CN118760022A - Processing monitoring method, device, equipment and storage medium based on digital twin - Google Patents

Abstract

The present application relates to the field of digital twin technology, and discloses a processing monitoring method, device, equipment and storage medium based on digital twins, the method comprising: performing dynamic modeling and tool wear evolution analysis on processing process parameters to obtain an initial digital twin model; performing virtual-real information fusion and parameter calibration on the initial digital twin model to obtain a target digital twin model; performing multi-scenario simulation and data generation on the target digital twin model to obtain a processing simulation data set; performing comparative analysis on the processing simulation data set with real-time physical processing data to construct a comparative error feature set; performing feature importance analysis on the comparative error feature set by superimposing sparse autoencoders using information gain to generate an intelligent monitoring indicator set; performing multi-objective optimization calculation on the intelligent monitoring indicator set to obtain a comprehensive processing decision instruction, thereby achieving continuous improvement and adaptive adjustment of the processing process, greatly improving the accuracy and reliability of processing monitoring.

Description

Processing monitoring method, device, equipment and storage medium based on digital twin

Technical Field

The present application relates to the field of digital twin technology, and in particular to a processing monitoring method, device, equipment and storage medium based on digital twin.

Background Art

Traditional machining monitoring methods rely on experience and simple data analysis, which makes it difficult to cope with complex and changing machining environments and increasing precision requirements. The emergence of digital twin technology provides a new approach to solving this problem, which can accurately simulate and predict the actual machining process in a virtual environment. However, how to effectively build a digital twin model and realize the fusion and dynamic update of virtual and real data is still a difficult problem that needs to be solved.

In addition, there are a lot of uncertain factors in the processing, such as tool wear, material property fluctuations and environmental interference, which can lead to fluctuations in processing quality and reduced efficiency. Traditional monitoring methods often find it difficult to fully consider the impact of these factors, resulting in insufficient accuracy and reliability of monitoring results. At the same time, how to extract target features from massive processing data and make intelligent processing decisions based on these features is also a major challenge facing current research.

Summary of the invention

The present application provides a processing monitoring method, device, equipment and storage medium based on digital twins, thereby realizing continuous improvement and adaptive adjustment of the processing process, greatly improving the accuracy and reliability of processing monitoring.

The first aspect of the present application provides a processing monitoring method based on digital twins, and the processing monitoring method based on digital twins includes:

Perform dynamic modeling of machining process parameters and tool wear evolution analysis to obtain the initial digital twin model;

Performing virtual-real information fusion and parameter calibration on the initial digital twin model to obtain a target digital twin model;

Performing multi-scenario simulation and data generation on the target digital twin model to obtain a processing simulation data set;

Comparing and analyzing the processing simulation data set with the real-time physical processing data to construct a comparative error feature set;

Performing feature importance analysis on the contrast error feature set by superimposing information gain and sparse autoencoder to generate an intelligent monitoring indicator set;

A multi-objective optimization calculation is performed on the intelligent monitoring index set to obtain a comprehensive processing decision instruction.

A second aspect of the present application provides a processing monitoring device based on digital twins, and the processing monitoring device based on digital twins includes:

The modeling module is used to perform dynamic modeling of machining process parameters and tool wear evolution analysis to obtain the initial digital twin model;

A calibration module, used to perform virtual-real information fusion and parameter calibration on the initial digital twin model to obtain a target digital twin model;

A simulation module, used to perform multi-scenario simulation and data generation on the target digital twin model to obtain a processing simulation data set;

A comparison module, used to compare and analyze the processing simulation data set with the real-time physical processing data, and construct a comparison error feature set;

An analysis module, used for performing feature importance analysis on the contrast error feature set by superimposing an information gain sparse autoencoder to generate an intelligent monitoring indicator set;

The calculation module is used to perform multi-objective optimization calculation on the intelligent monitoring indicator set to obtain comprehensive processing decision instructions.

The third aspect of the present application provides an electronic device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the electronic device executes the above-mentioned digital twin-based processing monitoring method.

The fourth aspect of the present application provides a computer-readable storage medium, which stores instructions. When the computer-readable storage medium is run on a computer, it enables the computer to execute the above-mentioned digital twin-based processing monitoring method.

Compared with the prior art, the present application has the following beneficial effects: the combination of dynamic modeling and tool wear evolution analysis improves the accuracy and reliability of the initial digital twin model. The application of virtual-real information fusion and parameter calibration technology enables the digital twin model to dynamically adapt to changes in the actual processing process, improving the accuracy and real-time performance of the model. The introduction of multi-scenario simulation and data generation has greatly expanded the processing data set and enhanced the model's adaptability and prediction capabilities to various processing conditions. The comparative analysis of processing simulation data and real-time physical processing data effectively identifies the virtual-real differences. The application of information gain superimposed sparse autoencoders realizes the intelligent screening and importance ranking of the contrast error feature set, improving the effectiveness and representativeness of the monitoring indicators. The introduction of multi-objective optimization calculation enables the comprehensive processing decision-making instructions to take into account multiple objectives such as processing quality, efficiency and cost at the same time, improving the scientificity and practicality of the decision. It can achieve continuous improvement and adaptive adjustment of the processing process, greatly improving the accuracy and reliability of processing monitoring.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

The structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with the technology. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantive technical significance. Any structural modification, change in proportion or adjustment of size shall still fall within the scope of the technical contents disclosed in the present invention without affecting the effects and purposes that can be achieved by the present invention.

FIG1 is a schematic flow chart of a processing monitoring method based on digital twins provided in an embodiment of the present invention;

FIG2 is a schematic block diagram of the structure of a processing monitoring device based on digital twins provided in an embodiment of the present invention.

DETAILED DESCRIPTION

The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations/steps, nor must they be executed in the order described. For example, some operations/steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

It should be further understood that the term "and/or" used in the present specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes these combinations. Referring to FIG. 1 , an embodiment of a processing monitoring method based on digital twins in an embodiment of the present application includes:

Step 100: Perform dynamic modeling and tool wear evolution analysis on machining process parameters to obtain an initial digital twin model;

It is understandable that the execution subject of the present application can be a processing monitoring device based on digital twins, or a terminal or a server, which is not limited here. The embodiment of the present application is described by taking the server as the execution subject as an example.

Specifically, the structural parameters and material properties of the machining system are analyzed by finite element analysis. By establishing a geometric model of the machining system, defining material properties and boundary conditions, and applying corresponding external forces, the system is discretized using the finite element method to obtain the stiffness matrix and mass matrix of the machining system, reflecting the deformation and mass distribution characteristics of the machining system under different states. According to the stiffness matrix and mass matrix, the machining system is modally analyzed to obtain the natural frequency and vibration mode of the system. The natural frequency represents the frequency of the natural vibration of the system without external excitation, while the vibration mode reflects the vibration mode of the system at these frequencies. The cutting force in the machining process is decomposed in the time domain to identify the characteristics of the cutting force changing with time, while the frequency domain conversion can reveal the frequency components and harmonic distribution characteristics of the cutting force. Combining the natural frequency, vibration mode and the dynamic characteristic model of the cutting force, the state space modeling of the machining system is carried out to obtain the machining dynamics equation, which describes the dynamic behavior of the machining system under different cutting conditions. At the same time, the wear mechanism of the tool material is microscopically analyzed to establish a relationship model between the wear rate and the cutting parameters. By combining experiments with simulations, the wear law of tool materials under different cutting conditions is studied, and a mathematical model is established to describe the relationship between tool wear rate and cutting parameters (such as cutting speed, feed speed and cutting depth). Based on this relationship model, the initial equation of tool wear evolution is preliminarily obtained. According to the initial equation, the processing data is regressed and the wear correction coefficient is extracted from the actual processing data to improve the accuracy of the model. The processing dynamics equation and the tool wear evolution correction coefficient are coupled to obtain the dynamic model of tool wear. It reflects the wear of the tool in the actual processing process. Based on the dynamic model, the time step integral calculation of the processing process is performed to obtain the time domain response of the processing process, which reflects the dynamic change of the processing system over time under external excitation and reveals the transient behavior of the system. The time domain response of the processing process is fast Fourier transformed to obtain the frequency domain characteristics of the processing process, revealing the main vibration modes and frequency components of the system. Combining the time domain response and frequency domain characteristics, multi-scale digital twin analysis is performed to digitally simulate and analyze the processing process from different scales and angles. The initial digital twin model is obtained, which can accurately reflect the dynamic behavior of the processing system and the evolution of tool wear.

Step 200: Perform virtual-real information fusion and parameter calibration on the initial digital twin model to obtain a target digital twin model;

Specifically, the processing system is comprehensively arranged with sensors and data is collected to obtain real-time physical processing data, which reflects the dynamic behavior of the processing system during actual operation. The real-time physical processing data is preprocessed and denoised to eliminate possible noise and outliers, and a cleaned measured data set is obtained. The initial digital twin model is subjected to parameter sensitivity analysis to determine which parameters have the greatest impact on the model output and obtain the target model parameter set. A Kalman filter is constructed based on the cleaned measured data set and the target model parameter set. The Kalman filter is a commonly used dynamic system state estimation tool. By combining the model prediction value with the measured data, the system state is iteratively optimized to obtain the optimal estimate of the model state variable. The parameters of the initial digital twin model are updated according to the optimal estimate to generate a preliminary calibration model. Residual analysis and error compensation are performed on the preliminary calibration model and the cleaned measured data set. By comparing the output of the preliminary calibration model with the cleaned measured data set, the systematic errors in the model are identified, and the model is adjusted through the error compensation strategy to obtain the compensated digital twin model. Bayesian inference is performed based on the compensated digital twin model. By using measured data to update the prior distribution of model parameters, a posterior probability model that is more in line with the actual situation is obtained. Bayesian inference allows the uncertainty of parameters to be quantified and incorporated into the model to enhance the model's expressiveness and predictive performance. According to the posterior probability model, the compensated digital twin model is probabilistically expressed and optimized. By considering the uncertainty of model parameters, the accuracy and reliability of the model are improved, and the target digital twin model is finally obtained.

Step 300: Perform multi-scenario simulation and data generation on the target digital twin model to obtain a processing simulation data set;

It should be noted that the processing parameters are discretized to construct the processing parameter space. By dividing the continuous processing parameters into several discrete intervals, the possible value range of the parameters is divided into a multi-dimensional processing parameter space. Hypercube sampling is performed on the processing parameter space to obtain a multi-scenario processing parameter combination set. Hypercube sampling is an efficient sampling method that can ensure a wide coverage of parameters in a high-dimensional space, so that subsequent simulations can represent a variety of actual processing scenarios. Orthogonal experimental design is performed on the multi-scenario processing parameter combination set to optimize these scenario parameter combinations. Orthogonal experimental design is a statistical method that can obtain as much information as possible with a small number of experiments by reasonably selecting experimental combinations to form an optimized scenario parameter matrix. Based on the optimized scenario parameter matrix, the target digital twin model is parallel calculated to generate an initial simulation result set. The initial simulation result set is statistically analyzed to construct a response surface model. The response surface model is a mathematical expression used to approximate a complex simulation model, which can quickly predict the response of the system under different parameter combinations. In this way, a fast simulation proxy model is generated, which can generate a large amount of simulation data in a short time. Based on the fast simulation proxy model, sampling is performed to generate an extended simulation data set to enrich the coverage of the simulation data. Outlier detection and analysis are performed on the extended simulation data set. By identifying and removing outliers, cleaned simulation data are obtained. Kernel density estimation is performed on the cleaned simulation data to obtain the probability distribution function of the processing process. Kernel density estimation is a non-parametric method used to estimate the probability distribution of random variables. Conditional sampling and interpolation are performed on the probability distribution function to generate a continuous simulation data stream, thereby generating more simulation data points while maintaining the overall trend of the data to fill the gaps in the original data. Time series correlation analysis is performed on the continuous simulation data stream to capture the dependencies and dynamic characteristics between each time point in the processing process to obtain a processing simulation data set.

Step 400, compare and analyze the processing simulation data set with the real-time physical processing data to construct a comparison error feature set;

Specifically, the real-time physical processing data is segmented into time series. According to the time characteristics in the processing process, the data is divided into several time windows, which correspond to the time periods in the processing simulation data set. According to the time windows, the processing simulation data set and the real-time physical processing data are synchronized in time to obtain a paired data set. Multi-dimensional feature extraction is performed on the paired data set. By extracting features in various dimensions, such as speed, acceleration, temperature, pressure, etc., a high-dimensional feature vector set is formed. The high-dimensional feature vector set is reduced in dimension. By principal component analysis or other dimensionality reduction techniques, the high-dimensional feature set is simplified into a more easily processed reduced dimension feature matrix while retaining as much original information as possible. Cluster analysis is performed on the reduced dimension feature matrix to identify the pattern categories in the data distribution. By dividing the data in the feature matrix into several categories, the inherent structure and regularity of the data are revealed. Based on the pattern categories, the processing simulation data and the real-time physical processing data are compared one by one to generate a preliminary error vector. The preliminary error vector reflects the difference between the simulation data and the physical data in each category. The preliminary error vector is regularized and normalized to obtain a standardized error feature. Dynamic time warping is performed on the standardized error features, and the data that are not completely aligned in time series are adjusted to ensure that the time points of the error series can correspond correctly, so as to obtain a time-aligned error series. Wavelet transform and multi-scale decomposition are performed on the time-aligned error series to extract multi-level information in the error series and form a multi-level error feature set. Feature space is constructed for the multi-level error feature set to integrate information at each level and obtain the final comparison error feature set.

Step 500: Perform feature importance analysis on the contrast error feature set by superimposing the information gain sparse autoencoder to generate an intelligent monitoring indicator set;

Specifically, the contrast error feature set is normalized by superimposing the sparse autoencoder with information gain to ensure the uniformity of the data scale, eliminate the dimensional differences between different features, and obtain standardized feature inputs. According to the standardized feature inputs, the information entropy and conditional entropy of each feature are calculated to obtain the initial information gain value. The information gain reflects the importance of each feature to the target variable. The feature with a higher initial information gain value has a greater impact on the system. The initial information gain values are sorted to construct a feature importance sorting table, from which the most representative features are selected to form a preliminary screening feature set. Based on the preliminary screening feature set, the first sparse autoencoder is reconstructed. The sparse autoencoder extracts the first layer of hidden feature representation by compressing and reconstructing the data. The first layer of hidden feature representation is optimized by sparse constraints. By introducing sparse constraints, the model is prompted to retain only the most important features, reduce redundancy, and obtain the first layer of compressed features. Based on the first layer of compressed features, the second sparse autoencoder is used to reconstruct features to obtain the second layer of hidden feature representation. The second layer of hidden feature representation is also optimized by sparse constraints to generate more concise and critical second layer of compressed features. The second layer of compressed features are reversely reconstructed to reconstruct the original data and calculate the reconstruction error matrix. By analyzing the reconstruction error matrix, the contribution of each original feature in the reconstruction process is evaluated and the importance score of each feature is calculated. The feature importance score is thresholded and a portion of the features with the highest scores are selected to form a target feature subset, which represents the most critical factors for system monitoring and prediction. Based on the target feature subset, combined with the comparison error feature set, the feature mapping relationship is constructed to generate an intelligent monitoring indicator set.

Step 600: Perform multi-objective optimization calculation on the intelligent monitoring index set to obtain comprehensive processing decision instructions.

Specifically, the intelligent monitoring index set is normalized. By standardizing each index to form a standardized index vector, the dimensional differences between different indicators are eliminated, so that they can be compared and optimized under the same standard. Based on the standardized index vector, the objective function set of the multi-objective optimization problem is constructed to reflect the key objectives in the machining process, such as machining quality, efficiency and cost. These objective functions constitute the optimization model and define the direction and objectives of the optimization. After the optimization model is constructed, constraints are set for it. These constraints usually include restrictions on machining quality requirements, efficiency requirements and cost control. These conditions ensure that the optimization results can not only improve a certain goal, but also take into account other important factors to form a comprehensive constrained optimization problem. The non-dominated sorting algorithm is used to solve the multi-objective problem of the constrained optimization problem. The algorithm can find a balance between multiple objectives and finally obtain a set of Pareto optimal solutions, which represent the optimal trade-offs between different objectives. The Pareto optimal solution set is comprehensively evaluated. In the comprehensive evaluation process, by analyzing the advantages and disadvantages of each solution, an optimal solution is determined, which best meets the comprehensive requirements of the machining process. Based on the optimal solution, the corresponding machining parameter combination is extracted to obtain a preliminary machining plan. Conduct a process feasibility analysis on the preliminary processing plan to determine the feasibility of the preliminary processing plan in actual operation and clarify the feasible processing parameter range. Based on the feasible processing parameter range and combined with the processing experience knowledge base, generate specific parameter adjustment suggestions to obtain a more complete optimized processing plan. Evaluate the safety and stability of the optimized processing plan to ensure that the optimized plan will not generate potential risks in actual application and can be implemented stably for a long time. Through the evaluation, a safety factor is obtained to reflect the reliability and safety of the plan. Combine the safety factor and the optimized processing plan to generate a comprehensive processing decision instruction containing specific parameter settings and operation instructions.

In the embodiment of the present application, dynamic modeling is combined with tool wear evolution analysis to improve the accuracy and reliability of the initial digital twin model. The application of virtual-real information fusion and parameter calibration technology enables the digital twin model to dynamically adapt to changes in the actual processing process, improving the accuracy and real-time performance of the model. The introduction of multi-scenario simulation and data generation has greatly expanded the processing data set and enhanced the model's adaptability and prediction capabilities to various processing conditions. The comparative analysis of processing simulation data and real-time physical processing data effectively identifies the virtual-real difference. The application of information gain superimposed sparse autoencoders realizes the intelligent screening and importance ranking of the contrast error feature set, improving the effectiveness and representativeness of the monitoring indicators. The introduction of multi-objective optimization calculation enables the comprehensive processing decision-making instructions to take into account multiple goals such as processing quality, efficiency and cost at the same time, improving the scientificity and practicality of the decision. It can achieve continuous improvement and adaptive adjustment of the processing process, greatly improving the accuracy and reliability of processing monitoring.

In a specific embodiment, the process of executing step 100 may specifically include the following steps:

Conduct finite element analysis on the structural parameters and material properties of the processing system to obtain the stiffness matrix and mass matrix of the processing system. Perform modal analysis on the processing system based on the stiffness matrix and mass matrix to obtain the natural frequency and vibration mode.

The cutting force in the machining process is decomposed in the time domain and transformed in the frequency domain to obtain the dynamic characteristic model of the cutting force. Based on the natural frequency, vibration mode and the dynamic characteristic model of the cutting force, the machining system is modeled in state space to obtain the machining dynamics equation.

Conduct microscopic analysis on the wear mechanism of tool materials, establish a relationship model between wear rate and cutting parameters, obtain the initial equation of tool wear evolution, and perform regression analysis on processing data based on the initial equation of tool wear evolution to obtain the correction coefficient of tool wear evolution;

The machining dynamics equation and the tool wear evolution correction coefficient are coupled to obtain the tool wear dynamics model, and the machining process time step integral calculation is performed based on the tool wear dynamics model to obtain the time domain response of the machining process;

The time domain response of the machining process is subjected to fast Fourier transform to obtain the frequency domain characteristics of the machining process. Multi-scale digital twin analysis is performed based on the time domain response and frequency domain characteristics of the machining process to obtain the initial digital twin model.

Specifically, the structural parameters and material properties of the processing system are analyzed by finite element analysis. By establishing a geometric model of the processing system and defining the corresponding material properties and boundary conditions, the system is discretized and a finite element model of the system is constructed. In this model, the relationship between nodes is described by the stiffness matrix and mass matrix obtained by the mechanical properties and geometric shapes of the materials. describes the force response of the system under unit displacement, and the mass matrix It reflects the inertial characteristics of the system. These matrices can be obtained by solving the following linear equations:

;

in, is the displacement vector, is the external force vector. By solving this set of equations, the deformation response of the system under different load conditions is obtained. According to the stiffness matrix and the mass matrix , perform modal analysis on the processing system, determine the natural frequency and vibration mode of the system, and reflect the behavior of the system under different natural vibration states. The core of modal analysis is to solve the following eigenvalue problem:

;

in, is the natural frequency, is the corresponding vibration mode. By solving this equation, we can get the natural frequency of the system And the corresponding vibration mode . For example, in a typical machining system, several main natural frequencies may be found, which determine the resonance of the system under specific excitation. The cutting force in the machining process is decomposed in the time domain and transformed in the frequency domain. The cutting force is an important dynamic parameter generated in the machining process. It changes continuously over time, and its frequency component has an important influence on the dynamic response of the machining system. Through time domain decomposition, the cutting force is expressed as a function that changes with time, while the frequency domain transformation can reveal the main frequency components of the cutting force. The frequency domain transformation is usually achieved through Fourier transform, which can convert the time domain signal into the frequency domain signal to obtain the spectral characteristics of the cutting force. Based on the natural frequency, vibration mode and the dynamic characteristics model of the cutting force, the state space modeling of the machining system is carried out to obtain the machining dynamics equation. The machining dynamics equation is usually expressed as a state space model in the form of:

;

in, is the time derivative of the system state vector, is the state vector, is the system matrix, is the input matrix, is the input vector. In the processing system, the state vector Can include variables such as displacement, velocity, etc., input vector represents external excitations such as cutting force. By solving the state space model, the dynamic response of the system at different time points is predicted. After the state space modeling is completed, the wear mechanism of the tool material is microscopically analyzed. Wear is an inevitable phenomenon of tools during processing. There is a complex relationship between the wear rate and cutting parameters (such as cutting speed, feed rate, cutting depth, etc.). Through experimental and theoretical analysis, a relationship model between the wear rate and these cutting parameters is established. For example, assuming the wear rate And cutting speed The following relationship exists:

;

in, is the material constant, is an exponential. Based on this model, the initial equation of tool wear evolution is obtained. The initial equation of tool wear evolution is subjected to machining data regression analysis, and the correction coefficient is extracted from the actual data. In this way, the parameters in the model are adjusted to more accurately predict the wear behavior of the tool. The machining dynamics equation and the tool wear evolution correction coefficient are coupled to obtain the dynamic model of tool wear. This model combines the dynamic response of the machining system and the evolution of tool wear, and is used to predict the wear of the tool under specific machining conditions. Based on this dynamic model, the time step integral calculation of the machining process is performed to obtain the time domain response of the machining process. The time domain response describes the change of the system over time under external excitation, and can reveal the dynamic behaviors such as vibration and deformation during the machining process. The time domain response is subjected to fast Fourier transform to obtain the frequency domain characteristics of the machining process. The frequency domain characteristics reflect the response of the system at different frequencies, which helps to identify resonance problems and other vibration characteristics that may exist in the machining process. Multi-scale digital twin analysis is performed based on the time domain response and frequency domain characteristics of the machining process. Multi-scale analysis can simultaneously consider the microscopic and macroscopic characteristics of the machining process to build a more accurate digital twin model. By combining this information, the initial digital twin model is finally obtained, reflecting the dynamic behavior of the machining system under different conditions.

In a specific embodiment, the process of executing step 200 may specifically include the following steps:

Arrange sensors and collect data on the processing system to obtain real-time physical processing data, and pre-process and filter the real-time physical processing data to obtain a cleaned measured data set;

Perform parameter sensitivity analysis on the initial digital twin model to obtain the target model parameter set, and construct a Kalman filter based on the cleaned measured data set and the target model parameter set;

Iterate the Kalman filter to obtain the optimal estimate of the model state variables, and update the parameters of the initial digital twin model based on the optimal estimate to obtain a preliminary calibration model.

Perform residual analysis and error compensation on the preliminary calibration model and the cleaned measured data set to obtain the compensated digital twin model;

Perform Bayesian inference based on the compensated digital twin model, update the prior distribution of model parameters, and obtain the posterior probability model;

According to the posterior probability model, the compensated digital twin model is probabilistically expressed and optimized to obtain the target digital twin model.

Specifically, sensors are arranged and data is collected for the processing system. The arrangement of sensors is optimized according to the specific characteristics of the processing system. For example, in the processing of a CNC machine tool, it may be necessary to install multiple sensors such as vibration sensors, temperature sensors and strain gauges on the spindle, tool, worktable and feed system. These sensors collect the physical data generated during the processing in real time, including vibration signals, temperature changes, strains, etc. The real-time physical processing data is preprocessed and denoised to remove the noise components. Commonly used denoising methods include low-pass filtering, Gaussian filtering and wavelet transform, which can effectively remove noise, retain the main features of the signal, and obtain the cleaned measured data set. Perform parameter sensitivity analysis on the initial digital twin model to determine which parameters have a significant impact on the model output. These parameters are used as the focus of subsequent optimization. By perturbing the input parameters of the model one by one and observing their impact on the output results, a set of target model parameter sets is obtained. Assume that the parameters of the initial model are ,in Representative parameters, and through sensitivity analysis, identify the set of parameters that are most sensitive to the model output , these parameters are calibrated in the Kalman filter process. A Kalman filter is constructed based on the cleaned measured data set and the target model parameter set. The Kalman filter is a recursive algorithm for state estimation. It iteratively calculates the optimal estimate of the system state by combining model predictions with actual measurement data. The state update equation of the Kalman filter can be expressed as:

;

in, It is The state estimate of the step, is the current state predicted based on the state at the previous moment. It is The observed value of the step, is the observation matrix, is the Kalman gain matrix, which is calculated based on the covariance matrix of the prediction error and the measurement error. Through multiple iterations, the Kalman filter can gradually approach the true state of the system and obtain the optimal estimate of the model state variables. The parameters of the initial digital twin model are updated according to the optimal estimate to form a preliminary calibration model. Residual analysis is performed on the preliminary calibration model and the cleaned measured data set to identify and quantify the systematic errors of the model. By analyzing the residuals, that is, the difference between the model prediction value and the measured value, the model is error compensated to obtain the compensated digital twin model. The prior distribution of the model parameters is updated through Bayesian inference. Bayesian inference updates the prior distribution by incorporating the measured data into the probability model to obtain the posterior probability model. The Bayesian formula can be expressed as:

;

in, is the posterior probability distribution, indicating that Under the condition of The probability distribution of is the likelihood function, which means that given the parameters Next, the observation data Probability of occurrence; is the prior probability distribution, which means that before there is any observation data, Initial assumptions of is the marginal likelihood, which represents the total probability of the observed data. Through Bayesian inference, the distribution of model parameters is gradually updated from prior to posterior, which is more in line with the characteristics of the actual system. According to the posterior probability model, the compensated digital twin model is probabilistically expressed and optimized. The probabilistic expression not only considers the point estimate of the parameter, but also includes the uncertainty of the parameter, making the model more flexible and robust. On this basis, the model parameters are adjusted through the optimization algorithm, and finally the target digital twin model is obtained.

In a specific embodiment, the process of executing step 300 may specifically include the following steps:

Discretize the processing parameters to obtain the processing parameter space, and perform hypercube sampling on the processing parameter space to obtain a combination set of multi-scenario processing parameters;

Orthogonal experimental design is performed on the combination set of multi-scenario processing parameters to obtain the optimized scenario parameter matrix. Based on the optimized scenario parameter matrix, the target digital twin model is parallel calculated to obtain the initial simulation result set.

Perform statistical analysis on the initial simulation result set, construct a response surface model, obtain a fast simulation proxy model, and perform sampling based on the fast simulation proxy model to obtain an extended simulation data set;

Perform outlier detection and analysis on the extended simulation data set to obtain cleaned simulation data, and perform kernel density estimation based on the cleaned simulation data to obtain the probability distribution function of the machining process;

The probability distribution function is subjected to conditional sampling and interpolation to obtain a continuous simulation data stream, and the continuous simulation data stream is subjected to time series correlation analysis to obtain a processing simulation data set.

Specifically, the machining process parameters are discretized. Each process parameter is divided into multiple discrete levels. For example, for a cutting speed parameter , divide it into several discrete points, such as , each point represents a specific cutting speed value. Through discretization operation, a discrete machining parameter space is constructed. ,in, represents the cutting speed, represents the feed rate, Represents the cutting depth. Each point in the parameter space represents a specific machining scenario. The hypercube sampling method is used to sample the machining parameter space and generate a combination set of multi-scenario machining parameters. Hypercube sampling is an efficient sampling method suitable for multi-dimensional parameter space. By dividing the discrete level of each parameter into multiple intervals and randomly selecting sample points in these intervals, a sample set covering the entire parameter space is generated. Assume there are three parameters , , Each parameter has five discrete levels, and a parameter set containing multiple processing scenario combinations is generated through hypercube sampling. , these combination sets represent possible scenarios under various machining conditions. Orthogonal experimental design is performed on the multi-scenario machining parameter combination set. Orthogonal experimental design is a statistical method used to select a representative part of experimental combinations in multi-factor, multi-level experiments, so as to obtain as much information as possible with a smaller number of experiments. Through orthogonal experimental design, the optimized scenario parameter matrix is screened out from the initial multi-scenario parameter combination set. Assuming there are three factors (such as cutting speed, feed rate and cutting depth), each factor has three levels, and an orthogonal experimental design is used to obtain a The orthogonal table of contains the optimized scenario parameter matrix. Based on the optimized scenario parameter matrix, the target digital twin model is parallelized to generate the initial simulation result set. Parallel computing can make full use of computing resources and significantly improve the simulation speed. Each scenario parameter combination is used as input and simulated through the digital twin model to generate the corresponding output results, such as cutting force, surface roughness, temperature field distribution, etc. These output results constitute the initial simulation result set. . Perform statistical analysis on the initial simulation result set to construct a response surface model. A response surface model is a mathematical model used to approximate the multidimensional functional relationship of simulation or experimental data, through which the behavior of the system under different parameter combinations can be quickly predicted. Response surface models are usually constructed using regression analysis or interpolation methods. Assume that the output result Is an input parameter , , The function can be expressed as:

;

in, is the regression coefficient to be estimated, is the error term. By Regression analysis is performed to obtain a fast simulation proxy model that can quickly predict system responses without complex simulation. Using the fast simulation proxy model, a wider range of parameter sampling is performed to generate an extended simulation data set. By further sampling the proxy model, the diversity of the data is increased to cover more possible processing scenarios. Outlier detection and analysis are performed on the extended simulation data set to identify and remove data points that deviate from the normal pattern to improve the quality and reliability of the data set. Through outlier detection, a cleaned simulation data set is obtained. Kernel density estimation is performed on the cleaned simulation data set to obtain the probability distribution function of the processing process. Kernel density estimation is a non-parametric method used to estimate the probability density function of data. Suppose there is a set of cleaned simulation data , the formula for kernel density estimation is:

;

in, is the estimated probability density function, is the kernel function, is the bandwidth parameter, is a data point. Through kernel density estimation, the probability distribution function of the machining process is obtained. The obtained probability distribution function is subjected to conditional sampling and interpolation to generate a continuous simulation data stream. Conditional sampling extracts samples from the probability distribution function under specific conditions, and interpolation is used to fill the gaps between data to generate a continuous data stream. The continuous simulation data stream is subjected to time series correlation analysis to obtain the final machining simulation data set.

In a specific embodiment, the process of executing step 400 may specifically include the following steps:

The real-time physical processing data is segmented into time series to obtain a time window corresponding to the processing simulation data set, and the processing simulation data set and the real-time physical processing data are synchronized according to the time window to obtain a paired data set;

Perform multi-dimensional feature extraction on the paired data set to obtain a high-dimensional feature vector set, and perform dimensionality reduction processing on the high-dimensional feature vector set to obtain a feature matrix after dimensionality reduction;

Perform cluster analysis on the feature matrix after dimension reduction to obtain the pattern category of data distribution, and compare the processing simulation data and real-time physical processing data category by category according to the pattern category to obtain the preliminary error vector;

Regularize and normalize the preliminary error vector to obtain a standardized error feature, and perform dynamic time warping on the standardized error feature to obtain a time-aligned error sequence;

Wavelet transform and multi-scale decomposition are performed on the time-aligned error sequence to obtain a multi-level error feature set, and feature space is constructed on the multi-level error feature set to obtain a comparative error feature set.

Specifically, the real-time physical processing data is segmented into time series. During the processing, the real-time physical processing data is usually recorded in the form of a continuous time series, covering various dynamic changes in the entire processing process. These data include information in multiple dimensions such as cutting force, temperature, vibration, etc. In order to match these data with the simulation data set, the time series is segmented to obtain time windows corresponding to the processing simulation data set, which define the start and end time of key events or stages in the processing process. The basis for segmentation can be a specific time point, a characteristic event (such as a sharp change in cutting force), or a predetermined time interval. For example, assuming that the processing process is divided into three stages: roughing, semi-finishing, and finishing, the time series can be segmented into three corresponding time windows. Time synchronization is performed on the processing simulation data set and the real-time physical processing data. The processing simulation data set is aligned with the real-time physical processing data to ensure that the two have the same time reference in the same time window, so as to obtain an accurately paired data set. Multi-dimensional feature extraction is performed on the paired data set, and meaningful features are extracted from the paired data set to describe the dynamic behavior in the processing process. The feature extraction methods include time domain analysis and frequency domain analysis. For example, the mean, variance, frequency components of cutting force, temperature rise rate, vibration amplitude, etc. can all be used as features. These extracted features form a high-dimensional feature vector set, which contains multi-dimensional and multi-type data information. The high-dimensional feature vector set is subjected to dimensionality reduction processing to reduce the number of features and simplify the model while retaining important information. Common dimensionality reduction methods include principal component analysis and linear discriminant analysis. Through dimensionality reduction, the high-dimensional feature vector set is converted into a low-dimensional feature matrix. Cluster analysis is performed on the reduced-dimensional feature matrix to identify pattern categories in the data, which represent different states or stages in the machining process. Common clustering algorithms include K-means clustering and hierarchical clustering. Through cluster analysis, the reduced-dimensional feature matrix is divided into multiple categories, each of which represents a typical machining state or data pattern. According to the pattern category, the machining simulation data and the real-time physical machining data are compared one by one to obtain a preliminary error vector. The preliminary error vector reflects the difference between the simulation data and the actual data in each pattern category. The preliminary error vector is regularized and normalized. The purpose of regularization is to control the range of the error vector to avoid the influence of some excessively large or small error values on the overall analysis. Normalization is to convert each element of the error vector to the same dimensional range, such as standardizing its value to between 0 and 1. Through regularization and normalization, standardized error characteristics are obtained. Dynamic time warping is performed on the standardized error characteristics. Dynamic time warping is an algorithm used to align two time series on the time axis, even if their time scales are different. Through dynamic time warping, the error feature sequence is aligned in time to obtain a time-aligned error sequence. The time-aligned error sequence is subjected to wavelet transform and multi-scale decomposition. Wavelet transform is a signal processing method used to analyze the local characteristics of a signal at different frequencies. Through wavelet transform, the error sequence is decomposed into multiple sub-sequences of different scales, each sub-sequence corresponding to a different frequency range. The multi-scale decomposition method can reveal implicit patterns in the error sequence, such as low-frequency trends, periodic components, and high-frequency fluctuations. Assume that the error sequence is , subsequences on different scales can be obtained through wavelet transform:

;

in, are the wavelet coefficients, is the wavelet basis function, Represents the scale. By combining the multi-scale subsequences, a multi-level error feature set is constructed. The feature space of the multi-level error feature set is constructed to obtain the final comparison error feature set.

In a specific embodiment, the process of executing step 500 may specifically include the following steps:

The contrast error feature set is normalized by superimposing the information gain with the sparse autoencoder to obtain the standardized feature input, and the information entropy and conditional entropy of each feature are calculated based on the standardized feature input to obtain the initial information gain value;

The initial information gain values are sorted, and a feature importance sorting table is constructed to obtain a preliminary screening feature set. Based on the preliminary screening feature set, the first sparse autoencoder is reconstructed to obtain the first layer of hidden feature representation;

Perform sparse constraint optimization on the first layer hidden feature representation to obtain the first layer compressed features;

According to the first layer of compressed features, the second sparse autoencoder is reconstructed to obtain the second layer of hidden feature representation, and the second layer of hidden feature representation is sparsely constrained optimized to obtain the second layer of compressed features;

The second layer compressed features are reversely reconstructed to obtain a reconstruction error matrix, and the reconstruction contribution of each original feature is calculated based on the reconstruction error matrix to obtain the feature importance score;

The feature importance scores are threshold segmented to obtain the target feature subset, and based on the target feature subset and the comparison error feature set, a feature mapping relationship is constructed to obtain the intelligent monitoring indicator set.

Specifically, the contrast error feature set is normalized by superimposing the information gain with a sparse autoencoder, and all feature data are converted to the same dimension and range. Usually, each feature value is scaled to the range of [0, 1] or [-1, 1] to obtain a standardized feature input. Based on the standardized feature input, the information entropy and conditional entropy of each feature are calculated. Information entropy is a measure of the random variable The uncertainty index is calculated by the following formula:

;

in, It is a feature Value The probability of . Conditional entropy This means that in the known features Under the condition of The uncertainty is calculated as:

;

in, yes and The probability of simultaneous occurrence, is the conditional probability. By calculation, we get the initial information gain value of each feature

;

The information gain value reflects the influence of each feature on the target variable. Importance. By sorting the initial information gain values, a feature importance ranking table is constructed, and the features are arranged according to the size of the information gain value to screen out the most representative features to form a preliminary screened feature set. The preliminary screened feature set is input into the first sparse autoencoder for feature reconstruction. A sparse autoencoder is a neural network that encodes the input data into a sparse hidden layer representation and then decodes it back to the original input. The goal of a sparse autoencoder is to minimize the reconstruction error while maintaining the sparsity of the hidden layer. The objective function of the autoencoder can be expressed as:

;

in, is the input feature vector, is the reconstructed eigenvector, is the activation value of the hidden layer, is the weight of the sparsity penalty term, is the hidden layer Regularization term. By minimizing this loss function, the sparse autoencoder can learn a low-dimensional, sparse feature representation. After completing the training of the first sparse autoencoder, the first layer of hidden feature representation is obtained. After these hidden feature representations are optimized with sparse constraints, the most important information is extracted to obtain the first layer of compressed features. The first layer of compressed features are input into the second sparse autoencoder for feature reconstruction to generate the second layer of hidden feature representation. Through similar sparse constraint optimization, the second layer of compressed features is obtained. The second layer of compressed features are reversely reconstructed to generate a reconstruction error matrix. Reconstruction error matrix is the difference between the input feature and the reconstructed feature, and is calculated as:

;

in, is the original input feature matrix, is the reconstructed feature matrix. By analyzing the reconstruction error matrix, the reconstruction contribution of each original feature is calculated. The features with higher reconstruction contribution have a greater role in the reconstruction process, and the importance scores of these features will also be higher. The feature importance scores are thresholded and a target feature subset is selected. The target feature subset and the comparison error feature set are combined to construct a feature mapping relationship and generate an intelligent monitoring indicator set.

In a specific embodiment, the process of executing step 600 may specifically include the following steps:

Normalize the intelligent monitoring indicator set to obtain a standardized indicator vector, and construct an objective function set of the multi-objective optimization problem based on the standardized indicator vector to obtain an optimization model;

The processing quality, efficiency and cost constraints are set for the optimization model to obtain the constrained optimization problem. Based on the constrained optimization problem, the non-dominated sorting algorithm is used to solve the multi-objective problem and obtain the Pareto optimal solution set.

Comprehensively evaluate the Pareto optimal solution set to obtain the optimal solution, and extract the corresponding processing parameter combination based on the optimal solution to obtain a preliminary processing plan;

Conduct process feasibility analysis on the preliminary processing plan to obtain the feasible processing parameter range, and generate specific parameter adjustment suggestions based on the feasible processing parameter range and the processing experience knowledge base to obtain the optimized processing plan;

The safety and stability of the optimized processing plan are evaluated to obtain the safety factor, and based on the safety factor and the optimized processing plan, a comprehensive processing decision instruction including specific parameter settings and operation guidance is generated.

Specifically, the intelligent monitoring indicator set is normalized to eliminate the dimensional differences between different indicators and obtain a standardized indicator vector. Construct an objective function set for multi-objective optimization problems. Multi-objective optimization problems usually involve multiple conflicting objectives, such as improving processing quality, improving efficiency, and reducing costs. Each objective function can be expressed as a function of the normalized indicator vector, for example:

;

Maximize Quality ;

Maximize Efficiency ;

These sets of objective functions constitute the optimization model. The core of the optimization model is to find one or more solutions that achieve the optimal balance between these objective functions. In order to make the optimization problem closer to the actual processing process, constraints on processing quality, efficiency and cost are set. For example, quality may require that certain indicators must meet specific standards, costs cannot exceed budget limits, and efficiency must be completed within a specified time. These constraints can be expressed as inequality constraints:

;

in, is the constraint function, is the threshold of the constraint. In order to solve the multi-objective optimization problem, the non-dominated sorting algorithm is used. The non-dominated sorting algorithm is a multi-objective optimization algorithm that identifies and retains diverse solution sets through non-dominated sorting and crowding comparison. Non-dominated sorting divides the solution set into different levels. The first-level solution is the non-dominated solution, which represents the Pareto optimal solution set. The solution process converges to a Pareto optimal solution set through iterative optimization. The Pareto optimal solution set contains solutions that cannot be further optimized on all objective functions, and these solutions achieve the best balance between different objectives. The Pareto optimal solution set is comprehensively evaluated to obtain the optimal solution. Comprehensive evaluation is performed based on specific weights or priorities. For example, if processing quality is more important than cost, a higher weight can be given to the quality objective function, and an optimal solution is finally selected. According to the optimal solution, the corresponding processing parameter combination is extracted to obtain a preliminary processing plan. The preliminary plan is subjected to a process feasibility analysis to ensure that the selected processing parameters are feasible in actual operation, which involves factors such as equipment capacity, material characteristics, and operating process. A feasible range of processing parameters is obtained through simulation or experimental verification. After determining the feasible range of processing parameters, specific parameter adjustment suggestions are generated in combination with the processing experience knowledge base. The processing experience knowledge base usually contains historical data, expert experience and process specifications. By applying this knowledge, the processing plan is optimized to make it more suitable for specific production environments and needs. For example, if historical data shows that a specific material has better processing performance within a certain temperature range, then the parameters can be adjusted based on this experience to obtain a better processing plan. The optimized processing plan is evaluated for safety and stability to ensure that the processing plan will not cause safety problems in actual applications, such as tool damage, equipment failure, etc. The stability assessment verifies the long-term feasibility of the processing plan, that is, whether the stable processing quality and efficiency can be maintained under different processing conditions. The safety of the processing plan is quantified by calculating the safety factor. The safety factor is usually defined as the ratio of the system's ultimate bearing capacity to the actual load, and the formula is:

;

After calculating the safety factor, if the safety factor is too low, the processing plan needs to be further adjusted to ensure the reliability of the plan. Based on the safety factor and the optimized processing plan, a comprehensive processing decision instruction containing specific parameter settings and operation instructions is generated. These instructions provide clear guidance for actual production operations, including specific equipment settings, process flows, operating precautions, etc. For example, the instruction may include adjusting the cutting speed within a certain temperature range during the processing process to ensure the best processing effect while avoiding excessive wear of the tool.

The above describes the processing monitoring method based on digital twin in the embodiment of the present application. The following describes the processing monitoring device 10 based on digital twin in the embodiment of the present application. Please refer to Figure 2. An embodiment of the processing monitoring device 10 based on digital twin in the embodiment of the present application includes:

A modeling module 11 is used to perform dynamic modeling of machining process parameters and tool wear evolution analysis to obtain an initial digital twin model;

The calibration module 12 is used to perform virtual and real information fusion and parameter calibration on the initial digital twin model to obtain a target digital twin model;

The simulation module 13 is used to perform multi-scenario simulation and data generation on the target digital twin model to obtain a processing simulation data set;

A comparison module 14 is used to compare and analyze the processing simulation data set with the real-time physical processing data to construct a comparison error feature set;

An analysis module 15 is used to perform feature importance analysis on the contrast error feature set by superimposing the information gain sparse autoencoder to generate an intelligent monitoring indicator set;

The calculation module 16 is used to perform multi-objective optimization calculation on the intelligent monitoring index set to obtain comprehensive processing decision instructions.

Through the synergy of the above components, the beneficial effects

The present application also provides an electronic device, which includes a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the digital twin-based processing monitoring method in the above-mentioned embodiments.

The present application also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the processing monitoring method based on digital twins.

Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable an electronic device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims (10)

1. A processing monitoring method based on digital twins, characterized in that the processing monitoring method based on digital twins includes: Perform dynamic modeling of machining process parameters and tool wear evolution analysis to obtain the initial digital twin model; Performing virtual-real information fusion and parameter calibration on the initial digital twin model to obtain a target digital twin model; Performing multi-scenario simulation and data generation on the target digital twin model to obtain a processing simulation data set; Comparing and analyzing the processing simulation data set with the real-time physical processing data to construct a comparative error feature set; Performing feature importance analysis on the contrast error feature set by superimposing information gain and sparse autoencoder to generate an intelligent monitoring indicator set; A multi-objective optimization calculation is performed on the intelligent monitoring index set to obtain a comprehensive processing decision instruction. 2. The processing monitoring method based on digital twin according to claim 1 is characterized in that the dynamic modeling of the processing process parameters and the tool wear evolution analysis are performed to obtain the initial digital twin model, including: Performing finite element analysis on structural parameters and material properties of a processing system to obtain a stiffness matrix and a mass matrix of the processing system, and performing modal analysis on the processing system based on the stiffness matrix and the mass matrix to obtain natural frequencies and vibration modes; Decomposing the cutting force in the machining process in the time domain and transforming it in the frequency domain to obtain a dynamic characteristic model of the cutting force, and performing state space modeling on the machining system according to the natural frequency, the vibration mode and the dynamic characteristic model of the cutting force to obtain a machining dynamics equation; Conduct a microscopic analysis on the wear mechanism of the tool material, establish a relationship model between the wear rate and the cutting parameters, obtain an initial equation for tool wear evolution, and perform a regression analysis on the processing data based on the initial equation for tool wear evolution to obtain a correction coefficient for tool wear evolution; The machining dynamics equation and the tool wear evolution correction coefficient are coupled to obtain a dynamics model of tool wear, and a machining process time step integral calculation is performed according to the dynamics model of tool wear to obtain a time domain response of the machining process; A fast Fourier transform is performed on the time domain response of the machining process to obtain the frequency domain characteristics of the machining process, and a multi-scale digital twin analysis is performed based on the time domain response and frequency domain characteristics of the machining process to obtain an initial digital twin model. 3. The processing monitoring method based on digital twin according to claim 2 is characterized in that the initial digital twin model is subjected to virtual-real information fusion and parameter calibration to obtain a target digital twin model, comprising: Arrange sensors and collect data on the processing system to obtain real-time physical processing data, and perform preprocessing and noise reduction filtering on the real-time physical processing data to obtain a cleaned measured data set; Performing parameter sensitivity analysis on the initial digital twin model to obtain a target model parameter set, and constructing a Kalman filter based on the cleaned measured data set and the target model parameter set; Iteratively calculating the Kalman filter to obtain an optimal estimate of a model state variable, and updating parameters of the initial digital twin model according to the optimal estimate to obtain a preliminary calibration model; Performing residual analysis and error compensation on the preliminary calibration model and the cleaned measured data set to obtain a compensated digital twin model; Performing Bayesian inference based on the compensated digital twin model, updating the prior distribution of model parameters, and obtaining a posterior probability model; According to the posterior probability model, the compensated digital twin model is probabilistically expressed and optimized to obtain a target digital twin model. 4. The processing monitoring method based on digital twin according to claim 3 is characterized in that the multi-scenario simulation and data generation of the target digital twin model to obtain a processing simulation data set includes: Discretizing the processing parameters to obtain a processing parameter space, and performing hypercube sampling on the processing parameter space to obtain a multi-scenario processing parameter combination set; Performing orthogonal experimental design on the multi-scenario processing parameter combination set to obtain an optimized scenario parameter matrix, and performing parallel calculation on the target digital twin model according to the optimized scenario parameter matrix to obtain an initial simulation result set; Performing statistical analysis on the initial simulation result set, constructing a response surface model, obtaining a fast simulation proxy model, and performing sampling based on the fast simulation proxy model to obtain an extended simulation data set; Performing outlier detection and analysis on the extended simulation data set to obtain cleaned simulation data, and performing kernel density estimation based on the cleaned simulation data to obtain a probability distribution function of the machining process; Conditional sampling and interpolation are performed on the probability distribution function to obtain a continuous simulation data stream, and time series correlation analysis is performed on the continuous simulation data stream to obtain a processing simulation data set. 5. The processing monitoring method based on digital twin according to claim 4 is characterized in that the comparison and analysis of the processing simulation data set and the real-time physical processing data to construct a comparison error feature set includes: Performing time series segmentation on the real-time physical processing data to obtain a time window corresponding to a processing simulation data set, and performing time synchronization on the processing simulation data set and the real-time physical processing data according to the time window to obtain a paired data set; Performing multi-dimensional feature extraction on the paired data set to obtain a high-dimensional feature vector set, and performing dimensionality reduction processing on the high-dimensional feature vector set to obtain a feature matrix after dimensionality reduction; Performing cluster analysis on the feature matrix after dimension reduction to obtain pattern categories of data distribution, and comparing the processing simulation data and the real-time physical processing data category by category according to the pattern categories to obtain a preliminary error vector; Regularizing and normalizing the preliminary error vector to obtain a standardized error feature, and dynamically time-warping the standardized error feature to obtain a time-aligned error sequence; Wavelet transform and multi-scale decomposition are performed on the time-aligned error sequence to obtain a multi-level error feature set, and feature space is constructed on the multi-level error feature set to obtain a comparison error feature set. 6. The processing monitoring method based on digital twins according to claim 5 is characterized in that the feature importance analysis of the contrast error feature set is performed by superimposing sparse autoencoders with information gain to generate an intelligent monitoring indicator set, including: The contrast error feature set is normalized by using information gain and sparse autoencoder to obtain a standardized feature input, and the information entropy and conditional entropy of each feature are calculated based on the standardized feature input to obtain an initial information gain value; Sorting the initial information gain values, constructing a feature importance ranking table, obtaining a preliminary screening feature set, and reconstructing the first sparse autoencoder based on the preliminary screening feature set to obtain a first layer of hidden feature representation; Performing sparse constraint optimization on the first layer of hidden feature representation to obtain a first layer of compressed features; According to the first layer of compressed features, reconstructing the features of the second sparse autoencoder to obtain a second layer of hidden feature representation, and performing sparse constraint optimization on the second layer of hidden feature representation to obtain a second layer of compressed features; Reversely reconstruct the second layer of compressed features to obtain a reconstruction error matrix, and calculate the reconstruction contribution of each original feature according to the reconstruction error matrix to obtain a feature importance score; Threshold segmentation is performed on the feature importance score to obtain a target feature subset, and a feature mapping relationship is constructed based on the target feature subset in combination with the comparison error feature set to obtain an intelligent monitoring indicator set. 7. The processing monitoring method based on digital twin according to claim 6 is characterized in that the multi-objective optimization calculation of the intelligent monitoring indicator set is performed to obtain a comprehensive processing decision instruction, including: Normalizing the intelligent monitoring indicator set to obtain a standardized indicator vector, and constructing an objective function set of a multi-objective optimization problem based on the standardized indicator vector to obtain an optimization model; Setting processing quality, efficiency and cost constraints on the optimization model to obtain a constrained optimization problem, and using a non-dominated sorting algorithm to perform multi-objective solving based on the constrained optimization problem to obtain a Pareto optimal solution set; Comprehensively evaluate the Pareto optimal solution set to obtain an optimal solution, and extract corresponding processing parameter combinations based on the optimal solution to obtain a preliminary processing plan; Conducting a process feasibility analysis on the preliminary processing plan to obtain a feasible processing parameter range, and generating specific parameter adjustment suggestions based on the feasible processing parameter range and in combination with a processing experience knowledge base to obtain an optimized processing plan; The optimized processing scheme is evaluated for safety and stability to obtain a safety factor, and a comprehensive processing decision instruction including specific parameter settings and operation guidance is generated based on the safety factor and the optimized processing scheme. 8. A processing monitoring device based on digital twins, characterized in that the processing monitoring device based on digital twins comprises: The modeling module is used to perform dynamic modeling of machining process parameters and tool wear evolution analysis to obtain the initial digital twin model; A calibration module, used to perform virtual-real information fusion and parameter calibration on the initial digital twin model to obtain a target digital twin model; A simulation module, used to perform multi-scenario simulation and data generation on the target digital twin model to obtain a processing simulation data set; A comparison module, used to compare and analyze the processing simulation data set with the real-time physical processing data, and construct a comparison error feature set; An analysis module, used for performing feature importance analysis on the contrast error feature set by superimposing an information gain sparse autoencoder to generate an intelligent monitoring indicator set; The calculation module is used to perform multi-objective optimization calculation on the intelligent monitoring indicator set to obtain comprehensive processing decision instructions. 9. An electronic device, characterized in that the electronic device comprises: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory so that the electronic device executes the digital twin-based processing monitoring method as described in any one of claims 1 to 7. 10. A computer-readable storage medium having instructions stored thereon, wherein when the instructions are executed by a processor, a processing monitoring method based on digital twins as described in any one of claims 1 to 7 is implemented.