Search

CN-122018432-A - Dynamic regulation and control method, device, computer equipment and medium for advanced simulation and hybrid optimization of numerical control machining

CN122018432ACN 122018432 ACN122018432 ACN 122018432ACN-122018432-A

Abstract

The embodiment of the invention provides a dynamic regulation and control method, a device, computer equipment and a medium for advanced simulation and hybrid optimization of numerical control machining, which relate to the technical field of numerical control machining, wherein the method comprises the following steps of injecting synchronous driving data generated by an instruction data stream and a multi-source sensor data stream into a digital twin model; the method comprises the steps of calculating a numerical control code to generate a corresponding future tool motion trail set, determining a dynamic local attention domain based on the future tool motion trail set to generate quantitative prediction information of a future processing state, constructing a multidimensional mixed state vector, generating a feedforward optimization adjustment amount and a feedback optimization compensation amount, carrying out weighted fusion on the feedforward optimization adjustment amount and the feedback optimization compensation amount to generate an optimization decision adjustment amount, and feeding back the regulated effect to a feedforward feedback mixed optimization framework to carry out online learning and parameter updating. According to the scheme, the stability and the precision of complex part processing are improved through advanced simulation and hybrid optimization.

Inventors

  • GAO RENZHI
  • YANG HAILONG
  • LIU HAO
  • DU BAORUI

Assignees

  • 中国科学院工程热物理研究所

Dates

Publication Date
20260512
Application Date
20260410

Claims (10)

  1. 1. The dynamic regulation and control method for advanced simulation and hybrid optimization of numerical control machining is characterized by comprising the following steps: Acquiring an instruction data stream of a physical numerical control system and a multi-source sensor data stream of a plurality of sensors, performing time stamp alignment and data fusion processing on the instruction data stream and the multi-source sensor data stream, generating synchronous driving data, and injecting the synchronous driving data into a digital twin model, wherein the digital twin model is used for driving virtual and real synchronous simulation of a physical machining process; Pre-reading a numerical control code in a future time window, resolving the numerical control code to generate a corresponding future tool motion track set, determining a dynamic local attention domain on a workpiece model of the digital twin model based on the future tool motion track set, performing local material removal simulation in the dynamic local attention domain, updating the workpiece model, and generating quantitative prediction information of a future processing state, wherein the dynamic local attention domain is a local area of expected interference of a future tool envelope body and the workpiece model; constructing a multidimensional mixed state vector containing mechanism characteristics, real-time data characteristics and the quantized prediction information, constructing a feedforward feedback mixed optimization framework, and generating feedforward optimization adjustment quantity and feedback optimization compensation quantity through the feedforward feedback mixed optimization framework based on the multidimensional mixed state vector; And dynamically calculating a fusion weight according to the prediction confidence and the real-time noise level, carrying out weighted fusion on the feedforward optimization adjustment quantity and the feedback optimization compensation quantity to generate an optimization decision adjustment quantity, transmitting the optimization decision adjustment quantity to a physical numerical control system for execution, regulating and controlling a processing process, and feeding back the regulated and controlled effect to the feedforward feedback hybrid optimization framework for online learning and parameter updating.
  2. 2. The dynamic regulation and control method for advanced simulation and hybrid optimization of numerical control machining according to claim 1, wherein obtaining an instruction data stream of a physical numerical control system and a multi-source sensor data stream of a plurality of sensors, performing time stamp alignment and data fusion processing on the instruction data stream and the multi-source sensor data stream, and generating synchronous driving data, comprises: Acquiring instruction data stream output by physical numerical control system in real time through industrial communication protocol Wherein t is the time, and the instruction data stream comprises a preparation function code, an auxiliary function code and actual position instructions of all motion axes; Multi-source sensing data stream output by various sensors is obtained in real time through data acquisition interface , wherein, , The sampling value of the ith sensor at the time t is obtained, and n is the number of the sensors; For the instruction data stream And the multi-source sensory data stream Adding a global unified time stamp to each frame of data, and aligning the instruction data stream and the multi-source sensing data stream on a time axis based on the global unified time stamp to obtain aligned instruction data and aligned sensing data; inputting the aligned sensing data into a pre-calibrated mapping model Calculating an equivalent virtual cutting force applied to a tool-workpiece contact area of the digital twin model , wherein, ; Aligning the aligned instruction data with the equivalent virtual cutting force Performing data fusion to generate synchronous driving data; injecting the synchronous drive data into the digital twin model, driving a machine tool virtual axis to move through the aligned instruction data, and applying the equivalent virtual cutting force to the tool-workpiece contact area of the digital twin model.
  3. 3. The dynamic tuning method for advanced simulation and hybrid optimization of numerical control machining according to claim 1, wherein pre-reading numerical control codes within a future time window, resolving the numerical control codes to generate a corresponding future tool motion trajectory set, comprises: Creating a lead calculation thread independent of a real-time synchronous simulation thread in the digital twin model; continuously pre-reading the current moment from an instruction buffer area of the physical numerical control system through the advanced calculation thread After a preset time period Numerical control codes within a future time window of (2); Extracting a motion instruction and a feed speed instruction of the numerical control code, performing interpolation pretreatment on the motion instruction based on the kinematic configuration of the current machine tool, and calculating to obtain a displacement time sequence of each axis of the machine tool in the future time window by combining the feed speed instruction; converting the displacement time sequence of each axis into a workpiece coordinate system, and calculating to obtain a discrete track point set of the tool relative to the workpiece surface in the future time window; Curve fitting is carried out on the discrete track point set to generate a continuous future tool motion track set , wherein, 。
  4. 4. The dynamic tuning method for advanced simulation and hybrid optimization of numerical control machining according to claim 1, wherein determining a dynamic local attention domain on a workpiece model of the digital twin model based on the set of future tool motion trajectories, performing a local material removal simulation in the dynamic local attention domain and updating the workpiece model, generating quantized prediction information of future machining states, comprises: based on the future tool motion trail set Generating a set of future tool motion trajectories Determined tool envelope , wherein, The time is the moment; acquiring a whole voxel set of a workpiece blank Traversing the whole voxel set of the workpiece Calculating the coordinates of the central point of each voxel i To the future tool envelope And construct a dynamic local attention domain based on the minimum distance , , wherein, As a function of the euclidean distance, For a preset safety margin threshold value, For the current moment of time, For a short step size of the forward prediction, The time is the moment; In the dynamic local attention domain In which the coordinates for each center point are Updating the material state corresponding to the voxel , wherein, In order to indicate the function, The time is preset; Based on the dynamic local attention domain Voxels with changed inner material state in the dynamic local focus region Internal increment type workpiece surface triangular mesh updating ; Based on the dynamic local attention domain Voxel with changed internal material state, calculating instantaneous material removal rate at future time , wherein, A volume that is a single voxel unit; based on the future tool motion trail set Triangular mesh with the surface of the workpiece Is calculated to obtain the contact arc length of the tool and the workpiece Tool-workpiece contact arc length rate of change ; The instantaneous material removal rate And the rate of change of the tool-workpiece contact arc length As quantized prediction information for future process states 。
  5. 5. The dynamic regulation method for advanced simulation and hybrid optimization of numerical control machining according to claim 1, wherein constructing a multidimensional hybrid state vector including a mechanism feature, a real-time data feature and the quantized prediction information, constructing a feed-forward feedback hybrid optimization framework, and generating a feed-forward optimization adjustment amount and a feed-forward optimization compensation amount by the feed-forward feedback hybrid optimization framework based on the multidimensional hybrid state vector, comprises: based on the technological parameters and physical model at the current moment, calculating to obtain mechanism characteristics Wherein the mechanical feature comprises an instantaneous cutting force component estimated from a cutting force model, Time is; Performing time-frequency domain analysis on the aligned sensing data, and extracting real-time data characteristics Wherein the real-time data characteristic comprises the energy of the vibration signal in a preset characteristic frequency band; Acquiring the quantized prediction information Characterizing the mechanism The real-time data feature And said quantized prediction information Splicing to generate multi-dimensional mixed state vector , wherein, ; Constructing a feedforward and feedback hybrid optimization framework comprising a feedforward optimization path and a feedback optimization path, and quantizing the prediction information Inputting to the feedforward optimization path, outputting feedforward optimization adjustment quantity, and mixing the multidimensional mixed state vector And inputting the feedback optimization path and outputting the feedback optimization compensation quantity.
  6. 6. The dynamic tuning method for advanced simulation and hybrid optimization of numerical control machining according to claim 5, wherein a feed-forward feedback hybrid optimization framework including a feed-forward optimization path and a feedback optimization path is constructed, and the quantized prediction information is obtained Inputting to the feedforward optimization path, outputting feedforward optimization adjustment quantity, and mixing the multidimensional mixed state vector Inputting to the feedback optimization path and outputting a feedback optimization compensation quantity, comprising: quantizing the prediction information Rate of change of tool-workpiece contact arc length Input to the feedforward optimization path, and output feedforward optimization adjustment quantity , wherein, , For the feed-forward gain to be the same, For a nominal contact arc length, Is the rated feed speed; Construction of lightweight neural networks Jacobian matrix by simplifying physical model Approximately initializing partial weights of a first layer of the lightweight neural network ; Integrating an attention layer in the lightweight neural network, and calculating the attention weight of the attention layer ; Through the lightweight neural network Constructing a feedback optimization path; mixing the multi-dimensional hybrid state vectors Input to the feedback optimization path, output feedback optimization compensation quantity , wherein, 。
  7. 7. The dynamic regulation and control method for advanced simulation and hybrid optimization of numerical control machining according to any one of claims 1 to 6, wherein dynamically calculating a fusion weight according to a prediction confidence and a real-time noise level, performing weighted fusion on the feedforward optimization adjustment amount and the feedback optimization compensation amount to generate an optimization decision adjustment amount, issuing the optimization decision adjustment amount to a physical numerical control system for execution, regulating and controlling a machining process, and feeding back the regulated and controlled effect to the feedforward feedback hybrid optimization framework for online learning and parameter updating, and the method comprises: Based on the quantized prediction information Confidence evaluation model of (2), calculating the prediction confidence of the current time t Wherein the prediction confidence is used for representing the reliability degree of the digital twin model on a future processing state prediction result; Calculating a real-time noise level at the current time t based on the noise level estimate in the post-alignment sensor data ; Based on the prediction confidence And the real-time noise level Dynamic calculation of feedforward fusion weights Fusing weights with feedback , wherein, , ; Optimizing feedforward adjustment And feedback optimizing compensation quantity According to the feedforward fusion weight And the feedback fusion weights Weighting and fusing to generate optimal decision adjustment quantity , wherein, , Is a reference process parameter; adjusting the optimized decision Converting the command into a command which can be identified by a numerical control system, issuing the command to the physical numerical control system through an industrial communication protocol and executing the command; Continuously collecting the regulated instruction data stream and the multisource sensor data stream, and constructing a multi-objective rewarding function based on the regulated processing effect , wherein, , In order for the cutting force to fluctuate, For the surface roughness estimated based on the vibration characteristics, For the tool wear rate estimated based on the cutting power model, In order for the actual material removal rate to be the same, For a configurable multi-target weight coefficient, ∈{1,2,3,4}; Applying the multi-objective rewards function As a reward signal for reinforcement learning, the lightweight neural network in the feedback optimization path is updated online by using an asynchronous dominant actor-critique algorithm Network parameters of (a) are provided.
  8. 8. A dynamic regulation and control device for advanced simulation and hybrid optimization of numerical control machining is characterized by comprising: The digital twin model driving module is used for acquiring an instruction data stream of a physical numerical control system and multi-source sensor data streams of various sensors, performing time stamp alignment and data fusion processing on the instruction data stream and the multi-source sensor data streams to generate synchronous driving data, and injecting the synchronous driving data into the digital twin model, wherein the digital twin model is used for driving virtual and real synchronous simulation of a physical machining process; the quantitative prediction information generation module is used for prereading numerical control codes in a future time window, resolving the numerical control codes to generate a corresponding future tool motion track set, determining a dynamic local attention domain on a workpiece model of the digital twin model based on the future tool motion track set, performing local material removal simulation in the dynamic local attention domain and updating the workpiece model to generate quantitative prediction information of a future processing state, wherein the dynamic local attention domain is a local area of expected interference of a future tool envelope body and the workpiece model; The parameter optimization module is used for constructing a multidimensional mixed state vector containing mechanism characteristics, real-time data characteristics and the quantized prediction information, constructing a feedforward feedback mixed optimization framework, and generating feedforward optimization adjustment quantity and feedback optimization compensation quantity through the feedforward feedback mixed optimization framework based on the multidimensional mixed state vector; And the dynamic regulation and control processing module is used for dynamically calculating a fusion weight according to the prediction confidence and the real-time noise level, carrying out weighted fusion on the feedforward optimization adjustment quantity and the feedback optimization compensation quantity, generating an optimization decision adjustment quantity, transmitting the optimization decision adjustment quantity to a physical numerical control system for execution, regulating and controlling the processing process, and feeding back the regulated and controlled effect to the feedforward feedback hybrid optimization framework for online learning and parameter updating.
  9. 9. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the dynamic regulation method for advanced simulation and hybrid optimization of numerical control machining according to any one of claims 1 to 7 when executing the computer program.
  10. 10. A computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program for executing the dynamic regulation method for advanced simulation and hybrid optimization of numerical control machining according to any one of claims 1 to 7.

Description

Dynamic regulation and control method, device, computer equipment and medium for advanced simulation and hybrid optimization of numerical control machining Technical Field The invention relates to the technical field of numerical control machining, in particular to a dynamic regulation and control method, a device, computer equipment and a medium for advanced simulation and hybrid optimization of numerical control machining. Background In the field of high-end numerical control machining, particularly when machining parts with complex curved surfaces and high-strength material characteristics, the stability, precision and efficiency of the machining process are important. At present, the main flow of the processing control method has the following defects: The first prior art is processing based on fixed parameters or off-line optimization. The technological parameters are preset according to experience or off-line simulation and remain unchanged in the whole processing process. The method cannot respond to dynamic changes which occur in real time in the machining process, such as gradual abrasion of a cutter, local hardness difference of workpiece materials, abrupt change of cutting force and the like, and often causes fluctuation of machining quality, unexpected damage of the cutter or forced adoption of conservative parameters for safety, and efficiency is sacrificed. In the second prior art, the passive adjustment is based on real-time sensing feedback. The method monitors the processing state through sensors such as vibration, current and the like, and when abnormality (such as flutter and overload) is detected, post-regulation (such as speed reduction and shutdown) is carried out. This "sense-react" mode has a response lag, cannot intervene before the problem occurs, and the tuning process may cause secondary disturbances, which are rough for high precision machining. In the prior art, pure data driving online optimization is performed. And mapping the optimization parameters directly according to the real-time data by adopting a machine learning model. The method is severely dependent on a large amount of high-quality training data, decision reliability is suddenly reduced when working conditions exceed a training range, model interpretation is poor, and trust barriers and risks exist in actual industrial application. In summary, the prior art lacks an effective means for predicting the process state change in advance and performing real-time, accurate and steady parameter regulation based on an interpretable hybrid model. Disclosure of Invention In view of the above, the embodiment of the invention provides a dynamic regulation and control method for advanced simulation and hybrid optimization of numerical control machining, so as to solve the technical problem that a regulation and control method capable of foresight prediction of machining state change is lacking in the prior art. The method comprises the following steps: Acquiring an instruction data stream of a physical numerical control system and a multi-source sensor data stream of a plurality of sensors, performing time stamp alignment and data fusion processing on the instruction data stream and the multi-source sensor data stream, generating synchronous driving data, and injecting the synchronous driving data into a digital twin model, wherein the digital twin model is used for driving virtual and real synchronous simulation of a physical machining process; Pre-reading a numerical control code in a future time window, resolving the numerical control code to generate a corresponding future tool motion track set, determining a dynamic local attention domain on a workpiece model of the digital twin model based on the future tool motion track set, performing local material removal simulation in the dynamic local attention domain, updating the workpiece model, and generating quantitative prediction information of a future processing state, wherein the dynamic local attention domain is a local area of expected interference of a future tool envelope body and the workpiece model; constructing a multidimensional mixed state vector containing mechanism characteristics, real-time data characteristics and the quantized prediction information, constructing a feedforward feedback mixed optimization framework, and generating feedforward optimization adjustment quantity and feedback optimization compensation quantity through the feedforward feedback mixed optimization framework based on the multidimensional mixed state vector; And dynamically calculating a fusion weight according to the prediction confidence and the real-time noise level, carrying out weighted fusion on the feedforward optimization adjustment quantity and the feedback optimization compensation quantity to generate an optimization decision adjustment quantity, transmitting the optimization decision adjustment quantity to a physical numerical control system for execution, regulating and con