propose an impedance control method based on an AI Agent and MCP architecture. This method aims to solve the technical problems of strong coupling of software modules, low efficiency of manual data conversion, and especially the inability to achieve dynamic and safe impedance control under multi-head collaborative operation conditions in the prior art.

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: The present invention provides an impedance control method based on AI Agent and MCP architecture.

[0007] The impedance control method based on AI Agent and MCP architecture includes:

[0008] Step S10: Obtain user-side multimodal test requirement data and circuit board engineering data. Based on the user-side multimodal test requirement data and circuit board engineering data, use a multimodal intent understanding and requirement normalization mechanism to perform requirement structuring tasks and output a standardized requirement constraint set, a target network semantic set, and an initial candidate test point set.

[0009] Step S20: Based on the standardized set of requirement constraints, the target network semantic set, and the initial candidate test point set, construct a three-level atomic capability service architecture, which includes a basic perception layer, an execution control layer, and a cognitive analysis layer. Then, based on this three-level atomic capability service architecture, use a layered invocation orchestration mechanism to execute the test context construction task and output a standardized test context. The execution control layer includes an atomic service set S2, and the cognitive analysis layer includes an atomic service set S3.

[0010] Step S30: Based on standardized test context The preprocessing task is performed using a context resolution and action sequence generation mechanism driven by an adaptive secure execution agent, outputting candidate action sequences for concurrent testing of multiple test heads. and candidate action sequences The corresponding initial action parameter set ;

[0011] Step S40: Based on candidate action sequences With the initial set of motion parameters The system employs a multi-sub-agent cross-verification mechanism and a dynamic anti-collision pre-simulation mechanism to perform security verification tasks and outputs a safe pass action sequence. and safety action parameter set ;

[0012] Step S50: According to the safety passage action sequence and safety action parameter set The atomic service set S2 of the execution control layer and the atomic service set S3 of the cognitive analysis layer are invoked to perform impedance control physical execution. During the impedance control physical execution process, feedback data is collected, and closed-loop adaptive parameter tuning is performed based on the feedback data, outputting the impedance test result set O.

[0013] Preferably, in step S10, the user-side multimodal test requirement data includes natural language requirement text, screenshot annotation images, and non-standard table attachments; the circuit board engineering data includes Gerber files, ODB++ layout files, stack-up information, and historical test logs.

[0014] Preferably, in step S20, the basic perception layer provides an atomic service set S1 for circuit board data parsing and multimodal requirement understanding; the atomic service set S1 includes layout file streaming loading service, semantic layer parsing service, network topology construction service, PCB data parsing atomic service, and requirement normalization service; the execution control layer provides an atomic service set S2 for machine tool precision motion control and visual alignment interaction; the atomic service set S2 includes S-shaped trajectory planning service, flexible contact service, fine-tuning compensation service, and Mark point positioning service; the cognitive analysis layer provides an atomic service set S3 for TDR signal processing and process optimization decision-making; the atomic service set S3 includes waveform noise reduction service, impedance curve fitting service, layer stripping service, collision prediction service, and process optimization suggestion service.

[0015] Preferably, in step S20, a three-level atomic capability service architecture is constructed based on the standardized set of requirement constraints, the target network semantic set, and the initial candidate test point set. This architecture includes a basic perception layer, an execution control layer, and a cognitive analysis layer. Furthermore, based on this three-level atomic capability service architecture, a layered invocation orchestration mechanism is used to execute the test context construction task, outputting a standardized test context. The steps specifically include:

[0016] Step S201: Call the PCB data parsing class atomic service to perform streaming loading and semantic layer parsing on Gerber file and ODB++ layout file, identify differential line pair engineering semantic features and BGA area engineering semantic features, and establish board layer mapping based on differential line pair engineering semantic features and BGA area engineering semantic features, and output layout semantic feature set and board layer mapping information.

[0017] Step S202: Based on the layout semantic feature set and board layer mapping information, call the network topology construction service to construct the electrical connection netlist and generate network topology information. The initial candidate test point set is subjected to constraint filtering processing by combining the standardized requirement constraint set and the target network semantic set, and the confirmed test point set is output.

[0018] Step S203: Encapsulate the confirmed test point set, standardized requirement constraint set, and target network semantic set into a standardized test context. .

[0019] Preferably, in step S30, the candidate action sequence Includes alignment actions, approach actions, flexible contact actions, TDR sampling actions, and retreat actions, with an initial set of action parameters. This includes parameters for velocity and acceleration along each axis, S-shaped trajectory planning parameters, Z-axis compression depth parameters, contact force threshold parameters, and sampling rate parameters.

[0020] Preferably, in step S40, based on the candidate action sequence With the initial set of motion parameters The system employs a multi-sub-agent cross-verification mechanism and a dynamic anti-collision pre-simulation mechanism to perform security verification tasks and outputs a safe pass action sequence. and safety action parameter set The steps specifically include:

[0021] Step S401: Preset a decision sub-agent, which is responsible for making decisions based on candidate action sequences. Generate a multi-test-head motion command stream and set heartbeat synchronization and timestamp constraints for concurrent execution windows for the multi-test-head motion command stream;

[0022] Step S402: A security verification sub-agent is preset. The security verification sub-agent calls the preset collision pre-simulation atomic service to perform digital twin rapid pre-simulation of the multi-test head motion command stream, and outputs the multi-test head dynamic bounding box overlap rate and security confidence. Simultaneously, the security verification sub-agent calls the preset collision pre-simulation atomic service to perform hardware travel limit verification and multi-axis interlock deadlock verification, outputting a set of conflict constraints. ;

[0023] Step S403: When the security confidence level Below the preset security confidence threshold or conflict constraint set When not empty, block the multi-test-head motion command flow and base it on the conflict constraint set. Trigger path replanning and update to obtain a safe passage action sequence. and safety action parameter set .

[0024] Preferably, step S50, which involves performing closed-loop adaptive parameter tuning based on feedback data, specifically includes: obtaining needle mark depth data from the feedback data; automatically adjusting the Z-axis compression depth or acceleration parameters when the needle mark depth data does not meet the preset needle mark threshold constraint; obtaining current waveform quality data from the feedback data; and automatically adjusting the sampling rate and repeating the TDR sampling action and backoff action when the current waveform quality data does not meet the preset current waveform template data.

[0025] This invention also provides an impedance control system based on an AI Agent and MCP architecture, comprising:

[0026] The requirement structuring module is used to acquire user-side multimodal test requirement data and circuit board engineering data. Based on the user-side multimodal test requirement data and circuit board engineering data, it uses a multimodal intent understanding and requirement normalization mechanism to perform requirement structuring tasks and outputs a standardized set of requirement constraints, a target network semantic set, and an initial candidate test point set.

[0027] The layered atomic capability architecture building module is used to construct a three-tiered atomic capability service architecture based on a standardized set of requirement constraints, a target network semantic set, and an initial candidate test point set. This three-tiered atomic capability service architecture includes a basic perception layer, an execution control layer, and a cognitive analysis layer. Based on this three-tiered atomic capability service architecture, a layered invocation orchestration mechanism is used to execute the test context construction task and output a standardized test context. The execution control layer includes an atomic service set S2, and the cognitive analysis layer includes an atomic service set S3.

[0028] Execute the preprocessing generation module for use based on the standardized test context. The preprocessing task is performed using a context resolution and action sequence generation mechanism driven by an adaptive secure execution agent, outputting candidate action sequences for concurrent testing of multiple test heads. and candidate action sequences The corresponding initial action parameter set ;

[0029] The secure cross-validation module is used for candidate action sequences. With the initial set of motion parameters The system employs a multi-sub-agent cross-verification mechanism and a dynamic anti-collision pre-simulation mechanism to perform security verification tasks and outputs a safe pass action sequence. and safety action parameter set ;

[0030] The closed-loop execution and adaptive parameter tuning module is used to determine the safe passage action sequence. and safety action parameter set The atomic service set S2 of the execution control layer and the atomic service set S3 of the cognitive analysis layer are invoked to perform impedance control physical execution. During the impedance control physical execution process, feedback data is collected, and closed-loop adaptive parameter tuning is performed based on the feedback data, outputting the impedance test result set O.

[0031] The present invention also provides an impedance control device based on AI Agent and MCP architecture, comprising: a memory, a processor, and an impedance control program based on AI Agent and MCP architecture stored in the memory and executable on the processor. When the impedance control program based on AI Agent and MCP architecture is executed by the processor, it implements an impedance control method based on AI Agent and MCP architecture.

[0032] The present invention also provides a computer program product, including an impedance control program based on an AI Agent and MCP architecture, wherein the impedance control program based on the AI ​​Agent and MCP architecture implements the impedance control method based on the AI ​​Agent and MCP architecture when executed by a processor.

[0033] The beneficial effects of this invention are as follows: By constructing a three-tiered atomic capability service architecture consisting of a basic perception layer, an execution control layer, and a cognitive analysis layer, and combining it with a dual AI Agent collaboration mechanism, this invention decouples "requirement understanding," "data generation," and "physical execution" from the traditional tightly coupled software structure. Compared to existing technologies that rely on manual data conversion and fixed file format-driven control, this invention can directly perform structured parsing of multimodal requirements and generate standardized test contexts, thereby avoiding the problems of software version upgrade linkage and manual transfer errors. Under complex process constraints or new test parameters, only the corresponding atomic service or context parameters need to be adjusted, without reconstructing the underlying control logic, significantly improving the system's flexibility, scalability, and cross-machine reuse capabilities.

[0034] This invention introduces a multi-sub-agent cross-verification mechanism and a dynamic anti-collision pre-simulation mechanism at the execution layer. Through digital twin pre-simulation and dynamic bounding box prediction, it achieves real-time safety verification during concurrent operation of multiple test heads, effectively reducing the risk of path conflicts and physical collisions. Simultaneously, by introducing a closed-loop adaptive parameter tuning mechanism with force feedback, visual feedback, and waveform quality feedback during impedance control physical execution, it achieves dynamic correction of Z-axis pressing parameters, sampling rate, and contact strategy, enabling the equipment to automatically converge to a stable parameter range under different material types and process requirements. Therefore, compared to traditional hard-coded execution methods, this invention significantly improves the safety, reliability, parameter stability, and adaptability to complex operating conditions during the testing process. Attached Figure Description

[0035] 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.

[0036] Figure 1 This is a flowchart illustrating the first embodiment of an impedance control method based on an AI Agent and MCP architecture according to the present invention.

[0037] Figure 2 This is a schematic diagram of the engineering data semantic extraction process of the first embodiment of the impedance control method based on AI Agent and MCP architecture of the present invention.

[0038] Figure 3 This is a schematic diagram of the three-level architecture construction and test context generation of the first embodiment of the impedance control method based on AI Agent and MCP architecture of the present invention.

[0039] Figure 4 This is a schematic diagram of the distributed multi-agent action planning process of the first embodiment of the impedance control method based on AI Agent and MCP architecture of the present invention.

[0040] Figure 5 This is a schematic diagram of the multi-sub-agent cross-verification and security verification process in the first embodiment of an impedance control method based on AI Agent and MCP architecture according to the present invention.

[0041] Figure 6 This is a schematic diagram of the closed-loop adaptive execution and correction process of the first embodiment of the impedance control method based on AI Agent and MCP architecture of the present invention.

[0042] Figure 7 This is a schematic diagram of an impedance control method based on an AI Agent and MCP architecture according to the present invention. Detailed Implementation

[0043] 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.

[0044] Example 1: As Figure 1The diagram shown is a flowchart of the first embodiment of the impedance control method based on AI Agent and MCP architecture of the present invention, which presents the first embodiment of the impedance control method based on AI Agent and MCP architecture of the present invention.

[0045] In the first embodiment, the impedance control method based on AI Agent and MCP architecture includes:

[0046] Step S10: Obtain user-side multimodal test requirement data and circuit board engineering data. Based on the user-side multimodal test requirement data and circuit board engineering data, use a multimodal intent understanding and requirement normalization mechanism to perform requirement structuring tasks and output a standardized requirement constraint set, a target network semantic set, and an initial candidate test point set.

[0047] It should be noted that the MCP (Model Context Protocol) mentioned in the text can be understood as a standardized API interface description specification (similar to OpenAI's Function Calling Schema), allowing large models to know what tools are available and what the parameters are. The multimodal intent understanding and requirement normalization mechanism refers to a processing mechanism that performs unified semantic mapping and parameter quantization transformation on heterogeneous input information such as natural language text, screenshot annotations, and non-standard table attachments. This includes sub-processes such as semantic word segmentation and parsing, engineering terminology recognition, spatial coordinate mapping, numerical range standardization, and physical constraint parameter extraction. For example, using semantic extraction operators based on a large language model, feature alignment is performed on the multimodal test requirement data and circuit board engineering data to extract the impedance constraint parameters and geometric coordinate mapping relationship, thereby achieving requirement structuring. Semantic mapping refers to transforming colloquial or unstructured expressions such as "differential line," "power integrity area," "lighter," and "key testing" into a computable set of parameters such as impedance target range, Z-axis undervoltage depth limit, and priority weight. Requirement normalization refers to uniformly mapping test indicators with different expressions to a standardized data structure, enabling subsequent control layers to directly call them.

[0048] Understandably, by introducing the MCP layer and combining it with an atomic service architecture, software functions are no longer closed code blocks, but rather independent services that are easy to install and share. When adding hardware functions or parsing algorithms, only a new MCP interface tool needs to be released, without refactoring Agent A or Agent B, solving the "one change affects the whole system" problem of version upgrades and reducing development costs. By building a unified and collaborative perception data foundation, the AI ​​Agent can call atomic services on demand, avoiding the resource waste caused by full loading. The fusion of multi-source data enables the system to achieve a globally optimal perception process, improving testing efficiency and accuracy.

[0049] For example, such as Figure 2 As shown, this invention utilizes RAG (Retrieval Enhanced Generation) technology to overcome the limitations of traditional manual parsing of Gerber files. By transforming unstructured engineering drawings into a standardized set of target network semantics, it provides a high-precision mathematical foundation for subsequent path planning by the AI ​​Agent.

[0050] Step S20: Based on the standardized set of requirement constraints, the target network semantic set, and the initial candidate test point set, construct a three-level atomic capability service architecture, which includes a basic perception layer, an execution control layer, and a cognitive analysis layer. Then, based on this three-level atomic capability service architecture, use a layered invocation orchestration mechanism to execute the test context construction task and output a standardized test context. The execution control layer includes an atomic service set S2, and the cognitive analysis layer includes an atomic service set S3.

[0051] It should be noted that the three-tier atomic capability service architecture refers to the structured layering of system functions according to the "basic perception layer—execution control layer—cognitive analysis layer." The basic perception layer includes services such as layout streaming loading, semantic layer parsing, and network topology construction. The execution control layer includes services such as trajectory planning, flexible contact, and visual alignment. Specifically, the flexible contact service is used to adjust the Z-axis feed step size in real time based on feedback pressure sensor data to control the probe contact pressure within a preset millinewton (mN) range. The cognitive analysis layer includes services such as waveform noise reduction, impedance curve fitting, collision pre-simulation, and process optimization suggestion. The layered call orchestration mechanism refers to dynamically selecting and sorting the atomic service call chain according to the test task context.

[0052] Understandably, the MPC atomic three-layer architecture transforms the traditional "function stacking" into "capability service." The perception layer standardizes data input, the execution layer ensures the accuracy and safety of physical actions, and the cognition layer provides data analysis capabilities similar to those of a senior engineer. This architecture allows the AI ​​Agent to flexibly combine atomic services according to different testing scenarios, like building blocks, achieving true software-defined manufacturing.

[0053] For example, such as Figure 3As shown, the three-level atomic capability service architecture constructed in this invention achieves deep decoupling between logical control and physical execution. The execution control layer S2 is responsible for encapsulating the underlying motion control primitives, while the cognitive analysis layer S3 is responsible for high-level task decisions. Through a layered call orchestration mechanism, the system can dynamically bind abstract test requirements to physical test heads and impedance analyzers, and the resulting standardized test context contains all environmental parameters and constraints required to execute the task.

[0054] Step S30: Based on standardized test context The preprocessing task is performed using a context resolution and action sequence generation mechanism driven by an adaptive secure execution agent, outputting candidate action sequences for concurrent testing of multiple test heads. and candidate action sequences The corresponding initial action parameter set ;

[0055] It should be noted that the context parsing and action sequence generation mechanism refers to parsing the standardized test context and generating a sequence combination of alignment actions, approach actions, flexible contact actions, sampling actions, and retreat actions based on the spatial distribution of test points, priority weights, and the current state of the equipment. At the same time, it assigns initial velocity, acceleration, pressure depth, and sampling rate parameters to each action.

[0056] Understandably, by breaking down test tasks into executable sequences of physical actions, the control logic is transformed from abstract test objectives into concrete motion commands, providing a foundation for subsequent safety verification and physical execution, while also enabling the rational allocation of concurrent tasks from multiple test heads. Compared to traditional fixed-path or static sequential execution methods, this step can generate different strategy combinations based on real-time task status, such as concurrent execution or staggered execution, improving the collaborative efficiency of multiple test heads.

[0057] For example, such as Figure 4 As shown, the "Adaptive Safe Execution Agent" is the core role in MAS, responsible for transforming Fctx (static context) into A0 (dynamic candidate action). The diagram illustrates the parallel relationship between LLM and heuristic algorithms, which addresses the technical limitation that AI-planned paths may not conform to geometrically optimal solutions.

[0058] Step S40: Based on candidate action sequences With the initial set of motion parameters The system employs a multi-sub-agent cross-verification mechanism and a dynamic anti-collision pre-simulation mechanism to perform security verification tasks and outputs a safe pass action sequence. and safety action parameter set ;

[0059] It should be noted that in step S40, based on the candidate action sequence and the initial action parameter set, a multi-sub-agent cross-verification mechanism and a dynamic anti-collision pre-simulation mechanism are used to perform the security verification task, outputting a safe pass action sequence and a safe action parameter set. Specifically, this includes: Step S401: Pre-setting a decision sub-agent Agent B, which generates a multi-test head motion command stream based on the candidate action sequence, and sets heartbeat synchronization and timestamp constraints for the concurrent execution window of the multi-test head motion command stream; Step S402: Pre-setting a security verification sub-agent Agent. Step A: The safety verification agent calls the preset collision pre-simulation atomic service to perform a rapid digital twin pre-simulation of the multi-test head motion command flow, and outputs the multi-test head dynamic bounding box overlap rate and safety confidence. Simultaneously, the safety verification agent calls the preset collision pre-simulation atomic service to perform hardware travel limit verification and multi-axis interlock deadlock verification, outputting a conflict constraint set. Step S403: When the safety confidence is lower than the preset safety confidence threshold or the conflict constraint set is not empty, the multi-test head motion command flow is blocked, and path replanning is triggered based on the conflict constraint set to update the safe passage action sequence and the safe action parameter set. The path replanning is not simply blocking commands; it also uses dynamic programming or heuristic search algorithms to reallocate execution time windows in the multi-head overlapping area, thereby updating the safe passage action sequence and the safe action parameter set.

[0060] Understandably, Agent B, with its multimodal understanding capabilities, replaces the traditional "after-sales personnel + manual documentation" process. It directly understands the customer's original needs (including verbal descriptions and image annotations), reducing information loss and formatting errors caused by manual translation. Agent A changes the traditional "rigid execution" mode of automated machines. Faced with special process requirements (such as special needle marks or special materials), Agent A can automatically find the optimal physical parameters through a closed-loop logic of "trial-feedback-adjustment," much like an experienced engineer. This allows the equipment to handle unknown testing scenarios after leaving the factory, without the need for original manufacturer development intervention. The focus of maintenance shifts from "writing business code" to "maintaining the MCP tool library" and "fine-tuning the AI ​​model." The tuning experience of excellent engineers can be compiled into Prompts or knowledge bases and injected into Agent A, achieving the digital transfer of process experience.

[0061] For example, such as Figure 5As shown, after step S30 generates the initial action parameter set and candidate action sequence, both are first input into the dynamic anti-collision pre-simulation module. Within this module, a multi-sub-agent cross-verification mechanism is first triggered. The decision sub-agent generates a multi-test head motion command stream based on the candidate action sequence and maps the initial action parameter set into specific velocity curves, acceleration curves, and time-series control signals, forming a set of multi-axis motion trajectories to be executed. Subsequently, the security verification sub-agent constructs a virtual physical constraint space in the digital twin environment. This virtual physical constraint space includes the following: spatial envelope models of each test head; travel limit constraints for each axis; multi-axis interlocking logic constraints; current device attitude and real-time status parameters; and predicted trajectory data within a future time window. After constructing the virtual physical constraint space, the system performs an unfolded simulation calculation of the motion trajectories of the multiple test heads within a future preset time window, calculates the dynamic bounding box overlap rate between each test head, and simultaneously calculates the comprehensive security confidence level. Simultaneously, it performs logical verification on whether the motion trajectory exceeds the hardware physical limits and whether it triggers multi-axis deadlock constraints, generating a conflict constraint set. The system then proceeds to the physical safety constraint verification node. If the verification passes, the system outputs a safe action sequence and the corresponding set of safety action parameters, which is then passed to the subsequent physical execution module. If the verification fails, the system triggers the feedback replanning module. Based on the conflict constraint set, the feedback replanning module modifies the action sequence or parameter set, such as adjusting time delays, reducing velocity curve peaks, reordering test head execution, or reducing motion amplitude. The updated action sequence re-enters the dynamic collision avoidance pre-simulation process until verification passes. This forms a closed-loop safety decision-making process of "generation—pre-simulation—verification—replanning—re-pre-simulation".

[0062] Step S50: According to the safety passage action sequence and safety action parameter set The atomic service set S2 of the execution control layer and the atomic service set S3 of the cognitive analysis layer are invoked to perform impedance control physical execution. During the impedance control physical execution process, feedback data is collected, and closed-loop adaptive parameter tuning is performed based on the feedback data, outputting the impedance test result set O.

[0063] It should be noted that the closed-loop adaptive parameter tuning mechanism refers to the real-time acquisition of needle mark depth, contact force, current waveform and TDR reflection waveform data during the impedance control physical execution process, and comparison with preset thresholds. When a parameter deviation is detected, the Z-axis pressing depth, motion acceleration or sampling rate parameters are automatically adjusted, and the test action is repeated.

[0064] Understandably, this step, by establishing a closed-loop control chain of execution-feedback-parameter update, enables the equipment to automatically optimize parameters under different board thicknesses, copper foil hardnesses, or ambient temperatures, thereby improving test consistency. Compared to traditional manual machine adjustment methods, this step can achieve parameter convergence without manual intervention, significantly reducing test fluctuations caused by differences in operating experience.

[0065] For example, such as Figure 6 As shown, firstly, the hierarchical atomic service invocation mechanism is triggered. The atomic service set is invoked in the order of "execution control layer priority, cognitive analysis layer coordination," where execution control layer atomic service S2 is responsible for generating specific physical motion control signals, and cognitive analysis layer atomic service S3 is responsible for real-time analysis and quality assessment of the impedance signal. During the execution control layer invocation, the system generates an S-shaped trajectory planning curve to drive the multi-axis motion unit to complete alignment, approach, and flexible contact actions; subsequently, TDR sampling is performed to complete the physical impedance test process. During the physical test and impedance execution, the system collects feedback data in real time, including: needle mark depth data during the Z-axis contact process; force sensing data at the moment of contact; TDR current reflection waveform data; waveform signal-to-noise ratio and impedance curve fitting error indicators. Then, the RLAF closed-loop correction mechanism stage is entered. The system inputs the real-time collected pressure data and waveform data into the AI ​​feedback reinforcement learning evaluation module to score the quality of the current action parameter combination and determine whether it meets the preset impedance control target. When the evaluation result is determined to be "meets the standard," the system outputs the optimal impedance test result set O. When the evaluation result is determined to be "parameter deviation," the system enters the adaptive parameter correction stage. Differentiated correction strategies are implemented based on the type of deviation: when the needle mark depth exceeds a preset threshold, the system automatically reduces the Z-axis pressing depth or decreases the pressing acceleration; when insufficient contact force leads to high waveform noise, the system appropriately increases the contact force threshold; when the current waveform signal-to-noise ratio is insufficient or the impedance curve fitting error exceeds the limit, the system automatically increases the sampling rate or adjusts the sampling time window; when changes in ambient temperature cause impedance drift, the system corrects based on historical model compensation parameters. The corrected action parameter set is then input back into the execution control layer and cognitive analysis layer to re-execute the impedance test until the evaluation module determines it to be "optimal" or meets the target threshold.

[0066] Example 2: Furthermore, the present invention provides an impedance control system based on an AI Agent and MCP architecture, employing an impedance control method based on an AI Agent and MCP architecture as described in the above embodiments, which can solve a technical problem related to impedance control based on an AI Agent and MCP architecture. The beneficial effects of the impedance control system based on an AI Agent and MCP architecture provided by the present invention are the same as those of the impedance control method based on an AI Agent and MCP architecture provided in the above embodiments, and other technical features in the impedance control system based on an AI Agent and MCP architecture are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0067] Example 3: This invention provides an impedance control device based on an AI Agent and MCP architecture. Please refer to... Figure 7An impedance control device based on an AI Agent and MCP architecture 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, when executed by the at least one processor, enable the at least one processor to perform an impedance control method based on an AI Agent and MCP architecture as described in Embodiment 1 above. An impedance control device based on an AI Agent and MCP architecture in this embodiment of the invention 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), and in-vehicle terminals (e.g., in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. An impedance control device based on an AI Agent and MCP architecture is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the invention. An impedance control device based on an AI Agent and MCP architecture may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.) that 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. The random access memory 1004 also stores various programs and data required for the operation of the impedance control device based on the AI ​​Agent and MCP architecture. The processing unit 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An I / O interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touchscreen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 1003 including, for example, magnetic tape, hard disk, etc.; and communication devices 1009. Communication device 1009 allows an impedance control device based on an AI Agent and MCP architecture to communicate wirelessly or wiredly with other devices to exchange data. While the figure shows an impedance control device based on an AI Agent and MCP architecture with various systems, it should be understood that implementation or possession of all the systems shown is not required. More or fewer systems may be implemented alternatively.

[0068] 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 impedance control method based on an AI Agent and MCP architecture as described above. The computer program product provided by this invention can solve a technical problem related to impedance control based on an AI Agent and MCP architecture. Compared with the prior art, the beneficial effects of the computer program product provided by this invention are the same as those of the impedance control method based on an AI Agent and MCP architecture provided in the above embodiments, and will not be repeated here.

[0069] 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.

[0070] 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.

[0071] 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 impedance control method based on AI Agent and MCP architecture, characterized in that, The methods include: Step S10: Obtain user-side multimodal test requirement data and circuit board engineering data. Based on the user-side multimodal test requirement data and circuit board engineering data, use a multimodal intent understanding and requirement normalization mechanism to perform requirement structuring tasks and output a standardized requirement constraint set, a target network semantic set, and an initial candidate test point set. Step S20: Based on the standardized set of requirement constraints, the target network semantic set, and the initial candidate test point set, construct a three-level atomic capability service architecture, which includes a basic perception layer, an execution control layer, and a cognitive analysis layer. Then, based on this three-level atomic capability service architecture, use a layered invocation orchestration mechanism to execute the test context construction task and output a standardized test context. The execution control layer includes an atomic service set S2, and the cognitive analysis layer includes an atomic service set S3. Step S30: Based on standardized test context The preprocessing task is performed using a context resolution and action sequence generation mechanism driven by an adaptive secure execution agent, outputting candidate action sequences for concurrent testing of multiple test heads. and candidate action sequences The corresponding initial action parameter set ; Step S40: Based on candidate action sequences With the initial set of motion parameters The system employs a multi-sub-agent cross-verification mechanism and a dynamic anti-collision pre-simulation mechanism to perform security verification tasks and outputs a safe pass action sequence. and safety action parameter set ; Specifically, this includes: A decision sub-agent is predefined, which is responsible for making decisions based on candidate action sequences. Generate a multi-test-head motion command stream and set heartbeat synchronization and timestamp constraints for concurrent execution windows for the multi-test-head motion command stream; A security verification sub-agent is pre-defined. This sub-agent calls a pre-defined collision pre-simulation atomic service to perform rapid digital twin pre-simulation of the multi-test head motion command stream, and outputs the multi-test head dynamic bounding box overlap rate and security confidence. Simultaneously, the security verification sub-agent calls the preset collision pre-simulation atomic service to perform hardware travel limit verification and multi-axis interlock deadlock verification, outputting a set of conflict constraints. ; When the safety confidence level Below the preset security confidence threshold or conflict constraint set When not empty, block the multi-test-head motion command flow and base it on the conflict constraint set. Trigger path replanning and update to obtain a safe passage action sequence. and safety action parameter set ; Step S50: According to the safety passage action sequence and safety action parameter set The atomic service set S2 of the execution control layer and the atomic service set S3 of the cognitive analysis layer are invoked to perform impedance control physical execution. During the impedance control physical execution process, feedback data is collected, and closed-loop adaptive parameter tuning is performed based on the feedback data, outputting the impedance test result set O.

2. The impedance control method based on AI Agent and MCP architecture as described in claim 1, characterized in that, In step S10, the user-side multimodal test requirement data includes natural language requirement text, screenshot annotation images, and non-standard table attachments; the circuit board engineering data includes Gerber files, ODB++ layout files, stack-up information, and historical test logs.

3. The impedance control method based on AI Agent and MCP architecture as described in claim 1, characterized in that, In step S20, the basic perception layer provides an atomic service set S1 for circuit board data parsing and multimodal requirement understanding; the atomic service set S1 includes layout file streaming loading service, semantic layer parsing service, network topology construction service, PCB data parsing atomic services, and requirement normalization service; the execution control layer provides an atomic service set S2 for machine tool precision motion control and visual alignment interaction; the atomic service set S2 includes S-shaped trajectory planning service, flexible contact service, fine-tuning compensation service, and Mark point positioning service; the cognitive analysis layer provides an atomic service set S3 for TDR signal processing and process optimization decision-making; the atomic service set S3 includes waveform noise reduction service, impedance curve fitting service, layer stripping service, collision prediction service, and process optimization suggestion service.

4. The impedance control method based on AI Agent and MCP architecture as described in claim 3, characterized in that, Step S20 specifically includes: Step S201: Call the PCB data parsing class atomic service to perform streaming loading and semantic layer parsing on Gerber file and ODB++ layout file, identify differential line pair engineering semantic features and BGA area engineering semantic features, and establish board layer mapping based on differential line pair engineering semantic features and BGA area engineering semantic features, and output layout semantic feature set and board layer mapping information. Step S202: Based on the layout semantic feature set and board layer mapping information, call the network topology construction service to construct the electrical connection netlist and generate network topology information. The initial candidate test point set is subjected to constraint filtering processing by combining the standardized requirement constraint set and the target network semantic set, and the confirmed test point set is output. Step S203: Encapsulate the confirmed test point set, standardized requirement constraint set, and target network semantic set into a standardized test context. .

5. The impedance control method based on AI Agent and MCP architecture as described in claim 1, characterized in that, In step S30, the candidate action sequence Includes alignment actions, approach actions, flexible contact actions, TDR sampling actions, and retreat actions, with an initial set of action parameters. This includes parameters for velocity and acceleration along each axis, S-shaped trajectory planning parameters, Z-axis compression depth parameters, contact force threshold parameters, and sampling rate parameters.

6. The impedance control method based on AI Agent and MCP architecture as described in claim 1, characterized in that, In step S50, the step of performing closed-loop adaptive parameter tuning based on feedback data specifically includes: obtaining needle mark depth data from the feedback data; when the needle mark depth data does not meet the preset needle mark threshold constraint, automatically adjusting the Z-axis pressing depth or acceleration parameter; obtaining current waveform quality data from the feedback data; when the current waveform quality data does not meet the preset current waveform template data, automatically adjusting the sampling rate and repeating the TDR sampling action and backoff action.

7. An impedance control system based on an AI Agent and MCP architecture, applied to the impedance control method based on an AI Agent and MCP architecture as described in any one of claims 1 to 6, characterized in that, The impedance control system based on AI Agent and MCP architecture includes: The requirement structuring module is used to acquire user-side multimodal test requirement data and circuit board engineering data. Based on the user-side multimodal test requirement data and circuit board engineering data, it uses a multimodal intent understanding and requirement normalization mechanism to perform requirement structuring tasks and outputs a standardized set of requirement constraints, a target network semantic set, and an initial candidate test point set. The layered atomic capability architecture building module is used to construct a three-tiered atomic capability service architecture based on a standardized set of requirement constraints, a target network semantic set, and an initial candidate test point set. This three-tiered atomic capability service architecture includes a basic perception layer, an execution control layer, and a cognitive analysis layer. Based on this three-tiered atomic capability service architecture, a layered invocation orchestration mechanism is used to execute the test context construction task and output a standardized test context. The execution control layer includes an atomic service set S2, and the cognitive analysis layer includes an atomic service set S3. Execute the preprocessing generation module for use based on the standardized test context. The preprocessing task is performed using a context resolution and action sequence generation mechanism driven by an adaptive secure execution agent, outputting candidate action sequences for concurrent testing of multiple test heads. and candidate action sequences The corresponding initial action parameter set ; The secure cross-validation module is used for candidate action sequences. With the initial set of motion parameters The system employs a multi-sub-agent cross-verification mechanism and a dynamic anti-collision pre-simulation mechanism to perform security verification tasks and outputs a safe pass action sequence. and safety action parameter set ; Specifically, this includes: A decision sub-agent is predefined, which is responsible for making decisions based on candidate action sequences. Generate a multi-test-head motion command stream and set heartbeat synchronization and timestamp constraints for concurrent execution windows for the multi-test-head motion command stream; A security verification sub-agent is pre-defined. This sub-agent calls a pre-defined collision pre-simulation atomic service to perform rapid digital twin pre-simulation of the multi-test head motion command stream, and outputs the multi-test head dynamic bounding box overlap rate and security confidence. Simultaneously, the security verification sub-agent calls the preset collision pre-simulation atomic service to perform hardware travel limit verification and multi-axis interlock deadlock verification, outputting a set of conflict constraints. ; When the safety confidence level Below the preset security confidence threshold or conflict constraint set When not empty, block the multi-test-head motion command flow and base it on the conflict constraint set. Trigger path replanning and update to obtain a safe passage action sequence. and safety action parameter set ; The closed-loop execution and adaptive parameter tuning module is used to determine the safe passage action sequence. and safety action parameter set The atomic service set S2 of the execution control layer and the atomic service set S3 of the cognitive analysis layer are invoked to perform impedance control physical execution. During the impedance control physical execution process, feedback data is collected, and closed-loop adaptive parameter tuning is performed based on the feedback data, outputting the impedance test result set O.

8. An impedance control device based on an AI Agent and MCP architecture, characterized in that, The impedance control device based on AI Agent and MCP architecture includes: a memory, a processor, and an impedance control program based on AI Agent and MCP architecture stored in the memory and executable on the processor. When the impedance control program based on AI Agent and MCP architecture is executed by the processor, it implements an impedance control method based on AI Agent and MCP architecture as described in any one of claims 1 to 6.

9. A computer program product, characterized in that, The computer program product includes an impedance control program based on an AI Agent and MCP architecture. When the impedance control program based on the AI ​​Agent and MCP architecture is executed by the processor, it implements an impedance control method based on an AI Agent and MCP architecture as described in any one of claims 1 to 6.

Images