Method for adjusting process parameters in a metal working process

Abstract

The application provides a process parameter adjustment method in a metal processing process, and belongs to the technical field of metal processing. The cross-scale digital twin model covering the macro process layer, mesoscopic organization layer and performance prediction layer is deeply coupled with the closed-loop adjustment mechanism to construct a planning, execution, perception and evolution self-loop system throughout the whole processing process. The inherent architecture of traditional offline optimization and online regulation is broken, so that the process parameter adjustment is no longer a one-time static setting or passive lag correction, but a dynamic intelligent closed loop that continuously approaches the optimal control strategy as the processing process continues. Therefore, while solving the problem of offline or online disconnection, the control effect is improved from passive compensation to proactive preview, which can simultaneously ensure the optimal matching of part size precision, microstructure uniformity and service performance.

Description

Technical Field

[0001] This invention relates to the field of metal processing technology, and more specifically to a method for adjusting process parameters during metal processing. Background Technology

[0002] In metal processing parameter adjustment technology, the common approach is to use a combination of manual experience-based trial cutting and adjustment with single-level feedback control. The drawback of this approach is that there is a disconnect between offline process optimization and dynamic actual working conditions.

[0003] Specifically, the digital twin model or process optimization scheme built in the offline stage cannot effectively map the dynamic time-varying factors such as tool wear, batch fluctuations of workpiece material, changes in cooling conditions, and thermal deformation of machine tools in actual processing, resulting in a significant decrease in the effectiveness of the optimal parameter combination obtained offline during actual execution. On the other hand, online feedback control only makes passive corrections after quality deviations occur, lacking the ability to predict and adjust based on forward-looking forecasts. Furthermore, there is severe nonlinear coupling between multi-source sensor signals, making it difficult for online control to accurately identify the real physical field state, which easily leads to adjustment lag and overshoot.

[0004] Therefore, a method for adjusting process parameters in metal processing is proposed. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a method for adjusting process parameters in metal processing. This method systematically solves the problem of disconnect between offline and online processes. By replacing traditional direct feedback adjustment with a cross-scale prospective prediction model, it achieves a leapfrog improvement in control effectiveness from reactive compensation to proactive pre-planning. This method can simultaneously ensure the optimal matching of part dimensional accuracy, microstructure uniformity, and service performance.

[0006] The technical solution adopted in this invention is as follows: A method for adjusting process parameters in metal processing includes the following steps: Step S1: Obtain the initial process parameter set and workpiece material property data of the metal processing process, and preprocess the initial process parameter set and workpiece material property data to obtain standardized initial process parameters and standardized material property data. Step S2: Construct a cross-scale digital twin model based on the standardized initial process parameters and the standardized material property data. The cross-scale digital twin model includes a macroscopic process layer sub-model, a mesoscopic structure layer sub-model, and a performance prediction layer sub-model. The macroscopic process layer sub-model is used to simulate the multiphysics field in the metal processing process to obtain physical field distribution data. The mesoscopic structure layer sub-model is used to simulate the microstructure evolution process based on the physical field distribution data to obtain microstructure characteristic data. The performance prediction layer sub-model is used to predict the service performance of the part based on the microstructure characteristic data to obtain performance prediction data. Step S3: Based on the cross-scale digital twin model, perform offline material matching pre-optimization to obtain the optimal combination of offline process parameters; Step S4: Control the metal processing equipment to perform processing operations according to the offline optimal process parameter combination, and collect multi-source sensor signals in real time during the processing; Step S5: Perform signal decoupling processing on the multi-source sensing signals to obtain multiple decoupled single-physical field feature signals; Step S6: Based on the multiple decoupled single-physics field characteristic signals, perform staged tool wear compensation to obtain the tool wear compensation amount; Step S7: Based on the multiple decoupled single-physics field feature signals, the tool wear compensation amount, and real-time working condition data, the cross-scale digital twin model is used as the prediction engine to execute model predictive control and generate online process parameter fine-tuning instructions. Step S8: According to the online process parameter fine-tuning command, a feedforward-feedback composite control strategy is adopted to generate a control drive signal, and the control drive signal is output to the actuator of the metal processing equipment to realize the online adjustment of process parameters; the feedforward-feedback composite control strategy includes: using the feedforward control quantity in the online process parameter fine-tuning command as the basic control quantity, calculating the feedback correction quantity by the deviation between the real-time collected actual process parameter value and the target value in the online process parameter fine-tuning command through a PID controller, superimposing the basic control quantity and the feedback correction quantity, and after compensating for the inertial lag phase of the actuator after synthesizing the control quantity, the control drive signal is generated; the control drive signal includes an analog frequency setpoint signal for adjusting the spindle speed, a pulse direction signal for adjusting the feed speed, and a pulse width modulation signal for adjusting the cooling conditions; Step S9: After processing is completed, obtain the actual processing result data, perform data-driven self-update on the cross-scale digital twin model based on the actual processing result data, obtain the updated cross-scale digital twin model, and associate the process parameter combination of this processing with the actual processing result data and store it in the process parameter case library.

[0007] Preferably, the construction of the cross-scale digital twin model in step S2 specifically includes: Step S2-1: Based on the material constitutive parameters and thermophysical parameters in the standardized material property data, establish a macroscopic finite element simulation model as the macroscopic process layer sub-model; the macroscopic process layer sub-model takes the standardized initial process parameters as input and outputs the distribution data of temperature field, stress field, strain field and strain rate field as the physical field distribution data. Step S2-2: Based on the temperature history curve and strain history curve in the physical field distribution data, establish a mesoscale simulation model based on cellular automata or phase field method as the mesoscale sub-model; the mesoscale sub-model takes the spatiotemporal distribution of temperature field, stress field and strain rate field in the physical field distribution data as input, and outputs the distribution data of grain size distribution, phase composition ratio and recrystallization fraction as the microstructure characteristic data. Step S2-3: Based on the microstructure characteristic data, establish a performance prediction model based on a physical constitutive model or a data-driven model as the performance prediction sub-model; the performance prediction sub-model takes the grain size distribution and phase composition ratio in the microstructure characteristic data as input, and outputs the predicted values ​​of hardness, tensile strength, yield strength and fatigue life as the performance prediction data. Step S2-4: Establish a data transfer interface between the macroscopic process layer sub-model, the mesoscopic organization layer sub-model, and the performance prediction layer sub-model to realize the bidirectional mapping transfer of physical field distribution data from the macroscopic process layer sub-model to the mesoscopic organization layer sub-model, and the unidirectional mapping transfer of microscopic organization feature data from the mesoscopic organization layer sub-model to the performance prediction layer sub-model, thereby forming the cross-scale digital twin model.

[0008] The beneficial effects of this invention are: This invention deeply couples a cross-scale digital twin model, encompassing macroscopic process, mesoscopic microstructure, and performance prediction layers, with a closed-loop control mechanism to construct a self-circulating system that integrates planning, execution, perception, and evolution throughout the entire manufacturing process. It breaks away from the traditional fragmented architecture of offline optimization and online control, transforming process parameter adjustment from a one-time static setting or passive, delayed correction into a dynamic, intelligent closed loop that autonomously evolves and continuously approaches the optimal control strategy as the manufacturing process continues and actual data accumulates. This systematically solves the problem of disconnect between offline and online control while replacing traditional direct feedback control with a cross-scale forward-looking prediction model. This achieves a leapfrog improvement in control effectiveness, moving from reactive compensation to proactive pre-emptive planning, simultaneously ensuring optimal matching of part dimensional accuracy, microstructure uniformity, and service performance. Attached Figure Description

[0009] Figure 1 This is a flowchart of the process parameter adjustment method of the present invention; Figure 2 This is a flowchart of the offline modeling and pre-optimization process of the present invention; Figure 3 This is a flowchart of the online sensing and control process of the present invention; Figure 4 This is a flowchart of the composite control drive execution of the present invention. Detailed Implementation

[0010] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0011] Example 1 like Figures 1-4 As shown, a method for adjusting process parameters in a metal processing procedure according to an embodiment of the present invention is implemented according to the following process.

[0012] Before machining begins, the CNC system or host computer acquires the initial set of process parameters and workpiece material property data input by the user. The initial set of process parameters includes at least adjustable parameters such as cutting speed, feed rate, depth of cut, coolant flow rate, and coolant temperature; the workpiece material property data includes at least physical and mechanical parameters such as material grade, chemical composition, initial grain size, thermal conductivity, specific heat capacity, density, elastic modulus, Poisson's ratio, yield strength, and work hardening index.

[0013] The obtained raw data were preprocessed. The preprocessing included: filling missing values ​​using statistical imputation methods (such as mean imputation or regression imputation); detecting and removing outliers using the Laida criterion (3σ criterion) or box plot method; and mapping parameters of each dimension to the same order of magnitude using min-max normalization or Z-score normalization methods to eliminate the influence of dimensions, resulting in standardized initial process parameters and standardized material property data. This preprocessing process ensured the numerical stability and convergence efficiency of subsequent cross-scale modeling and optimization calculations.

[0014] Based on the constitutive parameters (such as the corresponding coefficients of the Johnson-Cook constitutive model or the Zerilli-Armstrong constitutive model) and thermophysical parameters in the standardized material property data, a macroscopic thermo-mechanical coupled finite element simulation model is established in the finite element analysis platform as a macroscopic process layer sub-model. This sub-model uses standardized initial process parameters as input boundary conditions and outputs the spatiotemporal distribution data of temperature field, stress field, strain field, and strain rate field within the processing area by solving the governing equations of mass conservation, momentum conservation, and energy conservation.

[0015] Based on the temperature history curve and equivalent strain history curve in the physical field distribution data, a mesoscale simulation model based on cellular automata or phase field method is established as a sub-model of the mesoscale structure. This sub-model uses the spatiotemporal distribution of temperature field, stress field, and strain rate field in the physical field distribution data as driving conditions to simulate the microstructure evolution process such as dynamic recrystallization, grain growth, and phase transformation under the thermo-mechanical coupling effect of processing, and outputs the distribution data of grain size distribution, phase composition ratio, and recrystallization fraction.

[0016] Based on microstructure characteristic data, a performance prediction model is established using a physical constitutive model (such as Hall-Petch relation, Bailey-Hirsch model) or a data-driven model (such as artificial neural network, support vector regression, or Gaussian process regression) as a sub-model for performance prediction. This sub-model takes the grain size distribution and phase composition ratio in the microstructure characteristic data as input features and outputs predicted values ​​of part service performance, including hardness, tensile strength, yield strength, and fatigue life.

[0017] Establish a data transfer interface between the three sub-models. The data transfer interface includes a spatial interpolation mapping module and a time series alignment module, which are used to realize the bidirectional mapping transfer of physical field distribution data from the macroscopic process layer sub-model to the mesoscopic organization layer sub-model, and the unidirectional mapping transfer of microscopic organization feature data from the mesoscopic organization layer sub-model to the performance prediction layer sub-model, and finally integrate them to form a complete cross-scale digital twin model.

[0018] Based on the material grade and type information in the standardized material property data, a similarity-based search is performed from a pre-stored process parameter case library to retrieve historical process parameter records that match the material grade. If a matching record is found, a candidate process parameter range is determined based on the parameter value distribution in the historical process parameter records (e.g., the range from the minimum to the maximum value of each type of parameter in the historical records). If no matching record is found, a candidate process parameter range is determined based on the empirical parameter range recommended for this material type in the general material processing manual.

[0019] Within the candidate process parameter range, a cross-scale digital twin model is used as the fitness evaluation function. The optimization objective is to minimize the weighted comprehensive deviation between the performance prediction data output by the performance prediction layer sub-model and the preset target performance index. Simultaneously considering processing efficiency and cost constraints, a multi-objective optimization algorithm (such as NSGA-II, MOEA / D, or a decomposition-based multi-objective evolutionary algorithm) is employed to perform a global iterative search in the parameter space, generating a Pareto optimal solution set. Finally, based on the current processing equipment's capacity constraints (such as spindle power limit, maximum cutting force limit, etc.) and quality priority configuration, an optimal solution is selected from the Pareto optimal solution set as the offline optimal process parameter combination output.

[0020] The optimal offline process parameters are distributed to each execution unit of the metal processing equipment via the CNC system control bus, initiating the processing operation. During processing, multi-source sensor signals are simultaneously acquired by various types of sensors pre-deployed on the machine tool spindle, tool post, worktable, and cooling system. These sensors include at least force gauges, accelerometers, thermocouples, acoustic emission sensors, and power sensors. The sampling frequency is pre-set based on the processing speed and signal characteristics to ensure complete capture of dynamically changing information.

[0021] Each signal in the acquired multi-source sensor signals is preprocessed by analog bandpass filtering and programmable amplification to suppress power frequency interference and high-frequency noise, thus obtaining the preprocessed multi-source sensor signals.

[0022] The preprocessed multi-source sensor signals are subjected to time-frequency analysis (such as short-time Fourier transform or Wigner-Ville distribution), cross-correlation analysis between channels, and partial coherence analysis in sequence to quantitatively identify the common frequency components and coupling strength coefficients of each frequency component generated by physical field coupling between the sensor signals. The above analysis can distinguish which spectral components belong to the independent contribution of a single physical field and which belong to the contribution of cross-coupling of multiple physical fields.

[0023] Based on the identified coupling frequency components and coupling strength, a decoupling mapping matrix for signal spatial projection is constructed. This decoupling mapping matrix is ​​constructed based on blind source separation principles such as independent component analysis, principal component analysis, or sparse component analysis. By multiplying the observation vector composed of preprocessed multi-source sensor signals with the decoupling mapping matrix, the coupled observation signals of the superimposed physical fields are separated into multiple independent single-physical field characteristic signal components, resulting in multiple decoupled single-physical field characteristic signals. This simplified processing flow does not require layer-by-layer iterative decomposition of the signal, has low computational complexity, and is suitable for online real-time applications.

[0024] Cutting force and vibration signals are extracted from multiple decoupled single-physics field characteristic signals, and tool wear characteristic indices are calculated based on their amplitude, spectrum, and statistical characteristics. Tool wear characteristic indices include at least the ratio of the current cutting force to the initial cutting force, the offset of characteristic frequencies in the vibration spectrum, and the rate of energy change of the acoustic emission signal within a specific frequency band.

[0025] The calculated tool wear characteristic indicators are compared with the tool wear stage thresholds determined in advance through calibration experiments to determine the current wear stage of the tool. The wear stage is divided into at least the initial wear stage, the normal wear stage, and the rapid wear stage.

[0026] Based on the determined wear stage, the corresponding compensation strategy is retrieved from the pre-stored phased compensation strategy library: if it is in the initial wear stage, a small-step incremental compensation strategy is adopted, with each compensation amount not exceeding a preset small percentage of the total compensation amount; if it is in the normal wear stage, a linear interpolation compensation strategy is adopted, with the compensation amount having a linear relationship with the wear characteristic index; if it is in the rapid wear stage, a large-step rapid compensation strategy is adopted, and a tool change warning signal is triggered simultaneously. Based on the compensation strategy and the quantified value of the current wear characteristic index, the tool wear compensation amount (including cutting speed correction, feed rate correction, and depth of cut correction) is calculated.

[0027] The current state vector of process parameters is constructed from the actual values ​​of the process parameters acquired at the current sampling moment, multiple decoupled single-physics field characteristic signals, and tool wear compensation. This state vector comprehensively reflects the dynamic thermodynamic state and tool state of the current machining process.

[0028] Using the process parameter state vector as the initial state and a cross-scale digital twin model as the internal prediction model, the process parameter state at multiple future times is recursively predicted in multiple steps within a preset prediction time domain (such as Np sampling periods), thereby obtaining the performance prediction sequence within the prediction time domain.

[0029] The objective function is to minimize the weighted sum of squared deviations between the predicted performance values ​​in the performance prediction sequence and the user-defined target performance index. This constrained quadratic or nonlinear programming problem is solved within a preset control time domain to obtain the optimal control quantity sequence. A hierarchical priority parameter tuning rule is introduced as a constraint during this solution process.

[0030] The first control variable in the optimal control variable sequence (i.e. the optimal adjustment variable that should be executed at the current moment) is taken as the online process parameter fine-tuning instruction and output to the execution stage.

[0031] The online process parameter fine-tuning commands in digital form (such as target speed adjustment value and target feed rate adjustment value) are converted into electrical signals that drive the physical actuators to perform actual actions, and the adjustment is executed with high precision and low hysteresis.

[0032] First, the industrial control computer or programmable automation controller receives online process parameter fine-tuning instructions. Based on the current actuator's base range, the controller converts the target adjustment amount in the fine-tuning instruction into the electrical signal specifications required by the corresponding actuator input port. Specifically: for spindle speed adjustment, the controller converts the speed adjustment amount into an analog frequency command signal of 0~10V or 4~20mA through the analog output channel; for feed rate and displacement adjustment, the controller converts the feed adjustment amount into a pulse direction signal containing pulse frequency changes and direction levels through the digital output channel; for cooling condition adjustment, the controller generates a pulse width modulation signal with an adjustable duty cycle through the pulse width modulation output channel.

[0033] Secondly, at the control strategy level, this invention employs a feedforward-feedback composite control strategy to balance rapid response and steady-state accuracy. Specifically, the target adjustment amount in the online process parameter fine-tuning command is directly output as the feedforward control quantity (basic control quantity). This feedforward control quantity acts directly on the actuator without waiting for feedback deviation to appear, undertaking the main coarse adjustment task. Simultaneously, photoelectric encoders or rotary transformers installed at the ends of the actuators (such as the spindle and feed axis) provide real-time feedback of the actual response value. The controller compares the actual response value with the target value in the online process parameter fine-tuning command and calculates the deviation. This deviation is input to the PID controller, and after proportional, integral, and derivative operations, the feedback correction quantity is obtained. Subsequently, the basic control quantity and the feedback correction quantity are linearly superimposed and synthesized within the controller to form the total control output. This superposition method allows the feedforward to undertake most of the basic adjustment task (e.g., 80%), while the feedback only undertakes the fine correction task (e.g., 20%), thereby eliminating static deviation while ensuring response speed.

[0034] Furthermore, considering the first- or second-order inertial hysteresis characteristics of actuators (especially spindle inverters and servo drives), directly applying the synthesized control quantity to the actuators would result in significant response lag. Therefore, this invention connects a phase lead compensator in series before outputting the synthesized control quantity. The transfer function of the phase lead compensator is Gc(s) = (1 + α·T·s) / (1 + T·s), where T is the inertial time constant of the corresponding actuator, and α is the lead coefficient. This compensator effectively counteracts the lag phase of the actuator by increasing the rate of change of the control quantity in advance, enabling the actual output to quickly follow the command. In implementation, multiple sets of different phase lead compensation parameters are configured according to the respective inertial time constants of the spindle inverter, servo drive, and coolant regulating valve to match the dynamic characteristics of each actuator. After phase lead compensation is completed, the final control drive signal is generated and output to the corresponding actuator.

[0035] Finally, after outputting the control drive signal, closed-loop verification is continuously performed to prevent integral saturation. The controller acquires the actual response value of the actuator in real time and compares it with the target value. When the response deviation exceeds the preset response deviation threshold, it is determined that the integral term of the PID controller has saturated. At this time, integral clamping correction is immediately triggered: the integral accumulation value in the PID controller is clamped to a preset positive and negative limit value. The feedback correction amount is recalculated using the clamped integral value, and then the feedback correction amount is superimposed and synthesized with the basic control amount to generate a corrected control drive signal that is re-output to the actuator. This process is repeated iteratively until the deviation between the actual response value and the target value is less than the preset convergence threshold, thereby ensuring that the adjustment process is smooth without overshoot and ultimately without steady-state error.

[0036] After all machining tasks are completed, the actual machining result data is obtained by online measuring equipment or offline testing instruments. The actual machining result data includes at least the actual dimensional accuracy data, actual surface roughness data, actual microstructure test data (such as average grain size, phase ratio), and actual service performance test data (such as hardness, strength, fatigue life).

[0037] The actual processing results are compared with the performance prediction data output by the performance prediction layer sub-model in step S2, and the prediction deviation is calculated item by item. Each calculated prediction deviation is then compared with its corresponding preset deviation threshold. If one or more prediction deviations exceed the corresponding deviation threshold, the model self-update process is triggered.

[0038] During the model self-update process, actual processing results data are used as labeled real samples to incrementally train the performance prediction sub-model. For neural network-based performance prediction models, online stochastic gradient descent or mini-batch gradient descent algorithms are used to update network weights; for models based on Gaussian process regression, their hyperparameter posterior distribution is updated. Simultaneously, based on the deviation between the actual microstructure detection data and the microstructure feature data output by the mesostructure sub-model, key physical parameters in the mesostructure sub-model (such as recrystallization kinetic coefficients and grain growth activation energy) are Bayesian-updated to correct the model's accuracy in describing the physical process. After incremental training and Bayesian updates, the updated performance prediction sub-model, the updated mesostructure sub-model, and the unchanged macroscopic process sub-model together constitute the updated cross-scale digital twin model.

[0039] Simultaneously, the initial set of process parameters, the offline optimal combination of process parameters, the time sequence of online process parameter fine-tuning instructions, the actual processing result data, and the version identifier of the updated cross-scale digital twin model involved in the entire processing process are associated and packaged to form a complete processing case data. This processing case data is then scored for quality (based on the degree of conformity between the actual results and the target indicators). When the score exceeds a preset entry threshold, a multi-level classification index is established according to material type, grade, and processing procedure type. The complete case data is stored in the process parameter case library, and the index structure of the case library is updated synchronously for subsequent offline material matching pre-optimization retrieval for new tasks.

[0040] Example 2 like Figures 1-4 As shown, the core of the process parameter adjustment method proposed in this embodiment of the invention lies in establishing a quantitative mapping of the entire chain from macroscopic process parameters to the service performance of parts, and realizing the deep coupling of offline optimization and online control based on this mapping.

[0041] First, a multi-scale digital twin model is constructed, consisting of a macroscopic process layer, a mesoscopic structure layer, and a performance prediction layer connected in series. The macroscopic process layer takes standardized initial process parameters as input and outputs physical field distribution data such as temperature, stress, and strain rate fields through thermo-mechanical coupled finite element analysis. This physical field data is then spatially interpolated and transmitted to the mesoscopic structure layer, driving the recrystallization kinetic equation-based model. Simulation of microstructure evolution, in which This refers to the volume fraction of dynamic recrystallization. For the current equivalent change, The critical strain. For peak strain, Let be the recrystallization rate constant. The Avrami index, combined with a grain growth model. ( This refers to the grain size after recrystallization. This is the initial grain size. For growth index, The growth rate constant is denoted by . For the duration of high temperature maintenance, For recrystallization activation energy, The gas constant is... (At absolute temperature), outputting microstructure characteristics such as grain size distribution and phase composition ratio; these microstructure characteristics are further transferred to the performance prediction layer based on Hall-Petch relationships. Calculate yield strength (in For lattice friction stress, The Hall-Petch slope coefficient. To extract the strengthening increment, and from this, predictive data on properties such as hardness and fatigue life are derived, thereby completing the establishment of a full-chain cross-scale model.

[0042] Based on this model, offline material matching pre-optimization is performed: candidate parameter ranges are determined by retrieving them from the case library according to the material grade, and the cross-scale model is used as the fitness function to perform multi-objective search to obtain the optimal combination of offline process parameters and then issue them for execution.

[0043] After entering the online processing stage, the multi-source sensor signals undergo decoupling processing to eliminate the mutual coupling between physical fields, i.e., the observation vector. Solving the unmixing matrix Obtain independent source signal estimation ,in It is a mixed matrix. As an independent source signal, For noise, the unmixing matrix maximizes the negative entropy of each component. Iterative solution For the first Each separated component It is a non-quadratic comparison function. (As a standard Gaussian variable), thus obtaining a pure single-physics characteristic signal for subsequent wear identification and state perception.

[0044] Based on this, the system extracts the cutting force ratio. Vibration spectrum shift and the rate of change of acoustic emission energy This constitutes a wear feature vector, which is then compared with a preset threshold. , The wear stage of the tool is determined by comparison (early, normal, or rapid), and the tool wear compensation is calculated using piecewise linear interpolation based on the stage. (such as during normal wear and tear) ,in This is the initial compensation base value. (This is the maximum compensation range), and the compensation amount, together with the current actual process parameters and the decoupled sensing characteristics, constitutes the state vector. This information is then input into the model predictive control framework.

[0045] In model predictive control, a multi-scale digital twin model is used as the predictive model for multi-step recursion. (For prediction in the time domain), and by rolling the solution of the cost function. (y) For the target performance vector, , This is the weight matrix. To control the time domain, To control the minimum value of the increment, and applying hierarchical priority constraints (hard safety constraints, high weight for quality parameters, and maximizing material removal rate for efficiency parameters after satisfying the first two), the first component of the optimal control sequence is taken. As an online fine-tuning instruction output.

[0046] Finally, to achieve high-precision response of the physical actuator, a feedforward-feedback composite control strategy is adopted with in-series phase lead compensation: the fine-tuning command is used as the feedforward quantity. Simultaneously, the actual feedback deviation is calculated using PID control to obtain the feedback correction amount. ( For proportional, integral, and differential gains, The Laplace transform of the bias, (For the Laplace operator), the two are added together to form the total control quantity. ; Regarding the first-order inertial hysteresis characteristics of the actuator ( (Inertial time constant), series phase lead compensator ( (As a lead factor) to improve the phase margin, the final output It converts the signal into analog frequency command, pulse direction signal, and pulse width modulation signal to drive the spindle inverter, servo driver, and coolant regulating valve. After output, it continuously monitors the response deviation, and performs clamping correction on the PID integral term when the deviation exceeds a threshold. ( The sampling period is (This is the integral limit value) to resist saturation until the deviation converges.

[0047] After processing, the measured results are compared with the model predictions. If the deviation exceeds the limit, the model is automatically updated. Incremental learning is then performed on the weights of the performance prediction layer neural network. ( For learning rate, For mean square error loss, (This is based on measured performance data), and Bayesian updates are performed on the physical parameters of the mesoscopic layer. ( These are the recrystallization kinetic parameters. (This is for actual measured microstructure data), and complete processing cases will be included in the database after quality assessment.

[0048] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which embodiments of the invention pertain.

[0049] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0050] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0051] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0052] Furthermore, the functional units in the various embodiments of the present invention can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0053] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for adjusting process parameters in a metal processing procedure, characterized in that, Includes the following steps: Step S1: Obtain the initial process parameter set and workpiece material property data of the metal processing process, and preprocess the initial process parameter set and workpiece material property data to obtain standardized initial process parameters and standardized material property data. Step S2: Construct a cross-scale digital twin model based on the standardized initial process parameters and the standardized material property data. The cross-scale digital twin model includes a macroscopic process layer sub-model, a mesoscopic structure layer sub-model, and a performance prediction layer sub-model. The macroscopic process layer sub-model is used to simulate the multiphysics field in the metal processing process to obtain physical field distribution data. The mesoscopic structure layer sub-model is used to simulate the microstructure evolution process based on the physical field distribution data to obtain microstructure characteristic data. The performance prediction layer sub-model is used to predict the service performance of the part based on the microstructure characteristic data to obtain performance prediction data. Step S3: Based on the cross-scale digital twin model, perform offline material matching pre-optimization to obtain the optimal combination of offline process parameters; Step S4: Control the metal processing equipment to perform processing operations according to the offline optimal process parameter combination, and collect multi-source sensor signals in real time during the processing; Step S5: Perform signal decoupling processing on the multi-source sensing signals to obtain multiple decoupled single-physical field feature signals; Step S6: Based on the multiple decoupled single-physics field characteristic signals, perform staged tool wear compensation to obtain the tool wear compensation amount; Step S7: Based on the multiple decoupled single-physics field feature signals, the tool wear compensation amount, and real-time working condition data, the cross-scale digital twin model is used as the prediction engine to execute model predictive control and generate online process parameter fine-tuning instructions. Step S8: According to the online process parameter fine-tuning instruction, a feedforward-feedback composite control strategy is used to generate a control drive signal, and the control drive signal is output to the actuator of the metal processing equipment to realize the online adjustment of the process parameters; The feedforward-feedback composite control strategy includes: using the feedforward control quantity in the online process parameter fine-tuning command as the basic control quantity; calculating the feedback correction quantity by a PID controller based on the deviation between the real-time acquired actual process parameter value and the target value in the online process parameter fine-tuning command; superimposing the basic control quantity and the feedback correction quantity; and generating the control drive signal after compensating for the inertial lag phase of the actuator after applying the composite control quantity; the control drive signal includes an analog frequency setpoint signal for adjusting the spindle speed, a pulse direction signal for adjusting the feed rate, and a pulse width modulation signal for adjusting the cooling conditions. Step S9: After processing is completed, obtain the actual processing result data, perform data-driven self-update on the cross-scale digital twin model based on the actual processing result data, obtain the updated cross-scale digital twin model, and associate the process parameter combination of this processing with the actual processing result data and store it in the process parameter case library.

2. The method for adjusting process parameters in metal processing according to claim 1, characterized in that, The construction of the cross-scale digital twin model in step S2 specifically includes: Step S2-1: Based on the material constitutive parameters and thermophysical parameters in the standardized material property data, establish a macroscopic finite element simulation model as the macroscopic process layer sub-model; the macroscopic process layer sub-model takes the standardized initial process parameters as input and outputs the distribution data of temperature field, stress field, strain field and strain rate field as the physical field distribution data. Step S2-2: Based on the temperature history curve and strain history curve in the physical field distribution data, establish a mesoscale simulation model based on cellular automata or phase field method as the mesoscale sub-model; the mesoscale sub-model takes the spatiotemporal distribution of temperature field, stress field and strain rate field in the physical field distribution data as input, and outputs the distribution data of grain size distribution, phase composition ratio and recrystallization fraction as the microstructure characteristic data. Step S2-3: Based on the microstructure characteristic data, establish a performance prediction model based on a physical constitutive model or a data-driven model as the performance prediction sub-model; the performance prediction sub-model takes the grain size distribution and phase composition ratio in the microstructure characteristic data as input, and outputs the predicted values ​​of hardness, tensile strength, yield strength and fatigue life as the performance prediction data. Step S2-4: Establish a data transfer interface between the macroscopic process layer sub-model, the mesoscopic organization layer sub-model, and the performance prediction layer sub-model to realize the bidirectional mapping transfer of physical field distribution data from the macroscopic process layer sub-model to the mesoscopic organization layer sub-model, and the unidirectional mapping transfer of microscopic organization feature data from the mesoscopic organization layer sub-model to the performance prediction layer sub-model, thereby forming the cross-scale digital twin model.

3. The method for adjusting process parameters in metal processing according to claim 1, characterized in that, The offline material matching pre-optimization performed in step S3 specifically includes: Step S3-1: Determine the material type and grade based on the standardized material property data, and retrieve historical process parameter records that match the material type and grade from the pre-stored process parameter case library; Step S3-2: If a matching historical process parameter record is found, the candidate process parameter range is determined based on the historical process parameter record; if no matching historical process parameter record is found, the candidate process parameter range is determined based on the recommended value in the material processing manual corresponding to the material type and grade. Step S3-3: Within the range of candidate process parameters, using the cross-scale digital twin model as the fitness evaluation function and minimizing the comprehensive deviation between the performance prediction data and the target performance index as the optimization objective, a multi-objective optimization algorithm is used for global search to obtain the Pareto optimal solution set. Step S3-4: Select the optimal solution that matches the current processing constraints from the Pareto optimal solution set as the offline optimal process parameter combination.

4. The method for adjusting process parameters in metal processing according to claim 1, characterized in that, Step S5 involves decoupling the multi-source sensing signals, specifically including: Step S5-1: Filter and amplify each of the multi-source sensing signals to obtain the preprocessed multi-source sensing signals; the multi-source sensing signals include at least cutting force signals, vibration signals, temperature signals, acoustic emission signals, and spindle power signals; Step S5-2: Perform time-frequency analysis, cross-correlation analysis and partial coherence analysis on the preprocessed multi-source sensor signals in sequence to identify the coupling frequency components and coupling strength between each sensor signal; Step S5-3: Construct a decoupling mapping matrix based on the coupling frequency components and the coupling strength. Use the decoupling mapping matrix to separate the coupling signals of each physical field in the preprocessed multi-source sensing signal into independent single physical field feature signals, thereby obtaining the multiple decoupled single physical field feature signals.

5. The method for adjusting process parameters in metal processing according to claim 1, characterized in that, The step S6 involves performing the staged compensation for tool wear, specifically including: Step S6-1: Extract tool wear characteristic indicators from the cutting force signal and vibration signal in the multiple decoupled single-physics field characteristic signals; the tool wear characteristic indicators include the cutting force ratio, the vibration spectrum characteristic frequency offset, and the acoustic emission signal energy change rate; Step S6-2: Compare the tool wear characteristic index with the preset tool wear stage threshold to determine the current wear stage of the tool; the wear stage includes the initial wear stage, the normal wear stage, and the rapid wear stage; Step S6-3: Based on the wear stage, select the compensation strategy corresponding to the wear stage from the pre-stored stage compensation strategy library; the compensation strategy corresponding to the initial wear stage is small step increment compensation, the compensation strategy corresponding to the normal wear stage is linear interpolation compensation, and the compensation strategy corresponding to the rapid wear stage is large step fast compensation and triggering tool change warning. Step S6-4: Calculate the tool wear compensation amount based on the compensation strategy and the tool wear characteristic index.

6. The method for adjusting process parameters in metal processing according to claim 1, characterized in that, The model prediction control performed in step S7 specifically includes: Step S7-1: Using the actual value of the process parameters at the current moment, the multiple decoupled single-physical field feature signals, and the tool wear compensation amount as inputs, construct the process parameter state vector at the current moment; Step S7-2: Using the process parameter state vector as the initial state and the cross-scale digital twin model as the prediction model, perform multi-step prediction of the future process parameter state within the preset prediction time domain to obtain the performance prediction sequence within the prediction time domain. Step S7-3: Using minimizing the weighted sum of squared deviations between the performance prediction sequence and the target performance index as the objective function, solve the optimization problem in the preset control time domain to obtain the optimal control quantity sequence in the control time domain. Step S7-4: Take the first control quantity in the optimal control quantity sequence as the online process parameter fine-tuning instruction output.

7. The method for adjusting process parameters in metal processing according to claim 6, characterized in that, The model predictive control introduces a hierarchical priority parameter tuning rule to impose constraints on the solution of the optimization problem. The hierarchical priority parameter tuning rule includes: First priority rule: Set process parameters that affect processing safety as hard constraint parameters, and the hard constraint parameters shall not exceed their corresponding safety thresholds under any circumstances; Second priority rule: Set the process parameters that affect the key indicators of the final quality of the parts as high-quality priority parameters. The adjustment priority of the high-quality priority parameters is higher than that of general quality parameters. The third priority rule is to set the process parameters that affect processing efficiency as efficiency priority parameters. Under the premise that both the hard constraint parameters and the high quality priority parameters meet the constraints, the efficiency priority parameters are optimized with the goal of maximizing processing efficiency.

8. The method for adjusting process parameters in metal processing according to claim 1, characterized in that, After outputting the control drive signal in step S8, the method further includes a step of performing closed-loop verification of the adjustment effect: The actual response value of the actuator is acquired in real time, and the actual response value is compared with the target value in the online process parameter fine-tuning instruction to calculate the response deviation; When the response deviation exceeds a preset response deviation threshold, the integral term of the PID controller is clamped and corrected, and the feedback correction after clamping and correcting is superimposed and synthesized with the basic control quantity to generate a corrected control drive signal and output it to the actuator again until the deviation between the actual response value and the target value is less than the preset convergence threshold.

9. The method for adjusting process parameters in metal processing according to claim 1, characterized in that, Step S9, which involves data-driven self-updating of the cross-scale digital twin model based on the actual processing result data, specifically includes: Step S9-1: Obtain the actual processing result data; the actual processing result data includes actual dimensional accuracy data, actual surface roughness data, actual microstructure detection data, and actual service performance test data; Step S9-2: Compare the actual processing result data with the performance prediction data, and calculate the prediction deviation; Step S9-3: When the prediction deviation exceeds the preset deviation threshold, the actual processing result data is used as the labeled sample to perform incremental learning training on the performance prediction layer sub-model in the cross-scale digital twin model, and the model parameters of the mesoscopic tissue layer sub-model are updated using Bayesian method to obtain the updated cross-scale digital twin model.

10. The method for adjusting process parameters in a metal processing procedure according to any one of claims 1-9, characterized in that, The actuator in step S8 includes a spindle frequency converter, a servo driver, and a coolant regulating valve; the analog frequency command signal is output to the spindle frequency converter to adjust the spindle rotation speed, the pulse direction signal is output to the servo driver to adjust the motion speed and displacement of the feed axis, and the pulse width modulation signal is output to the coolant regulating valve to adjust the coolant flow rate; Furthermore, in the feedforward-feedback composite control strategy, when performing the mechanism inertial lag phase advance compensation on the synthesized control quantity, the corresponding phase advance compensation parameters are configured according to the respective inertial time constants of the spindle frequency converter, the servo driver, and the coolant regulating valve.

Images