Numerical control agent execution device driven by large model

Abstract

The invention discloses a numerical control agent execution device driven by a large model, which belongs to the technical field of intelligent manufacturing and numerical control and comprises an agent subsystem, a real-time actuator subsystem, a virtual machine tool subsystem and a physical machine tool. The agent subsystem generates a verified instruction sequence through simulation interaction with the virtual machine tool subsystem; the real-time actuator subsystem is responsible for analyzing and executing the instruction generated by the intelligent agent and driving the physical machine tool to move; the virtual machine tool subsystem is used for verifying and optimizing instructions of the intelligent agent subsystem and providing simulation verification support; and the physical machine tool is driven by the real-time actuator subsystem to execute final machining operation. According to the numerical control agent execution device driven by the large model, intelligent optimization of a machine tool machining process is achieved through driving of the cutting force model, the control model and the error model in the agent, and the machining efficiency and the machining quality are improved under the condition that the machining dynamic precision is guaranteed.

Description

Technical Field

[0001] This invention relates to the fields of intelligent manufacturing and numerical control technology, and in particular to a large-model driven numerical control intelligent agent execution device. Background Technology

[0002] Existing large-model-driven CNC systems are based on traditional instruction modes such as G-code, relying heavily on fixed CAM, decoding, trajectory interpolation, and servo loops. Their function chains are rigid, and their compensation methods are limited, making them ill-suited for complex process environments and dynamic disturbances. Furthermore, their input / output designs cannot meet the interactive requirements of modern large-model intelligent agents. Existing large-model-driven CNC systems suffer from the following pain points: The contradiction between the real-time requirements of CNC devices and the non-real-time output of large models; How to influence and adjust the control of real-time systems based on non-real-time data; How to combine intelligent agents with real-time control to effectively improve the accuracy, quality and efficiency of CNC devices; How to replace manual optimization processes based on intelligent agent applications. Summary of the Invention

[0003] The purpose of this invention is to provide a large-model driven CNC intelligent agent execution device, which realizes intelligent optimization of machine tool processing technology through the driving of cutting force model, control model and error model in the intelligent agent, thereby improving processing efficiency and processing quality while ensuring processing dynamic accuracy.

[0004] To achieve the above objectives, the present invention provides a large model-driven CNC intelligent agent execution device, including an intelligent agent subsystem, a real-time actuator subsystem, a virtual machine tool subsystem, and a physical machine tool; The intelligent agent subsystem generates a verified instruction sequence through simulation interaction with the virtual machine tool subsystem. The intelligent agent subsystem includes a path generator, a cutting force model functional module, a control model functional module, a trajectory planner, and an error compensator. The path generator uses reinforcement learning to convert the process planning or part geometry information generated by the large model layer into an initial toolpath. The path generator extracts the machining area, layer depth, tool entry method, and tool type. Based on material properties, tool geometry parameters, and machining strategy, it automatically generates roughing, semi-finishing, and finishing paths. The path planning for multi-axis motion is shown below: ; in, For the generated path points, 1~ Indicates the axis number for multi-axis linkage. and Indicates the first axis and the second axis. The motion function of the axis, For feed rate, For the tool radius, This refers to the tool length; The real-time actuator subsystem is responsible for parsing and executing the instructions generated by the agent to drive the movement of the physical machine tool. The real-time actuator subsystem includes an instruction sequence executor, a motion control module, a PLC and I / O control module, a low-level interpolator, and a feedback module. The instruction sequence executor uses a circular buffer to receive and buffer the instruction sequence of the agent, parses the instruction sequence generated by the agent, and converts it into executable motion control commands. It is responsible for the real-time execution of the commands, controlling the movement of the machine tool, and feeding back key information to the agent. The virtual machine bed subsystem is used to verify and optimize the instructions of the intelligent agent subsystem, providing simulation verification support; The physical machine tool is driven by a real-time actuator subsystem to perform the final machining operation.

[0005] Preferably, the cutting force model function module is used to predict the distribution of instantaneous cutting force, torque and thermal effects during the machining process in order to evaluate the rationality of the path and feed parameters; The control model functional module is used for agent optimization and iteration. It uses a neural network algorithm to fit the control characteristics of the actual machine tool, simulates the output of the agent, and further obtains evaluation data. The trajectory planner is used to convert the discrete path points output by the path generator into continuous executable trajectory instructions while ensuring smoothness, machining accuracy and real-time performance. The error compensator corrects geometric and dynamic errors generated during machining online based on the prediction results of the cutting force model function module and the control model function module.

[0006] Preferably, the motion control module includes multi-axis control, interpolation algorithm and real-time control to achieve precise control of motion trajectory; The PLC and I / O control module enable real-time logic processing of the machine tool, ensure safe logic execution, and provide a standardized instruction sequence interface. The low-level interpolator performs fine interpolation on the input coarse interpolation trajectory data to improve control accuracy; The feedback module switches to the traditional motion control mode based on the abnormal information driven by the intelligent agent, thereby adjusting and restoring the machine tool status.

[0007] Preferably, the virtual machine tool subsystem consists of a machine tool model, a control model, and a cutting force model. It performs virtual modeling of the machine tool by collecting motion state data of the actual machine tool and fits the model state of the actual machine tool through a neural network.

[0008] Preferably, the cutting force model module calculates the cutting force, feed force, and normal component force based on the tool-workpiece contact geometry and material properties, simulating the load variation trend under different spindle speeds and feed rates. The specific calculation formula is shown below: ; in, For cutting force, This is the correction value for the unit cutting force. For cutting depth, For cutting speed, For the tool radius, Main spindle speed , , The index determined by the experiment.

[0009] Preferably, the control model function module simulates the motion process of the physical machine tool by calculating the periodic target current. The calculation process of the periodic target current is as follows: ; in, For maximum speed, For maximum acceleration, For the target location, This is the proportionality coefficient. The integral coefficient is... The differential coefficients are... For a period of time, The target time is the periodic target time.

[0010] Preferably, the trajectory planner calculates the periodic target position to generate a smooth trajectory that satisfies acceleration and jerk constraints. This trajectory, in conjunction with the control model functional module, enables segmented dynamic optimization of the trajectory. The periodic target position for multi-axis motion is shown below: ; in, For maximum jerk, This represents the error value caused by the cutting force.

[0011] Preferably, the error compensator calculates the position error during the machining process, models and compensates for the trajectory error caused by each axis control and cutting force, and outputs a dynamic compensation vector to correct the trajectory planner or control model output. The formula for calculating the position error is as follows: ; in, For speed coefficient, For quality coefficient, This refers to the speed deviation.

[0012] Therefore, the present invention adopts the above-mentioned large model driven CNC intelligent agent execution device, which realizes intelligent optimization of machine tool processing technology through the driving of cutting force model, control model and error model in intelligent agent, thereby improving processing efficiency and processing quality while ensuring processing dynamic accuracy.

[0013] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the overall architecture of a large-model driven numerical control intelligent agent execution device according to the present invention. Figure 2 This is a schematic diagram of the architecture of the real-time actuator subsystem of a large-model driven numerical control intelligent agent execution device according to the present invention. Detailed Implementation

[0015] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0016] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0017] Example 1 like Figures 1 to 2 As shown, the present invention provides a large model-driven CNC intelligent agent execution device, including an intelligent agent subsystem, a real-time actuator subsystem, a virtual machine tool subsystem, and a physical machine tool; the intelligent agent subsystem generates a verified instruction sequence through simulation interaction with the virtual machine tool subsystem; The intelligent agent subsystem includes a path generator, a cutting force model functional module, a control model functional module, a trajectory planner, and an error compensator; The path generator, based on reinforcement learning and other technologies, is responsible for converting the process planning or part geometry information generated from the large model layer into initial toolpaths that conform to the kinematic constraints of the machine tool. Its main functions include: extracting information such as machining area, layer depth, tool entry method, and tool type from the CAD / CAM model; automatically generating roughing, semi-finishing, and finishing paths based on material properties, tool geometry parameters, and machining strategies; supporting multi-axis linkage (3-axis / 5-axis) and toolpath planning for complex curved surfaces; and outputting a unified format path data stream for subsequent modules to call.

[0018] The path generator extracts the machining area, layer depth, tool entry method, and tool type. Based on material properties, tool geometry parameters, and machining strategy, it automatically generates roughing, semi-finishing, and finishing paths. The path planning for multi-axis motion is shown below: ; in, For the generated path points, 1~ Indicates the axis number for multi-axis linkage. and Indicates the first axis and the second axis. The motion function of the axis, For feed rate, For the tool radius, This is the length of the cutting tool.

[0019] The cutting force model module is used to predict the instantaneous cutting force, torque, and thermal effect distribution during the machining process to evaluate the rationality of the path and feed parameters. Its main functions include: calculating the cutting force, feed force, and normal component force based on the tool-workpiece contact geometry and material properties; and simulating the load variation trend under different spindle speeds and feed rates.

[0020] The cutting force model module calculates cutting force, feed force, and normal component force based on tool-workpiece contact geometry and material properties, simulating load variation trends under different spindle speeds and feed rates. The specific calculation formulas are shown below: ; in, For cutting force, This is the correction value for the unit cutting force. For cutting depth, For cutting speed, For the tool radius, Main spindle speed , , The index determined by the experiment.

[0021] The control model module is used for agent optimization iteration. It fits the control characteristics of the actual machine tool through a neural network algorithm, simulates the output of the agent, and further obtains evaluation data. The neural network algorithm adopts existing technology, only the input and output are different. Its main functions include: fitting the control characteristics of the target machine tool based on the neural network and reproducing it with the model; and predicting the dynamic error of each axis due to the control characteristics based on the fitted model.

[0022] The control model module simulates the motion process of a physical machine tool by calculating a periodic target current. The calculation process of the periodic target current is as follows: ; in, For maximum speed, For maximum acceleration, For the target location, This is the proportionality coefficient. The integral coefficient is... The differential coefficients are... For a period of time, The target time is the periodic target time.

[0023] The trajectory planner is used to convert discrete path points output by the path generator into continuous executable trajectory instructions while ensuring smoothness, machining accuracy, and real-time performance. Its main functions include: generating smooth trajectories that meet acceleration and jerk constraints; supporting S-curve, B-spline, and time-optimal trajectory planning algorithms; considering multi-axis synchronization, velocity look-ahead, and path error feedback; and working in conjunction with the control model to achieve dynamic optimization of trajectory segments.

[0024] The trajectory planner calculates the periodic target position and generates a smooth trajectory that satisfies acceleration and jerk constraints. The periodic target position for multi-axis motion is shown below: ; in, For maximum jerk, This represents the error value caused by the cutting force.

[0025] The error compensator, based on the prediction results of the cutting force model function module and the control model function module, corrects the geometric and dynamic errors generated during the machining process online. Its main functions include: modeling and compensating for trajectory errors caused by each axis control and cutting force; using existing error compensation algorithms to output dynamic compensation vectors to directly correct the output of the trajectory planner or control model; and supporting model update and self-learning mechanisms to achieve self-evolution of compensation strategies.

[0026] The error compensator calculates the position error during the machining process, models and compensates for trajectory errors caused by each axis control and cutting force, and outputs a dynamic compensation vector to correct the trajectory planner or control model output. The formula for calculating the position error is as follows: ; in, For speed coefficient, For quality coefficient, This refers to the speed deviation.

[0027] As the core execution layer of the agent-driven CNC system, the real-time actuator subsystem maintains the basic capabilities of traditional interpolation, multi-axis servo, PLC logic and safety protection, while adding an instruction sequence actuator layer and a virtual machine tool model for process closed loop. The agent-based functions and traditional functions are integrated and unified through the instruction sequence actuator layer, and the intelligentization and process-in-the-loop characteristics are realized based on the virtual machine tool model and agent instruction sequence.

[0028] In terms of instruction processing, the real-time actuator subsystem can parse and execute high-level machining instruction sequences generated by the intelligent agent, including motion control, I / O operations, and parameter tuning, and ensures determinism and continuity of execution through high-speed buffering and priority scheduling. At the trajectory control layer, the real-time actuator subsystem not only possesses high-precision interpolation and smooth acceleration / deceleration algorithms, but also introduces an observer-based multi-axis trajectory error feedforward compensation mechanism to correct complex disturbances such as thermal errors, tool wear, and structural deformation in real time, thus overcoming the limitations of traditional systems relying on single PID or static feedforward.

[0029] Meanwhile, the real-time actuator subsystem integrates process-in-the-loop closed-loop capabilities: on the one hand, it integrates the prediction results of the virtual machine tool, and on the other hand, it combines the feedback data from the physical machine tool to achieve online closed-loop trajectory optimization and error compensation, ensuring that machining accuracy remains stable under dynamic conditions. In terms of safety, it is designed with a multi-level backoff mechanism, which can quickly switch to traditional CNC mode or safely shut down the machine in the event of agent or observer failure, preventing damage to the machine tool and workpiece.

[0030] The real-time actuator subsystem transforms from a "fixed-logic motion controller" to a "bridge between intelligent agents and physical execution," ensuring both the determinism and security of hard real-time operation and endowing the system with intelligent optimization and adaptive capabilities. It is a key innovation of the next generation of intelligent CNC systems.

[0031] The real-time actuator subsystem is responsible for parsing and executing the instructions generated by the intelligent agent to drive the physical machine tool to move. The real-time actuator subsystem includes an instruction sequence executor, a motion control module, a PLC and I / O control module, a low-level interpolator, and a feedback module. The instruction sequence executor integrates the traditional motion control kernel with the execution requirements of the intelligent agent, and designs a standard instruction sequence format to ensure the real-time execution of control commands. It uses a circular buffer to receive and cache the intelligent agent's instruction sequence, parses the instruction sequence generated by the intelligent agent, and converts it into executable motion control commands. This buffer is responsible for the real-time execution of the commands, controlling the machine tool's movements, and feeding back key information to the intelligent agent to improve the accuracy of the intelligent agent model.

[0032] The motion control module includes multi-axis control, interpolation algorithms, and real-time control, enabling precise control of the motion trajectory; it also provides traditional control functions such as manual and MDI; and the motion control output is compatible with the standard format of instruction sequences.

[0033] The PLC and I / O control module realize the real-time logic processing of the machine tool, such as sensor input, equipment switching, status monitoring, etc., to ensure safe logic execution and provide a standardized instruction sequence interface; The low-level interpolator performs fine interpolation on the input coarse interpolation trajectory data, thereby improving control accuracy.

[0034] When an abnormality occurs in the intelligent agent drive, the feedback module switches to the traditional motion control mode based on the abnormal information of the intelligent agent drive, so as to realize the adjustment and recovery of the machine tool status.

[0035] The virtual machine tool subsystem is used to verify and optimize the instructions of the intelligent agent subsystem and provides simulation verification support. The virtual machine tool subsystem consists of a machine tool model, a control model and a cutting force model. It performs virtual modeling of the machine tool by collecting motion state data of the actual machine tool and fits the model state of the actual machine tool through a neural network.

[0036] The physical machine tool is driven by a real-time actuator subsystem to perform the final machining operation.

[0037] Therefore, the present invention adopts the above-mentioned large model driven CNC intelligent agent execution device, which realizes intelligent optimization of machine tool processing technology through the driving of cutting force model, control model and error model in intelligent agent, thereby improving processing efficiency and processing quality while ensuring processing dynamic accuracy.

[0038] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A large-model-driven numerical control intelligent agent execution device, characterized in that: It includes an intelligent agent subsystem, a real-time actuator subsystem, a virtual machine tool subsystem, and a physical machine tool; The intelligent agent subsystem generates a validated sequence of instructions through simulation interaction with the virtual machine bed subsystem; The intelligent agent subsystem includes a path generator, a cutting force model functional module, a control model functional module, a trajectory planner, and an error compensator; The path generator uses reinforcement learning to convert process plans or part geometry information generated from large model layers into initial toolpaths; The path generator extracts the machining area, layer depth, tool entry method, and tool type. Based on material properties, tool geometry parameters, and machining strategy, it automatically generates roughing, semi-finishing, and finishing paths. The path planning for multi-axis motion is shown below: ； in, For the generated path points, 1~ Indicates the axis number for multi-axis linkage. and Indicates the first axis and the second axis. The motion function of the axis, For feed rate, For the tool radius, This refers to the tool length; The real-time actuator subsystem is responsible for parsing and executing the instructions generated by the intelligent agent to drive the physical machine tool to move. The real-time actuator subsystem includes an instruction sequence executor, a motion control module, a PLC and I / O control module, a low-level interpolator, and a feedback module. The instruction sequence executor uses a circular buffer to receive and cache the instruction sequence of the agent, parses the instruction sequence generated by the agent, and converts it into executable motion control commands. It is responsible for the real-time execution of the commands, controls the movement of the machine tool, and feeds back key information to the agent. The virtual machine bed subsystem is used to verify and optimize the instructions of the intelligent agent subsystem, providing simulation verification support; The physical machine tool is driven by a real-time actuator subsystem to perform the final machining operation.

2. The large-model-driven numerical control intelligent agent execution device according to claim 1, characterized in that: The cutting force model module is used to predict the distribution of instantaneous cutting force, torque and thermal effects during the machining process in order to evaluate the rationality of the path and feed parameters; The control model functional module is used for agent optimization and iteration. It uses a neural network algorithm to fit the control characteristics of the actual machine tool, simulates the output of the agent, and further obtains evaluation data. The trajectory planner is used to convert the discrete path points output by the path generator into continuous executable trajectory instructions while ensuring smoothness, machining accuracy and real-time performance. The error compensator corrects geometric and dynamic errors generated during machining online based on the prediction results of the cutting force model function module and the control model function module.

3. The large-model-driven numerical control intelligent agent execution device according to claim 2, characterized in that: The motion control module includes multi-axis control, interpolation algorithms, and real-time control, enabling precise control of the motion trajectory; The PLC and I / O control module enable real-time logic processing of the machine tool, ensure safe logic execution, and provide a standardized instruction sequence interface. The low-level interpolator performs fine interpolation on the input coarse interpolation trajectory data to improve control accuracy; The feedback module switches to the traditional motion control mode based on the abnormal information driven by the intelligent agent, thereby adjusting and restoring the machine tool status.

4. The large-model-driven numerical control intelligent agent execution device according to claim 3, characterized in that: The virtual machine tool subsystem consists of a machine tool model, a control model, and a cutting force model. It performs virtual modeling of the machine tool by collecting motion state data of the actual machine tool and fits the model state of the actual machine tool through a neural network.

5. The large-model-driven numerical control intelligent agent execution device according to claim 4, characterized in that: The cutting force model module calculates cutting force, feed force, and normal component force based on tool-workpiece contact geometry and material properties, simulating load variation trends under different spindle speeds and feed rates. The specific calculation formulas are shown below: ； in, For cutting force, This is the correction value for the unit cutting force. For cutting depth, For cutting speed, For the tool radius, Main spindle speed , , The index determined by the experiment.

6. The large-model-driven numerical control intelligent agent execution device according to claim 5, characterized in that: The control model module simulates the motion process of a physical machine tool by calculating a periodic target current. The calculation process of the periodic target current is as follows: ； in, For maximum speed, For maximum acceleration, For the target location, This is the proportionality coefficient. The integral coefficient is... The differential coefficients are... For a period of time, The target time is the periodic target time.

7. The large-model-driven numerical control intelligent agent execution device according to claim 6, characterized in that: The trajectory planner calculates the periodic target position and generates a smooth trajectory that satisfies acceleration and jerk constraints. Working in conjunction with the control model module, it achieves dynamic optimization of the trajectory segments. The periodic target position for multi-axis motion is shown below: ； in, For maximum jerk, This represents the error value caused by the cutting force.

8. The large-model-driven numerical control intelligent agent execution device according to claim 7, characterized in that: The error compensator calculates the position error during the machining process, models and compensates for trajectory errors caused by each axis control and cutting force, and outputs a dynamic compensation vector to correct the trajectory planner or control model output. The formula for calculating the position error is as follows: ； in, For speed coefficient, For quality coefficient, This refers to the speed deviation.

Images