Dynamic regulation method and device for lookahead simulation and hybrid optimization of numerical control machining, computer equipment and medium

Abstract

Embodiments of the present application provide a kind of for numerical control processing's dynamic regulation and control method, device, computer equipment and medium of advance simulation and mixed optimization, it is related to numerical control processing technical field, wherein, the method includes the following steps: the synchronous driving data of instruction data stream and the synchronous driving data of multi-source sensor data stream generation are injected into digital twin model;Solving numerical control code generates corresponding future tool movement trajectory set, based on future tool movement trajectory set, determine dynamic local focus domain, generate the quantitative prediction information of future processing state;Multi-dimensional mixed state vector is constructed, and feedforward optimization adjustment and feedback optimization compensation are generated;The feedforward optimization adjustment and feedback optimization compensation are weighted and fused, and optimization decision adjustment is generated, and the effect after regulation and control is fed back to feedforward feedback mixed optimization framework and is carried out online learning and parameter update.The scheme improves the stability and precision of complex part processing by advance simulation and mixed optimization.

Description

Technical Field

[0001] This invention relates to the field of CNC machining technology, and in particular to a dynamic control method, device, computer equipment, and medium for advanced simulation and hybrid optimization in CNC machining. Background Technology

[0002] In the field of high-end CNC machining, especially when machining parts with complex curved surfaces and high-strength material properties, the stability, accuracy, and efficiency of the machining process are crucial. Currently, mainstream machining process control methods have the following shortcomings:

[0003] Existing technology 1: Machining based on fixed parameters or offline optimization. Process parameters are preset based on experience or offline simulation and remain constant throughout the machining process. This method cannot respond to dynamic changes that occur in real time during machining, such as gradual tool wear, local hardness differences in the workpiece material, and sudden changes in cutting force. This often leads to fluctuations in machining quality, unexpected tool damage, or the need to adopt conservative parameters for safety, sacrificing efficiency.

[0004] Existing technology 2: Passive adjustment based on real-time sensor feedback. This method monitors the processing status through sensors such as vibration and current. When an abnormality is detected (such as chatter or excessive load), it performs post-processing adjustments (such as speed reduction or shutdown). This "sensor-response" mode has a response lag, cannot intervene before problems occur, and the adjustment process may cause secondary disturbances, resulting in coarse control for high-precision machining.

[0005] Existing technology three: Purely data-driven online optimization. This method uses machine learning models to directly map optimization parameters from real-time data. However, it heavily relies on large amounts of high-quality training data. When operating conditions exceed the training range, the reliability of decisions drops sharply, and the model has poor interpretability, posing trust barriers and risks in practical industrial applications.

[0006] In summary, existing technologies lack an effective means to proactively predict changes in processing status and to perform real-time, accurate, and stable parameter control based on an interpretable hybrid model. Summary of the Invention

[0007] In view of this, embodiments of the present invention provide a dynamic control method for CNC machining using advanced simulation and hybrid optimization, to solve the technical problem of the lack of control methods in the prior art capable of predicting changes in machining state. The method includes:

[0008] The system acquires the instruction data stream of the physical numerical control system and the multi-source sensor data stream of various sensors. It performs time stamp alignment and data fusion processing on the instruction data stream and the multi-source sensor data stream to generate synchronous driving data. The synchronous driving data is then injected into a digital twin model, wherein the digital twin model is used to drive the virtual-real synchronous simulation of the physical processing process.

[0009] The system reads the CNC code within a future time window, solves the CNC code to generate a set of corresponding future tool motion trajectories, and, based on the set of future tool motion trajectories, determines a dynamic local area of ​​interest on the workpiece model of the digital twin model. Within the dynamic local area of ​​interest, it performs local material removal simulation and updates the workpiece model, generating quantitative prediction information for future machining states. The dynamic local area of ​​interest is the local region where the future tool envelope and the workpiece model are expected to interfere.

[0010] A multidimensional hybrid state vector containing mechanistic features, real-time data features, and the quantized prediction information is constructed. A feedforward-feedback hybrid optimization framework is constructed. Based on the multidimensional hybrid state vector, feedforward optimization adjustment and feedback optimization compensation are generated through the feedforward-feedback hybrid optimization framework.

[0011] The fusion weights are dynamically calculated based on the prediction confidence and real-time noise level. The feedforward optimization adjustment amount and the feedback optimization compensation amount are then weighted and fused to generate an optimization decision adjustment amount. This optimization decision adjustment amount is then sent to the physical numerical control system for execution to regulate the machining process. The effect of the regulation is then fed back to the feedforward-feedback hybrid optimization framework for online learning and parameter updates.

[0012] This invention also provides a dynamic control device for advanced simulation and hybrid optimization in CNC machining, addressing the technical problem of the lack of control methods in the prior art capable of predicting changes in machining states. The device includes:

[0013] The digital twin model module is used to acquire the instruction data stream of the physical numerical control system and the multi-source sensor data stream of various sensors. It performs time stamp alignment and data fusion processing on the instruction data stream and the multi-source sensor data stream to generate synchronous driving data. The synchronous driving data is then injected into the digital twin model, wherein the digital twin model is used to drive the virtual-real synchronous simulation of the physical processing process.

[0014] A quantitative prediction information generation module is used to pre-read the CNC code within a future time window, solve the CNC code to generate a corresponding set of future tool motion trajectories, and based on the set of future tool motion trajectories, determine a dynamic local interest region on the workpiece model of the digital twin model, perform local material removal simulation within the dynamic local interest region and update the workpiece model to generate quantitative prediction information for future machining states. The dynamic local interest region is the local area where the expected interference between the future tool envelope and the workpiece model is expected.

[0015] The parameter optimization module is used to construct a multidimensional hybrid state vector containing mechanistic features, real-time data features and the quantized prediction information, construct a feedforward-feedback hybrid optimization framework, and generate feedforward optimization adjustment amount and feedback optimization compensation amount based on the multidimensional hybrid state vector through the feedforward-feedback hybrid optimization framework.

[0016] The dynamic control processing module is used to dynamically calculate the fusion weight based on the prediction confidence and real-time noise level, weight and fuse the feedforward optimization adjustment amount and the feedback optimization compensation amount to generate the optimization decision adjustment amount, and send the optimization decision adjustment amount to the physical numerical control system for execution to control the processing process, and feed back the effect of the control to the feedforward-feedback hybrid optimization architecture for online learning and parameter updates.

[0017] This invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements any of the above-mentioned dynamic control methods for advanced simulation and hybrid optimization of CNC machining, thereby solving the technical problem in the prior art of lacking a control method capable of predicting changes in machining state in advance.

[0018] This invention also provides a computer-readable storage medium storing a computer program that executes any of the above-described dynamic control methods for CNC machining involving advanced simulation and hybrid optimization, in order to solve the technical problem in the prior art of lacking control methods capable of predicting changes in machining state in advance.

[0019] Compared with the prior art, the beneficial effects that at least one technical solution adopted in the embodiments of this specification can achieve include at least:

[0020] Achieving high-fidelity, millisecond-level synchronous digital twins of the machining process, and possessing short-term advanced simulation capabilities, allows for the prediction of upcoming geometric and physical state changes during machining. A hybrid-driven optimization model integrating physical mechanisms and real-time data is constructed, utilizing advanced predictive information for feedforward optimization and combining it with real-time feedback for compensation, achieving collaborative dynamic optimization of multiple objectives (quality, efficiency, and stability). Ultimately, a closed-loop control system combining "predictive feedforward" and "adaptive feedback" is formed, fundamentally improving the stability, accuracy, and overall efficiency of complex part machining. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. 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 of a dynamic control method for advanced simulation and hybrid optimization in CNC machining, provided by an embodiment of the present invention.

[0023] Figure 2 This is a flowchart of a dynamic control method for advanced simulation and hybrid optimization for CNC machining, provided by an embodiment of the present invention;

[0024] Figure 3 This is a flowchart illustrating the definition of the tool's future path and dynamic domain of interest, provided in an embodiment of the present invention.

[0025] Figure 4 This is a flowchart of local voxel state update and mesh reconstruction provided in an embodiment of the present invention;

[0026] Figure 5 This is a flowchart illustrating the core working principle and data flow of the hybrid-driven process parameter dynamic optimization module provided in this embodiment of the invention.

[0027] Figure 6 This is a structural block diagram of a computer device provided in an embodiment of the present invention;

[0028] Figure 7 This is a structural block diagram of a dynamic control device for advanced simulation and hybrid optimization in CNC machining, provided in an embodiment of the present invention. Detailed Implementation

[0029] The embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0030] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. This application can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0031] In this embodiment of the invention, a dynamic control method for advanced simulation and hybrid optimization in CNC machining is provided, such as... Figure 1 As shown, the method includes:

[0032] Step S101: Obtain the instruction data stream of the physical numerical control system and the multi-source sensor data stream of various sensors; perform timestamp alignment and data fusion processing on the instruction data stream and the multi-source sensor data stream to generate synchronous driving data; inject the synchronous driving data into the digital twin model, wherein the digital twin model is used to drive the virtual-real synchronous simulation of the physical processing process.

[0033] Step S102: Pre-read the CNC code within the future time window, solve the CNC code to generate the corresponding set of future tool motion trajectories, and based on the set of future tool motion trajectories, determine the dynamic local interest region on the workpiece model of the digital twin model. Perform local material removal simulation within the dynamic local interest region and update the workpiece model to generate quantitative prediction information for future machining states. The dynamic local interest region is the local area where the expected interference between the future tool envelope and the workpiece model is expected.

[0034] Step S103: Construct a multidimensional hybrid state vector containing mechanistic features, real-time data features and the quantized prediction information; construct a feedforward-feedback hybrid optimization framework; and generate feedforward optimization adjustment amount and feedback optimization compensation amount based on the multidimensional hybrid state vector and the feedforward-feedback hybrid optimization framework.

[0035] Step S104: Dynamically calculate the fusion weight based on the prediction confidence and real-time noise level, and perform weighted fusion of the feedforward optimization adjustment amount and the feedback optimization compensation amount to generate the optimization decision adjustment amount. Send the optimization decision adjustment amount to the physical numerical control system for execution to regulate the machining process, and feed back the effect of the regulation to the feedforward-feedback hybrid optimization framework for online learning and parameter updates.

[0036] In specific implementation, the following steps are used to acquire the instruction data stream of the physical numerical control system and the multi-source sensor data stream from various sensors; the instruction data stream and the multi-source sensor data stream are then time-stamp aligned and fused to generate synchronous drive data:

[0037] Real-time acquisition of instruction data streams output by the physical numerical control system via industrial communication protocols. Where t is time, the instruction data stream includes preparation function code, auxiliary function code, and actual position instructions for each motion axis; multi-source sensor data streams output by various sensors are acquired in real time through the data acquisition interface. ,in, , The sampled value of the i-th sensor at time t; for the instruction data stream and the multi-source sensor data stream A globally unified timestamp is added to each frame of data. Based on the globally unified timestamp, the command data stream and the multi-source sensor data stream are aligned on the time axis to obtain aligned command data and aligned sensor data. The aligned sensor data is then input into a pre-calibrated mapping model. Calculate the equivalent virtual cutting force applied to the tool-workpiece contact area of ​​the digital twin model. ,in, The aligned instruction data and the equivalent virtual cutting force are then compared. Data fusion is performed to generate synchronous driving data; the synchronous driving data is injected into the digital twin model, and the virtual axis of the machine tool is driven to move through the aligned command data, and the equivalent virtual cutting force is applied to the tool-workpiece contact area of ​​the digital twin model.

[0038] In practice, the following steps are used to pre-read the CNC code within a future time window and calculate the CNC code to generate a corresponding set of future tool motion trajectories:

[0039] Create an advanced computing thread in the digital twin model that is independent of the real-time synchronous simulation thread;

[0040] The aforementioned advance calculation thread continuously reads the current moment from the instruction buffer of the physical numerical control system. Then preset duration The CNC code within the future time window is used to extract motion commands and feed rate commands from the CNC code. Based on the current kinematic configuration of the machine tool, the motion commands are preprocessed by interpolation, and the displacement time sequence of each axis of the machine tool within the future time window is calculated in combination with the feed rate commands. The displacement time sequence of each axis is transformed to the workpiece coordinate system, and the discrete trajectory point set of the tool relative to the workpiece surface within the future time window is calculated. The discrete trajectory point set is then curve-fitted to generate a continuous set of future tool motion trajectories. ,in, .

[0041] In specific implementation, the following steps are used to determine a dynamic local interest region on the workpiece model of the digital twin model based on the set of future tool motion trajectories, perform local material removal simulation within the dynamic local interest region, update the workpiece model, and generate quantitative prediction information for future machining states:

[0042] Based on the set of future tool motion trajectories Generate the set of future tool movement trajectories. Determined tool envelope ,in, For time; obtain the set of all voxels of the workpiece blank. Traverse the entire set of voxels of the workpiece For all voxels, calculate the center point coordinates of each voxel i. To the future tool envelope The minimum distance is calculated, and a dynamic local interest region is constructed based on the minimum distance. , ,in, It is the Euclidean distance function. The preset safety margin threshold, For the current moment, For forward prediction, the short time step, For time; in the dynamic local region of interest Inside, for each center point, the coordinates are... The voxels are updated to reflect the material state of the corresponding voxels. ,in, For indicator functions, The preset duration; based on the dynamic local interest region. Voxels whose internal material states change, within the dynamic local region of interest Incremental update of workpiece surface triangular mesh Based on the dynamic local area of ​​interest For voxels showing changes in the internal material state, calculate the instantaneous material removal rate at future moments. ,in, The volume of a single voxel unit; based on the set of future tool motion trajectories. With the triangular mesh on the surface of the workpiece Based on the geometric relationship, the tool-workpiece contact arc length is calculated. Tool-workpiece contact arc length variation rate The instantaneous material removal rate and the rate of change of the contact arc length between the tool and the workpiece As a quantitative predictive information for future processing status .

[0043] In specific implementation, the following steps are used to construct a multidimensional hybrid state vector containing mechanistic features, real-time data features, and the quantized prediction information; construct a feedforward-feedback hybrid optimization framework; and generate feedforward optimization adjustment and feedback optimization compensation amounts based on the multidimensional hybrid state vector and the feedforward-feedback hybrid optimization framework:

[0044] Based on the current process parameters and physical model, the mechanism characteristics are calculated. The mechanistic features include instantaneous cutting force components estimated based on the cutting force model. For time; perform time-frequency domain analysis on the aligned sensor data to extract real-time data features. The real-time data features include the energy of the vibration signal within a preset characteristic frequency band; the quantization prediction information is obtained. The mechanism features The real-time data features and the quantized prediction information The vectors are concatenated to generate a multidimensional mixed state vector. ,in, Construct a hybrid feedforward-feedback optimization framework that includes feedforward optimization pathways and feedback optimization pathways, and integrate the quantized prediction information. The input is given to the feedforward optimization path, and the output is the feedforward optimization adjustment amount, which is used to adjust the multidimensional mixed state vector. The input is fed into the feedback optimization path, and the feedback optimization compensation amount is output.

[0045] In specific implementation, the following steps are used to construct a feedforward-feedback hybrid optimization framework that includes feedforward optimization paths and feedback optimization paths, and to integrate the quantized prediction information. The input is given to the feedforward optimization path, and the output is the feedforward optimization adjustment amount, which is used to adjust the multidimensional mixed state vector. The input is fed into the feedback optimization path, and the feedback optimization compensation amount is output:

[0046] The quantized prediction information Tool-workpiece contact arc length variation rate Input to the feedforward optimization path, output feedforward optimization adjustment amount ,in, , For feedforward gain, For the rated contact arc length, For the rated feed rate; construct a lightweight neural network. By simplifying the Jacobian matrix of the physical model Approximate initialization of partial weights of the first layer of the lightweight neural network An attention layer is integrated into the lightweight neural network, and the attention weights of the attention layer are calculated. ; through the lightweight neural network Construct a feedback optimization path; convert the multidimensional mixed state vector The input is fed into the feedback optimization path, and the output is the feedback optimization compensation amount. ,in, .

[0047] In specific implementation, the following steps are used to dynamically calculate the fusion weight based on the prediction confidence and real-time noise level; weight the feedforward optimization adjustment amount and the feedback optimization compensation amount to generate the optimization decision adjustment amount; send the optimization decision adjustment amount to the physical numerical control system for execution to regulate the machining process; and feed back the effect of the regulation to the feedforward-feedback hybrid optimization framework for online learning and parameter updates.

[0048] Based on the quantitative prediction information The confidence assessment model calculates the current time. t Prediction confidence The prediction confidence level is used to characterize the reliability of the digital twin model's prediction of future processing states; the current time is calculated based on the noise level estimate in the aligned sensor data. t Real-time noise level Based on the predicted confidence level and the real-time noise level Dynamically calculate feedforward fusion weights Weights fused with feedback ,in, , Adjust the feedforward optimization amount With feedback to optimize compensation amount According to the feedforward fusion weight and the feedback fusion weight Perform weighted fusion to generate optimized decision adjustment quantities. ,in, , The baseline process parameters are used; the optimization decision adjustment amount is used. The commands are converted into instructions recognizable by the CNC system, sent to the physical CNC system via an industrial communication protocol, and executed. The regulated command data stream and the multi-source sensor data stream are continuously collected, and a multi-objective reward function is constructed based on the regulated machining effect. ,in, , For cutting force fluctuation, For surface roughness estimated based on vibration characteristics, The tool wear rate is estimated based on the cutting power model. This represents the actual material removal rate. Configurable multi-objective weighting coefficients, ∈{1,2,3,4}; the multi-objective reward function As a reward signal for reinforcement learning, the lightweight neural network in the feedback optimization pathway is updated online using the asynchronous advantage actor-critic algorithm. Network parameters.

[0049] The core technology of the system and method proposed in this invention is embodied in two collaborative modules: a high-fidelity synchronous digital twin module based on advanced computing and a hybrid-driven dynamic optimization module for process parameters. These two modules form a closed loop of "prediction-optimization-execution," and the overall technical principle of the system is as follows: Figure 2 As shown, the process begins with the physical machining system and the CNC system, whose status and commands are collected by a sensor array and data interface. This real-time data drives the digital twin module to perform calculations and predictions. The predicted information is sent to the hybrid optimization module, which integrates the real-time data to generate optimal control commands. The commands are sent back to the CNC system for execution via the issuing unit, thereby adjusting the physical machining process to form a closed loop. Simultaneously, the execution effect of the commands is fed back to the optimization module for online learning and adjustment.

[0050] 1. A high-fidelity synchronous digital twin module based on advanced computing.

[0051] This module serves as a real-time mirror and predictor of the physical processing in the information space. Its core technical principle lies in achieving "advanced simulation" that surpasses the real-time clock through NC code (numerical control code) pre-reading, dynamic local domain calculation, and efficient local update algorithms.

[0052] 1.1 Virtual-Real Synchronization and Data Injection:

[0053] The module acquires the instruction data stream from the physical numerical control system in real time via industrial communication protocols. (in time) The instruction data stream acquired in real time from the physical numerical control system (including G-code, M-code, target position information for each axis, etc.) and multi-source sensor data streams. , It is a multi-dimensional vector composed of raw data collected in real time by various sensors installed on CNC machine tools, cutting tools, workpieces, or the machining environment, used to comprehensively perceive the dynamic state of the machining process. These sensors cover a variety of physical quantities that can reflect the machining state, including but not limited to: vibration, current, power, temperature, acoustic emission, cutting force, displacement, strain, pressure, and rotational speed. The sensor data stream can be represented as: ,in This represents the sampled value of the i-th sensor at time t (e.g., at time t). Acquired instantaneous value of spindle current This reflects the load. In time... spindle speed vibration amplitude Temperature T(t), etc. Data synchronization is ensured through timestamp alignment and interpolation algorithms.

[0054] To establish a mapping between the physical world and the virtual model, a pre-calibrated mapping model is used. Transforming real-time sensor data streams into equivalent cutting forces in virtual space:

[0055]

[0056] The mapping model It can be based on analytical formulas, neural networks, or a combination of both, with all relevant sensor data as input and the equivalent cutting force applied to the tool-workpiece contact area of ​​the virtual twin model as output, thereby driving the virtual model to maintain dynamic synchronization with the physical machining process.

[0057] 1.2 Advance Computation and Dynamic Local Interest Domain Delineation:

[0058] To achieve prediction, the module creates a "forward calculation thread" that runs parallel to the real-time thread. This thread continuously reads the future. NC code for a time window (e.g., 2 seconds), and calculate the corresponding set of future tool motion trajectories. .

[0059] Based on this future trajectory, instead of updating the entire workpiece model, the system intelligently calculates the dynamic local interest region of the workpiece that may only interfere with the tool in the near future. This significantly reduces the computational load. This domain is the future tool envelope. With workpiece blank model The representation of the expected interference region in discrete voxel space:

[0060]

[0061] in, This represents the coordinates of the center point of the i-th voxel, used to locate and calculate the state of that voxel. The set of all voxels of the workpiece. Env(T(t)) is a distance function used to calculate the Euclidean distance between two points or between a point and an entity. Env(T(t)) is the tool envelope determined by the future tool trajectory T(t), i.e., the spatial region swept by the tool during its movement. As a safety margin threshold, This refers to the short time step for forward prediction.

[0062] like Figure 3 As shown, traditional simulation requires calculating the entire workpiece model, resulting in a huge computational load. This invention pre-reads the NC code to calculate the future tool trajectory and its envelope in advance. Then, based on distance judgment, it selects only the extremely small regions that may interfere with the tool in the future—the dynamic local region of interest.

[0063] This step, which confines large-scale geometric calculations to tiny, localized areas where changes are imminent, is crucial for meeting real-time requirements.

[0064] 1.3 Localized Efficient Material Removal and Model Update:

[0065] exist Within the domain, the system performs efficient material removal decisions and geometric updates. For each center point within the domain, the coordinates are... The voxels, their material state The update rule (1 indicates material exists, 0 indicates material has been removed) is as follows:

[0066]

[0067] in, This is an indicator function; its value is 1 when the voxel is not within the tool envelope. It only applies to voxels whose state changes (i.e.,...). Perform storage updates.

[0068] Subsequently, the local moving cube algorithm is invoked, based solely on... Changes in the internal voxel state incrementally update the triangular mesh on the workpiece surface in that region. This is used for high-fidelity visualization. This local update strategy reduces computational complexity from... (Global Update) Reduced to ,in .

[0069] like Figure 4As shown, during simulation updates, only the voxels within the specified region are checked and updated for "material presence / removal," and mesh reconstruction is performed only in regions where the state has changed. This reduces the computational load from total to local, enabling millisecond-level real-time synchronous simulation.

[0070] 1.4 Prediction Information Output:

[0071] Based on the results of advanced simulation, the module calculates and outputs quantitative prediction information for future processing states. For example, the instantaneous material removal rate at future moments. and the rate of change of the contact arc length between the tool and the workpiece :

[0072]

[0073]

[0074] in, For the volume of a single voxel, This refers to the instantaneous contact area between the tool and the workpiece. As a key feedforward signal, it is sent to the optimization module in real time.

[0075] 2. Hybrid-driven dynamic optimization module for process parameters

[0076] This module is the core of the system's intelligent decision-making. Its technical principle lies in building a hybrid model that integrates personality determination, data-driven learning, and attention mechanisms to collaboratively optimize decisions based on predictive information from the twin module and real-time sensor information.

[0077] like Figure 5 As shown, the core working principle and data flow of the hybrid-driven process parameter dynamic optimization module demonstrate the process from multi-source information input to intelligent decision output. Its design embodies the combination of "feedforward rapid response" and "feedback precise compensation," and achieves an innovative architecture of complementary advantages through "dynamic fusion."

[0078] 2.1 Characterization of mixed states:

[0079] The system constructs a multidimensional hybrid state vector. , serving as a unified input to the optimization model:

[0080]

[0081] Mechanism characteristics Calculations based on physical models, such as cutting force components estimated from mechanical models. ,in To compare cutting force, For cutting depth, For feed rate, This is the instantaneous contact angle.

[0082] Real-time data characteristics Time-frequency domain features extracted from sensor signals, such as vibration signals. In the flutter characteristic frequency band Internal energy

[0083] Predicted features Prediction information from the twin module, such as .

[0084] The constructed multidimensional hybrid state vector integrates theoretical estimates based on physical laws, the actual status captured by sensors, and future predictions provided by digital twins. This ensures that decision-making respects physical laws, fits the actual processing conditions, and takes future changes into account.

[0085] 2.2 Feedforward-Feedback Hybrid Optimization Architecture:

[0086] Optimization of decision-making adjustment (such as feed rate adjustment amount) It is generated by the coordinated action of two pathways:

[0087]

[0088] Feedforward optimization adjustment amount of feedforward optimization path Based on established physical rules or empirical models, it directly responds to predicted information. For example, when predicting the contact arc length... When the load is about to increase, feed forward and reduce the feed rate according to the principle of "constant cutting load":

[0089]

[0090] in, For feedforward gain, and This is the rated value. Feedforward optimization is the system's fast response mechanism. Once the contact arc length is predicted... The cutting load will increase, according to the proportional relationship. Immediately and proactively calculate a way to reduce the feed rate. The instructions are designed to counteract foreseeable disturbances, and their advantage is that they have an extremely fast response time.

[0091] Feedback optimization compensation amount of the feedback optimization path : Composed of a lightweight neural network This implementation is used to learn and compensate for inaccuracies and unmodeled perturbations in the feedforward model. The network employs mechanism-guided initialization and attention mechanisms.

[0092] (1) Mechanism-guided initialization: Partial weights of the first layer of the network Instead of random assignment, the values ​​are derived from the Jacobian matrix of the simplified physical model. Approximate initialization allows the network to learn from a "close to correct" starting point. It is a partial differential. It is a hybrid state vector, a multidimensional input composed of real-time data features, physical mechanism features, and twin prediction features. The ideal optimal action vector represents the theoretically optimal adjustment of process parameters calculated based on classical physical and mechanical models (such as cutting force balance and constant cutting load principles) under a given state S. The Jacobian matrix J describes how the ideal optimal action Aideal changes when the state vector S undergoes a small change.

[0093] (2) Attention mechanism: An attention layer is integrated inside the network, which is a state vector. Dynamically calculate weights for different feature dimensions ,in, For query vector, For key vectors, For value vectors, To query the key similarity matrix, Scaling factor The function is a normalized exponential function. This allows the network to automatically focus on key features (such as sudden vibrations) while suppressing irrelevant noise.

[0094] (3) The output of the feedback path is: .

[0095] Feedback optimization acts as the system's self-learning and fine-tuning brain. Neural networks handle complex, nonlinear perturbations (such as subtle material inhomogeneities or minor tool breakage) that feedforward rules cannot accurately model. Ultimately, it outputs a compensation amount. It is used to correct the shortcomings of feedforward instructions.

[0096] 2.3 Adaptive Fusion Decision-Making:

[0097] The outputs of the feedforward and feedback paths are controlled by adaptive weights. Fusion. Weights are based on prediction confidence. and real-time noise level Dynamic adjustment:

[0098]

[0099] When the prediction confidence is high and the ambient noise is low, the system places more trust in the deterministic regulation of the feedforward ( Conversely, when the working conditions are complex and the prediction uncertainty is high, the adaptive learning capability of the feedback path becomes more important. Adaptive fusion decision-making is the system's decision arbitration mechanism, determining whether feedforward or feedback has a greater impact.

[0100] 2.4 Multi-objective rewards and online learning:

[0101] Optimize the long-term performance of the model through a multi-objective reward function. To guide and fine-tune online:

[0102]

[0103] in, For cutting force fluctuation, For surface roughness estimated based on vibration characteristics, The tool wear rate is estimated based on the cutting power model. This represents the actual material removal rate. (Coefficient) It can be configured according to the processing strategy ("quality priority" or "efficiency priority").

[0104] The system utilizes online reinforcement learning algorithms such as the asynchronous advantage actor-critic algorithm to continuously optimize network parameters and maximize long-term cumulative rewards. Large fluctuations in processing force, poor surface roughness, and rapid tool wear result in negative rewards, while high material removal rates result in positive rewards. The coefficient λ represents the degree of importance placed on different objectives such as quality, efficiency, and cost. By guiding the neural network to pursue the maximization of long-term cumulative rewards, it adjusts its parameters to find the optimal solution among multiple mutually constraining objectives—that is, the optimal solution for multi-objective optimization.

[0105] Through the close coupling of these two modules, a complete intelligent control system with predictive, optimization, and adaptive capabilities is formed. The digital twin module provides... This creates a valuable "foresight" window for the optimization module, and the dynamic decisions of the optimization module, in turn, react on the physical world through the CNC system, forming an enhanced closed loop that improves machining performance.

[0106] The embodiments of the present invention will be further described in detail based on the processing examples of blade parts.

[0107] 1. System deployment.

[0108] The system is deployed on a five-axis CNC machining center used for machining blades made of difficult-to-machine materials. A high-response accelerometer is installed on the machine tool spindle, and a high-precision current transformer is installed on the main circuit. An industrial edge computing gateway is deployed, running the software system of this invention. The sensor, edge gateway, and machine tool CNC system communicate via a network using a protocol.

[0109] 2. Operating procedures.

[0110] 2.1 Initialization: Import the CAD model of the blade part and the 3D model of the tool into the system, and initialize the voxel model of the workpiece blank. Load the NC program for this machining.

[0111] 2.2 Synchronous Operation: Machine tool processing begins. The digital twin module starts working, reading the actual position of the machine tool in real time and driving the virtual machine tool to move synchronously; it continuously pre-reads the NC code, and when the system predicts that it is about to enter a high-curvature corner area, it immediately calculates the "dynamic attention voxel domain" of that area; it performs local material removal simulation within this area, predicting that the tool-workpiece contact area will increase by 80% within the next 0.5 seconds, and then... Send to the optimization module.

[0112] 2.3 Hybrid Optimization Decision: The hybrid optimization module receives... back:

[0113] (1) Feedforward path: Based on the rule of "significant increase in contact area", immediately output a It is recommended to reduce the feed rate by 15%.

[0114] (2) Feedback path: At the same time, the real-time vibration data shows that the current state is stable, and the neural network outputs a small compensation amount. .

[0115] (3) Given the high prediction confidence, the fusion decision-maker assigns a high weight to the feedforward path (α=0.8, β=0.2) and finally generates the instruction: adjust the feed rate to 87% of the original value.

[0116] In this embodiment, a computer device is provided, such as... Figure 6 As shown, it includes a memory 601, a processor 602, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements any of the above-mentioned dynamic control methods for advanced simulation and hybrid optimization of CNC machining.

[0117] Specifically, the computer device can be a computer terminal, a server, or a similar computing device.

[0118] In this embodiment, a computer-readable storage medium is provided, which stores a computer program that executes any of the above-described dynamic control methods for advanced simulation and hybrid optimization of CNC machining.

[0119] Specifically, computer-readable storage media include both permanent and non-permanent, removable and non-removable media, which can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer-readable storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable storage media do not include transient media, such as modulated data signals and carrier waves.

[0120] Based on the same inventive concept, this invention also provides a dynamic control device for advanced simulation and hybrid optimization in CNC machining, as described in the following embodiments. Since the principle of the dynamic control device for advanced simulation and hybrid optimization in CNC machining is similar to that of the dynamic control method for advanced simulation and hybrid optimization in CNC machining, the implementation of the dynamic control device for advanced simulation and hybrid optimization in CNC machining can refer to the implementation of the dynamic control method for advanced simulation and hybrid optimization in CNC machining, and will not be repeated. As used below, the terms "unit" or "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0121] Figure 7 This is a structural block diagram of a dynamic control device for advanced simulation and hybrid optimization in CNC machining according to an embodiment of the present invention, such as... Figure 7 As shown, it includes: a digital twin model driving module 701, a quantitative prediction information generation module 702, a parameter optimization module 703, and a dynamic control processing module 704. The structure is described below.

[0122] The digital twin model module 701 is used to acquire the instruction data stream of the physical numerical control system and the multi-source sensor data stream of various sensors, perform timestamp alignment and data fusion processing on the instruction data stream and the multi-source sensor data stream to generate synchronous driving data, and inject the synchronous driving data into the digital twin model, wherein the digital twin model is used to drive the virtual-real synchronous simulation of the physical processing process.

[0123] A quantitative prediction information generation module 702 is used to pre-read the CNC code within a future time window, solve the CNC code to generate a corresponding set of future tool motion trajectories, and based on the set of future tool motion trajectories, determine a dynamic local interest region on the workpiece model of the digital twin model, perform local material removal simulation within the dynamic local interest region and update the workpiece model to generate quantitative prediction information for future machining states. The dynamic local interest region is the local area where the expected interference between the future tool envelope and the workpiece model is expected.

[0124] The parameter optimization module 703 is used to construct a multidimensional hybrid state vector containing mechanistic features, real-time data features and the quantized prediction information, construct a feedforward-feedback hybrid optimization framework, and generate feedforward optimization adjustment amount and feedback optimization compensation amount based on the multidimensional hybrid state vector and the feedforward-feedback hybrid optimization framework.

[0125] The dynamic control processing module 704 is used to dynamically calculate the fusion weight based on the prediction confidence and real-time noise level, weight and fuse the feedforward optimization adjustment amount and the feedback optimization compensation amount to generate the optimization decision adjustment amount, send the optimization decision adjustment amount to the physical numerical control system for execution, control the processing process, and feed back the effect of the control to the feedforward-feedback hybrid optimization architecture for online learning and parameter updates.

[0126] In one embodiment, driving the digital twin model module includes:

[0127] The instruction data stream acquisition unit is used to acquire the instruction data stream output by the physical numerical control system in real time via industrial communication protocols. Where t is time, the instruction data stream includes preparation function code, auxiliary function code and actual position instructions for each motion axis;

[0128] The sensor data stream acquisition unit is used to acquire multi-source sensor data streams output by various sensors in real time through the data acquisition interface. ,in, , Let be the sampled value of the i-th sensor at time t;

[0129] A unified timestamp unit is used for the instruction data stream. and the multi-source sensor data stream A globally unified timestamp is added to each frame of data. Based on the globally unified timestamp, the instruction data stream and the multi-source sensor data stream are aligned on the time axis to obtain aligned instruction data and aligned sensor data.

[0130] A unit for calculating the equivalent virtual cutting force is used to input the aligned sensing data into a pre-calibrated mapping model. Calculate the equivalent virtual cutting force applied to the tool-workpiece contact area of ​​the digital twin model. ,in, ;

[0131] Generate a synchronous drive data unit for integrating the aligned command data with the equivalent virtual cutting force. Perform data fusion to generate synchronous driving data;

[0132] The twin model injection unit is used to inject the synchronous drive data into the digital twin model, drive the virtual axis of the machine tool to move through the aligned command data, and apply the equivalent virtual cutting force to the tool-workpiece contact area of ​​the digital twin model.

[0133] In one embodiment, the module for generating quantitative prediction information includes:

[0134] Create thread units to create advanced computation threads in the digital twin model that are independent of the real-time synchronous simulation threads;

[0135] The advance read control code unit is used to continuously read the current time from the instruction buffer of the physical numerical control system through the advance calculation thread. Then preset duration Numerical control code within the future time window;

[0136] The displacement time sequence calculation unit is used to extract the motion commands and feed speed commands of the CNC code, perform interpolation preprocessing on the motion commands based on the current kinematic configuration of the machine tool, and calculate the displacement time sequence of each axis of the machine tool within the future time window in combination with the feed speed commands;

[0137] A discrete trajectory point set unit is constructed to transform the displacement time series of each axis to the workpiece coordinate system and solve for the discrete trajectory point set of the tool relative to the workpiece surface within the future time window.

[0138] The motion trajectory set generation unit is used to perform curve fitting on the discrete trajectory point set to generate a continuous set of future tool motion trajectories. ,in, .

[0139] In one embodiment, the module for generating quantitative prediction information further includes:

[0140] Generate tool envelope units for use based on the set of future tool motion trajectories. Generate the set of future tool movement trajectories. Determined tool envelope ,in, For a specific moment;

[0141] Construct dynamic local interest units to obtain the total set of voxels of the workpiece blank. Traverse the entire set of voxels of the workpiece For all voxels, calculate the center point coordinates of each voxel i. To the future tool envelope The minimum distance is calculated, and a dynamic local interest region is constructed based on the minimum distance. , ,in, It is the Euclidean distance function. The preset safety margin threshold, For the current moment, For forward prediction, the short time step, For a specific moment;

[0142] Update material state unit for use in the dynamic local region of interest Inside, for each center point, the coordinates are... The voxels are updated to reflect the material state of the corresponding voxels. ,in, For indicator functions, Preset duration;

[0143] Update the triangular mesh elements on the workpiece surface for use based on the dynamic local region of interest. Voxels whose internal material states change, within the dynamic local region of interest Incremental update of workpiece surface triangular mesh ;

[0144] The instantaneous material removal rate calculation unit is used to calculate the instantaneous material removal rate based on the dynamic local region of interest. For voxels showing changes in the internal material state, calculate the instantaneous material removal rate at future moments. ,in, The volume of a single voxel unit;

[0145] The unit for calculating the rate of change of arc length is used to calculate the future tool motion trajectory set. With the triangular mesh on the surface of the workpiece Based on the geometric relationship, the tool-workpiece contact arc length is calculated. Tool-workpiece contact arc length variation rate ;

[0146] Generate a quantitative prediction information unit for the instantaneous material removal rate. and the rate of change of the contact arc length between the tool and the workpiece As a quantitative predictive information for future processing status .

[0147] In one embodiment, the parameter optimization module includes:

[0148] The computer-defined feature unit is used to calculate the mechanism features based on the current process parameters and physical model. The mechanistic features include instantaneous cutting force components estimated based on the cutting force model. For time;

[0149] The real-time data feature extraction unit is used to perform time-frequency domain analysis on the aligned sensor data and extract real-time data features. The real-time data features include the energy of the vibration signal within a preset characteristic frequency band;

[0150] Generate multidimensional hybrid state vector units to obtain the quantized prediction information. The mechanism features The real-time data features and the quantized prediction information The vectors are concatenated to generate a multidimensional mixed state vector. ,in, ;

[0151] The data output unit is used to construct a feedforward-feedback hybrid optimization architecture, including feedforward optimization paths and feedback optimization paths, and to output the quantized prediction information. The input is given to the feedforward optimization path, and the output is the feedforward optimization adjustment amount, which is used to adjust the multidimensional mixed state vector. The input is fed into the feedback optimization path, and the feedback optimization compensation amount is output.

[0152] In one embodiment, the data output unit is further configured to output the quantized prediction information. Tool-workpiece contact arc length variation rate Input to the feedforward optimization path, output feedforward optimization adjustment amount ,in, , For feedforward gain, For the rated contact arc length, For the rated feed rate; construct a lightweight neural network. By simplifying the Jacobian matrix of the physical model Approximate initialization of partial weights of the first layer of the lightweight neural network An attention layer is integrated into the lightweight neural network, and the attention weights of the attention layer are calculated. ; through the lightweight neural network Construct a feedback optimization path; convert the multidimensional mixed state vector The input is fed into the feedback optimization path, and the output is the feedback optimization compensation amount. ,in, .

[0153] In one embodiment, the dynamic control processing module includes:

[0154] Calculate prediction confidence unit, used to calculate prediction confidence based on the quantized prediction information. The confidence assessment model calculates the current time. t Prediction confidence The prediction confidence level is used to characterize the reliability of the digital twin model's prediction results for future processing states;

[0155] The real-time noise level calculation unit is used to calculate the current time based on the noise level estimate in the aligned sensor data. t Real-time noise level ;

[0156] The weight calculation unit is used to calculate the weight based on the prediction confidence level. and the real-time noise level Dynamically calculate feedforward fusion weights Weights fused with feedback ,in, , ;

[0157] The weighted fusion unit is used to combine the feedforward optimization adjustment amount. With feedback to optimize compensation amount According to the feedforward fusion weight and the feedback fusion weight Perform weighted fusion to generate optimized decision adjustment quantities. ,in, , These are the baseline process parameters;

[0158] The processing instruction unit is used to adjust the optimization decision amount. The instructions are converted into commands that the CNC system can recognize, and then sent to the physical CNC system via an industrial communication protocol for execution.

[0159] Construct a multi-objective reward function unit to continuously collect the regulated instruction data stream and the multi-source sensor data stream, and construct a multi-objective reward function based on the regulated processing effect. ,in, , For cutting force fluctuation, For surface roughness estimated based on vibration characteristics, The tool wear rate is estimated based on the cutting power model. This represents the actual material removal rate. Configurable multi-objective weighting coefficients, ∈{1,2,3,4};

[0160] The feedback optimization unit is used to optimize the multi-objective reward function. As a reward signal for reinforcement learning, the lightweight neural network in the feedback optimization pathway is updated online using the asynchronous advantage actor-critic algorithm. Network parameters.

[0161] The embodiments of the present invention achieve the following technical effects:

[0162] Through the advanced calculations of the digital twin module, the system has the ability to "predict" and can adjust process parameters early and smoothly before drastic changes in cutting conditions (such as corners or entry points), turning passive response into active prevention and fundamentally avoiding machining oscillations or quality defects caused by response lag.

[0163] The hybrid-driven optimization model innovatively integrates the interpretability and reliability of physical mechanisms with the adaptive capabilities of data-driven approaches. The feedforward path handles predictable changes based on deterministic knowledge, while the feedback path utilizes neural networks to compensate for random disturbances. The combination of the two makes the optimization decisions both accurate and robust, overcoming the performance instability of pure data-driven models under unfamiliar conditions.

[0164] The digital twin module employs a "dynamic local area of ​​concern" and "local update" strategy, which reduces the simulation computation load by more than 70% and ensures that the entire system's response time from data acquisition and simulation prediction to optimization decision-making is less than 100 milliseconds, fully meeting the real-time control requirements of high-end CNC machining.

[0165] Through optimization driven by a multi-objective reward function, the system can dynamically balance efficiency and tool life while ensuring the core premise of machining surface quality and accuracy. Validated on blade parts, it can achieve a machining efficiency improvement of more than 15% while reducing tool wear rate by more than 20%.

[0166] Obviously, those skilled in the art should understand that the modules or steps of the above-described embodiments of the present invention can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented here, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the embodiments of the present invention are not limited to any particular hardware and software combination.

[0167] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, various modifications and variations can be made to the embodiments of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A dynamic control method for advanced simulation and hybrid optimization in CNC machining, characterized in that, include: The system acquires the instruction data stream of the physical numerical control system and the multi-source sensor data stream of various sensors. It performs time stamp alignment and data fusion processing on the instruction data stream and the multi-source sensor data stream to generate synchronous driving data. The synchronous driving data is then injected into a digital twin model, wherein the digital twin model is used to drive the virtual-real synchronous simulation of the physical processing process. The system reads the CNC code within a future time window, solves the CNC code to generate a set of corresponding future tool motion trajectories, and, based on the set of future tool motion trajectories, determines a dynamic local area of ​​interest on the workpiece model of the digital twin model. Within the dynamic local area of ​​interest, it performs local material removal simulation and updates the workpiece model, generating quantitative prediction information for future machining states. The dynamic local area of ​​interest is the local region where the future tool envelope and the workpiece model are expected to interfere. A multidimensional hybrid state vector containing mechanistic features, real-time data features, and the quantized prediction information is constructed. A feedforward-feedback hybrid optimization framework is constructed. Based on the multidimensional hybrid state vector, feedforward optimization adjustment and feedback optimization compensation are generated through the feedforward-feedback hybrid optimization framework. The fusion weights are dynamically calculated based on the prediction confidence and real-time noise level. The feedforward optimization adjustment amount and the feedback optimization compensation amount are then weighted and fused to generate an optimization decision adjustment amount. This optimization decision adjustment amount is then sent to the physical numerical control system for execution to regulate the machining process. The effect of the regulation is then fed back to the feedforward-feedback hybrid optimization framework for online learning and parameter updates.

2. The dynamic control method for advanced simulation and hybrid optimization in CNC machining as described in claim 1, characterized in that, Acquire instruction data streams from a physical numerical control system and multi-source sensor data streams from various sensors; perform timestamp alignment and data fusion processing on the instruction data streams and the multi-source sensor data streams to generate synchronous drive data, including: Real-time acquisition of instruction data streams output by the physical numerical control system via industrial communication protocols. Where t is time, the instruction data stream includes preparation function code, auxiliary function code and actual position instructions for each motion axis; Real-time acquisition of multi-source sensor data streams from various sensors via data acquisition interface. ,in, , Let be the sampled value of the i-th sensor at time t. n The number of sensors; For the instruction data stream and the multi-source sensor data stream A globally unified timestamp is added to each frame of data. Based on the globally unified timestamp, the instruction data stream and the multi-source sensor data stream are aligned on the time axis to obtain aligned instruction data and aligned sensor data. The aligned sensing data is input into a pre-calibrated mapping model. Calculate the equivalent virtual cutting force applied to the tool-workpiece contact area of ​​the digital twin model. ,in, ; The aligned instruction data and the equivalent virtual cutting force Perform data fusion to generate synchronous driving data; The synchronous drive data is injected into the digital twin model, the virtual axis of the machine tool is driven to move by the aligned command data, and the equivalent virtual cutting force is applied to the tool-workpiece contact area of ​​the digital twin model.

3. The dynamic control method for advanced simulation and hybrid optimization in CNC machining as described in claim 1, characterized in that, Preview the CNC code within a future time window, and calculate the CNC code to generate a corresponding set of future tool motion trajectories, including: Create an advanced computing thread in the digital twin model that is independent of the real-time synchronous simulation thread; The aforementioned advance calculation thread continuously reads the current moment from the instruction buffer of the physical numerical control system. Then preset duration Numerical control code within the future time window; The motion commands and feed rate commands of the CNC code are extracted. Based on the current kinematic configuration of the machine tool, the motion commands are preprocessed by interpolation. The displacement time sequence of each axis of the machine tool within the future time window is calculated by combining the feed rate commands. The displacement time series of each axis is transformed into the workpiece coordinate system, and the discrete trajectory point set of the tool relative to the workpiece surface within the future time window is calculated. Curve fitting is performed on the discrete trajectory point set to generate a continuous set of future tool motion trajectories. ,in, .

4. The dynamic control method for advanced simulation and hybrid optimization in CNC machining as described in claim 1, characterized in that, Based on the set of future tool motion trajectories, a dynamic local interest region is determined on the workpiece model of the digital twin model. Local material removal simulation is performed within the dynamic local interest region, and the workpiece model is updated to generate quantitative prediction information for future machining states, including: Based on the set of future tool motion trajectories Generate the set of future tool movement trajectories. Determined tool envelope ,in, For a specific moment; Obtain the set of all voxels of the workpiece blank. Traverse the entire set of voxels of the workpiece For all voxels, calculate the center coordinates of each voxel i. To the future tool envelope The minimum distance is calculated, and a dynamic local interest region is constructed based on the minimum distance. , ,in, It is the Euclidean distance function. The preset safety margin threshold, For the current moment, For forward prediction, the short time step, For a specific moment; In the dynamic local area of ​​interest Inside, for each center point, the coordinates are... The voxels are updated to reflect the material state of the corresponding voxels. ,in, For indicator functions, Preset duration; Based on the dynamic local interest region Voxels whose internal material states change, within the dynamic local region of interest Incremental update of workpiece surface triangular mesh ; Based on the dynamic local interest region For voxels showing changes in the internal material state, calculate the instantaneous material removal rate at future moments. ,in, The volume of a single voxel unit; Based on the set of future tool motion trajectories With the triangular mesh on the surface of the workpiece Based on the geometric relationship, the tool-workpiece contact arc length is calculated. Tool-workpiece contact arc length variation rate ; The instantaneous material removal rate and the rate of change of the contact arc length between the tool and the workpiece As a quantitative predictive information for future processing status .

5. The dynamic control method for advanced simulation and hybrid optimization in CNC machining as described in claim 1, characterized in that, A multidimensional hybrid state vector containing mechanistic features, real-time data features, and the quantized prediction information is constructed. A feedforward-feedback hybrid optimization framework is then constructed. Based on the multidimensional hybrid state vector, feedforward optimization adjustment and feedback optimization compensation are generated through the feedforward-feedback hybrid optimization framework, including: Based on the current process parameters and physical model, the mechanism characteristics are calculated. The mechanistic features include instantaneous cutting force components estimated based on the cutting force model. For time; Time-frequency domain analysis was performed on the aligned sensor data to extract real-time data features. The real-time data features include the energy of the vibration signal within a preset characteristic frequency band; Obtain the quantization prediction information The mechanism features The real-time data features and the quantized prediction information The vectors are concatenated to generate a multidimensional mixed state vector. ,in, ; Construct a hybrid feedforward-feedback optimization framework that includes feedforward optimization pathways and feedback optimization pathways, and integrate the quantized prediction information. The input is given to the feedforward optimization path, and the output is the feedforward optimization adjustment amount, which is used to adjust the multidimensional mixed state vector. The input is fed into the feedback optimization path, and the feedback optimization compensation amount is output.

6. The dynamic control method for advanced simulation and hybrid optimization in CNC machining as described in claim 5, characterized in that, Construct a hybrid feedforward-feedback optimization framework that includes feedforward optimization pathways and feedback optimization pathways, and integrate the quantized prediction information. The input is given to the feedforward optimization path, and the output is the feedforward optimization adjustment amount, which is used to adjust the multidimensional mixed state vector. The input to the feedback optimization path and the output of the feedback optimization compensation amount include: The quantized prediction information Tool-workpiece contact arc length variation rate Input to the feedforward optimization path, output feedforward optimization adjustment amount ,in, , For feedforward gain, For the rated contact arc length, This is the rated feed rate; Building lightweight neural networks By simplifying the Jacobian matrix of the physical model Approximate initialization of partial weights of the first layer of the lightweight neural network ; An attention layer is integrated into the lightweight neural network, and the attention weights of the attention layer are calculated. ; Through the lightweight neural network Construct feedback optimization pathways; The multidimensional mixed state vector The input is fed into the feedback optimization path, and the output is the feedback optimization compensation amount. ,in, .

7. The dynamic control method for advanced simulation and hybrid optimization of CNC machining as described in any one of claims 1 to 6, characterized in that, The fusion weights are dynamically calculated based on the predicted confidence level and real-time noise level. The feedforward optimization adjustment and the feedback optimization compensation are then weighted and fused to generate an optimization decision adjustment. This optimization decision adjustment is sent to the physical numerical control system for execution to regulate the machining process. The effect of the regulation is then fed back to the feedforward-feedback hybrid optimization framework for online learning and parameter updates, including: Based on the quantitative prediction information The confidence assessment model calculates the current time. t Prediction confidence The prediction confidence level is used to characterize the reliability of the digital twin model's prediction results for future processing states; Calculate the current time based on the noise level estimate in the aligned sensor data. t Real-time noise level ; Based on the predicted confidence level and the real-time noise level Dynamically calculate feedforward fusion weights Weights fused with feedback ,in, , ; Adjust the feedforward optimization amount With feedback to optimize compensation amount According to the feedforward fusion weight and the feedback fusion weight Perform weighted fusion to generate optimized decision adjustment quantities. ,in, , These are the baseline process parameters; Adjust the optimization decision amount The instructions are converted into commands that the CNC system can recognize, and then sent to the physical CNC system via an industrial communication protocol for execution. Continuously collect the regulated instruction data stream and the multi-source sensor data stream, and construct a multi-objective reward function based on the regulated processing effect. ,in, , For cutting force fluctuation, For surface roughness estimated based on vibration characteristics, The tool wear rate is estimated based on the cutting power model. This represents the actual material removal rate. Configurable multi-objective weighting coefficients, ∈{1,2,3,4}; The multi-objective reward function As a reward signal for reinforcement learning, the lightweight neural network in the feedback optimization pathway is updated online using the asynchronous advantage actor-critic algorithm. Network parameters.

8. A dynamic control device for advanced simulation and hybrid optimization in CNC machining, characterized in that, include: The digital twin model module is used to acquire the instruction data stream of the physical numerical control system and the multi-source sensor data stream of various sensors. It performs time stamp alignment and data fusion processing on the instruction data stream and the multi-source sensor data stream to generate synchronous driving data. The synchronous driving data is then injected into the digital twin model, wherein the digital twin model is used to drive the virtual-real synchronous simulation of the physical processing process. A quantitative prediction information generation module is used to pre-read the CNC code within a future time window, solve the CNC code to generate a corresponding set of future tool motion trajectories, and based on the set of future tool motion trajectories, determine a dynamic local interest region on the workpiece model of the digital twin model, perform local material removal simulation within the dynamic local interest region and update the workpiece model to generate quantitative prediction information for future machining states. The dynamic local interest region is the local area where the expected interference between the future tool envelope and the workpiece model is expected. The parameter optimization module is used to construct a multidimensional hybrid state vector containing mechanistic features, real-time data features and the quantized prediction information, construct a feedforward-feedback hybrid optimization framework, and generate feedforward optimization adjustment amount and feedback optimization compensation amount based on the multidimensional hybrid state vector through the feedforward-feedback hybrid optimization framework. The dynamic control processing module is used to dynamically calculate the fusion weight based on the prediction confidence and real-time noise level, weight and fuse the feedforward optimization adjustment amount and the feedback optimization compensation amount to generate the optimization decision adjustment amount, and send the optimization decision adjustment amount to the physical numerical control system for execution to control the processing process, and feed back the effect of the control to the feedforward-feedback hybrid optimization architecture for online learning and parameter updates.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the dynamic control method for advanced simulation and hybrid optimization for CNC machining as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that executes the dynamic control method for advanced simulation and hybrid optimization of CNC machining as described in any one of claims 1 to 7.

Images