Method, system and device for verifying consistency of synthetic data with physical laws, and medium

Abstract

The method, system, device and medium for verifying the consistency of synthetic data with physical laws, the method comprising: obtaining real data in a target equipment manufacturing process, verifying based on a pre-constructed physical rule library and a parameter library corresponding to the physical rule library, processing data that passes the verification as negative samples and labeling; obtaining synthetic data of the target equipment, screening samples that violate the physical rules in the physical rule library from the synthetic data, processing as positive samples and labeling; converting the physical rules in the physical rule library and the parameters in the parameter library into multi-dimensional feature vectors, training an SVM model using the negative samples and the positive samples; the SVM model is used for consistency verification to determine whether the synthetic data to be verified conforms to the physical law; the synthetic data to be verified is pre-processed and converted into a multi-dimensional feature vector, which is input into the trained SVM model to determine whether the synthetic data conforms to the physical law and output the verification result. The present application ensures the effectiveness and reliability of synthetic data in high-end equipment manufacturing.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method, system, device, and medium for verifying the consistency between synthetic data and physical laws. Background Technology

[0002] In high-end equipment manufacturing, data-driven approaches are playing an increasingly crucial role in product design, production optimization, and quality control. While real-world data accurately reflects actual production conditions, its acquisition is costly and time-consuming, making it difficult to meet the demands of rapidly evolving manufacturing. Therefore, synthetic data has emerged. Through simulation and algorithmic generation, it can quickly supplement large amounts of data, demonstrating enormous potential in accelerating model training and expanding data diversity. For example, in the aerospace field, synthetic data can simulate the operational data of aircraft under extreme conditions, assisting engineers in performance evaluation and optimization design; in the automotive manufacturing industry, it can generate virtual test data for various components, identifying potential design flaws in advance; and in precision machine tool manufacturing, it can simulate force, heat, and vibration data during the cutting process, optimizing machining parameters to improve machining accuracy.

[0003] The core value of synthetic data lies in "fidelity," that is, its consistency with the physical laws of real data. This directly determines its usability in subsequent model training, performance simulation, and quality prediction. For example, in the generation of synthetic data for aero-engine blades, if the stress distribution data of the blades violates the laws of material mechanics, the fault prediction model trained based on this data will fail, leading to serious safety hazards. In precision bearing manufacturing, if the synthetic rolling element contact stress data does not conform to Hertzian contact theory, it will mislead the direction of process optimization and increase production costs. In the defect detection of electronic components, if the synthetic solder joint defect data does not conform to the actual physical conditions, it will not play any positive role in model training; on the contrary, it will reduce the accuracy of model recognition and greatly reduce the yield rate.

[0004] In related technologies, methods for generating synthetic data include: Generative methods based on deep learning large models (such as Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs) generate synthetic data with similar features by learning from and extracting patterns from massive amounts of data. Methods that expand and synthesize based on existing data (such as data augmentation techniques) increase the diversity of data by performing operations such as translation, rotation, scaling, and noise addition on the original data; The synthesis method based on causal reasoning generates synthetic data that meets specific conditions from the perspective of causal relationships, based on physical or business logic. And compositing methods based on professional software platforms such as UE5, which utilize their powerful graphics rendering and simulation capabilities to generate realistic composite data.

[0005] Regarding the "fidelity" of synthetic data, these methods start from the data generation stage, embedding physical models or rule constraints into the generation algorithm to generate synthetic data that naturally conforms to physical laws. For example, in simulating the motion of a mechanical system, mechanical principles such as Newton's laws of motion are incorporated into the generation algorithm, making the generated motion trajectory data mechanically plausible; in simulating heat treatment, Fourier's heat conduction equation is embedded into the generation model to constrain the plausibility of the temperature field distribution data. However, synthetic data generation methods that embed physical rules into the generation algorithm still have significant limitations: On the one hand, in complex multiphysics coupling environments, such as the internal environment of an aero-engine where multiple physical fields such as heat, force, and flow interact simultaneously, accurately embedding all relevant physical laws and solving them in real time is extremely challenging. On the other hand, even if physical rules are embedded in the generation algorithm, due to the complexity and uncertainty of actual production scenarios (such as fluctuations in raw material performance, deviations in processing technology, and changes in ambient temperature), it is still impossible to completely avoid the generated data violating physical laws under certain special circumstances (such as following pre-defined thermodynamic laws but generating data that violates other undefined physical laws). Once an oversight occurs in the generation process, data that does not conform to physical laws may still lead to serious consequences in subsequent application stages, such as product design failures, quality control failures, and production accidents. Summary of the Invention

[0006] To address the shortcomings of the existing technologies, this invention provides a method, system, device, and medium for verifying the consistency between synthetic data and physical laws, employing the following technical solution: Firstly, a method for verifying the consistency between synthetic data and physical laws is provided, including: Real data from the target equipment manufacturing process is acquired and validated based on a pre-built physical rule base and a parameter base corresponding to the physical rule base. Validated data is processed as negative samples and labeled. The physical rule base is used to store physical rules in the form of standardized mathematical formulas and / or logical constraints. The parameter base is used to store physical parameters, process parameters, and / or performance parameters related to the target equipment manufacturing and corresponding to the physical rules in the physical rule base. Acquire the synthetic data of the target equipment, filter out samples that violate the physical rules in the physical rule base, process them as positive samples and label them, and at the same time construct an error synthesis database; The physical rules in the physical rule base and the parameters in the parameter base are transformed into multi-dimensional feature vectors, and the negative and positive samples are used to train an SVM model; the SVM model is used for consistency verification to determine whether the synthetic data to be verified conforms to physical laws. The synthetic data to be verified is preprocessed and transformed into the multi-dimensional feature vector, which is then input into the trained SVM model to determine whether the synthetic data conforms to physical laws and output the verification result.

[0007] Secondly, a system for verifying the consistency between synthetic data and physical laws is provided, including: The data source module is used to acquire real data during the manufacturing process of the target equipment, and to verify the data based on a pre-built physical rule library and a parameter library corresponding to the physical rule library. Data that passes the verification is processed as negative samples and labeled. The physical rule library is used to store physical rules in the form of standardized mathematical formulas and / or logical constraints. The parameter library is used to store physical parameters, process parameters and / or performance parameters related to the manufacturing of the target equipment that correspond to the physical rules in the physical rule library. The data processing module is used to acquire the synthetic data of the target equipment, filter out samples that violate the physical rules in the physical rule base, process them as positive samples and label them, and at the same time construct an error synthesis database. The core verification module is used to transform the physical rules in the physical rule base and the parameters in the parameter base into multi-dimensional feature vectors, and to train an SVM model using the negative and positive samples; the SVM model is used for consistency verification to determine whether the synthetic data to be verified conforms to physical laws. The optimization and output module is used to preprocess the synthetic data to be verified and transform it into the multi-dimensional feature vector, input it into the trained SVM model, determine whether the synthetic data conforms to physical laws, and output the verification result.

[0008] Thirdly, an electronic device is provided, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, performs the steps of the method as described in the first aspect.

[0009] Fourthly, a readable storage medium is provided, on which a program or instructions are stored, which, when executed by a processor, implement the steps of the method as described in the first aspect.

[0010] The method for verifying the consistency between synthetic data and physical laws in this application has the following beneficial effects: This paper presents a comprehensive, efficient, and accurate method for verifying the consistency between synthetic data and physical laws. By constructing a systematic physical rule system, optimizing the verification model design, and establishing a closed-loop optimization mechanism, it ensures the effectiveness and reliability of synthetic data in various stages of high-end equipment manufacturing, avoiding various potential risks caused by inconsistencies between data and physical laws. At the same time, it improves verification efficiency, reduces verification costs, and enables rapid, batch verification of synthetic data, adapting to scenarios in high-end equipment manufacturing with large data volumes and high verification timeliness requirements. Furthermore, the modular design enhances the universality of the method, enabling it to adapt to the synthetic data verification needs of different high-end equipment manufacturing fields such as aerospace, automotive manufacturing, and precision machine tools.

[0011] The apparatus, electronic equipment, and readable storage medium corresponding to the method for verifying the consistency between synthetic data and physical laws in this application can achieve the same technical effect, and will not be described in detail here to avoid duplication. Attached Figure Description

[0012] Figure 1 A schematic flowchart illustrating a method for verifying the consistency between synthetic data and physical laws, provided in an embodiment of this application; Figure 2 A schematic flowchart illustrating another method for verifying the consistency between synthetic data and physical laws provided in this application embodiment; Figure 3 A schematic diagram of a model architecture provided for an embodiment of this application; Figure 4 A system device architecture diagram provided for an embodiment of this application; Figure 5 This is a timing diagram for verifying the consistency between synthetic data and physical laws, provided in an embodiment of this application. Detailed Implementation

[0013] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the technical solutions in the embodiments of this application are clearly described. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art are within the scope of protection of this application.

[0014] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such terms can be used interchangeably where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0015] The steps described in this application and the flowcharts in the accompanying drawings are not necessarily strictly executed according to the step numbers; the execution order of the steps can be changed. Furthermore, certain steps can be omitted, multiple steps can be combined into one step, and / or one step can be broken down into multiple steps. The following detailed description, in conjunction with the accompanying drawings and preferred embodiments, will illustrate each step in detail.

[0016] First, we will further explain the limitations of existing methods that embed physical rules into generative algorithms. Taking the generation of motion synthesis data for mechanical systems as an example, existing technical solutions typically embed Newton's second law into the generative model. (in To combine external forces, For the mass of the object, To avoid generating synthetic data with abrupt velocity changes or mismatches between force and motion state, the algorithm constrains the proportional relationship between acceleration and net external force. On the one hand, in the complex scenarios of multi-physics coupling and multi-parameter correlation in high-end equipment manufacturing, accurately embedding all relevant physical laws and solving them in real time is extremely challenging. Taking the combustion chamber of an aero-engine as an example, it simultaneously contains multi-physics coupling relationships such as fluid dynamics (Navier-Stokes equations), heat conduction (Fourier's law), and chemical reaction kinetics. These physical laws influence and constrain each other, and existing technologies cannot comprehensively and accurately embed all coupling relationships into the generation algorithm, leading to potential discrepancies between the generated synthetic data and the physical laws. The simplified form of the Navier-Stokes equations is... ( For fluid density, For fluid velocity, For time, For pressure, For dynamic viscosity, (As for volume forces), solving this equation in real time requires significant computational resources, making it difficult to embed into conventional synthetic data generation models. On the other hand, the complexity and uncertainty of real-world production scenarios present challenges: for example, when generating contact stress data for the tooth surfaces of precision gears, even embedding the Hertzian contact stress formula... ( For contact load, For contact width, Let be the radius of curvature of the tooth surfaces of the two gears. The elastic modulus of the two gear materials, (This is Poisson's ratio). However, due to uncertainties such as minor wear and tooth profile deviation on the gear tooth surface in actual production, the generated contact stress data may still exceed the reasonable range. Therefore, it is still impossible to completely avoid the generated data violating physical laws under certain special circumstances.

[0017] Based on this, this specification provides a system for verifying the consistency between synthetic data and physical laws, as well as a method, a device, a computer, and a computer-readable storage medium. The aim is to provide a comprehensive verification system and means for verifying the consistency between synthetic data and physical laws in high-end equipment manufacturing. This ensures the effectiveness and reliability of synthetic data in all aspects of high-end equipment manufacturing, avoiding various potential risks (affecting product quality and production efficiency, increasing production risks and costs) caused by data that does not conform to physical laws entering practical applications. Furthermore, the accuracy and comprehensiveness of the verification are particularly capable of covering complex scenarios involving multi-physical field coupling and multi-parameter correlation in high-end equipment manufacturing.

[0018] Please see Figure 1-5 This application provides a method for verifying the consistency between synthetic data and physical laws, including: Step S1: Obtain real data from the target equipment manufacturing process, verify it based on the pre-built physical rule base and the parameter base corresponding to the physical rule base, and process the verified data as negative samples and label them; The physical rule base is used to store physical rules in the form of standardized mathematical formulas and / or logical constraints; The parameter library is used to store physical parameters, process parameters, and / or performance parameters related to the manufacturing of target equipment and corresponding to the physical rules in the physical rule library.

[0019] The real-world data acquired in this step comes from sources including: real-time monitoring data from various sensors on the production site (such as force sensors, temperature sensors, vibration sensors, and displacement sensors), laboratory simulation test data, product performance testing data, and historical production data. This data comprehensively reflects the actual conditions that conform to physical laws and production requirements, serving as the core reference benchmark for verifying the synthesized data.

[0020] This step involves building a comprehensive physical rule base covering the core physical domains involved in high-end equipment manufacturing. Rules for each domain are stored in the form of standardized mathematical formulas or logical constraints to ensure the accuracy and operability of the verification. Specifically, this includes: In the field of mechanics: Incorporating Newton's three laws of motion ( Hooke's Law ,in The stiffness coefficient, For deformation variables), Hertzian contact stress formula ( ), Bending normal stress formula ( ,in For bending moment, Let be the distance from a point on the cross section to the neutral axis. (e.g., the moment of inertia of the cross section about the neutral axis); In the field of thermodynamics: Fourier's law of heat conduction is incorporated ( ,in For heat flux density, Thermal conductivity, For temperature), heat balance equation ( )wait; In the field of fluid mechanics: incorporating the simplified form of the Navier-Stokes equations ( Bernoulli's equation ,in It is the acceleration due to gravity. For height, (as constants), etc. In the field of optics: incorporating the law of refraction of light ( ,in Let be the refractive index of the two media. (for the angle of incidence and the angle of refraction), the law of reflection ( ,in Angle of incidence (e.g., reflection angle).

[0021] The established parameter library collects parameters closely related to high-end equipment manufacturing, including: Physical (material) parameters, such as elastic modulus, Poisson's ratio, thermal conductivity, density, etc.; Process parameters, such as machining temperature, cutting speed, feed rate, welding current, etc. Performance parameters, such as component dimensional tolerances, surface roughness, fatigue strength, load-bearing capacity, etc. In practice, the parameter library supports dynamic updates, allowing for the addition or modification of relevant parameters based on the needs of different high-end equipment manufacturing scenarios (such as aerospace and precision machine tools), thereby improving the method's versatility. Similarly, the physical rule library can be dynamically updated with rules as needed.

[0022] Furthermore, the verification includes: for each set of real data, the corresponding parameters in the parameter library... The actual value must satisfy the parameter verification formula: in, For the minimum value of the parameter, This represents the maximum value of the parameter.

[0023] A parameter validation formula is introduced to validate any parameter in the parameter library. Parameters that exceed the range will be marked as abnormal and require manual validation to ensure the accuracy of the rules and parameters.

[0024] In practice, the collected real data undergoes standardized preprocessing: First, remove outliers, using... Criteria, that is, for a set of data If data points satisfy ,in The mean of the data. If the value is less than the standard deviation, it is considered an outlier and removed. Then, data normalization is performed, mapping the data to the [0,1] interval. The normalization formula is: in, The original data, The minimum value of the data. The maximum value of the data; Finally, data alignment is performed to ensure that real data from different sources and dimensions have a unified timestamp and data format.

[0025] After preprocessing, the real data is labeled and explicitly marked as negative samples (label 0), representing samples that do not have problems in terms of physical laws and production requirements, providing a reliable reference benchmark for subsequent model training.

[0026] During the annotation process, each set of real data is verified by combining the rules in the physical rule base and the parameters in the parameter base. Only after confirming that it conforms to the relevant physical laws and parameter ranges is the annotation completed to ensure the accuracy of negative samples.

[0027] Step S2: Obtain the synthetic data of the target equipment, filter out samples that violate the physical rules in the physical rule base, process them as positive samples and label them, and construct an error synthesis database.

[0028] The synthetic data obtained in this step is generated by existing synthetic data generation methods (GAN, VAE, data augmentation, UE5 simulation, etc.). Samples that clearly violate physical laws are carefully screened, processed as positive samples, and labeled. Simultaneously, an error synthesis database is constructed. The screening criteria are based on core rules in the physics rule base, for example: In the field of mechanics: when simulating the motion of an object, sudden changes in velocity occur (contrary to Newton's first law), and there is a mismatch between the force and acceleration (contrary to...). Samples with contact stress exceeding a reasonable range (not conforming to the Hertzian contact stress formula); In the field of thermal science: when simulating heat conduction processes, temperature distributions may not conform to Fourier's law, and thermal equilibrium may not be satisfied. (sample); In the field of fluid mechanics: when simulating fluid flow, samples appear where the relationship between flow velocity and pressure violates Bernoulli's equation, and the fluid trajectory does not conform to the Navier-Stokes equation.

[0029] These erroneous synthetic data samples undergo the same standardized preprocessing as real data, including outlier removal, normalization, and data alignment. They are then labeled in detail as positive samples (label 1) and the specific physical rules they violate are noted. This serves as an important basis for the consistency verification model (see below) to learn and identify data that does not conform to physical laws, helping the model to grasp the characteristic patterns of data that does not conform to physical laws.

[0030] Step S3: Convert the physical rules in the physical rule base and the parameters in the parameter base into multi-dimensional feature vectors, and train an SVM model using the negative and positive samples; the SVM model is used for consistency verification to determine whether the synthetic data to be verified conforms to physical laws.

[0031] This step uses the Support Vector Machine (SVM) algorithm to construct the consistency verification model. SVM is a supervised machine learning algorithm that aims to find a hyperplane that separates positive samples (i.e., erroneous synthetic data) and negative samples (i.e., real data) to the greatest extent possible in the feature space. Its core advantages lie in its ability to process high-dimensional data and its good generalization performance, making it adaptable to complex verification scenarios with multiple physical rules and parameters. Addressing the characteristics of linearly inseparable high-dimensional features in synthetic data and the presence of some operational noise in the samples, this application trains the SVM model with a Radial Basis Function (RBF) kernel function. The model training process includes two core adjustable hyperparameters: the kernel function parameters... and penalty parameters These two parameters work together to regulate the model's fitting effect and generalization ability, providing core support for subsequent model parameter optimization and accurate classification. Simultaneously, the rules in the physical rule base and the parameters in the parameter base are transformed into multi-dimensional feature vectors. The dimension of the feature vector is determined by the number of physical rules and the number of parameters, with each dimension corresponding to a constraint value of a physical rule or the value of a key parameter. Specifically, in step S3, converting the physical rules in the physical rule base and the parameters in the parameter base into multi-dimensional feature vectors includes: Suppose that the physical rule base contains There are several physical rules, and the parameter library contains them. If there are 1 parameter, then the feature vector The dimension is ,Right now ,in: For physical rule constraint features, each ( The constraint deviation value corresponding to a physical rule is calculated as follows: in, The calculated values ​​related to this physical rule in the synthetic data to be verified. These are theoretical values ​​calculated based on physical rules. The smaller the value, the more the synthesized data conforms to the physical rule; For parameter comparison features, each ( The comparison deviation value corresponding to one parameter is calculated as follows: in, The parameter values ​​in the synthetic data to be verified. The value is the standard value for this parameter in the parameter library; The smaller the value, the better the synthetic data matches the actual operating parameters.

[0032] Furthermore, the method of training the SVM model using the negative and positive samples includes: Step S301: Divide the positive and negative samples into a training set and a test set according to a set ratio; Step S302: Use the radial basis function (RBF) as the kernel function. The expression for the RBF kernel function is: in, These are kernel function parameters used to control the distribution of samples after mapping to a high-dimensional space; Step S303: Adjust the penalty parameters using a grid search method and kernel function parameters ;in Used to balance the training error and generalization ability of the model; Used to control the influence of kernel functions; The larger the value, the better the model fits the training set, and the more prone it is to overfitting. The smaller the value, the stronger the model's generalization ability, but it may underfit. The larger the value, the smaller the influence range of a single sample, and the more prone the model is to overfitting. The smaller the value, the larger the influence range of a single sample, and the stronger the model's generalization ability. Step S304: With the optimization objective of maximizing the accuracy of the test set, determine the optimal parameter combination through cross-validation; Step S305: Substitute the optimal parameter combination into the SVM model for iterative training. The iteration termination condition is: The training set loss function value is less than a preset threshold or the number of iterations reaches a preset maximum value.

[0033] For example, a specific training step is as follows: The positive and negative samples are divided into training and test sets in a 7:3 ratio to ensure a consistent ratio between the two sets and avoid model bias caused by data imbalance. Since the feature vectors of the synthetic data are high-dimensional and the positive and negative samples may be linearly inseparable, a radial basis function (RBF) is used as the kernel function. The penalty parameter and kernel function parameter are adjusted using a grid search method. The optimization objective is to maximize the accuracy of the test set, and the optimal parameter combination is determined through cross-validation. The optimal parameter combination is then substituted into the SVM model for iterative training. The iteration terminates when the loss function value of the training set is less than a preset threshold (e.g., 1000 times) or the number of iterations reaches a preset maximum value (e.g., 1000 times). At this point, the model training is complete and it has the ability to verify the consistency of the synthetic data.

[0034] In some possible implementations, accuracy, precision, recall, and F1 score are used as model performance evaluation metrics. After the model is trained, the accuracy on the test set must reach a first set threshold, and the F1 score must reach a second set threshold.

[0035] The formula for calculating accuracy (Acc) is as follows: The formula for calculating Prec (Precision Rate) is as follows: The recall rate (Rec) is calculated using the following formula: The formula for calculating the F1 score is as follows: Wherein, TP (true positive) is the number of synthetic data samples that the model correctly judged as not conforming to physical laws, TN (true negative) is the number of synthetic data samples that the model correctly judged as conforming to physical laws, FP (false positive) is the number of synthetic data samples that the model incorrectly judged as not conforming to physical laws, and FN (false negative) is the number of synthetic data samples that the model incorrectly judged as conforming to physical laws.

[0036] Preferably, after the model training is completed, the accuracy on the test set should reach more than 95%, and the F1 score should reach more than 94% to ensure the accuracy of model validation.

[0037] Step S4: After preprocessing the synthetic data to be verified, convert it into the multi-dimensional feature vector, input it into the trained SVM model, determine whether the synthetic data conforms to physical laws, and output the verification result.

[0038] Furthermore, the SVM model is also used to: calculate the symbolic distance from the feature vector of the synthetic data to the optimal classification hyperplane of the SVM, and determine whether the synthetic data conforms to physical laws based on the distance value; if not, the model automatically transfers the synthetic data and its judgment result to manual judgment for forced review, and includes the reviewed samples into the retraining set.

[0039] In this step, the synthetic data to be verified is first processed according to the above-mentioned standardized preprocessing methods (outlier removal, normalization, and data alignment). Then, according to the above-mentioned feature vector construction method, the synthetic data to be verified is transformed into a multi-dimensional feature vector and input into the trained consistency verification model.

[0040] Based on learned physical laws and sample characteristics, the SVM model calculates the distance between the feature vector to be verified and the optimal hyperplane to determine whether the synthesized data is consistent with physical laws. If the feature vector is located on the negative sample side of the hyperplane, i.e., the label is 0, the synthesized data is determined to conform to physical laws and can be used for subsequent applications. If the feature vector is located on the positive sample side of the hyperplane, i.e., the label is 1, the synthesized data is determined to not conform to physical laws, and the result of rejection is output, along with the physical rules that it may have violated (based on the physical rules corresponding to the dimension with the largest deviation value in the feature vector), providing a reference for the optimization of the synthesized data generation method.

[0041] To improve verification efficiency, the model introduces a batch verification mechanism, which can simultaneously input multiple sets of synthetic data to be verified. The model processes the data in parallel, and the verification time for a single set of data does not exceed 0.1 seconds, meeting the needs of batch generation and rapid verification of synthetic data in high-end equipment manufacturing.

[0042] In some possible implementations, the method for verifying the consistency between the synthesized data and physical laws further includes: Step S5: Randomly sample the verification results. If an error is found, correct the labeling of the incorrectly labeled samples and retrain based on the corrected samples.

[0043] In practice, the judgment results output by the model are randomly sampled, with a sampling ratio of no less than 10%, and the accuracy of the model's judgment is manually checked. During the manual judgment process, staff members verify each sampled by combining the rules in the physical rule base and the parameters in the parameter base to confirm the correctness of the model's judgment.

[0044] When model errors are detected (including FP and FN), the erroneous samples are corrected: for FP samples (erroneously judged as not conforming to physical laws), they are relabeled as negative samples (label 0), and the bias value calculation in their feature vectors is corrected; for FN samples (erroneously judged as conforming to physical laws), they are relabeled as positive samples (label 1), the physical rules they violate are identified, and the feature vectors are corrected. Simultaneously, the causes of the errors are analyzed (e.g., insufficient learning of certain specific physical rules by the model, oversights in feature vector design, untimely updates to the parameter library, etc.), and relevant information is recorded.

[0045] The corrected samples are added back to the training set, and the model is retrained using the above training method (fine-tuning model parameters without re-performing the grid search) to further optimize the model's decision-making ability and improve its accuracy and stability under complex conditions. After every 1000 sets of validation data, the manual sampling and model retraining process is repeated, forming a closed-loop optimization mechanism to ensure continuous improvement in model performance.

[0046] Based on the above solution, this application has the following outstanding substantive effects and significant progress: This invention comprehensively and effectively solves the problem of verifying the consistency between synthetic data and physical laws in the field of high-end equipment manufacturing, providing a strong guarantee for improving product quality, optimizing production processes, and reducing production risks. Specifically: This application innovatively constructs a dual-database collaborative system of "physical rule base + parameter base", which stores the physical rules (mechanics, thermodynamics, fluid mechanics, etc.) involved in high-end equipment manufacturing in the form of standardized mathematical formulas, and establishes a dynamically updated parameter base. Combined with parameter verification formulas, it ensures the accuracy of rules and parameters, providing a comprehensive and systematic theoretical basis and comparison benchmark for synthetic data verification, making the verification process more scientific and reliable.

[0047] Based on the collection and labeling of positive and negative samples from real data and erroneous synthetic data, combined with standardized data preprocessing methods, the consistency verification model can fully learn data features that conform to and do not conform to physical laws. A multi-dimensional feature vector construction method based on physical rules and parameters is designed. By calculating the deviation between synthetic data and the theoretical values ​​of physical rules and standard values ​​of parameters, abstract physical rules are transformed into quantifiable feature indicators. Combined with the high-dimensional data processing advantages of the SVM algorithm, multi-dimensional and high-precision consistency verification of synthetic data is achieved. The model's test set accuracy can reach over 95%, effectively identifying synthetic data that does not conform to physical laws in multi-physics coupling scenarios. This overcomes the limitations of existing technologies that are difficult to adapt to multi-physics coupling scenarios and effectively improves the accuracy of verification.

[0048] A closed-loop optimization mechanism of "model validation - manual sampling - sample correction - model retraining" was established. The model error judgment was corrected by manual sampling, and the corrected samples were fed back to the model for retraining. The model performance was continuously optimized by combining model performance evaluation indicators (accuracy, precision, recall, F1 score) to ensure that the model maintains high accuracy and stability when facing complex and ever-changing synthetic data and production scenarios. This solves the problem that existing technologies lack effective optimization mechanisms and it is difficult to continuously improve the validation accuracy.

[0049] By combining the SVM algorithm with the RBF kernel function, we can fully leverage the advantages of SVM in high-dimensional data processing and complex pattern recognition, making it adaptable to complex verification scenarios with multiple physical rules and parameters. Furthermore, we can optimize the penalty parameters through a grid search method. and kernel function parameters This approach balances training error and generalization ability, preventing overfitting or underfitting. A closed-loop feedback mechanism between manual evaluation and model retraining allows the model to continuously optimize itself in response to new problems and scenarios encountered in real-world applications, constantly improving its performance and adapting to the ever-changing characteristics of synthetic data and physical requirements, thus ensuring the sustainability and stability of the method. Furthermore, the introduction of batch verification and containerized deployment technology enables rapid batch verification of synthetic data, with a single data set verification time not exceeding 0.1 seconds. This is suitable for scenarios in high-end equipment manufacturing where large volumes of synthetic data and high verification timeliness requirements are present. Simultaneously, it enhances the system's scalability and maintainability, reducing system deployment and maintenance costs.

[0050] Corresponding to the above embodiments of the method for verifying the consistency between synthetic data and physical laws, this application provides a system for verifying the consistency between synthetic data and physical laws, including: The data source module is used to acquire real data during the manufacturing process of the target equipment, and to verify the data based on a pre-built physical rule library and a parameter library corresponding to the physical rule library. Data that passes the verification is processed as negative samples and labeled. The physical rule library is used to store physical rules in the form of standardized mathematical formulas and / or logical constraints. The parameter library is used to store physical parameters, process parameters and / or performance parameters related to the manufacturing of the target equipment that correspond to the physical rules in the physical rule library. The data processing module is used to acquire the synthetic data of the target equipment, filter out samples that violate the physical rules in the physical rule base, process them as positive samples and label them, and at the same time construct an error synthesis database. The core verification module is used to transform the physical rules in the physical rule base and the parameters in the parameter base into multi-dimensional feature vectors, and to train an SVM model using the negative and positive samples; the SVM model is used for consistency verification to determine whether the synthetic data to be verified conforms to physical laws. The optimization and output module is used to preprocess the synthetic data to be verified and transform it into the multi-dimensional feature vector, input it into the trained SVM model, determine whether the synthetic data conforms to physical laws, and output the verification result.

[0051] Furthermore, the verification includes: for each set of real data, the corresponding parameters in the parameter library... The actual value must satisfy the parameter verification formula: in, For the minimum value of the parameter, This represents the maximum value of the parameter.

[0052] Furthermore, the step of converting the physical rules in the physical rule base and the parameters in the parameter base into multi-dimensional feature vectors includes: Suppose that the physical rule base contains There are several physical rules, and the parameter library contains them. If there are 1 parameter, then the feature vector The dimension is ,Right now ,in: For physical rule constraint features, each ( The constraint deviation value corresponding to a physical rule is calculated as follows: in, The calculated values ​​related to this physical rule in the synthetic data to be verified. These are theoretical values ​​calculated based on physical rules. For parameter comparison features, each （ ） The comparison deviation value corresponding to a single parameter is calculated as follows: in, The parameter values ​​in the synthetic data to be verified. This is the standard value for this parameter in the parameter library.

[0053] Furthermore, the method of training the SVM model using the negative and positive samples includes: The positive and negative samples are divided into training and test sets according to a set ratio; The radial basis function (RBF) is used as the kernel function. The expression for the RBF kernel function is as follows: in, These are kernel function parameters used to control the distribution of samples after mapping to a high-dimensional space; Adjusting penalty parameters using a grid search method and kernel function parameters ;in Used to balance the training error and generalization ability of the model; Used to control the influence of kernel functions; The optimal parameter combination is determined through cross-validation with the goal of maximizing the accuracy of the test set. The optimal parameter combination is substituted into the SVM model for iterative training. The iteration termination condition is: The training set loss function value is less than a preset threshold or the number of iterations reaches a preset maximum value.

[0054] Furthermore, accuracy, precision, recall, and F1 score are used as performance evaluation metrics for the model. After the model is trained, the accuracy on the test set must reach a first set threshold and the F1 score must reach a second set threshold. The formula for calculating accuracy (Acc) is as follows: ; The formula for calculating Prec (Precision Rate) is as follows: ; The recall rate (Rec) is calculated using the following formula: ; The formula for calculating the F1 score is as follows: ; Wherein, TP is the number of synthetic data samples that the model correctly identified as not conforming to physical laws, TN is the number of synthetic data samples that the model correctly identified as conforming to physical laws, FP is the number of synthetic data samples that the model incorrectly identified as not conforming to physical laws, and FN is the number of synthetic data samples that the model incorrectly identified as conforming to physical laws.

[0055] Furthermore, the SVM model is also used to: calculate the symbolic distance from the feature vector of the synthetic data to the optimal classification hyperplane of the SVM, and determine whether the synthetic data conforms to physical laws based on the distance value; if not, the model automatically transfers the synthetic data and its judgment result to manual judgment for forced review, and includes the reviewed samples into the retraining set.

[0056] Furthermore, the optimization and output module is also used for: The verification results are randomly sampled. If an error is found, the incorrectly labeled samples are corrected, and retraining is performed based on the corrected samples.

[0057] The above-mentioned system for verifying the consistency between synthetic data and physical laws implements the steps and processes of the above-mentioned method embodiment for verifying the consistency between synthetic data and physical laws, and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0058] For example, the following provides a system apparatus implementation with a specific configuration.

[0059] At the hardware level, the specific hardware configuration is as follows: Database server: A high-performance rack server is selected, configured with hardware such as CPU, memory, and hard disk, to store the physical rule base, parameter base, real database, error synthesis database, and model training data. Relational databases can be used to store structured physical rules, parameters, and labeled data, while non-relational databases can be used to store unstructured raw data. The selection is flexible according to the characteristics of the data and application requirements to ensure efficient data storage and fast retrieval.

[0060] Computing Server: A high-performance GPU server is selected, configured with hardware such as GPU, CPU, memory, and hard disk, for training of consistency verification models and batch verification of synthetic data to be verified. The parallel computing capabilities of GPUs are utilized to improve model training speed and verification efficiency.

[0061] Data acquisition equipment includes various high-precision (accuracy ≥ 0.1%FS) sensors (force sensors, temperature sensors, vibration sensors, etc.) and data acquisition cards, used to obtain real data from data sources such as production sites and experimental platforms.

[0062] Terminal equipment: High-performance personal computers are used for manual data annotation, manual sampling of model verification results, and correction of erroneous samples. Staff can access the model verification results and perform related operations through a browser or specially developed client software.

[0063] At the software level, each module is developed and implemented using programming languages ​​and related frameworks, as detailed below: The physical rule base and parameter base module is developed using the Django framework. It implements the functions of adding, modifying, deleting, and querying rules and parameters. It interacts with the database through a database connection library to ensure efficient reading and writing of rules and parameters.

[0064] Data collection and annotation module: Data preprocessing is achieved using libraries such as OpenCV and Pandas, a visual annotation interface is developed, manual annotation of positive and negative samples is supported, and annotation information is recorded.

[0065] Consistency Validation Model Module: The Scikit-learn library is used to build, train, and validate the SVM model. The NumPy and SciPy libraries are used to calculate feature vectors and process data. The Matplotlib library is used to plot model performance curves (such as ROC curves and PR curves) to facilitate model performance analysis.

[0066] Manual evaluation and optimization module: The web interface is developed using web development technologies (HTML, CSS, JavaScript), which allows staff to view model validation results, conduct sampling checks, and correct erroneous samples. It interacts with the model training module through the Flask framework, feeding back the corrected samples to the training set and triggering model retraining.

[0067] System integration: Docker containerization technology is used to package each module into an independent container, thereby decoupling the modules and facilitating system deployment, maintenance and upgrades. At the same time, Kubernetes is used to orchestrate and manage the containers, ensuring system stability and scalability.

[0068] See Figure 5The timing diagram for verifying the consistency between synthetic data and physical laws is explained as follows: This timing diagram takes the time axis as its core, clearly defining the time sequence, duration, and interaction relationship of each operation, involving five participating entities: the system main body, the rule parameter module, the data module, the model module, and the manual module. The time-series process is divided into four core stages: The first stage, t0-t1, involves the main system sending construction instructions to the rule parameter module, which then completes the input, verification, and storage of rules and parameters and feeds them back to the system. The second stage, t1-t3, involves the system initiating data collection and annotation operations. The data module performs parallel collection, preprocessing, and positive / negative sample annotation of real data and erroneous synthetic data, and feeds back to the system upon completion. The third stage, t3-t5, involves the system triggering model building and training. The model module completes feature vector construction, parameter optimization, model training, and performance verification, and feeds back to the system upon achieving the required standards. The fourth stage, after t5, involves the system continuously sending synthetic data verification instructions to the model module. The model module processes the data to be verified in real time and outputs the judgment results. Simultaneously, after each set of data verification is completed, the system can initiate a manual judgment phase. The manual module performs sampling checks. If errors are found, the samples are corrected and fed back to the model module for retraining. If no errors are found, the verification continues, forming a continuously optimized time-series closed loop.

[0069] Corresponding to the above embodiments of the method for verifying the consistency between synthetic data and physical laws, this application provides an electronic device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the steps and processes of the above embodiments of the method for verifying the consistency between synthetic data and physical laws, and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0070] Memory can be used to store software programs and various data. Memory can primarily include a first storage area for storing programs or instructions and a second storage area for storing data. The first storage area can store the operating system, application programs or instructions required for at least one function (such as sound playback, image playback, etc.). Furthermore, memory can include volatile memory or non-volatile memory, or both. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory can be random access memory (RAM), static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (Synchlink DRAM, SLDRAM), and direct memory bus RAM (DRRAM). The memory in the embodiments of this application includes, but is not limited to, these and any other suitable types of memory.

[0071] The processor may include one or more processing units; optionally, the processor integrates an application processor and a modem processor, wherein the application processor mainly handles operations related to the operating system, user interface, and applications, while the modem processor mainly handles wireless communication signals, such as a baseband processor. It is understood that the aforementioned modem processor may also not be integrated into the processor.

[0072] Corresponding to the above embodiments of the method for verifying the consistency between synthetic data and physical laws, this application also provides a readable storage medium storing a program or instructions. When the program or instructions are executed by a processor, they implement the steps and processes of the above embodiments of the method for verifying the consistency between synthetic data and physical laws, and achieve the same technical effect. To avoid repetition, these will not be described again here.

[0073] The processor is the processor in the electronic device described in the above embodiments of this application. The readable storage medium includes a computer-readable storage medium, such as a computer read-only memory (ROM), random access memory (RAM), a magnetic disk, or an optical disk.

[0074] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0075] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a computer software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0076] It is understood that the embodiments of this application have been described above in conjunction with the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. As those skilled in the art will know, various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of the invention. Furthermore, those skilled in the art, under the guidance or instruction of this application, can modify these features and embodiments to adapt to specific situations and materials without departing from the spirit and scope of the invention. Therefore, this invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the protection scope of this invention.

Claims

1. A method for verifying the consistency between synthetic data and physical laws, characterized in that, include: Real data from the target equipment manufacturing process is acquired and validated based on a pre-built physical rule base and a parameter base corresponding to the physical rule base. Validated data is processed as negative samples and labeled. The physical rule base is used to store physical rules in the form of standardized mathematical formulas and / or logical constraints. The parameter base is used to store physical parameters, process parameters, and / or performance parameters related to the target equipment manufacturing and corresponding to the physical rules in the physical rule base. Acquire the synthetic data of the target equipment, filter out samples that violate the physical rules in the physical rule base, process them as positive samples and label them, and at the same time construct an error synthesis database; The physical rules in the physical rule base and the parameters in the parameter base are transformed into multi-dimensional feature vectors, and the negative and positive samples are used to train an SVM model; the SVM model is used for consistency verification to determine whether the synthetic data to be verified conforms to physical laws. The synthetic data to be verified is preprocessed and transformed into the multi-dimensional feature vector, which is then input into the trained SVM model to determine whether the synthetic data conforms to physical laws and output the verification result.

2. The method for verifying the consistency between synthetic data and physical laws according to claim 1, characterized in that, The verification includes: for each set of real data, the corresponding parameters in the parameter library. The actual value must satisfy the parameter verification formula: in, For the minimum value of the parameter, This represents the maximum value of the parameter.

3. The method for verifying the consistency between synthetic data and physical laws according to claim 1, characterized in that, The process of converting the physical rules in the physical rule base and the parameters in the parameter base into multi-dimensional feature vectors includes: Suppose that the physical rule base contains There are several physical rules, and the parameter library contains them. If there are 1 parameter, then the feature vector The dimension is ,Right now ,in: For physical rule constraint features, each ( The constraint deviation value corresponding to a physical rule is calculated as follows: in, The calculated values ​​related to this physical rule in the synthetic data to be verified. These are theoretical values ​​calculated based on physical rules. For parameter comparison features, each （ ） The comparison deviation value corresponding to a single parameter is calculated as follows: in, The parameter values ​​in the synthetic data to be verified. This is the standard value for this parameter in the parameter library.

4. The method for verifying the consistency between synthetic data and physical laws according to claim 1, characterized in that, The method of training an SVM model using the negative and positive samples includes: The positive and negative samples are divided into training and test sets according to a set ratio; The radial basis function (RBF) is used as the kernel function. The expression for the RBF kernel function is as follows: in, These are kernel function parameters used to control the distribution of samples after mapping to a high-dimensional space; Adjusting penalty parameters using grid search method and kernel function parameters ;in Used to balance the training error and generalization ability of the model; Used to control the influence of kernel functions; The optimal parameter combination is determined through cross-validation with the goal of maximizing the accuracy of the test set. The optimal parameter combination is substituted into the SVM model for iterative training. The iteration termination condition is: The training set loss function value is less than a preset threshold or the number of iterations reaches a preset maximum value.

5. The method for verifying the consistency between synthetic data and physical laws according to claim 1, characterized in that, Accuracy, precision, recall, and F1 score are used as the performance evaluation metrics for the model. After the model is trained, the accuracy on the test set must reach above the first set threshold, and the F1 score must reach above the second set threshold. The formula for calculating accuracy (Acc) is as follows: The formula for calculating Prec (Precision Rate) is as follows: The recall rate (Rec) is calculated using the following formula: The formula for calculating the F1 score is as follows: Wherein, TP is the number of synthetic data samples that the model correctly identified as not conforming to physical laws, TN is the number of synthetic data samples that the model correctly identified as conforming to physical laws, FP is the number of synthetic data samples that the model incorrectly identified as not conforming to physical laws, and FN is the number of synthetic data samples that the model incorrectly identified as conforming to physical laws.

6. The method for verifying the consistency between synthetic data and physical laws according to claim 1, characterized in that, The SVM model is also used to: calculate the sign distance from the feature vector of the synthetic data to the optimal classification hyperplane of the SVM, and determine whether the synthetic data conforms to physical laws based on the distance value; if not, the model automatically transfers the synthetic data and its judgment result to manual judgment for forced review, and includes the reviewed samples into the retraining set.

7. The method for verifying the consistency between synthetic data and physical laws according to claim 1, characterized in that, The method for verifying the consistency between the synthesized data and physical laws also includes: The verification results are randomly sampled. If an error is found, the incorrectly labeled samples are corrected, and retraining is performed based on the corrected samples.

8. A system for verifying the consistency between synthetic data and physical laws, characterized in that, include: The data source module is used to acquire real data during the manufacturing process of the target equipment, and to verify the data based on a pre-built physical rule library and a parameter library corresponding to the physical rule library. Data that passes the verification is processed as negative samples and labeled. The physical rule library is used to store physical rules in the form of standardized mathematical formulas and / or logical constraints. The parameter library is used to store physical parameters, process parameters and / or performance parameters related to the manufacturing of the target equipment that correspond to the physical rules in the physical rule library. The data processing module is used to acquire the synthetic data of the target equipment, filter out samples that violate the physical rules in the physical rule base, process them as positive samples and label them, and at the same time construct an error synthesis database. The core verification module is used to transform the physical rules in the physical rule base and the parameters in the parameter base into multi-dimensional feature vectors, and to train an SVM model using the negative and positive samples; the SVM model is used for consistency verification to determine whether the synthetic data to be verified conforms to physical laws. The optimization and output module is used to preprocess the synthetic data to be verified and transform it into the multi-dimensional feature vector, input it into the trained SVM model, determine whether the synthetic data conforms to physical laws, and output the verification result.

9. An electronic device, characterized in that, The electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the computer program is executed by the processor, it implements the steps of the method for verifying the consistency between synthetic data and physical laws as described in any one of claims 1 to 7.

10. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of the method for verifying the consistency between synthetic data and physical laws as described in any one of claims 1 to 7.

Images