A Smart Control Method for Metal Material Processing Equipment Based on Digital Twin
By acquiring and analyzing the operating load data and sensor data of the processing equipment, and updating the digital twin model, the error compensation problem in multi-axis linkage processing was solved, real-time error compensation and precision control were realized, and the stability and accuracy of the processing equipment were improved.
Patent Information
- Application Number
- CN202611270195.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-20
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies in multi-axis linkage machining of metal materials suffer from problems such as dynamic error compensation delay and high-frequency distortion, long-term model drift, multi-axis interpolation timing misalignment, and instruction execution lag, which lead to decreased machining accuracy and mechanical interference, and cannot meet the error control requirements throughout the entire life cycle.
By acquiring the operating load data of the machining equipment, analyzing the friction coefficient evolution tensor and contact stiffness attenuation rate, and updating the boundary conditions of the dynamic model of the digital twin; using a dual-heat-source coupled Kalman observer to separate transient cutting heat, generating a spatial pose error matrix, and aligning it with the interpolation cycle of the CNC system to generate synchronous pose deviation data, which is then imported into the servo position loop for compensation displacement control.
It achieves real-time error compensation for the processing equipment throughout its entire life cycle, ensuring that the processing accuracy is within the target tolerance, avoiding motion interference and command execution lag in multi-axis linkage control, and guaranteeing the stability and continuity of the processing process.
Smart Images

Figure CN123113513A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of industrial control technology, specifically to an intelligent control method for metal material processing equipment based on digital twins. Background Technology
[0002] In multi-axis machining of metallic materials, the nonlinear time-varying physical field generated by the combined effects of cutting heat, frictional heat, and cutting force is the dominant factor causing spatial volume errors to deviate from the target tolerance. The control method based on digital twins aims to construct a virtual mapping network for the machine tool, achieving machining state tracking and error closed-loop control.
[0003] However, existing dynamic error compensation techniques are limited by the nonlinear physical evolution across multiple spatiotemporal scales, resulting in the following core defects: First, real-time calculation delays and high-frequency distortions of physical mechanisms. The iterative solution of the full-size finite element thermo-mechanical coupling equations takes longer than the basic interpolation cycle of the CNC system, failing to meet the dynamic compensation timeliness within a single interpolation cycle. If conventional matrix reduction methods are used to compress the solution cycle, the fixed reduction boundary will force the truncation of the high-frequency characteristics of nonlinear transient thermal shock in the local tool tip contact area, leading to severe distortion in local deformation prediction; while pure data-driven models lack boundary constraints of thermodynamic laws, easily outputting prediction values that violate physical common sense under extreme conditions. Second, time-varying degradation throughout the entire life cycle leads to long-term model drift. Existing twin models are mostly based on equipment calibration at the factory state. As the equipment service life progresses, physical wear of the guide rail pair and lead screw pair will cause irreversible slow drift of basic mechanical parameters such as the friction coefficient of the mating surface and the normal contact stiffness matrix. Because existing models lack an online adaptive update mechanism for fundamental physical parameters, the fixed weights of the data-driven network deviate from actual physical conditions over time, leading to a decline in long-term prediction accuracy. Thirdly, there are issues with multi-axis interpolation timing misalignment and underlying execution phase delay. Existing single-axis compensation mechanisms lack synchronous mapping scheduling for spatial affine transformations, making them prone to interpolation timing misalignment during multi-axis concurrent compensation, inducing transient interference stress within the mechanical transmission chain. Such additional mechanical loads and physical energy cannot be effectively absorbed or mitigated by the machine tool's main structure, leaving ripple defects on the workpiece surface when fed back to the cutting execution end. Furthermore, due to the closed proprietary protocol barriers at the bottom layer of the CNC system, the displacement compensation amount suffers from severe transmission phase delay, making it impossible to establish a deterministic synchronous communication link with the feedforward parameters of the underlying servo position loop.
[0004] In summary, improving the accuracy of dynamic spatial volume error compensation for processing equipment throughout its entire life cycle and under nonlinear time-varying conditions, and avoiding motion interference and command execution lag during multi-axis linkage control, are technical problems that urgently need to be solved in this field. Summary of the Invention
[0005] The purpose of this invention is to provide an intelligent control method for metal material processing equipment based on digital twins, which can realize real-time identification and error compensation of physical degradation of processing equipment.
[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for intelligent control of metal material processing equipment based on digital twins, comprising: The operating load data of the processing equipment is obtained and the friction coefficient evolution tensor and contact stiffness attenuation rate are extracted and analyzed to update the boundary condition constraints of the digital twin dynamics model. Acquire machining sensor data, and under boundary condition constraints, separate transient cutting heat through a dual-heat-source coupled Kalman observer. Output transient deformation vector after order reduction solution. Combine geometric tolerance matrix and transient deformation vector to generate spatial pose error matrix. Align spatial pose error matrix with the interpolation cycle of CNC system to generate synchronous pose deviation data. Based on the viscous friction component in the friction coefficient evolution tensor and the synchronous pose deviation data, an axial acceleration smoothing factor is generated. The axial acceleration smoothing factor is used as a dynamic feedforward parameter and imported into the interpolation trajectory planning module of the CNC system to smooth out transient mechanical loads. The synchronous pose deviation data is separated into attitude offset components and position offset components. The control timing sequence of first superimposing the rotation axis compensation amount and then superimposing the linear axis compensation amount is reconstructed into a communication message, which is injected into the servo position loop superposition node of the CNC system to control the servo motor to output the compensation displacement.
[0007] Preferably, the step of analyzing and extracting the friction coefficient evolution tensor and contact stiffness attenuation rate from the operating load data includes: obtaining the feed drive torque time series and servo follower error series under no-load and loaded operating conditions of the machining equipment; converting the feed drive torque time series to the frequency domain using discrete-time Fourier transform, and separating each order of harmonic friction resistance components according to the spindle speed fundamental frequency; substituting each order of harmonic friction resistance components into the LuGre friction model, analyzing and extracting the micro-stiffness parameters and Coulomb friction parameters, and assembling them to generate the friction coefficient evolution tensor; extracting the phase lag angle parameters of the servo follower error series in the backlash region, establishing a spatial mapping grid of the phase lag angle parameters and normal force deformation, and generating the contact stiffness attenuation rate by comparing with the factory rated stiffness reference value.
[0008] Preferably, the step of updating the boundary condition constraints of the digital twin dynamic model includes: extracting the set of discrete contact mesh nodes representing the mechanical assembly interface from the digital twin mesh model; using the contact stiffness attenuation rate as the independent variable of the stiffness degradation reduction coefficient, reconstructing the local contact stiffness matrix elements corresponding to the set of discrete contact mesh nodes using the spring-damping equivalent model, and assembling the local contact stiffness matrix elements into the global stiffness matrix of the system; decomposing the friction coefficient evolution tensor into nodal tangential viscous damping coefficients and nonlinear Coulomb force vectors according to the degrees of freedom of motion; reloading the nodal tangential viscous damping coefficients into the global damping matrix of the system, and merging the nonlinear Coulomb force vector as the reverse mechanical resistance boundary condition into the external excitation load vector of the dynamic state space equation, thereby completing the synchronous update of the boundary condition constraints.
[0009] Preferably, the step of separating transient cutting heat using a dual-heat-source coupled Kalman observer and outputting the transient deformation vector through order reduction solution includes: extracting the contact area spindle temperature rise gradient data and the effective value of the three-phase current of the spindle motor from the machining sensing data; inputting the contact area spindle temperature rise gradient data and the effective value of the three-phase current of the spindle motor into the dual-heat-source coupled Kalman observer; using the spindle temperature rise gradient and the effective value of the three-phase current as the observation vector, and the structural frictional heat generation component and the transient cutting heat flux density as the state vector, the Kalman filter equation is used to iteratively separate the structural frictional heat. The frictional heat component is used to extract the transient cutting heat flux density; the transient cutting heat flux density is input into the intrinsic orthogonal decomposition (POD) reduction module to truncate and reduce the dimension of the thermo-coupled mesh model inside the digital twin to generate a spatial basis function matrix; based on the projection residual of the POD reduction, physical constraint correction terms are constructed for the boundary nodes of the truncated region and coupled to the right-hand side of the reduced-order control equation to compensate for the truncation error; the physical constraint correction terms are coupled to the right-hand side of the reduced-order control equation using the spatial basis function matrix, and a step solution is performed within the reduced-order dimension to output the transient deformation vector.
[0010] Preferably, the step of generating synchronous pose deviation data includes: extracting error calibration data of each feed axis of the machining equipment and the worktable to construct a geometric tolerance matrix; using homogeneous coordinate transformation, performing tensor fusion product operation on the geometric tolerance matrix and the transient deformation vector to solve the spatial pose error matrix; listening to the fieldbus communication clock beat of the CNC system and extracting the reference synchronization trigger pulse issued by the interpolator; at the moment when the rising edge of the reference synchronization trigger pulse is captured, sampling and latching the spatial pose error matrix, eliminating inter-phase communication jitter data, and outputting synchronous pose deviation data with the same time interval attribute as the interpolation period.
[0011] Preferably, the step of generating an axial acceleration smoothing factor and importing it into the interpolation trajectory planning process to smooth out transient mechanical loads includes: reading the viscous friction component in the friction coefficient evolution tensor, calculating the transient mechanical interference load during multi-axis linkage by combining synchronous pose deviation data; determining the acceleration constraint threshold based on the damping ratio of the machine tool bed structure; obtaining the acceleration curves of each axis by differentiating the look-ahead velocity curves, superimposing the equivalent interference acceleration components, and then truncating the extreme values using the acceleration constraint threshold; multiplying the truncated difference by the damping gain coefficient to generate the axial acceleration smoothing factor, wherein the damping gain coefficient is determined by linear mapping based on the structural damping ratio within a preset range; and importing the smoothing factor into the interpolation trajectory planning module to smooth out acceleration extreme value jumps.
[0012] Preferably, the steps of reconstructing the communication message, injecting it into the servo position loop superposition node of the CNC system, and controlling the servo motor to output compensation displacement include: performing spatial decoupling calculation on the synchronous pose deviation data, separating the attitude offset component and the position offset component, and encapsulating them into a communication message; establishing a communication link with the CNC system kernel and completing clock phase alignment, locating the physical address of the position control register in the servo drive chip of each axis; within the current interpolation cycle synchronization window, according to the control timing sequence of first correcting the rotary axis attitude and then translating the linear axis position, feeding forward the pose offset component and the position offset component decoded and restored from the communication message into the corresponding servo position loop superposition node, updating the servo pulse equivalent, and performing the compensation displacement operation.
[0013] Preferably, after outputting the compensation displacement, the method further includes a parameter iteration update step for the digital twin, specifically including: obtaining the actual processing contour coordinate set after the processing equipment completes the processing cycle; calculating the difference between the actual processing contour coordinate set and the theoretically predicted contour coordinate set output by the digital twin to generate an error residual matrix; when the absolute value of the error residual matrix exceeds the allowable threshold of geometric tolerance, the friction coefficient evolution tensor is inverted and corrected using the least squares method, and fed back to the digital twin for parameter iteration reloading.
[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention acquires operational load data, extracts the friction coefficient evolution tensor and contact stiffness attenuation rate, reconstructs the contact stiffness matrix using stiffness degradation reduction coefficient and spring-damping model, and updates the underlying stiffness, damping, and heat flow conduction boundary conditions of the digital twin; it separates transient cutting heat through a Kalman filter with dual heat sources coupled to a Kalman observer, and outputs a transient deformation vector after intrinsic orthogonal decomposition; it fuses the geometric tolerance matrix and transient deformation vector to generate a spatial pose error matrix, and generates synchronous pose deviation data by aligning it with the interpolation cycle through a timing latch scheduler; it injects dynamic feedforward parameters into the interpolation planning module to smooth mechanical loads, and injects pose deviation data into the servo position loop to drive the motor output compensation displacement, thereby achieving real-time identification and error compensation of the physical degradation of the machining equipment. Specifically: 1. This invention analyzes operational load data based on the dynamic model of a digital twin, extracts the friction coefficient evolution tensor and contact stiffness attenuation rate, and can dynamically reload the boundary conditions of the underlying stiffness and damping matrix of the digital twin. This mechanism enables the virtual prediction model to perceive the natural wear and structural mechanical evolution of the machine tool's physical entity, ensuring the convergence stability of the error deduction logic over a long service life, and maintaining the spatial volume error under long-term machining conditions within the target tolerance range; 2. This invention utilizes a dual-heat-source coupled Kalman observer to accurately separate transient cutting heat from multi-source sensor data. Combined with a reduced-order solution module outputting a deformation vector, it achieves the tracking of local transient thermodynamic physical characteristics. Furthermore, a timing latch scheduler is used to strictly align the mapped spatial pose error matrix with the core interpolation cycle of the CNC system, ensuring that the externally generated spatial pose deviation data and the internal servo control cycle are synchronized at equal time intervals, eliminating the risk of instruction lag caused by interpolation timing misalignment. 3. In the spatial kinematics control dimension, the synchronous pose deviation data is separated into attitude offset components and position offset components. Following the control sequence of first driving the rotary axis to correct the spatial attitude and then driving the linear axis to translate the spatial position, the message is directly overwritten to the servo position loop as a feedforward pre-control parameter, thus streamlining the nonlinear affine transformation logic in five-axis concurrent operation. Simultaneously, the axial acceleration smoothing factor is integrated for pre-intervention trajectory planning, enabling the transient mechanical interference loads and cutting resistances generated within the mechanical transmission chain during multi-axis coordinated motion to be effectively absorbed and mitigated by the machine tool bed's damping network, ensuring the continuity of the cutting trajectory and the smooth operation of the machine tool body. Attached Figure Description
[0015] Figure 1 This invention relates to an intelligent control method for metal material processing equipment based on digital twins. Figure 2 This is a diagram of the main architecture of the double-ring nested control system in an embodiment of the present invention; Figure 3This is a flowchart of spatial error decoupling and servo injection in an embodiment of the present invention. Detailed Implementation
[0016] The technical solution of the present invention will now be clearly and completely described in conjunction with the embodiments and accompanying drawings.
[0017] like Figures 1 to 3 As shown, this invention provides an intelligent control method for metal material processing equipment based on digital twins, and the technical solution is as follows: Intelligent control methods for metal material processing equipment based on digital twins include: The operating load data of the processing equipment is obtained and the friction coefficient evolution tensor and contact stiffness attenuation rate are extracted and analyzed to update the boundary condition constraints of the digital twin dynamics model. Acquire machining sensor data, and under boundary condition constraints, separate transient cutting heat through a dual-heat-source coupled Kalman observer. Output transient deformation vector after order reduction solution. Combine geometric tolerance matrix and transient deformation vector to generate spatial pose error matrix. Align spatial pose error matrix with the interpolation cycle of CNC system to generate synchronous pose deviation data. Based on the viscous friction component in the friction coefficient evolution tensor and the synchronous pose deviation data, an axial acceleration smoothing factor is generated. The axial acceleration smoothing factor is used as a dynamic feedforward parameter and imported into the interpolation trajectory planning module of the CNC system to smooth out transient mechanical loads. The synchronous pose deviation data is separated into attitude offset components and position offset components. The control timing sequence of first superimposing the rotation axis compensation amount and then superimposing the linear axis compensation amount is reconstructed into a communication message, which is injected into the servo position loop superposition node of the CNC system to control the servo motor to output the compensation displacement.
[0018] Example 1: This embodiment focuses on the application scenario of five-axis continuous machining of typical thin-walled parts. In the CNC machining of titanium alloy impellers, as machining continues, the machine tool feed system and spindle structure undergo complex nonlinear thermal deformation due to continuous frictional heat generation. Furthermore, the machine tool guideways and lead screws experience irreversible natural wear after hundreds of hours of service, leading to dynamic drift of the spatial volume error between the rotary and linear axes during multi-axis linkage. Simultaneously, the traditional mixed error direct injection mechanism triggers timing conflicts in multi-axis interpolation, thereby generating transient mechanical interference within the mechanical transmission chain, affecting the workpiece surface roughness. This embodiment achieves adaptive convergence and mitigation of errors through a double-loop nested control system.
[0019] As one embodiment of the present invention, refer to Figure 1 Flowchart of intelligent control method for metal material processing equipment based on digital twin, referring to Figure 2 The main architecture diagram of the dual-ring nested control system is shown in the reference diagram. Figure 3 Flowchart of spatial error decoupling and servo injection.
[0020] In the pre-processing stage of data acquisition, a three-axis microelectromechanical system (MEMS) accelerometer attached to the mating surface between the spindle end face shell and the bed guide rail is used to collect multi-dimensional vibration acceleration signals; a non-contact infrared array temperature sensor arranged inside the cutting chamber is used to collect the radiation temperature field signal of the cutting contact area; and a current transformer inside the spindle servo driver is used to collect the effective value signal of the three-phase current of the spindle motor. A hardware synchronization clock chip is used to force the three signal streams to be timestamped at equal intervals through hardware pin triggering.
[0021] Considering the significant differences in the physical response bandwidths of the three signals—acceleration signal bandwidth in the kilohertz range, current signal bandwidth in the 100 Hz range, and temperature signal bandwidth in the Hz range—different wavelet packet decomposition strategies with different parameters are used for denoising the three signals. Specifically, a six-level decomposition is applied to the acceleration signal to cover its complete high-frequency vibration characteristic bandwidth; a four-level decomposition is applied to the current signal to match the frequency range of its main harmonic components; and a two-level decomposition is applied to the temperature signal to remove only high-frequency instrument noise while retaining its low-frequency temperature rise trend information. After denoising the three signals using wavelet packet decomposition with their respective parameters, the timestamps are aligned based on the sampling rate of the acceleration signal; linear interpolation is performed between adjacent sampling points for the temperature and current signals to make their time resolution consistent with that of the acceleration signal, thereby outputting a time-aligned processing sensing data stream.
[0022] Furthermore, in the subsequent Kalman filter state estimation stage, considering that the physical response time constant of the temperature rise gradient is much larger than the update period of the current effective value, the prediction step size of the Kalman filter is set based on the effective response period of the temperature signal, and the current effective value is preprocessed by mean filtering within this prediction step size. The current values sampled multiple times are aggregated into a slow variable that matches the time scale of the temperature rise gradient, thus avoiding numerical oscillations in the state estimation caused by the time scale difference between the two observation signals.
[0023] Further, the steps of analyzing and extracting the friction coefficient evolution tensor and contact stiffness attenuation rate from the operating load data include: obtaining the feed drive torque time series and servo follower error series under no-load and loaded operating conditions of the machining equipment; converting the feed drive torque time series to the frequency domain using discrete-time Fourier transform, and separating each order of harmonic friction resistance components according to the fundamental frequency of the spindle speed; substituting each order of harmonic friction resistance components into the LuGre friction model, and analytically extracting the micro-stiffness parameters of the pre-sliding stage and the Coulomb friction parameters of the macro-sliding stage, assembling and arranging the micro-stiffness parameters and the Coulomb friction parameters to generate the friction coefficient evolution tensor; extracting the phase lag angle parameters of the servo follower error series in the reverse clearance region, establishing a spatial mapping grid of the phase lag angle parameters and normal force deformation, and generating the contact stiffness attenuation rate by comparing with the factory rated stiffness reference value.
[0024] The process of generating a contact stiffness attenuation rate by comparing the factory-rated stiffness reference value involves: establishing a linear mapping relationship between the phase lag angle parameter and the normal force deformation of the corresponding feed axis; using the phase lag angle as the independent variable; and calculating the estimated value of the current normal force deformation using a pre-calibrated slope coefficient and offset; and then inferring the current contact stiffness based on the ratio of normal force to deformation. Using the factory-rated normal contact stiffness of each feed axis as a benchmark, the contact stiffness attenuation rate is calculated as the ratio of the current contact stiffness to the factory-rated contact stiffness. This value is dimensionless and ranges from 0 to 1, where 0 indicates no degradation and 1 indicates complete loss of stiffness.
[0025] It should be noted that the bristle elastic deformation assumption of the LuGre friction model applies to the viscous-slip transition range experienced by the transmission chain during frequent reversals. During contour cutting on a CNC machine tool, each feed axis experiences a zero-velocity interval near the curvature abrupt change point, and the friction behavior falls precisely at this transition stage. This scheme separates the harmonic friction resistance components from the feed drive torque time series using discrete-time Fourier transform. Their physical origin is the nonlinear force fluctuation caused by the alternation of viscous and slip within the reversal transition range. Substituting these harmonic components into the LuGre model identifies the microscopic stiffness parameters and Coulomb friction parameters. The applicability of this model holds true under the given conditions, and the identification results characterize the gradual degradation trend of the frictional characteristics in the reversal transition range during long-term service of the transmission chain.
[0026] Specifically, under standard test conditions, the control system acquires the multidimensional data stream from the feed servo drive and decouples it into a feed drive torque time series and a servo following error series with a sampling frequency of 1000 Hz. Based on the Nyquist theorem, this can cover the first five harmonic frictional resistance components of the spindle rotation fundamental frequency (100 Hz), meeting the frequency resolution requirements for LuGre model parameter identification. The feed drive torque time series is transformed to the frequency domain using discrete-time Fourier transform. Based on the spindle rotation fundamental frequency, the cutting component caused by workpiece resistance is extracted, thus separating the harmonic frictional resistance components. These harmonic frictional resistance components are then substituted into the LuGre friction model to calculate the Coulomb friction parameters and equivalent tangential stiffness parameters of the contact surface during the viscous-slip transition stage. These stiffness parameters and Coulomb friction parameters are then assembled and arranged according to the degrees of freedom of motion to generate a first-order friction coefficient evolution tensor.
[0027] The extracted microscopic tangential stiffness, microscopic damping coefficient, and Coulomb friction amplitude along each feed axis are arranged into row vectors according to the feed axis number. Each column corresponds to a set of friction parameters for one feed axis, and the number of columns for a five-axis machine tool is fifteen (5 axes × 3 parameters). These row vectors are stacked along the time axis for each sampling period to form a two-dimensional matrix with the sampling time as the row direction and the feed axis friction parameters as the column direction; this is the friction coefficient evolution tensor. Each row of this tensor records the instantaneous state of the friction parameters of each feed axis at the current sampling time, and each column records the evolution trend of a certain feed axis friction parameter over time. It should be noted that the first order mentioned in this embodiment refers to the first round of iterative updates, not the mathematical order of the tensor.
[0028] Simultaneously, the phase lag angle parameter of the servo following error sequence within the backlash region is extracted. Within the commutation frequency range of 0.5 to 2 Hz, the phase lag angle range for each feed axis is 9 degrees to 54 degrees (based on the relationship between phase lag angle, commutation frequency, and backlash time constant). The calculated time constant of the backlash is... The equivalent hysteresis time of the ball screw drive chain during the viscosity-slip transition phase is given by the pre-slip stiffness parameters of the LuGre model. With maximum static friction The values are determined by the combination of factors, with a typical value of 4 to 5 milliseconds; the commutation frequency range of 0.5 to 2 Hz corresponds to the conventional commutation cycle of the CNC system interpolator during contour cutting; the upper and lower limit combinations cover the typical feed conditions from low-speed heavy load to high-speed light load), thereby establishing a spatial mapping grid of phase lag angle and normal force deformation, and calculating the contact stiffness attenuation rate by comparing it with the rated stiffness reference value under factory conditions.
[0029] This invention, by decoupling the operating load to the frequency domain and combining it with mechanical model inversion, can quantify and separate the friction and stiffness degradation parameters of the microscopic interface of the transmission chain. This step provides the digital twin system with mechanical description features that accurately reflect the actual wear condition of the physical machine tool, solving the problem of deviation of basic physical parameters from the initial factory settings due to long-term service, and improving the accuracy of state identification at the data-driven level.
[0030] Further, the steps for updating the boundary condition constraints of the digital twin dynamics model include: extracting the set of discrete contact mesh nodes representing the mechanical assembly interface from the digital twin mesh model; using the contact stiffness attenuation rate as the independent variable of the stiffness degradation reduction coefficient, reconstructing the local contact stiffness matrix elements corresponding to the set of discrete contact mesh nodes using the spring-damping equivalent model, and assembling the local contact stiffness matrix elements into the global stiffness matrix of the system; decomposing the friction coefficient evolution tensor into nodal tangential viscous damping coefficients and nonlinear Coulomb force vectors according to the degrees of freedom of motion; reloading the nodal tangential viscous damping coefficients into the global damping matrix of the system, and merging the nonlinear Coulomb force vector as the reverse mechanical resistance boundary condition into the external excitation load vector of the dynamic state space equation, thereby completing the synchronous update of the boundary condition constraints.
[0031] The step of updating the boundary condition constraints of the digital twin dynamics model further includes: calculating the increase in local thermal contact impedance of the corresponding discrete contact mesh node set based on the contact stiffness attenuation rate; using the increase in local thermal contact impedance as an additional correction amount for boundary thermal resistance in the heat conduction partial differential equation, and simultaneously correcting the heat flow conduction matrix in the digital twin dynamics model to eliminate the distortion in heat flow density distribution calculation caused by contact stiffness degradation.
[0032] Specifically, the control system extracts a set of discrete contact mesh nodes representing the mating surfaces of the X-axis guide slider, Y-axis guide slider, and Z-axis lead screw nut from the digital twin dynamic model. This set typically contains, for example, 2000 to 5000 (e.g., 3200) spatial coordinate nodes, the specific number of which can be adaptively adjusted according to the geometric dimensions of the machine tool mating surfaces and the finite element mesh generation accuracy. To quantify and map macroscopic degradation to microscopic nodes, the attenuation ratio of the normal contact stiffness of the mating surface is first used. Construct stiffness degradation reduction coefficients as independent variables. :when hour This corresponds to a factory-degraded condition; when hour This corresponds to a severely degraded state; as the decay rate increases, It monotonically decreases in the interval from zero to one, realizing a continuous penalty mapping for stiffness degradation.
[0033] Subsequently, the stiffness coefficients of 3200 contact mesh nodes under normal load were recalculated using a nonlinear spring-damped equivalent model. The normal contact force was then applied to each discrete contact mesh node. Modeled as normal elastic deformation The nonlinear function is composed of two parts: the first part is the linear elastic restoring force term. ,in This is the factory-rated normal stiffness reference value for this node. The first part is the stiffness degradation reduction factor, which characterizes the linear elastic response of the foundation structure after wear degradation; the second part is the nonlinear hardening term. The nonlinear stiffness hardening effect caused by the contact area of the rough peaks in the surface microstructure increases with increasing load is characterized by the Hertzian contact constant. Comprehensive elastic modulus With equivalent contact radius of curvature The normal contact force is determined by the factory specifications of the ball screw diameter and the elastic modulus of the guide rail material. Combining the two parts, the expression for the normal contact force is: .
[0034] When solving, Regarding normal elastic deformation Calculate the first-order partial derivatives to obtain the local normal tangent stiffness coefficients of each mesh node after dynamic evolution. Since the 3200 discrete contact mesh nodes are sparsely distributed and independent in space, the cross-stiffness coupling term between nodes can be ignored. Therefore, the local normal tangent stiffness coefficient of each node is... As diagonal elements, the local contact stiffness matrix elements are reconstructed in diagonal sparse matrix form, and these local matrix elements are assembled into the global stiffness matrix of the system according to the node number. At the same time, the friction coefficient evolution tensor is orthogonally decomposed according to the degrees of freedom of motion of the linear axis and the rotation axis, and transformed into the nodal tangential viscous damping coefficient and nonlinear Coulomb force vector on each grid node.
[0035] The nodal tangential viscous damping coefficients are directly reloaded into the system's global damping matrix, overriding the original static damping allocation values. The nonlinear Coulomb force vector is used as the boundary condition for the reverse mechanical resistance and incorporated into the external excitation load vector of the dynamic state-space equation. This dynamic state-space equation consists of a state vector composed of the system's nodal displacement and velocity vectors, and a system matrix composed of the inverse of the system's global mass matrix, the global stiffness matrix, and the global damping matrix. The external excitation load vector includes three components: the cutting force vector, the nonlinear Coulomb force vector, and the thermal expansion equivalent force vector, describing the second-order dynamic response process of each grid node in the digital twin under external loads. Simultaneously, based on the generated contact stiffness attenuation rate, the control system calculates the increase in local thermal contact impedance for the discrete contact grid node sets corresponding to the X-axis, Y-axis, and Z-axis transmission interface. This increase in local thermal contact impedance is used as an additional correction for boundary thermal resistance in the thermo-coupling heat conduction partial differential equation, simultaneously correcting the heat flow conduction matrix in the digital twin's dynamic model. By synchronously updating the mechanical stiffness and thermal conductivity impedance, the distortion in the heat flux density distribution calculation caused by the degradation of contact stiffness is eliminated, and the cross-physical domain synchronous reconstruction and update of the boundary condition constraints of the digital twin dynamic model is completed.
[0036] This invention utilizes a stiffness degradation reduction factor and a spring-damping model to non-destructively map macroscopic performance degradation parameters to microscopic mesh nodes. Simultaneously, it introduces an adaptive boundary thermal resistance correction factor for local thermal contact impedance, eliminating computational distortion of the microscopic heat transfer matrix without the need for additional sensors. This method solves the problem of distorted heat transfer boundary conditions caused by mechanical wear, maintains dynamic consistency between the virtual predicted state and the physical entity behavior throughout the entire lifecycle, effectively suppresses spatial volume error divergence under nonlinear temperature variations, and ensures the long-term stability and reliability of the control closed loop.
[0037] Furthermore, the transient cutting heat is separated using a dual-heat-source coupled Kalman observer, and the transient deformation vector is output after order reduction. This includes: extracting the spindle temperature rise gradient data and the effective value of the three-phase current of the spindle motor from the machining sensing data; inputting the spindle temperature rise gradient data and the effective value of the three-phase current of the spindle motor into the dual-heat-source coupled Kalman observer; using the spindle temperature rise gradient and the effective value of the three-phase current as the observation vector, and the structural frictional heat generation component and the transient cutting heat flux density as the state vector, a Kalman filter equation is used to iteratively separate the structural frictional heat. The frictional heat component is used to extract the transient cutting heat flux density; the transient cutting heat flux density is input into the intrinsic orthogonal decomposition (POD) reduction module to truncate and reduce the dimension of the thermo-coupled mesh model inside the digital twin to generate a spatial basis function matrix; based on the projection residual of the POD reduction, physical constraint correction terms are constructed for the boundary nodes of the truncated region and coupled to the right-hand side of the reduced-order control equation to compensate for the truncation error; the physical constraint correction terms are coupled to the right-hand side of the reduced-order control equation using the spatial basis function matrix, and a step solution is performed within the reduced-order dimension to output the transient deformation vector.
[0038] Specifically, during the actual cutting process of the titanium alloy impeller, the control system extracts the spindle temperature rise gradient data in the contact area and the effective value of the three-phase current of the spindle motor from the machining sensor data in real time. This data is then input into a dual-heat-source coupled Kalman observer, and a first-order Kalman filter state equation is used to perform joint state estimation on the two input signals. The state vector is set to include the frictional heat generated by the bearing structure. With transient cutting heat flux density Two-dimensional column vector The observation vector is set as the principal axis temperature rise gradient. RMS value of three-phase current of spindle motor The constructed two-dimensional column vector State transition matrix Constructed based on the physical dynamics of the two components: Frictional Heat Generation Component It follows the first-order heat conduction law of the machine tool spindle, and its diagonal elements are set as ,in The sampling period is The heat transfer time constant inside the machine tool, with a value range of [value range missing]. The transient cutting heat flux density is determined by the machine tool's factory thermal calibration test. It approximates a constant random walk within a single-step time domain, with diagonal elements set as follows: The coupling coefficient between the two state components , i.e., matrix .
[0039] Observation matrix Constructing a temperature rise gradient based on the physical mapping relationship of multi-source heat transfer. Through the heat capacity equivalent coefficient Linear response to frictional heat component Based on the one-dimensional steady-state heat conduction model, the equivalent coefficient of heat capacity is determined. ,in Let λ be the annular contact area of the bearing, and λ be the thermal conductivity of the principal shaft material. This represents the equivalent conduction path length from the heat source to the temperature sensor. All parameters were determined from the machine tool's structural specifications and thermal calibration experiments. Dimensions are K·m / W; temperature gradient Simultaneously, through the thermal diffusivity of the cutting zone Linear response to transient cutting heat flux density , ,in The equivalent conduction path length from the cutting heat source to the temperature sensor is determined by the machine tool structure layout from the tool contact area to the sensor. Both have the same dimension of K·m / W, but the path lengths from the frictional heat source and the cutting heat source to the sensor are different. The two column vectors are linearly independent, ensuring the observation matrix is full rank. RMS current value. Through electrothermal conversion coefficient Linear response to frictional heat component The electrothermal conversion coefficient is determined based on the electrothermal conversion relationship. ,in This is the equivalent phase resistance of the motor stator winding. The effective value of the no-load reference current is determined by the motor's factory parameters and no-load calibration test. Dimensions are A·m² / W; RMS current value Simultaneously through cutting force Current mapping coefficient Linear response to transient cutting heat flux density , Spindle drive efficiency and cutting torque The current ratio calibration is determined, and the dimensions are all A·m² / W. Therefore, the observation matrix is: Both columns contain non-zero elements, ensuring that both the frictional heat generation component and the transient cutting heat flux density can be estimated in a closed loop through joint updates of the two observations. Process noise covariance matrix. Covariance matrix of observation noise All elements are set as diagonal matrices, and each element is determined by offline noise power spectrum estimation under machine tool static conditions. Within each sampling period, the frictional heat generation component of the bearing structure with exponential decay characteristics is dynamically extracted based on the standard Kalman prediction-update two-step iterative calculation. The transient cutting heat flux density with pulse step characteristics was accurately separated. .
[0040] The transient cutting heat flux density is used as a real-time load input to the intrinsic orthogonal decomposition and order reduction module. Before online truncation and dimensionality reduction, the system needs to pre-calculate experimental design matrices based on different spindle speeds and cutting depths. In offline mode, finite element analysis software is run to calculate and generate a global three-dimensional thermo-mechanical coupled mesh node deformation time series covering different thermal conditions. This time series is then extracted and constructed into a high-dimensional historical snapshot matrix. The experimental design matrix uses spindle speed and cutting depth as experimental factors, each with three to five levels, combined according to a full-factor design. The snapshot matrix uses the number of global mesh nodes as rows and the number of time sampling points as columns, with each column representing the global node deformation vector at the corresponding time. During online machining, the high-dimensional historical snapshot matrix is subjected to covariance matrix eigenvalue decomposition. The eigenvalues are sorted from largest to smallest, and the smallest order whose cumulative energy contribution rate reaches a preset threshold (this threshold is determined based on the balance between accuracy requirements and computational efficiency, typically ranging from 90% to 99%, and 95% in this embodiment) is selected as the retained order of the spatial basis function matrix. In the titanium alloy impeller cutting condition of this embodiment, offline training verification showed that the cumulative energy contribution rate of the first 5 eigenvectors reached 96%, and the truncation order was set to 5. This truncation order is determined by the specific working condition and the eigenvalue spectrum distribution of the mesh model. For more complex thermo-coupling conditions or finer mesh division, the required truncation order may increase. The control system automatically calculates the minimum order that meets the threshold during the offline training phase and stores it in the digital twin configuration parameters.
[0041] To address the deformation deviation at the boundary nodes of the truncated region caused by rapid changes in local temperature gradients, the contribution of the truncated mode at the boundary nodes is estimated based on the projection residuals of the reduced-order POD model. A physical constraint correction term is constructed and coupled to the right-hand side of the reduced-order POD governing equations to compensate for the truncation error at the boundary. The global stiffness matrix and global damping matrix only participate in the Galerkin projection calculation. The calculation and storage of the projection matrix are completed in advance during the offline stage; during the online solution stage, step calculations are performed within the reduced-order dimension, outputting the transient deformation vector.
[0042] This invention utilizes Kalman filtering and dual order reduction techniques to separate transient cutting heat while ensuring physical conservation. By truncating the full-size thermodynamic matrix and mapping it to low-dimensional basis functions, it achieves step-by-step solution of the reduced-order dynamic equations within the CNC interpolation cycle. Boundary correction terms based on projection residuals compensate for the loss of high-frequency transient features caused by modal truncation, ensuring the real-time performance and physical accuracy of deformation prediction in the critical tool tip region without increasing computational overhead.
[0043] Furthermore, the steps for generating synchronous pose deviation data specifically include: extracting the error calibration data of each feed axis of the machining equipment and the worktable to construct a geometric tolerance matrix; using homogeneous coordinate transformation, performing tensor fusion product operation on the geometric tolerance matrix and the transient deformation vector to solve the spatial pose error matrix; listening to the fieldbus communication clock beat of the CNC system and extracting the reference synchronization trigger pulse emitted by the interpolator; at the moment when the rising edge of the reference synchronization trigger pulse is captured, sampling and latching the spatial pose error matrix, eliminating inter-phase communication jitter data, and outputting synchronous pose deviation data with the same time interval attribute as the interpolation period.
[0044] Specifically, the control system pre-extracts positioning error, straightness error, pitch error, yaw error, and roll error data of each linear feed axis and rotary table of the five-axis machining center, calibrated by a laser interferometer in the entire space, and constructs a geometric tolerance matrix of 21 geometric errors. Using homogeneous coordinate transformation, the geometric tolerance matrix and the dynamically output transient deformation vector are fused and multiplied in the workpiece reference coordinate system to solve for the spatial pose error matrix generated by the coupling of each motion axis under the current spatial pose. The timing latch scheduler is then activated, and the onboard fieldbus control chip listens for the fieldbus communication clock beat that matches the current interpolation cycle of the CNC system, extracting the reference synchronization trigger pulse issued by the interpolator. The timing latch scheduler uses a field-programmable gate array (FPGA) as its hardware platform. It locks the clock phase of the current interpolation cycle by tracking the fieldbus synchronization frame messages of the CNC system. A hardware interrupt is triggered at the rising edge of the reference synchronization trigger pulse, latching the currently calculated spatial pose error matrix into a register. Simultaneously, jitter sampling data before and after the trigger moment is discarded, outputting synchronization pose deviation data strictly aligned with the current interpolation cycle. Upon capturing the rising edge of the reference synchronization trigger pulse, an interrupt request is immediately triggered to the analog-to-digital converter and the computing core. This forces sampling and hardware latching of the currently calculated spatial pose error matrix, eliminating clock jitter data caused by asynchronous fieldbus transmission, thereby outputting synchronization pose deviation data with equal time intervals to the CNC system's interpolation cycle.
[0045] To ensure the continuous equal time interval between the output data and the interpolation cycle, the timing latch scheduler internally maintains a phase tracking loop with the reference synchronization trigger pulse as the reference clock. Each time the rising edge of the reference synchronization trigger pulse is captured, this loop calculates the actual interpolation cycle time interval by subtracting the current system clock count from the clock count at the previous trigger edge, and stores this interval as the current interpolation cycle estimate. The digital twin's solver scheduler uses this interpolation cycle estimate as a reference, locking the start time of its solver task at a fixed offset time after the rising edge of the reference synchronization trigger pulse, ensuring that each solver result is completed and ready before the next rising edge. If a single solver operation of the digital twin takes more than one interpolation cycle (i.e., the solver fails to complete before the next rising edge), the latch scheduler discards the current solver result, uses the latched data from the previous cycle, and triggers an alarm flag to notify the upper-level control module to enter data retention mode. Through the aforementioned coordinated mechanism of phase tracking and solution scheduling, the time interval of the output data under normal operating conditions is guaranteed to strictly track the interpolation cycle of the CNC system, thereby achieving continuous time-distance synchronization.
[0046] This invention achieves spatial fusion of geometric-thermal coupling errors and temporal synchronization of virtual-real mappings by deeply coupling homogeneous coordinate transformation with the CNC bus hardware clock interrupt. The timing latch scheduler fundamentally eliminates the timing misalignment between the external calculation cycle and the internal hardware interpolation cycle, avoiding instruction backlog and control divergence, and achieving continuous convergence of spatial errors.
[0047] Further, the steps of generating an axial acceleration smoothing factor and importing it into the interpolation trajectory planning process to smooth out transient mechanical loads include: reading the viscous friction component in the friction coefficient evolution tensor and calculating the transient mechanical interference load during multi-axis linkage by combining synchronous pose deviation data; determining the acceleration constraint threshold based on the damping ratio of the machine tool bed structure; obtaining the acceleration curves of each axis by differentiating the look-ahead velocity curves, superimposing the equivalent interference acceleration components, and then truncating the extreme values using the acceleration constraint threshold; multiplying the truncated difference by the damping gain coefficient to generate the axial acceleration smoothing factor, wherein the damping gain coefficient is determined by linear mapping based on the structural damping ratio within a preset range; and importing the smoothing factor into the interpolation trajectory planning module to smooth out acceleration extreme value jumps.
[0048] The step of using an acceleration / deceleration profile reshaping device to perform a second derivative on the look-ahead velocity curve of the feed drive and outputting an axial acceleration smoothing factor includes: calculating the maximum transient excitation force threshold that the mechanical system can absorb without loss based on the structural damping ratio parameter of the machine tool bed material; using the maximum transient excitation force threshold as a dynamic constraint boundary, performing acceleration limit truncation processing on the look-ahead velocity curve of the feed drive, and outputting the axial acceleration smoothing factor that conforms to the constraint boundary to limit the sudden change in mechanical excitation force during multi-axis concurrent interpolation.
[0049] Specifically, the acceleration / deceleration profile reshaping unit reads the viscous friction component from the friction coefficient evolution tensor and uses it as the dynamic damping boundary condition of the transmission system. Combined with the current pose deviation of each axis reflected by the synchronous pose deviation data, it calculates the transient mechanical interference load vector excited within the feed screw and guide rail transmission chain during multi-axis concurrent interpolation. Simultaneously, the control system pre-calculates the maximum transient excitation force threshold that the current mechanical system can absorb without loss, based on the structural damping ratio parameters of the machine tool bed's cast iron material. Subsequently, the acceleration / deceleration profile reshaping unit performs a second time derivative on the look-ahead velocity curve currently being executed by the feed driver to obtain the acceleration curves of each axis. This acceleration curve is then vector-superimposed with the equivalent interference acceleration components obtained after inverse matrix transformation of the equivalent moving mass matrix and the equivalent rotating inertia matrix to obtain the corrected acceleration commands for each axis. Using the equivalent acceleration constraint threshold obtained after inverse transformation of the equivalent mass-inertia matrix from the maximum transient excitation force threshold as the boundary, extreme value truncation is performed on the corrected acceleration commands to ensure they do not exceed the acceleration range that the mechanical system can absorb. The difference in acceleration before and after the cutoff is inverted and multiplied by the damping gain coefficient to output the axial acceleration smoothing factor.
[0050] The damping gain coefficient ranges from [0.5, 0.8], determined as follows: a lower limit of 0.5 ensures the smoothing effect is no less than 50% of the interference acceleration, avoiding loss of smoothing significance due to excessive conservatism; an upper limit of 0.8 reserves a 20% control margin to prevent the introduction of reverse excitation in the transmission chain due to overcompensation. The specific value is determined by linear mapping within this range based on the actual calibrated value of the machine tool bed's damping ratio: a smaller value is used when the structural damping ratio is large, and a larger value is used when the structural damping ratio is small, to compensate for insufficient damping capacity of the machine tool body. Through a reserved shared memory area, this axial acceleration smoothing factor is imported as a dynamic feedforward parameter into the interpolation trajectory planning module of the CNC system to buffer and smooth extreme jumps in acceleration of each feed axis.
[0051] This invention combines viscous friction components to reshape the look-ahead velocity curve and transforms the machine tool bed damping ratio into an acceleration constraint boundary. This method controls the nonlinear interference load within the physical range that the machine body can absorb. It eliminates extreme jumps in axial acceleration from the interpolation pre-stage, effectively suppressing excess excitation energy through the damping network, preventing hardware resonance caused by sudden error changes, and ensuring the continuity of the control trajectory and the stability of the closed-loop system.
[0052] Further, the steps of reconstructing the communication message, injecting it into the servo position loop superposition node of the CNC system, and controlling the servo motor output compensation displacement include: performing spatial decoupling calculation on the synchronous pose deviation data, separating the attitude offset component and the position offset component, and encapsulating them into a communication message, wherein the attitude offset component corresponds to the angular displacement compensation amount of the rotary axis, and the position offset component corresponds to the linear displacement compensation amount of the linear axis; establishing a communication link with the CNC system kernel and completing clock phase alignment, locating the physical address of the position control register in the servo drive chip of each axis; within the current interpolation cycle synchronization window, according to the control timing sequence of first correcting the rotary axis attitude and then translating the linear axis position, feeding forward the pose offset component and the position offset component decoded and restored from the communication message into the corresponding servo position loop superposition node, updating the servo pulse equivalent, and executing the compensation displacement operation. The servo position loop superposition node is the external superposition entry point for the position command in the CNC system servo driver, allowing the external compensation amount to be incrementally superimposed onto the original interpolation position command. Specifically, after receiving the synchronous pose deviation data containing the axial acceleration smoothing factor, the control system performs a serial spatial decoupling calculation in the underlying kinematic dimension, first rotating and then straightening. The specific decoupling process is as follows: the control system first extracts the fourth-order homogeneous transformation matrix from the synchronous pose deviation data and separates the rotation sub-matrix in the upper left corner of the matrix; by calling the inverse trigonometric function library of the corresponding machine tool kinematic topology, the deflection angle compensation value for the rotation axis (such as the A-axis and C-axis) is solved from the rotation sub-matrix and used as an independent attitude bias component.
[0053] The control system extracts the factory-set structural vector parameters of the machine tool spindle tilting head or rotary table, substitutes the calculated attitude offset components into the positive kinematic model of the tool center point, and calculates the nonlinear spatial translation linkage of the tool tip in the X, Y, and Z axes, which is inevitably caused by the execution of this rotational compensation action. The control system extracts the translation vector from the above-mentioned fourth-order homogeneous transformation matrix, performs vector subtraction on the translation vector to completely eliminate the nonlinear spatial translation linkage, and solves the remaining pure translation deviation as the position offset component for the linear axis. Through the above-mentioned low-level inverse kinematics and vector stripping, the physical-level decoupling of the mixed error data is completed. The attitude offset component and the position offset component are converted to hexadecimal format, and a 16-bit cyclic redundancy check code, timestamp tag, and frame delimiter are added and encapsulated into a communication message conforming to the industrial real-time Ethernet bus standard. Through the direct memory access controller, a communication link is established with the CNC system's operating kernel to locate the absolute memory physical address of the position control register inside the servo drive control chip of each axis. After the communication link is established, a distributed clock synchronization handshake protocol is initiated with the CNC system's kernel to complete phase alignment between the external control terminal and the underlying phase-locked loop of the CNC system. Within the CNC system's interpolation cycle synchronization window, following the control timing sequence of first driving the rotary axis to correct spatial attitude and then driving the linear axis to translate spatial position, the pose offset component and position offset component decoded from the communication message are transmitted via the external compensation channel reserved on the industrial real-time Ethernet bus. During the feedforward injection process, the control system synchronously monitors abnormal states such as sensor data stream interruption, digital twin calculation value divergence, and continuous failures of communication link cyclic redundancy check. When any abnormality is detected, the injection of compensation offset amount into the servo position loop is immediately stopped, the compensation channel output is forcibly returned to zero, and a hold mode is triggered to maintain the command value of the previous normal compensation state in the servo system until the abnormality is eliminated and the system self-test passes, after which normal injection resumes. The feedforward bias is injected into the corresponding servo position loop superposition node in the form of a standard; the servo pulse equivalent of the corresponding axis is updated using the servo position loop superposition node, which drives the stator winding of the servo motor to generate compensating electromagnetic torque and output compensating displacement, thus completing the timing decoupling and cancellation of spatial volume error.
[0054] This invention streamlines the kinematic logic of nonlinear affine transformations in five-axis linkage by separating the timing of mixed spatial volume errors—first the rotational attitude, then the linear position—and avoids secondary geometric interference in space caused by concurrent injection. Combined with direct overwriting of the underlying layer, it eliminates the communication phase delay of the surface protocol, ensuring that the dynamic response frequency of the execution end covers the cutting characteristic frequency.
[0055] Furthermore, after outputting the compensation displacement, the process also includes a parameter iteration update step for the digital twin, specifically including: obtaining the actual machining contour coordinate set after the machining equipment completes the machining cycle; calculating the difference between the actual machining contour coordinate set and the theoretically predicted contour coordinate set output by the digital twin to generate an error residual matrix; when the absolute value of the error residual matrix exceeds the allowable threshold of geometric tolerance, the friction coefficient evolution tensor is inverted and corrected using the least squares method, and fed back to the digital twin for parameter iteration reloading.
[0056] Specifically, after the metal processing equipment completes a full cutting cycle, an in-machine measurement trigger signal is activated, controlling the trigger-type probe mounted on the machine tool to perform online measurement on the machined titanium alloy impeller workpiece, acquiring an in-machine measurement coordinate data report. The measurement point cloud in this report is converted into an actual machining contour coordinate set containing 10,000 spatial coordinate points. This coordinate set is then compared with the theoretically predicted contour coordinate set output by the virtual tool tip trajectory model within the digital twin, and the difference is calculated to generate an error residual matrix. The Euclidean norm of each component of the error residual matrix is taken and compared with the geometric tolerance allowable threshold. This threshold is set proportionally according to the workpiece design contour tolerance. In this embodiment, the part design contour tolerance is eight micrometers, and 60% of this is taken as the warning threshold, i.e., 5 micrometers, which can be adjusted according to the tolerance requirements of different workpieces. When the tolerance exceeds 5 micrometers, the boundary condition optimization module is triggered, using a nonlinear least squares method to invert and correct the friction coefficient evolution tensor. Specifically, using the deviation components of each measurement point in the error residual matrix as observed values, and each element in the friction coefficient evolution tensor as parameters to be corrected, the objective function is the sum of squared residuals between the observed values and the digital twin's predicted values. The corrected parameter values that minimize the objective function are then solved. The corrected friction coefficient evolution tensor updates the boundary conditions of the stiffness matrix, damping matrix, and heat transfer matrix, and regenerates the spatial basis function matrix during tool change intervals or downtime between processing batches, completing the synchronous refresh of the reduced-order model. Using the error residual matrix as observed values and each element of the friction coefficient evolution tensor as parameters to be corrected, the solution is iteratively solved until the convergence criterion is met or the maximum number of iterations reaches fifty. When the system terminates due to reaching the maximum number of iterations, it uses the current optimal estimate as the corrected value and triggers an alarm, notifying the operator for manual confirmation. The corrected tensor is fed back to the digital twin dynamics model for parameter reloading, completing adaptive closed-loop calibration.
[0057] This invention constructs a macroscopic post-adaptive closed loop at the system level by introducing long-cycle feedback from on-machine workpiece measurement. This step establishes a long-cycle mapping and correction relationship between the final forming quality of the workpiece and the base parameters of the digital twin model, which can effectively eliminate the cumulative deviation caused by unmodeled external disturbances and ensure the iterative stability of the entire control system under long-cycle operation.
[0058] Example 2: This embodiment abstracts away the specific processing parameters and elaborates on the core technical solution and workflow of the digital twin intelligent control method in this invention from the perspective of the underlying control architecture and algorithm execution flow. This embodiment focuses on revealing the reverse decoupling of physical parameters, multi-physics collaborative calculation, and feedforward injection mechanism of underlying servo control commands under nonlinear time-varying conditions.
[0059] During equipment operation, the control system first executes adaptive heavy load under the boundary condition constraints of the digital twin dynamic model. The control system acquires multi-dimensional operating load data streams from the feed servo drive of the metal processing equipment in real time and decouples them into a feed drive torque time series and a servo following error series. Using discrete-time Fourier transform, the feed drive torque time series is converted to the frequency domain. Based on the spindle speed fundamental frequency, the cutting resistance component generated by material removal is stripped away, separating the harmonic frictional resistance components characterizing the resistance of the mechanical transmission chain. Subsequently, these harmonic frictional resistance components are substituted into the tribological derivation logic to calculate the Coulomb friction parameters and equivalent tangential stiffness parameters in the viscous-slip transition stage, thereby generating the friction coefficient evolution tensor characterizing the current physical wear state. Simultaneously, the control system extracts the phase lag angle parameter of the servo following error series in the reverse commutation gap region, establishes a spatial mapping relationship between this parameter and the normal force deformation of the mating surface, and calculates the contact stiffness attenuation rate by comparing it with the factory rated stiffness reference value.
[0060] After obtaining the aforementioned macroscopic physical degradation parameters, the control system performs matrix reconstruction across the physical domain. The discrete contact mesh node set representing the mechanical assembly interface within the digital twin is extracted. The contact stiffness attenuation rate is used as the independent variable of the stiffness degradation reduction coefficient. The local contact stiffness matrix elements are reconstructed using a nonlinear spring-damping equivalent model and assembled into the system's global stiffness matrix. Simultaneously, the increase in local thermal contact impedance is calculated based on the contact stiffness attenuation rate and used as an additional correction to the boundary thermal resistance, synchronously correcting the heat flow conduction matrix. Furthermore, the friction coefficient evolution tensor is orthogonally decomposed according to the degrees of freedom of motion, and its tangential viscous damping coefficient is reloaded into the global damping matrix. The nonlinear Coulomb force vector is merged into the external excitation load vector of the dynamic state-space equation, completing the closed-loop mapping of the underlying physical boundary.
[0061] After completing the boundary condition overload, the system initiates real-time calculation of the nonlinear thermodynamic state. The control system extracts the spindle temperature rise gradient and the effective value of the three-phase motor current in the machining contact area and inputs them into a dual-heat-source coupled Kalman observer. Using the state transition equation based on Kalman filtering and the observation matrix, prediction-update iteration is performed to dynamically remove the structural frictional heat generation component with exponential decay characteristics, accurately separating the transient cutting heat flux density with pulse step characteristics. This transient cutting heat flux density is input into the intrinsic orthogonal decomposition reduction module to truncate and reduce the high-dimensional covariance matrix of the global thermo-coupled mesh model, generating a spatial basis function matrix. For the truncated region, the contribution of the truncated mode at the tool tip boundary node is estimated based on the projection residual. A physical constraint correction term is constructed and coupled to the right-hand side of the reduced-order control equation. The spatial basis function matrix and the boundary correction term are jointly substituted into the POD reduced-order equation. The global stiffness matrix and global damping matrix are calculated using Galerkin projection offline and solved stepwise within the reduced-order dimension online, outputting a transient deformation vector that compensates for the truncation error.
[0062] The system performs kinematic fusion and timing alignment of spatial errors. Using homogeneous coordinate transformation, it performs tensor fusion product operations on the machine tool's factory geometric tolerance matrix and the dynamically output transient deformation vector to solve for the spatial pose error matrix generated by the coupling of each feed axis in three-dimensional space. To eliminate phase drift caused by communication, a timing latch scheduler is activated to listen to the clock tick of the fieldbus communication at the bottom layer of the CNC system. Upon capturing the rising edge of the interpolator reference synchronization trigger pulse, a hardware interrupt mechanism is triggered to force sampling and latching of the spatial pose error matrix, eliminating clock jitter data caused by asynchronous transmission, and outputting synchronous pose deviation data with equal time intervals to the internal interpolation cycle.
[0063] In the pre-control command issuance stage, the system consists of two parallel control links: dynamics and kinematics. In the dynamics control link, the acceleration / deceleration profile reshaper reads the viscous friction component from the friction coefficient evolution tensor and calculates the transient mechanical interference loads excited during multi-axis concurrent linkage. Based on the structural damping ratio of the machine tool bed material, the maximum transient excitation force threshold that the mechanical system can absorb without loss is calculated. Using this threshold as the dynamic constraint boundary, the look-ahead velocity curve of the feed driver is subjected to acceleration limit truncation processing, and an axial acceleration smoothing factor that conforms to the constraint boundary and is in the opposite direction is output. This factor is then incorporated into the interpolation trajectory planning process of the CNC system to smooth out mechanical resonances caused by extreme acceleration jumps.
[0064] In the kinematic control link, the system performs serial spatial decoupling on the synchronous pose deviation data, first rotating and then linearly. The control system extracts the fourth-order homogeneous transformation matrix from the synchronous pose deviation data and separates the rotation submatrix in the upper left corner of the matrix; it then calls the inverse kinematic topology function of the machine tool to solve for the attitude offset component of the rotation axis. Subsequently, this attitude offset component is substituted into the forward kinematic model of the tool center point to calculate the nonlinear spatial translation linkage caused by rotation compensation. The translation vector in the fourth-order homogeneous transformation matrix is extracted, and the nonlinear spatial translation linkage is subtracted to solve the remaining deviation as the position offset component of the linear axis.
[0065] The system converts the extracted attitude and position offset components, adds checksums, and encapsulates them into communication packets conforming to the industrial real-time Ethernet protocol standard. By locating the physical address of the position control register within each axis servo drive chip via the direct memory access controller, and after initiating the distributed clock synchronization handshake protocol to complete phase-locked loop (PLL) alignment, within the current interpolation cycle synchronization window, it strictly follows the timing sequence of first driving the rotary axis to correct the spatial attitude, then driving the linear axis to translate the spatial position. The decoded compensation values are directly overwritten to the servo position loop superposition node via an external compensation channel. This node is used to update the servo pulse equivalent, driving the servo motor to perform compensated displacement operations, thus completing the low-level sub-interpolation cycle delay cancellation of the error.
[0066] It should be noted that the industrial real-time Ethernet bus transmission cycle, the direct memory access controller addressing cycle, and the servo drive chip register write cycle are all non-zero finite values, making overall injection latency unavoidable. This solution bypasses the operating system protocol stack through the direct memory access controller to shorten the transmission path, and uses a distributed clock synchronization protocol to control the phase deviation between the compensation command sending time and the servo sampling time within the bus transmission cycle. Simultaneously, through the prediction mechanism of the timing latch scheduler, the calculation of the spatial pose error matrix is completed ahead of the current interpolation cycle, ensuring that the compensation command is ready before the servo sampling window opens in the next interpolation cycle. This effectively converts the overall latency into a deterministic delay of one interpolation cycle, which can be eliminated through a one-step advance prediction compensation in the dynamic feedforward parameters.
[0067] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various substitutions and modifications can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A method for intelligent control of metal material processing equipment based on digital twins, characterized in that, include: The operating load data of the processing equipment is obtained and the friction coefficient evolution tensor and contact stiffness attenuation rate are extracted and analyzed to update the boundary condition constraints of the digital twin dynamics model. Acquire machining sensor data, and under boundary condition constraints, separate transient cutting heat through a dual-heat-source coupled Kalman observer. Output transient deformation vector after order reduction solution. Combine geometric tolerance matrix and transient deformation vector to generate spatial pose error matrix. The spatial pose error matrix is clock-aligned with the interpolation cycle of the CNC system to generate synchronous pose deviation data. Based on the viscous friction component in the friction coefficient evolution tensor and the synchronous pose deviation data, an axial acceleration smoothing factor is generated. The axial acceleration smoothing factor is used as a dynamic feedforward parameter and imported into the interpolation trajectory planning module of the CNC system to smooth out transient mechanical loads. The synchronous pose deviation data is separated into attitude offset components and position offset components. The control timing sequence of first superimposing the rotation axis compensation amount and then superimposing the linear axis compensation amount is reconstructed into a communication message, which is injected into the servo position loop superposition node of the CNC system to control the servo motor to output the compensation displacement.
2. The intelligent control method for metal material processing equipment based on digital twins according to claim 1, characterized in that, The steps for analyzing and extracting the friction coefficient evolution tensor and contact stiffness attenuation rate from the operating load data include: The feed drive torque time series and servo following error series of the machining equipment under no-load and loaded operation conditions are obtained. The feed drive torque time series is converted to the frequency domain using discrete-time Fourier transform, and the harmonic friction resistance components of each order are separated according to the fundamental frequency of the spindle speed. The harmonic friction resistance components of each order are substituted into the LuGre friction model to extract the micro-stiffness parameters and Coulomb friction parameters, and the friction coefficient evolution tensor is generated. The phase lag angle parameters of the servo following error series in the backlash region are extracted, and a spatial mapping grid of the phase lag angle parameters and normal force deformation is established. By comparing with the factory rated stiffness reference value, the contact stiffness attenuation rate is generated.
3. The intelligent control method for metal material processing equipment based on digital twins according to claim 1, characterized in that, The steps for updating the boundary condition constraints of a digital twin dynamics model include: Extract the discrete contact mesh node set representing the mechanical assembly interface from the digital twin mesh model; use the contact stiffness attenuation rate as the independent variable of the stiffness degradation reduction coefficient, and reconstruct the local contact stiffness matrix elements corresponding to the discrete contact mesh node set using the spring-damping equivalent model, and assemble the local contact stiffness matrix elements into the global stiffness matrix of the system; decompose the friction coefficient evolution tensor into nodal tangential viscous damping coefficients and nonlinear Coulomb force vectors according to the degrees of freedom of motion; reload the nodal tangential viscous damping coefficients into the global damping matrix of the system, and use the nonlinear Coulomb force vector as the reverse mechanical resistance boundary condition, and merge it into the external excitation load vector of the dynamic state space equation to complete the synchronous update of the boundary condition constraints.
4. The intelligent control method for metal material processing equipment based on digital twins according to claim 1, characterized in that, The steps of separating transient cutting heat using a dual-heat-source coupled Kalman observer and outputting the transient deformation vector through order reduction solution include: Extract the spindle temperature rise gradient data and the effective value of the three-phase current of the spindle motor from the machining sensing data. Input the spindle temperature rise gradient data and the effective value of the three-phase current of the spindle motor into a dual-heat-source coupled Kalman observer. Use the spindle temperature rise gradient and the effective value of the three-phase current as the observation vector and the structural frictional heat generation component and the transient cutting heat flux density as the state vector in the Kalman filter equation. Extract the transient cutting heat flux density by iteratively stripping the structural frictional heat generation component. Input the transient cutting heat flux density into the intrinsic orthogonal decomposition (POD) reduction module to truncate and reduce the dimension of the thermo-mechanical coupled mesh model inside the digital twin to generate a spatial basis function matrix. Based on the projection residual of the POD reduction, construct physical constraint correction terms for the boundary nodes of the truncated region and couple them to the right-hand side of the reduced-order control equation to compensate for the truncation error. Use the spatial basis function matrix to couple the physical constraint correction terms to the right-hand side of the reduced-order control equation and perform step-by-step solution within the reduced-order dimension to output the transient deformation vector.
5. The intelligent control method for metal material processing equipment based on digital twins according to claim 1, characterized in that, The steps for generating synchronized pose deviation data include: The geometric tolerance matrix is constructed by extracting the error calibration data of each feed axis and the worktable of the machining equipment; the spatial pose error matrix is solved by tensor fusion product operation of the geometric tolerance matrix and the transient deformation vector using homogeneous coordinate transformation; the fieldbus communication clock beat of the CNC system is monitored to extract the reference synchronization trigger pulse issued by the interpolator; at the moment when the rising edge of the reference synchronization trigger pulse is captured, the spatial pose error matrix is sampled and latched, the inter-phase communication jitter data is eliminated, and the synchronous pose deviation data with the same time interval attribute as the interpolation period is output.
6. The intelligent control method for metal material processing equipment based on digital twins according to claim 1, characterized in that, The steps for generating an axial acceleration smoothing factor and importing it into the interpolation trajectory planning process to mitigate transient mechanical loads include: The viscous friction component in the friction coefficient evolution tensor is read, and the transient mechanical interference load during multi-axis linkage is calculated by combining it with synchronous pose deviation data. The acceleration constraint threshold is determined based on the damping ratio of the machine tool bed structure. The acceleration curves of each axis are obtained by differentiating the look-ahead velocity curves. After superimposing the equivalent interference acceleration components, the extreme values are truncated using the acceleration constraint threshold. The truncated difference is multiplied by the damping gain coefficient to generate an axial acceleration smoothing factor, where the damping gain coefficient is determined by linear mapping based on the structural damping ratio within a preset range. This smoothing factor is then imported into the interpolation trajectory planning module to smooth out acceleration extreme value jumps.
7. The intelligent control method for metal material processing equipment based on digital twins according to claim 1, characterized in that, The steps of reconstructing communication messages, injecting them into the servo position loop overlay node of the CNC system, and controlling the servo motor to output compensating displacement include: Spatial decoupling calculations are performed on the synchronous pose deviation data to separate the attitude offset component and the position offset component, which are then encapsulated into a communication message. A communication link with the CNC system kernel is established and clock phase alignment is completed. The physical addresses of the position control registers in the servo drive chips of each axis are located. Within the synchronization window of the current interpolation cycle, the pose offset component and the position offset component decoded from the communication message are fed forward and injected into the corresponding servo position loop superposition node according to the control timing sequence of first correcting the attitude of the rotary axis and then translating the position of the linear axis. The servo pulse equivalent is updated, and the compensation displacement operation is performed.
8. The intelligent control method for metal material processing equipment based on digital twins according to claim 1, characterized in that, After outputting the compensated displacement, the process also includes a parameter iteration update step for the digital twin, specifically including: Obtain the actual machining contour coordinate set after the machining equipment completes the machining cycle; calculate the difference between the actual machining contour coordinate set and the theoretical predicted contour coordinate set output by the digital twin to generate the error residual matrix; when the absolute value of the error residual matrix exceeds the allowable threshold of geometric tolerance, use the least squares method to invert and correct the friction coefficient evolution tensor, and feed it back to the digital twin for parameter iteration and reloading.
