Parameter calibration method and device of object, electronic equipment and storage medium

Abstract

The application relates to the technical field of computer-aided engineering, and provides a parameter calibration method and device of an object, an electronic device and a storage medium.The method comprises the following steps: acquiring experimental time series data of an object to be measured under a target working condition and simulation time series data of the current iteration optimization; acquiring first frequency domain features corresponding to the experimental time series data and second frequency domain features corresponding to the simulation time series data; determining a target loss based on the frequency domain feature difference and the amplitude distribution difference; and adjusting the parameter set to be calibrated by using the target loss to obtain the parameter set to be calibrated for the next iteration optimization. The target loss is determined by combining the amplitude distribution difference and the frequency domain feature difference between the experimental time series data and the simulation time series data, parameter decoupling can be realized, and the accuracy of the calibrated parameters is improved.

Description

Technical Field

[0001] This invention relates to the field of computer-aided engineering technology, and in particular to a method and apparatus for calibrating the parameters of an object, an electronic device, and a storage medium. Background Technology

[0002] In engineering simulation, structural mechanics, fluid simulation, and physical prototype modeling, the input parameters of the simulation model are usually the parameter set of the physical prototype, and the accuracy of the input parameters determines the reliability and accuracy of the simulation results. To reduce the parameter deviation between the simulation model and the actual physical prototype, it is necessary to iteratively optimize and calibrate the set of parameters to be calibrated for the physical prototype, ultimately making the simulation response characteristics closely match the actual response characteristics of the real physical prototype.

[0003] Currently, most mainstream object parameter calibration methods in the industry are based on time-domain data for single-dimensional error calibration. Specifically, existing technologies mainly collect experimental data from the physical prototype after it reaches steady state and simulation data output by the simulation model. They directly compare the static amplitude deviation between the two types of data to construct a loss function, and iteratively adjust the set of parameters to be calibrated based on the constructed loss function to achieve optimized calibration of the input parameters of the simulation model, thereby fitting the macroscopic time-domain response characteristics of the physical prototype.

[0004] However, multiple parameters (such as thermal conductivity and specific heat capacity, viscosity and contact angle) are highly correlated with the static amplitude. If only the static amplitude is relied upon, it will be difficult to decouple different parameters and the optimization algorithm will easily get trapped in local optima, resulting in an inaccurate set of calibrated parameters. Summary of the Invention

[0005] This invention provides a method and apparatus for calibrating the parameters of an object, an electronic device, and a storage medium, to solve the defect in the prior art that the calibrated parameter set is inaccurate due to relying solely on static amplitude deviation, thereby achieving decoupling of the parameters in the parameter set and improving the accuracy of the calibrated parameter set.

[0006] This invention provides a parameter calibration method for an object, comprising iteratively executing the following steps until a preset cutoff condition is met, and using the set of parameters to be calibrated targeted by the last round of iteration as the parameter set of the object under test: The experimental time series data of the object under test under the target working condition and the simulation time series data of this round of iterative optimization are obtained. The simulation time series data is obtained by the simulation model based on the set of calibration parameters of the object under test. Obtain the first frequency domain features corresponding to the experimental time series data and the second frequency domain features corresponding to the simulation time series data; The target loss is determined based on the differences in frequency domain features between the first and second frequency domain features, as well as the differences in amplitude distribution between experimental time series data and simulated time series data. The set of parameters to be calibrated is adjusted using the target loss to obtain the set of parameters to be calibrated for the next round of iterative optimization.

[0007] According to the object parameter calibration method provided by the present invention, the target loss is determined based on the frequency domain feature difference between the first frequency domain feature and the second frequency domain feature, and the amplitude distribution difference between experimental time series data and simulated time series data, including: The first error term is determined based on the difference in the position of the resonant peak in the frequency domain characteristics, the difference in gain and / or phase at several sampling frequencies, and the second error term is determined based on the difference in amplitude distribution between experimental time series data and simulated time series data. By combining the first and second error terms, the target loss is obtained.

[0008] According to the parameter calibration method for an object provided by the present invention, the experimental time-series data includes experimental time-series sub-data of multiple location points in the object under test, and the simulation time-series data includes simulation time-series sub-data of multiple location points. A second error term is determined based on the amplitude distribution difference between the experimental time-series data and the simulation time-series data, including: The amplitude error term is obtained based on the difference in amplitude between the experimental time series data and the simulated time series data at each location point. The first spatial gradient data is obtained based on the difference in amplitude between the experimental time series sub-data at each adjacent location point. The second spatial gradient data is obtained based on the difference in amplitude between the simulation time series sub-data at each adjacent location point. The spatial gradient error term is determined based on the first spatial gradient data and the second spatial gradient data. By combining the magnitude error term and the spatial gradient error term, the second error term is obtained.

[0009] According to the object parameter calibration method provided by the present invention, the target loss is determined based on a first error term determined by the difference in frequency domain characteristics and a second error term determined by the difference in amplitude distribution. The set of parameters to be calibrated is adjusted using the target loss to obtain the set of parameters to be calibrated for the next round of iterative optimization, including: The difference between the first error terms during several rounds of iterative optimization is obtained to obtain the first change value, and the difference between the second error terms during several rounds of iterative optimization is obtained to obtain the second change value; Based on the relationship between the first and second changes, determine the adjustment strategy for the set of parameters to be calibrated; According to the adjustment strategy, the set of parameters to be calibrated is adjusted to obtain the set of parameters to be calibrated for the next round of iterative optimization.

[0010] According to the parameter calibration method for an object provided by the present invention, the adjustment strategy includes the adjustment priority of each parameter to be calibrated in the set of parameters to be calibrated, and determining the adjustment strategy for the set of parameters to be calibrated based on the magnitude relationship between a first change value and a second change value, including: The error term corresponding to the larger of the first and second change values ​​is taken as the target error term; Set the adjustment priority of the parameters in the set of parameters to be calibrated that are related to the target error term to the highest priority; Among them, the adjustment range of each parameter to be calibrated in the set of parameters to be calibrated is positively correlated with its adjustment priority.

[0011] According to the parameter calibration method for an object provided by the present invention, experimental time-series data of the object under test under target working conditions and simulation time-series data of the current round of iterative optimization are obtained, including: The experimental time series data of several locations on the test object are collected during the experimental test under the target working condition of applying a dynamic disturbance signal to the test object, and the experimental time series data is obtained based on the experimental time series data of several locations. After the dynamic disturbance signal is input to the input terminal of the simulation model, simulation calculations are performed to obtain simulation time series data.

[0012] According to the object parameter calibration method provided by the present invention, experimental time-series data is obtained based on experimental time-series sub-data from several location points, including: Construct a reconstruction loss function that includes the physical conservation equations as regularization terms; The collected experimental time series data were used as physical constraints. By minimizing the reconstruction loss function, experimental time-series data representing the entire domain of the object under test are obtained.

[0013] The present invention also provides a parameter calibration device for an object, which is used to iteratively optimize a set of parameters to be calibrated until a preset cutoff condition is met, and uses the set of parameters to be calibrated targeted in the last round of iterative optimization as the parameter set of the object to be tested. The parameter calibration device for the object includes: The data acquisition module is used to acquire experimental time-series data of the test object under target working conditions and simulation time-series data of this round of iterative optimization. The simulation time-series data is obtained by the simulation model based on the set of calibration parameters of the test object. The frequency domain feature acquisition module is used to acquire the first frequency domain feature corresponding to the experimental time series data and the second frequency domain feature corresponding to the simulation time series data. The target loss determination module is used to determine the target loss based on the frequency domain feature differences between the first and second frequency domain features, as well as the amplitude distribution differences between experimental time series data and simulated time series data. The parameter adjustment module is used to adjust the set of parameters to be calibrated using the target loss, so as to obtain the set of parameters to be calibrated for the next round of iterative optimization.

[0014] According to the object parameter calibration device provided by the present invention, the target loss determination module includes: The error term determination module is used to determine the first error term based on the difference in the position of the resonant peak in the frequency domain characteristics, the gain difference and / or phase difference at several sampling frequencies, and to determine the second error term based on the difference in amplitude distribution between experimental time series data and simulated time series data. The error term combination module is used to combine the first error term and the second error term to obtain the target loss.

[0015] According to the parameter calibration device for an object provided by the present invention, the experimental time series data includes experimental time series sub-data of multiple location points in the object under test, the simulation time series data includes simulation time series sub-data of multiple location points, and the error term determination module includes: The amplitude error term determination module is used to obtain the amplitude error term based on the difference in amplitude between the experimental time series sub-data and the simulation time series sub-data at each location point; The spatial gradient error term determination module is used to obtain first spatial gradient data based on the magnitude difference between experimental time series sub-data at each adjacent location point, obtain second spatial gradient data based on the magnitude difference between simulation time series sub-data at each adjacent location point, and determine the spatial gradient error term based on the first spatial gradient data and the second spatial gradient data. The second error term determination module is used to combine the magnitude error term and the spatial gradient error term to obtain the second error term.

[0016] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a parameter calibration method for any of the objects described above.

[0017] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a parameter calibration method for any of the objects described above.

[0018] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements a parameter calibration method for any of the objects described above.

[0019] The object parameter calibration method, device, electronic device, and storage medium provided by this invention determine the target loss by simultaneously utilizing the amplitude distribution difference and frequency domain characteristic difference between experimental time-series data and simulation time-series data. This makes the parameter calibration process not only focus on whether the time-domain values ​​are close, but also on the consistency of the dynamic response characteristics of the object under test and the simulation model. This can reduce the parameter coupling and local optimum problems caused by relying solely on static amplitude fitting, improve the accuracy of parameter calibration, and further enhance the predictive reliability of the calibrated simulation model under complex or unknown working conditions. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0021] Figure 1 This is one of the flowcharts illustrating the object parameter calibration method provided by the present invention.

[0022] Figure 2 This is a schematic diagram showing the distribution of experimental data collected from multiple sampling points on the object under test at a single moment.

[0023] Figure 3 This is a schematic diagram of the experimental data distribution over the entire domain of the object under test at a single moment.

[0024] Figure 4 This is a schematic diagram illustrating the spatial gradient difference between experimental and simulation data at a single moment.

[0025] Figure 5 This is the second flowchart illustrating the object parameter calibration method provided by the present invention.

[0026] Figure 6 This is one of the structural schematic diagrams of the object parameter calibration device provided by the present invention.

[0027] Figure 7 This is the second schematic diagram of the object parameter calibration device provided by the present invention.

[0028] Figure 8 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0030] The following is combined with Figures 1 to 5 This invention describes a method for calibrating the parameters of an object. The object parameter calibration method provided by this invention can be implemented by a computer device, server, industrial control equipment, etc. Figure 1 This is one of the flowcharts illustrating the object parameter calibration method provided by the present invention, such as... Figure 1 As shown, the method includes iteratively executing the following steps until a preset cutoff condition is met, and using the set of parameters to be calibrated targeted by the last round of iteration optimization as the parameter set of the object under test. Each round of iteration optimization includes the following steps: Step 101: Obtain the experimental timing data of the object under test under the target working condition and the simulation timing data of this round of iterative optimization.

[0031] The object under test can be a physical object for which a simulation model needs to be established or modified, or it can be a local structure, functional component, or process object within a physical object. For example, the object under test can be a glass component in an automotive glass heating system, a glass sample with heating circuitry, the frame and adhesive circuit structure in a mobile phone frame dispensing process, a fluid transport component, a thermal management component, or a structural load-bearing component. The heating circuitry can include, but is not limited to, heating circuitry designed with materials such as heating silver wire, nano-carbon crystal, metal alloy wire, indium tin oxide, graphene, or ceramic microparticle composite materials. In other words, the object under test is not limited to a specific product; as long as its physical quantity data changing over time can be obtained through experimental testing, and simulation calculations under corresponding working conditions can be performed using a simulation model, it can be used as the object under test in this embodiment.

[0032] The target operating condition refers to the operating conditions used when conducting experimental tests on the object under test and performing simulation calculations on the simulation model. The target operating condition can include one or more of the following: load conditions, temperature conditions, speed conditions, pressure conditions, electrical power conditions, boundary constraint conditions, contact conditions, fluid inlet conditions, and environmental conditions. For example, in an automotive glass heating system, the target operating condition can include heating power, ambient temperature, and heat dissipation conditions at the glass boundary; in the dispensing process for a mobile phone frame, the target operating condition can include dispensing speed, glue valve pressure, glue needle height, ambient temperature, and glue tank dimensions.

[0033] Experimental time-series data refers to experimental data obtained by conducting experiments on the object under test under target operating conditions, and data that changes over time. Experimental time-series data can be time-series data for temperature, displacement, stress, strain, pressure, velocity, adhesive path height, liquid surface profile, current, or other target physical quantities. Experimental time-series data can be acquired by testing equipment such as thermocouples, infrared thermal imagers, high-speed cameras, laser profilometers, pressure sensors, displacement sensors, or strain gauges. Experimental time-series data can be the time series of a single measuring point on the object under test, or it can be a collection of time series corresponding to multiple measuring points or multiple location points including the measuring points.

[0034] In this context, "current iteration optimization" refers to the current parameter update cycle in the parameter calibration process. Parameter calibration typically involves multiple iterations. Each iteration calculates the set of parameters to be calibrated by calling a simulation model, obtaining simulation time-series data, and adjusting the set of parameters to be calibrated based on the differences between the simulation time-series data and the experimental time-series data.

[0035] The simulation time-series data is obtained by simulating the set of parameters to be calibrated for the object under test using a simulation model. The simulation model is a computational model used to describe the physical response of the object under test under target operating conditions. The simulation model can be a finite element simulation model, a computational fluid dynamics simulation model, a heat conduction simulation model, a thermo-structural coupling simulation model, a fluid-structure interaction simulation model, or other multiphysics simulation models. The simulation time-series data corresponds to the experimental time-series data in terms of the type of physical quantities. For example, when the experimental time-series data is a temperature-time curve, the simulation time-series data can be the temperature-time curve calculated by the simulation model; when the experimental time-series data is a glue path height-time curve, the simulation time-series data can be the glue path height-time curve calculated by the simulation model. Exemplarily, the simulation time-series data and the experimental time-series data can be subjected to time alignment, sampling frequency unification, interpolation, normalization, or outlier removal.

[0036] The set of parameters to be calibrated refers to a group of parameters in the simulation model that need to be back-calibrated using experimental time-series data. The set of parameters to be calibrated may include one or more parameters. These parameters may include those that need to be considered during the actual use of the object under test. For example, in an automotive glass heating system, the set of parameters to be calibrated may include the thermal conductivity of silver wire, the specific heat capacity of glass, and the boundary heat transfer coefficient. In the dispensing process for a mobile phone frame, the set of parameters to be calibrated may include the adhesive viscosity, power-law exponent, contact angle, and surface tension coefficient. The set of parameters to be calibrated for the first round of iterative optimization can be determined based on material parameters provided by material suppliers, empirical parameters from publicly available information, nominal parameters of the product design, historical project calibration results, or manually preset values. The set of parameters to be calibrated for the second round of iterative optimization and subsequent rounds is the set of parameters adjusted using the target loss in the previous round of iteration.

[0037] Step 102: Obtain the first frequency domain features corresponding to the experimental time series data and the second frequency domain features corresponding to the simulation time series data.

[0038] For example, frequency domain transformation is performed on experimental time-series data and simulated time-series data to obtain the first frequency domain features corresponding to the experimental time-series data and the second frequency domain features corresponding to the simulated time-series data. The frequency domain transformation method can include, but is not limited to, Fourier transform, fast Fourier transform, short-time Fourier transform, wavelet transform, discrete cosine transform, or other time-frequency analysis methods. In some application scenarios, the frequency domain transformation method can be determined based on the physical response type of the object under test, the sampling length and sampling frequency of the experimental time-series data, and the sensitivity of the parameters to be calibrated to the dynamic response. In some application scenarios, the same frequency domain transformation method, the same sampling window, the same normalization method, and the same frequency range can be used for both experimental and simulated time-series data.

[0039] The first frequency domain feature can be used to characterize the dynamic response of the object under test under real experimental conditions. For example, the first frequency domain feature can include one or more of the following: spectral amplitude information, spectral phase information, band energy information, dominant frequency component information, and frequency domain response curve information. The second frequency domain feature can be used to characterize the prediction result of the simulation model of the dynamic response of the object under test under the current set of parameters to be calibrated.

[0040] Step 103: Determine the target loss based on the frequency domain feature differences between the first and second frequency domain features, as well as the amplitude distribution differences between the experimental time series data and the simulated time series data.

[0041] The frequency domain feature difference refers to the difference in frequency domain response characteristics between the first and second frequency domain features. For example, the frequency domain feature difference can be determined by Euclidean distance, cosine distance, correlation coefficient, frequency domain curve integral error, band energy difference, or other similarity measures. The smaller the frequency domain feature difference, the closer the dynamic response predicted by the simulation model under the current set of parameters to be calibrated is to the true dynamic response of the object under test.

[0042] Amplitude distribution difference refers to the difference in amplitude between experimental and simulated time-series data in the time domain. Amplitude distribution difference can be used to measure the accuracy of the simulation model in predicting the magnitude of the target physical quantity of the object under test and its trend over time. For example, for a glass heating system, amplitude distribution difference can reflect the deviation between the simulated temperature curve and the experimental temperature curve; while for the dispensing process of a mobile phone frame, amplitude distribution difference can reflect the deviation between the simulated adhesive path height curve and the experimental adhesive path height curve.

[0043] In this embodiment, the target loss can be determined jointly by the differences in frequency domain features and the differences in amplitude distribution. For example, the target loss can be obtained by weighted summing of the errors corresponding to the differences in frequency domain features and the errors corresponding to the differences in amplitude distribution. In some application scenarios, the weights can be preset according to the physical type of the object under test, the importance of the target operating condition, the quality of experimental data, or the parameter calibration requirements, or they can be adjusted in different application scenarios. The smaller the target loss, the better the current set of parameters to be calibrated enables the simulation model to reproduce the true physical response of the object under test under the target operating condition.

[0044] Step 104: Adjust the set of parameters to be calibrated using the target loss to obtain the set of parameters to be calibrated for the next round of iterative optimization.

[0045] For example, in each round of iterative optimization, after obtaining the target loss, it can be determined whether the iterative optimization stopping condition is met. The iterative optimization stopping condition may include at least one of the following: the target loss is less than a preset loss threshold, the decrease in the target loss obtained in several consecutive iterative optimization processes is less than a preset change threshold, or the number of iterative optimizations reaches a preset upper limit. If the iterative optimization stopping condition is not met, step 104 is executed. If the iterative optimization stopping condition is met, iterative optimization is determined to stop, and this round of iterative optimization is taken as the last round of iterative optimization, and the set of parameters to be calibrated in this round of iterative optimization is taken as the parameter set of the object under test. That is, after the iterative optimization stops, the set of parameters to be calibrated in the last round of iterative optimization is taken as the parameter set of the object under test.

[0046] For example, step 104 can be implemented as follows: using the target loss as the objective function value of the optimization algorithm, and employing gradient descent, genetic algorithm, differential evolution, Bayesian optimization, particle swarm optimization, sequential quadratic programming, or other parameter optimization algorithms to update the set of parameters to be calibrated. Optionally, during the update process, upper and lower bounds, step size constraints, physical rationality constraints, or engineering experience constraints can be set for each parameter to be calibrated to prevent obtaining a set of parameters that does not conform to the true physical meaning. For example, in an automotive glass heating system, if the current set of parameters to be calibrated includes the thermal conductivity of silver wire and the specific heat capacity of glass, this set of parameters to be calibrated can be input into the simulation model in each round of iterative optimization to obtain simulated temperature time-series data; then, the simulated temperature time-series data and the experimental temperature time-series data are compared from the perspectives of time-domain amplitude and frequency-domain response to obtain the target loss; finally, the thermal conductivity of silver wire and the specific heat capacity of glass are adjusted according to the target loss, so that the set of parameters to be calibrated for the next round of iterative optimization is closer to the true parameters. For example, in the dispensing process of mobile phone frames, if the current set of parameters to be calibrated includes glue viscosity and power law exponent, the glue viscosity and power law exponent can be corrected round by round by comparing the amplitude distribution differences and frequency domain characteristics between experimental glue path height time series data and simulated glue path height time series data.

[0047] In the above scheme, the target loss is determined by simultaneously utilizing the difference in amplitude distribution and frequency domain characteristics between experimental time series data and simulation time series data. This allows the parameter calibration process to focus not only on whether the time domain values ​​are close, but also on the consistency of the dynamic response characteristics between the test object and the simulation model. This can solve the parameter coupling and local optima problems caused by relying solely on static amplitude fitting, improve the accuracy of parameter calibration, and enhance the predictive reliability of the calibrated simulation model under complex or unknown working conditions.

[0048] In some embodiments, step 101 above may include the following steps: First, experimental time series data at several locations on the test object are acquired during the experimental test under the target working condition of applying a dynamic disturbance signal to the test object. Then, experimental time series data is obtained based on the experimental time series data at several locations.

[0049] The dynamic disturbance signal can be a controllable time-varying input signal superimposed on the steady-state condition. The dynamic disturbance signal can be a sine wave, square wave, pulse, step, swept frequency, random excitation, or other controllable input signal. For example, the dynamic disturbance signal can have preset amplitude, frequency, phase, duration, and waveform. For instance, the amplitude of the dynamic disturbance signal can be set to a small disturbance amplitude that does not disrupt the normal operating state of the object under test, ensuring that the experimental test is conducted within the engineering allowable range. In some application scenarios, the frequency of the dynamic disturbance signal can be determined based on the response speed of the object under test, the sampling frequency, and the sensitive frequency band of the parameter to be calibrated to the dynamic response.

[0050] For example, in thermal response experiments, the dynamic disturbance signal can be a small sinusoidal power fluctuation superimposed on the basic heating power, or it can be a temperature step signal; in fluid dispensing experiments, the dynamic disturbance signal can be a high-frequency small velocity pulse superimposed on the steady-state dispensing speed, or it can be a pressure pulse signal; in structural response experiments, the dynamic disturbance signal can be a periodic force load or displacement excitation.

[0051] The experimental time-series data at several locations can be acquired using sensors or image measurement devices. For example, thermocouples can be used to acquire temperature time-series data at multiple locations, infrared thermal imagers can be used to acquire temperature time-series data at multiple spatial locations, high-speed cameras can be used to acquire data on the change of adhesive path morphology over time, and laser profilometers can be used to acquire adhesive path height time-series data. The acquired experimental time-series data at several locations can be directly combined into experimental time-series data, or they can be preprocessed through filtering, interpolation, normalization, outlier removal, and time synchronization before being combined into experimental time-series data.

[0052] Then, the dynamic disturbance signal is input to the input terminal of the simulation model for simulation calculation to obtain simulation time series data.

[0053] During the simulation calculation, the dynamic disturbance signal input to the simulation model is consistent with the dynamic disturbance signal applied to the object under test during the experimental test. This ensures that the experimental time-series data and the simulation time-series data are generated under the same dynamic excitation conditions, thus making the differences in frequency domain characteristics and amplitude distribution between the two comparable.

[0054] In the above scheme, by applying the same dynamic disturbance signal in experimental testing and simulation calculation, interference can be increased, making the simulation model more accurate in unknown operating conditions based on the calibrated parameter set.

[0055] In some embodiments, the above-mentioned method of obtaining experimental time series data based on experimental time series sub-data of several location points may include the following steps: constructing a reconstruction loss function that includes physical conservation equations as regularization terms; using the collected experimental time series sub-data as physical constraints; and solving for experimental time series data representing the entire domain of the object under test by minimizing the reconstruction loss function.

[0056] Among them, the physical conservation equations are physical equations used to describe the evolution of the target physical quantities of the object under test. For example, physical conservation equations may include the heat conduction equation, mass conservation equation, momentum conservation equation, energy conservation equation, Navier-Stokes (NS) equations, structural equilibrium equations, diffusion equations, or other equations related to the physical processes of the object under test. The physical conservation equations can be determined according to the type of physical field to which the object under test belongs. For example, for a glass heating system, the heat conduction equation can be used as the physical conservation equation; for a dispensing fluid system, the mass conservation equation, momentum conservation equation, or fluid interface evolution constraints can be used as the physical conservation equation.

[0057] The reconstruction loss function can include a data fitting term and a physical regularization term. The data fitting term is used to constrain the reconstruction result to be consistent with the experimental time series subdata at the acquired location points; the physical regularization term is used to constrain the reconstruction result to satisfy the physical conservation equations or to minimize the residuals of the physical conservation equations. For example, the reconstruction loss function can include: the error term between the reconstructed value and the experimental measurement value at the acquired location point, and the residual term of the equation obtained after substituting the reconstruction result into the heat conduction equation, the fluid conservation equation, or the structural equilibrium equation. The reconstruction loss function can be obtained by weighted summation of the above terms.

[0058] Physical constraints refer to the constraints imposed on the reconstruction process by the experimental time-series sub-data actually collected during the experimental testing. The experimental time-series sub-data can serve as either hard or soft constraints. As a hard constraint, the reconstruction result must be strictly equal to the experimental time-series sub-data at the corresponding location and time point; as a soft constraint, a certain range of measurement error deviation is allowed between the reconstruction result and the experimental time-series sub-data.

[0059] In some application scenarios, the boundary conditions, initial conditions, geometric constraints, material physical range, or measurement error range of the object under test can also be used as constraints in the reconstruction process to improve the physical rationality of the reconstruction results.

[0060] Among them, the experimental time-series data of the entire domain of the object under test refers to the experimental time-series data covering the entire spatial region of the object under test or the target analysis region. This experimental time-series data can be a continuous spatial field or a time-series field discretely represented on a high-density grid or sampling nodes. For example... Figure 2 and Figure 3 As shown, Figure 2 The data records experimental data (physical quantity values) collected from multiple sampling points on the object under test at a single moment. The X and Y coordinates represent the horizontal and vertical coordinates of each sampling point, and different colors represent the physical quantity value (amplitude) collected at that sampling point. Figure 3 The X and Y coordinates represent the horizontal and vertical coordinates of each point in the global domain, with different colors indicating the physical quantity values ​​at those points. Figure 2 Compared to the experimental data shown, which only includes a few location points, Figure 3 The experimental time series data shown can reflect the distribution of target physical quantities of the object under test in the area where no sensors are deployed.

[0061] In some application scenarios, a physical information neural network can be used to reconstruct the experimental time-series data of the entire test object. Specifically, spatial coordinates and time are used as network inputs, and the target physical quantity is used as the network output; the experimental time-series sub-data at the collected location points are used as data fitting constraints; the residuals of the physical conservation equations are used as physical regularization terms; and the physical information neural network is trained by minimizing the reconstruction loss function to obtain a reconstruction model that can output the experimental time-series data of the entire test object.

[0062] In some application scenarios, Bayesian skrygian variants or other physically constrained interpolation algorithms can be used for reconstruction. Specifically, experimental time series data from several location points are used as observation points, and physical conservation equations or physical priors are introduced as covariance structures, trend terms, or regularization terms to solve for the experimental time series data and its confidence level of the entire domain of the object under test.

[0063] In the above scheme, by constructing a reconstruction loss function that includes the physical conservation equation as a regularization term and using the collected experimental time series sub-data as constraints, sparse experimental data can be reconstructed into experimental time series data that characterizes the entire domain of the object under test, thereby reducing the information loss caused by the sparse sensor layout. At the same time, the physical conservation equation can suppress and smooth experimental noise, improving the reliability of amplitude distribution differences, spatial gradient differences, and frequency domain feature differences in subsequent parameter calibration.

[0064] In some embodiments, step 103 above may include the following steps: First, a first error term is determined based on the difference in the position of the resonant peak in the frequency domain characteristics, the difference in gain and / or phase at several sampling frequencies, and a second error term is determined based on the difference in amplitude distribution between experimental time series data and simulated time series data.

[0065] The difference in resonance peak position refers to the difference in the frequency positions of the main peaks in the frequency domain response curves of the first and second frequency domain characteristics. The resonance peak position can reflect the inherent dynamic response characteristics of the test object under target operating conditions. For example, in thermal response experiments, the resonance peak position can reflect the response frequency band of the test object to periodic heat input; in fluid dispensing experiments, the resonance peak position can reflect the response frequency band of the colloid to periodic velocity or pressure disturbances.

[0066] The gain difference at several sampling frequencies refers to the difference in gain amplitude between the first and second frequency domain characteristics at one or more sampling frequencies. The gain spectrum characterizes the ability of the object under test to amplify or attenuate input disturbances. For example, in a glass heating system, when a periodic power disturbance is applied, the gain of the temperature response at the corresponding frequency reflects the thermal conductivity and thermal inertia characteristics; in a dispensing system, when a speed pulse is applied to the dispensing valve, the gain of the adhesive path height response at the corresponding frequency reflects the amplification or attenuation law of the colloidal rheological parameters to external disturbances.

[0067] The phase difference refers to the difference in phase lag between the first and second frequency domain features at one or more sampling frequencies. Phase characterizes the degree of lag in the output response of the object under test relative to the input disturbance.

[0068] For example, peak searches can be performed on the first frequency domain feature and the second frequency domain feature respectively to determine the position of the first resonant peak in the first frequency domain feature and the position of the second resonant peak in the second frequency domain feature, and the distance between the first resonant peak position and the second resonant peak position can be used as the resonant peak position difference. Alternatively, the gain value and phase value corresponding to the first frequency domain feature and the second frequency domain feature can be extracted from a preset set of sampling frequencies, and the gain difference and phase difference can be calculated respectively. The preset set of sampling frequencies may include the fundamental frequency, harmonics, or frequencies determined by engineering experience of the controlled dynamic disturbance signal.

[0069] In some applications, the first error term can be determined based on at least one of the differences in resonant peak position, gain, and phase. The second error term is determined based on the difference in amplitude distribution between experimental and simulated time-series data. This difference in amplitude distribution may include, but is not limited to, differences between experimental and simulated time-series data in terms of time-domain amplitude magnitude, amplitude variation trend, peak value, steady-state value, response rise process, or response decay process.

[0070] Then, by combining the first and second error terms, the target loss is obtained.

[0071] For example, the target loss can be obtained by weighted summation of the first and second error terms. Optionally, the weights of the first and second error terms can be fixed or adjusted according to the convergence state during the iterative optimization process.

[0072] In the above embodiments, by introducing the differences in resonance peak position, gain, and / or phase into the first error term and jointly determining the target loss with the second error term characterizing the differences in time-domain amplitude distribution, the parameter calibration process is simultaneously constrained by dynamic response characteristics and amplitude prediction accuracy. This effectively alleviates the problem of strong coupling of the parameters to be calibrated caused by relying solely on static amplitude fitting and improves the accuracy of the parameter set obtained from calibration.

[0073] In some embodiments, experimental time-series data includes experimental time-series sub-data of multiple locations on the object under test, and simulation time-series data includes simulation time-series sub-data of multiple locations. These locations can be sensor mounting positions on the object under test, feature positions obtained from image acquisition, laser profilometer sampling positions, simulation mesh node positions, simulation cell center positions, or spatial sampling positions pre-defined according to the geometry of the object under test. Multiple locations can be sparsely distributed on the object under test or densely distributed in local key areas. For example, in a glass heating system, multiple locations can be multiple thermocouple placement positions, multiple pixel positions in an infrared thermal image, or multiple sampling nodes on the glass surface; in a dispensing system, multiple locations can be multiple cross-sectional positions along the length of the adhesive path or multiple profile sampling points on the cross-section of the adhesive path. Experimental time-series sub-data can represent the experimental results of the target physical quantity changing over time at that location point. For example, a temperature-time curve, displacement-time curve, or adhesive path height-time curve at a certain location point. Simulation time-series sub-data refers to the simulation results of the target physical quantity changing over time output by the simulation model at the location point corresponding to the experimental time-series sub-data.

[0074] Based on this, the second error term, determined by the difference in amplitude distribution between experimental time-series data and simulated time-series data, includes: First, based on the difference in amplitude between the experimental time series data and the simulated time series data at each location point, the amplitude error term is obtained.

[0075] Optionally, to improve the accuracy of the amplitude error term calculation, the experimental time series data and the simulation time series data can be time-aligned, spatially mapped, and have their sampling frequencies unified. For example, if the experimental sampling location does not completely coincide with the simulation grid node, the simulation time series data of the corresponding location point can be obtained from the simulation model output results through interpolation.

[0076] This process involves first acquiring the amplitude differences between multiple locations at various time points in the experimental and simulated time-series data. Then, statistical analysis of these amplitude differences yields a sub-amplitude error term. Finally, these sub-amplitude error terms are accumulated over time to obtain the final amplitude error term. The sub-amplitude error term can be obtained by summing or weighted summing multiple amplitude differences. In some applications, higher weights can be assigned to critical locations of greater engineering interest. For example, areas with dense silver lines, large temperature gradients, or sensitive touch areas in glass heating systems can be assigned higher weights; similarly, areas prone to adhesive breakage, overflow, or wall-climbing defects in dispensing systems can be assigned higher weights.

[0077] Secondly, the first spatial gradient data can be obtained by differentiating the experimental time-series data, and the second spatial gradient data can be obtained by differentiating the simulation time-series data. Specifically, the first spatial gradient data is obtained based on the magnitude difference between the experimental time-series sub-data at each adjacent location point, and the second spatial gradient data is obtained based on the magnitude difference between the simulation time-series sub-data at each adjacent location point. The first spatial gradient data can be the temporal variation of the spatial gradient at each location point in the experimental time-series data, and the second spatial gradient data can be the temporal variation of the spatial gradient at each location point in the simulation time-series data.

[0078] For example, at each time step, the amplitude difference corresponding to that time step in the experimental time series sub-data of adjacent locations is acquired, and the ratio of this amplitude difference to the spatial distance between the two locations is used as the first spatial gradient at that time step. Combining the first spatial gradients at each time step yields the first spatial gradient data. In the heat conduction scenario, the first spatial gradient data reflects the spatial temperature gradient of the experimental temperature field, thereby further indicating the direction of heat flow transfer. In the dispensing scenario, the first spatial gradient data reflects the changing trend of the adhesive path height along the spatial direction, thereby indicating the edge morphology and cross-sectional slope of the adhesive path. The calculation method for the second spatial gradient data can be consistent with that for the first spatial gradient data, and will not be elaborated here.

[0079] The adjacent locations can be determined based on the geometric adjacency of the object under test, the mesh topology, the sampling point arrangement order, or a preset neighborhood radius. For example, in a two-dimensional glass surface, the closest location in the horizontal or vertical direction can be used as the adjacent location; in a glue path section, adjacent sampling points on the section contour line can be used as adjacent location points; in a three-dimensional simulation mesh, mesh nodes sharing edges or surfaces can be used as adjacent location points.

[0080] The spatial gradient error term is determined based on the first and second spatial gradient data. For example, the spatial gradient differences between the first and second spatial gradient data for multiple locations at each time point are first obtained. Then, the differences in spatial gradients are statistically analyzed to obtain sub-error terms at each time point. Finally, the sub-error terms are accumulated over time to obtain the spatial gradient error term. The spatial gradient error term may include differences in gradient magnitude, gradient direction, or a combination of both. For example, the Euclidean distance between the first and second spatial gradient data can be calculated, or the angle difference between them can be calculated. Figure 3 Based on this, the spatial gradient difference between experimental and simulation data at a single time point can be as follows: Figure 4 As shown, Figure 4 In the diagram, the X and Y coordinates represent the horizontal and vertical coordinates of each point in the global domain, with different colors indicating spatial gradient differences at corresponding locations. For heat conduction problems, the spatial gradient error term can be used to ensure that the heat flow path in the simulation model matches the experimental data; for dispensing problems, the spatial gradient error term can be used to ensure that the edge morphology and height variation trend of the dispensing path in the simulation model match the experimental data.

[0081] Then, by combining the magnitude error term and the spatial gradient error term, the second error term is obtained.

[0082] For example, the magnitude error term and the spatial gradient error term can be weighted to obtain a second error term.

[0083] In the above scheme, by simultaneously introducing an amplitude error term and a spatial gradient error term into the second error term, the parameter calibration process not only focuses on the numerical errors at multiple location points, but also on the change path and distribution pattern of the target physical quantity of the object under test in the spatial direction. This improves the simulation model's ability to reproduce the evolution mechanism of the real physical field and reduces the risk of similar numerical values ​​but inconsistent physical mechanisms.

[0084] As described above, the target loss is determined based on a first error term determined by differences in frequency domain features and a second error term determined by differences in amplitude distribution. In some embodiments, step 104 may include, for example... Figure 5 The following steps are shown: Step 201: Obtain the difference between the first error terms during several rounds of iterative optimization to obtain the first change value; and obtain the difference between the second error terms during several rounds of iterative optimization to obtain the second change value.

[0085] The iterative optimization process can be a series of consecutive iterative optimizations preceding the current iteration, or it can include the current iteration. For example, the most recent 2, 3, 5, or other number of iterative optimization processes can be selected. The specific number of iterations chosen can be determined based on the convergence speed of the target loss, the simulation computation cost, and the dimensionality of the set of parameters to be calibrated.

[0086] The first change value can be the difference between the first error terms in two adjacent rounds, the decrease in the first error terms over several rounds, the average rate of change of the first error terms over several rounds, or the degree of deviation of the first error term in this round of iterative optimization from the minimum value of the first error term in several rounds. That is, the first change value can reflect the convergence state or sensitivity of the frequency domain response matching process. The second change value can be the difference between the second error terms in two adjacent rounds, the decrease in the second error terms over several rounds, the average rate of change of the second error terms over several rounds, or the degree of deviation of the second error term in this round of iterative optimization from the minimum value of the second error term in several rounds. That is, the second change value can reflect the convergence state or sensitivity of the amplitude distribution matching process.

[0087] Step 202: Determine the adjustment strategy for the set of parameters to be calibrated based on the relationship between the first change value and the second change value.

[0088] The adjustment strategy may include one or more of the following: prioritization of parameter adjustments, adjustment of parameter search step size, adjustment of parameter search range, and selection of optimization direction. For example, when the first change value is greater than the second change value, it can be considered that the frequency domain response-related error needs to be improved more in the current stage, and the parameter search step size related to the frequency domain response can be increased. Or, when the second change value is greater than the first change value, it can be considered that the time domain amplitude-related error needs to be improved more in the current stage, and the parameter search step size related to the amplitude distribution can be increased.

[0089] As mentioned above, the second error term may include an amplitude error term and a spatial gradient error term, and the second change value may include the change value of the amplitude error term and / or the change value of the spatial gradient error term. Therefore, the adjustment strategy for the set of parameters to be calibrated can be determined based on the relationship between the magnitudes of the change values ​​of the amplitude error term, the spatial gradient error term, and the first change value.

[0090] Step 203: Adjust the set of parameters to be calibrated according to the adjustment strategy to obtain the set of parameters to be calibrated for the next round of iterative optimization.

[0091] For example, in the parameter calibration process of an automotive glass heating system, the first error term reflects the difference in dynamic response characteristics between experimental temperature time-series data and simulated temperature time-series data, while the second error term reflects the difference in temperature amplitude. If, during several rounds of iterative optimization, the first change value is significantly greater than the second change value, it indicates that the error change related to dynamic response characteristics is more prominent, and the adjustment of the parameters to be calibrated that affect the thermal inertial response can be strengthened in the next round of iterative optimization. Conversely, if the second change value is significantly greater than the first change value, it indicates that the error change related to temperature amplitude matching is more prominent, and the adjustment of the parameters to be calibrated that affect the temperature amplitude distribution can be strengthened in the next round of iterative optimization.

[0092] For example, in the parameter calibration process of the adhesive dispensing process for mobile phone frames, the first error term reflects the difference in frequency domain dynamic response between experimental adhesive path height time-series data and simulated adhesive path height time-series data, while the second error term reflects the difference in adhesive path height amplitude. By comparing the first and second changes, it is possible to determine whether the current iterative optimization should focus more on improving the prediction of adhesive dynamic response or adhesive path height amplitude, thereby updating the set of parameters to be calibrated accordingly.

[0093] In the above embodiments, the changes of the first error term and the second error term are monitored in several rounds of iterative optimization, and the adjustment strategy of the set of parameters to be calibrated is determined according to the relationship between the first change value and the second change value. This enables the parameter calibration process to adaptively adjust the optimization direction according to the actual convergence state, thereby improving the update efficiency and calibration stability of the set of parameters to be calibrated.

[0094] In some embodiments, the adjustment strategy includes the adjustment priority of each parameter in the set of parameters to be calibrated. The adjustment priority indicates the degree to which each parameter is prioritized or searched in the next round of iterative optimization. A higher adjustment priority results in a larger adjustment magnitude, search step size, perturbation intensity, or search range for that parameter; conversely, a lower adjustment priority results in a smaller adjustment magnitude, search step size, perturbation intensity, or search range. In some application scenarios, parameters with low adjustment priority can remain unchanged in the current round of iterative optimization, while only high-priority parameters are adjusted.

[0095] The above-mentioned method of determining the adjustment strategy of the set of parameters to be calibrated based on the relationship between the first change value and the second change value can be as follows: take the error term corresponding to the larger value of the first change value and the second change value as the target error term; set the adjustment priority of the parameters to be calibrated in the set of parameters to be calibrated that are related to the target error term as the highest priority; wherein, the adjustment range of each parameter to be calibrated in the set of parameters to be calibrated is positively correlated with its adjustment priority.

[0096] For example, if the first change value is greater than the second change value, then the target change value is the first change value, indicating that the first error term needs more attention at the current stage; if the second change value is greater than the first change value, then the target change value is the second change value, indicating that the second error term needs more attention at the current stage.

[0097] The correlation refers to the influence relationship between the parameter to be calibrated and the error term corresponding to the first or second change value. The correlation can be established in advance based on the physical mechanism, or it can be determined based on sensitivity analysis, historical calibration data, expert experience, parameter perturbation experiments, gradient information, or surrogate model analysis results.

[0098] In some application scenarios, if the target change value is the first change value, the adjustment priority of the parameter to be calibrated associated with the first error term can be set to the highest priority. For example, the parameter to be calibrated that is strongly related to the frequency domain response characteristics can include parameters that affect the system's inertia, hysteresis, damping, or dynamic transfer characteristics. For instance, in thermal response experiments, specific heat capacity and thermal inertia-related parameters may have a strong correlation with the frequency domain response; in colloidal flow experiments, viscosity and rheological index may have a strong correlation with the frequency domain response.

[0099] In some applications, if the target change value is the second change value, the adjustment priority of the parameter to be calibrated associated with the second error term can be set to the highest priority. Parameters to be calibrated that are strongly correlated with the time-domain amplitude distribution can include parameters that affect the steady-state amplitude, spatial amplitude distribution, or overall response intensity. For example, in a glass heating system, thermal conductivity and boundary heat transfer coefficient may be strongly correlated with the temperature amplitude distribution; in a dispensing process, contact angle, surface tension coefficient, and geometric parameters may be strongly correlated with the amplitude distribution of the adhesive path cross-section.

[0100] As mentioned above, the adjustment range of each parameter in the set of parameters to be calibrated is positively correlated with its adjustment priority. This adjustment range can include the search step size. That is, the higher the adjustment priority of a parameter, the larger the adjustment range; and the lower the adjustment priority, the smaller the adjustment range. For example, a larger search step size can be set for the highest priority parameter, a medium search step size for the medium priority parameter, and a smaller search step size for the lower priority parameter. Alternatively, the search range can be expanded for the highest priority parameter, while maintaining the original search range or narrowing the search range for the lower priority parameters.

[0101] Taking an automotive glass heating system as an example, if the set of parameters to be calibrated includes silver wire thermal conductivity and glass specific heat capacity, and the current first change value is large, then the adjustment priority of the glass specific heat capacity related to the first error term can be set to the highest priority, so that the search step size of the glass specific heat capacity is relatively larger; if the current second change value is large, then the adjustment priority of the silver wire thermal conductivity related to the second error term can be set to the highest priority, so that the search step size of the silver wire thermal conductivity is relatively larger.

[0102] Alternatively, taking the dispensing process of a mobile phone frame as an example, if the set of parameters to be calibrated includes adhesive viscosity, power law exponent, and contact angle, and the current first change value is large, then the adjustment priority of adhesive viscosity or power law exponent related to the first error term can be set to the highest priority; if the current second change value is large, then the adjustment priority of contact angle related to the second error term can be set to the highest priority.

[0103] In the above embodiments, by taking the error term corresponding to the larger of the first and second change values ​​as the target error term, and increasing the adjustment priority of the parameters to be calibrated that are related to the target error term, the update of the set of parameters to be calibrated is no longer an indiscriminate search, but can make targeted adjustments to different parameters to be calibrated according to the source of error. This can improve the decoupling ability of strongly coupled parameters to be calibrated, reduce the number of invalid searches, and accelerate the convergence of the target loss.

[0104] In some application scenarios, during the initial iterative optimization phase of step 104 above, several candidate parameters can be selected from the search space of the set of parameters to be calibrated, and a high-fidelity simulation model can be invoked to obtain the corresponding simulation time-series data and second frequency domain features. Based on the candidate parameters and the corresponding target loss, a surrogate model can be trained. The surrogate model can be a Gaussian process regression model, a radial basis function model, a neural network surrogate model, or other response surface models. The surrogate model is used to quickly predict the target loss or frequency domain feature differences under different sets of parameters to be calibrated.

[0105] In the middle or late stages of iterative optimization, a surrogate model can be used to quickly screen parameter regions that may have lower target losses. Then, only a small set of candidate parameters to be calibrated within that parameter region can be accurately verified using a high-fidelity simulation model. This significantly reduces the number of times the high-fidelity simulation model can be called.

[0106] In addition, when the difference in frequency domain features or the decrease in target loss are small in several consecutive rounds of iterative optimization, it indicates that the optimization process may be trapped in a local optimum. In this case, the search space of the set of parameters to be calibrated can be expanded, or random perturbations can be introduced into some of the parameters to be calibrated, so that the optimization process can jump out of the local optimum region.

[0107] After obtaining the parameter set of the object under test, experiments can be conducted on the object under different environmental conditions to obtain corresponding experimental time series data. The simulation model performs simulations based on the obtained parameter set under corresponding environmental conditions, which can verify whether the simulation results of the parameter set under different environmental conditions are consistent with the experimental results of the object under test. For example, experiments and simulations can be conducted at different ambient temperatures to verify whether the parameter set is accurate.

[0108] Finally, the calibrated parameter set and simulation model can be output. After outputting the parameter set, based on the measurement error range of the pre-calibrated experimental time series sub-data and the confidence level of the reconstructed experimental time series data, the confidence interval of each parameter in the parameter set can be calculated using polynomial chaotic expansion, and the prediction uncertainty of the calibrated simulation model under unknown disturbance conditions can be evaluated.

[0109] In some application scenarios, the object under test is a car B-pillar glass heating system, and the set of parameters to be calibrated includes the thermal conductivity of the silver wire and the specific heat capacity of the glass. In a car B-pillar glass heating system, the glass surface is equipped with silver wire structures for heating, defrosting, or defogging. Due to differences in the thickness of the silver wire printing, batches of silver paste materials, glass lamination structures, and boundary heat dissipation conditions, directly using empirical parameters provided by the supplier to build a simulation model may result in an inaccurate prediction of the temperature distribution on the glass surface. Especially in areas with dense silver wires, the traditional method of fitting using only a few thermocouple temperature points is prone to local temperature field prediction distortion, and the thermal conductivity of the silver wire and the specific heat capacity of the glass are strongly coupled in the static temperature response, making unique determination difficult. To address this, the parameter calibration method for the object provided in this invention can first construct an initial high-fidelity simulation model of the object under test and define the set of parameters to be calibrated.

[0110] Specifically, a heat conduction simulation model of the automotive B-pillar glass heating system is established. This simulation model includes the glass substrate, the silver wire heating region, the boundary heat transfer region, and the electrical power input boundary. The set of parameters to be calibrated may include the thermal conductivity of the silver wire and the specific heat capacity of the glass. The set of parameters to be calibrated in the first round of iterative optimization can be derived from material handbooks provided by material suppliers, public databases, or historical project experience values, and upper and lower bounds and initial values ​​for the thermal conductivity of the silver wire and the specific heat capacity of the glass are set.

[0111] Secondly, experimental tests were conducted under the target operating conditions, and experimental time series data were collected from multiple location points.

[0112] Specifically, the automotive B-pillar glass heating system was placed in a temperature-controlled experimental environment, and heating tests were conducted on the system under basic heating input. A small sinusoidal power disturbance can be superimposed on the basic heating input; for example, a small periodic power fluctuation with a frequency of 0.1Hz can be superimposed near the basic heating power. This sinusoidal power disturbance can be achieved by modulating the input power or equivalent duty cycle through a power controller. In other words, the glass heating system, under the target operating condition, exhibits both basic heating and a small dynamic disturbance input.

[0113] During the experimental testing, thermocouples can be placed at multiple key locations on the glass surface to collect temperature time-series data from various points. Key locations may include areas with dense silver lines, glass edges, areas with large temperature gradients, and functional areas of interest to the user. If an infrared thermal imager is used, temperature time-series image data over a larger area of ​​the glass surface can also be acquired.

[0114] Then, the physical field is reconstructed from the sparse experimental time series data.

[0115] Specifically, a reconstruction loss function can be constructed that includes the residuals of the heat conduction equation as a regularization term. Temperature time-series data collected by each thermocouple can be used as physical constraints, and a physical information neural network can be used to solve for the global temperature time-series data of the glass surface. This global temperature time-series data is used to characterize the actual temperature evolution process of the glass surface during experimental testing. In this way, temperature data from only a few measuring points can be expanded into a continuous or high-density experimental temperature field.

[0116] In addition, simulation calculations are performed and simulation timing data is obtained.

[0117] Specifically, the set of parameters to be calibrated for this round of iterative optimization is input into the heat conduction simulation model of the automotive B-pillar glass heating system, and the same basic heating input and sinusoidal power perturbation as in the experimental test are applied to the power input end of the simulation model. After the simulation model runs, the simulation temperature time series data of each corresponding point on the glass surface or the global grid nodes are output.

[0118] Then, frequency domain transformation and target loss calculation are performed.

[0119] Specifically, frequency domain transformation is performed on the experimental temperature time-series data and the simulated temperature time-series data to obtain a first frequency domain feature and a second frequency domain feature. The first and second frequency domain features may include the gain and phase of each location point at a 0.1Hz perturbation frequency. The gain at each location point at the 0.1Hz perturbation frequency can constitute a gain spectrum, and the phase at each location point at the 0.1Hz perturbation frequency can constitute a phase spectrum. Alternatively, in other embodiments, frequency domain response features at other sensitive frequencies may also be included. A first error term is determined based on the gain and phase differences between the first and second frequency domain features. An amplitude error term is determined based on the magnitude difference between the experimental and simulated temperature time-series data; a spatial gradient error term is determined based on the spatial gradient difference between the experimental and simulated temperature fields; and the amplitude error term and the spatial gradient error term are combined to obtain a second error term. The target loss is further obtained by combining the first and second error terms.

[0120] Adjust the set of parameters to be calibrated and output the calibration results.

[0121] Specifically, differential evolution algorithm, Bayesian optimization algorithm or other adaptive optimization algorithm can be used to adjust the thermal conductivity of the silver wire and the specific heat capacity of the glass according to the target loss.

[0122] When the target loss is less than a preset loss threshold, or when the number of iterations reaches a preset limit, the iteration optimization stops, and the set of parameters to be calibrated in the last iteration is used as the parameter set for the automotive B-pillar glass heating system. The calibrated parameter set can be used to update the simulation model of the automotive B-pillar glass heating system.

[0123] The calibration parameters are then verified. This allows for the comparison of the temperature field distribution predicted by the simulation model with the experimental data measured by the infrared thermal imager at different power levels. Tests have shown that the parameter calibration method provided by this approach is superior to the traditional method of parameter calibration using sparse points.

[0124] In some application scenarios, the object under test is the adhesive path forming system in the dispensing process of a mobile phone frame, and the set of parameters to be calibrated includes adhesive viscosity, power law exponent, and contact angle.

[0125] In the adhesive dispensing and sealing process of smartphone frames, the adhesive needs to form a continuous and stable adhesive path along a predetermined adhesive groove. During high-speed automated dispensing, defects such as adhesive breakage, overflow, or wall climbing may occur. Adhesive breakage leads to seal failure, overflow may contaminate the appearance or affect subsequent assembly, and wall climbing may cause uneven adhesive path height. Traditional computational fluid dynamics simulation models typically rely on theoretical rheological parameters or empirical values ​​provided by suppliers. However, due to the non-Newtonian fluid properties of adhesives and the micro-roughness and wetting differences on the surface of the smartphone frame, directly using theoretical parameters often fails to accurately reproduce actual defect boundaries. Therefore, the parameter calibration method for objects provided by this invention may include the following steps: First, an initial high-fidelity simulation model of the object under test is constructed, and a set of parameters to be calibrated is defined.

[0126] Specifically, a computational fluid dynamics simulation model is established, including the glue needle, glue tank, middle frame wall, and free interface of the colloid. The simulation model can use the Volume of Fluid (VOF) method or other interface tracking methods to describe the flow and shaping process of the colloid in the glue tank. The set of parameters to be calibrated can include glue viscosity, power law exponent, contact angle, surface tension coefficient, etc. The set of parameters to be calibrated in the first round of iterative optimization can be derived from rheological test data from the glue supplier, publicly available empirical values, or manually preset values.

[0127] Secondly, active perturbation experiments were conducted under the target operating conditions to collect experimental time-series data.

[0128] Specifically, a base dispensing speed and base glue valve pressure are set on the dispensing equipment for the mobile phone frame to bring the dispensing system to the target operating condition. High-frequency micro-velocity pulses can be superimposed on the base dispensing speed. For example, a square wave velocity disturbance with a frequency of 10Hz can be input to the dispensing valve control terminal to cause small periodic changes in the glue flow rate or the speed of the glue needle movement.

[0129] During the experiment, a high-speed camera can be used to acquire images of the adhesive path morphology during the flow of the colloid, and a laser profilometer can be used to collect time-series data on the adhesive path height at multiple locations. These multiple locations can be distributed along the length of the adhesive path or along its cross-sectional direction. In this way, experimental time-series data on the changes in adhesive path height, width, and edge climb height over time can be obtained.

[0130] Next, the experimental time series sub-data was reconstructed based on physical constraints.

[0131] Specifically, a reconstruction loss function containing mass conservation constraints, momentum conservation constraints, or colloidal interface evolution constraints can be constructed. The temporal sub-data of the adhesive path height obtained by high-speed cameras and laser profilometers can be used as physical constraints. By minimizing the reconstruction loss function, the temporal data of the global height or the temporal data of the global interface morphology within the target area of ​​the adhesive path can be solved.

[0132] In addition, simulation calculations are performed and simulation timing data is obtained.

[0133] Specifically, the set of parameters to be calibrated for this round of iterative optimization is input into the computational fluid dynamics simulation model of the dispensing process in the mobile phone frame. The same base velocity and high-frequency micro-velocity pulses as in experimental testing are applied to the inlet velocity boundary or glue valve flow boundary of the simulation model. After the simulation model runs, it outputs simulation timing data such as glue path height, glue path width, and glue interface position.

[0134] Then, frequency domain transformation and target loss calculation are performed.

[0135] Specifically, a Fast Fourier Transform or other frequency domain transformation is performed on the experimental and simulated adhesive path height time-series data to obtain the first frequency domain characteristics on the experimental side and the second frequency domain characteristics on the simulated side. The first and second frequency domain characteristics can include the gain and phase spectra at the 10Hz velocity pulse frequency, or the response characteristics at harmonics. A first error term is determined based on the gain difference, phase difference, or resonance peak position difference between the first and second frequency domain characteristics. An amplitude error term is determined based on the amplitude difference between the experimental and simulated adhesive path height time-series data; a spatial gradient error term is determined based on the spatial gradient difference between the experimental and simulated adhesive path cross-sectional shapes; the amplitude error term and the spatial gradient error term are then combined to obtain the second error term. The target loss is then obtained by further combining the first and second error terms.

[0136] Next, adjust the set of parameters to be calibrated and output the calibration results.

[0137] Specifically, Bayesian optimization algorithms, differential evolution algorithms, or other adaptive optimization algorithms can be used to adjust the glue viscosity, power law exponent, and contact angle.

[0138] Once the target loss meets the convergence condition, the calibrated set of parameters to be calibrated is output and used to update the simulation model of the dispensing process for the mobile phone frame. The updated simulation model can be used to predict the risks of glue breakage, glue overflow, or wall climbing under different dispensing speeds, temperatures, and glue tank sizes.

[0139] The output calibration parameters are then verified, which can be performed under extreme conditions of low temperature and high speed. The verification results show that the simulation results accurately predicted the wall climbing height and the glue breakage event.

[0140] In this embodiment, by reconstructing experimental time series sub-data and applying gradient constraints at each sampling point, the model is ensured to be not only numerically accurate, but also to have a physical field distribution (such as heat flow path) consistent with the real situation. This makes the simulation model a reliable digital twin of the object under test, improving the predictive ability of the simulation model under unknown working conditions.

[0141] In one application scenario, taking the multiphysics parameter calibration of a car rear windshield heating system as an example, the object under test is the rear windshield of the heating system, and its heating circuit uses iron-chromium-aluminum alloy wire laid using a fused wire process. Considering the high coupling between the temperature coefficient of resistance of the alloy wire and the contact thermal resistance of the glass in static steady-state temperature data, traditional methods struggle to distinguish whether the heat transfer is hindered due to uneven heating of the alloy wire itself or poor contact between the alloy wire and the glass. Furthermore, the thermal expansion coefficients of the metal alloy wire and the glass differ significantly. If the contact stiffness or interface bonding strength parameters in the simulation model are inaccurate, the simulation cannot accurately predict the risk of microcracks during the heating process.

[0142] In current common tests, only a small number of thermocouples are placed at the edge of the glass. This method cannot capture the local overheating at the intersection of heating wires and the resulting heat flow distortion.

[0143] In this embodiment, a multilayer coupled high-fidelity simulation model comprising a glass substrate, an adhesive layer, and a metal alloy wire is first constructed. A parameter vector to be calibrated is defined, specifically including the following three key parameters: the temperature coefficient of resistance of the alloy wire (affecting the characteristic of heat generation changing with temperature), the contact thermal resistance between the metal alloy wire and the glass interface (affecting the efficiency of heat diffusion from the wire to the glass), and the equivalent shear stiffness at the interface (affecting the distribution of thermal stress, indirectly affecting the change in contact state caused by local deformation). Simultaneously, search upper and lower bounds and initial values ​​are set for each parameter, where the initial values ​​can be set to nominal values ​​from the material handbook.

[0144] Next, experimental data acquisition and physical field reconstruction were performed. The experiment was designed as follows: the rear windshield was placed in a constant temperature chamber. A rated voltage was applied for heating, and a dual-frequency controlled dynamic disturbance signal was superimposed before steady state: the fundamental frequency was a sinusoidal power fluctuation used to excite thermal inertia to distinguish heat capacity-related parameters; the high frequency was a small-amplitude current pulse used to excite the system's rapid thermal response to distinguish contact thermal resistance from volume thermal conductivity. Then, temperature time-series data of sparse points were collected using embedded thermocouples and surface infrared thermometers. Through an inversion model that included the energy conservation equation (considering the Joule heat source term) and the heat conduction equation, the sparse temperature measurement point data were used as hard constraints, and the heat flux continuity equation was introduced as a regularization term. The reconstructed experimental time-series data was output, which could include a continuous temperature data field across the entire surface. The heat flux density vector field was then determined using this experimental time-series data.

[0145] Next, considering that high-frequency phase lag is extremely sensitive to contact thermal resistance, while low-frequency gain is more sensitive to the temperature coefficient of resistance, the temperature gain amplitude and phase lag information at the fundamental frequency and high frequency can be extracted during the frequency domain analysis of the reconstructed field.

[0146] During the simulation, the current parameter vector is input into the simulation model, such as a multiphysics solver (coupled electric field, thermal field), and the same voltage perturbation signal as in the experimental design is applied. The simulated predicted temperature data field and the corresponding heat flux field are calculated. The temperature gain amplitude and phase hysteresis information under the simulation conditions are extracted.

[0147] Then, a composite loss function is constructed to quantify the difference between the reconstructed experimental field and the simulated predicted field. This difference is composed of a weighted sum of an amplitude error term, a spatial gradient error term, and a frequency domain error term (i.e., the first error term mentioned above). The amplitude error term is obtained by calculating the root mean square error between the reconstructed temperature data field and the simulated temperature data field, reflecting the deviation in the overall temperature level. The spatial gradient error term introduces a heat flow path consistency constraint and is determined by calculating the difference in heat flow convergence patterns near the heating wire intersection nodes between the reconstructed experimental field and the simulated field. If the heat flow is linearly conducted in the simulation but diffuses in the experiment, it indicates a large deviation in the contact thermal resistance parameter, and this error will be significant. The frequency domain error term is obtained by calculating the Euclidean distance or correlation coefficient between the gain amplitude and phase lag information obtained from frequency domain analysis of the experimental time-series data and the temperature gain amplitude and phase lag information under simulation. Considering that high-frequency phase matching mainly constrains contact thermal resistance and low-frequency gain matching mainly constrains the resistance temperature coefficient, decoupling the gain amplitude and phase lag information allows for more targeted subsequent parameter adjustments.

[0148] Then, an adaptive Bayesian optimization algorithm is used for parameter iteration. If high-frequency phase error is found to dominate in the dynamic fingerprint matching term, the resistance temperature coefficient is automatically locked, the search step size of the contact thermal resistance is significantly reduced, and the interface thermal resistance is calibrated first. If the heat flow pattern at the intersection is found to be mismatched in the spatial gradient error term, the interface shear stiffness is adjusted to correct the contact area change model caused by thermal deformation.

[0149] It can also be combined with surrogate models to assist in adjusting parameters. For example, Gaussian process regression can be used to establish the mapping relationship between the above-mentioned information such as gain amplitude and phase lag and parameters, and quickly eliminate parameter combinations that cause simulation divergence (such as numerical instability caused by contact separation).

[0150] After multiple iterations, once the composite loss function converges to the preset accuracy threshold, the calibrated temperature coefficient of the alloy wire resistance, average contact thermal resistance, and interface equivalent shear stiffness distribution map are output.

[0151] Then, comparative experiments were conducted under extreme low-temperature start-up conditions. The calibrated simulation model showed a high degree of overlap between the predicted location of the maximum thermal stress and the measured initiation point of the glass microcracks, while the uncalibrated simulation model misjudged the high-stress zone. Simultaneously, the temperature field prediction error was reduced from a large percentage to a smaller percentage, significantly improving the accuracy of the simulation model.

[0152] By introducing spatial gradient residuals and dynamic fingerprints, the difficult-to-measure thermal resistance of the wire-glass contact was successfully decoupled, solving the unique contact interface problem of metal wires, which is something that traditional point fitting cannot achieve.

[0153] Furthermore, by calibrating the interface stiffness parameters, the thermal stress coupling simulation becomes more realistic, enabling more accurate early warning of the risk of glass thermal shock breakage caused by metal heating wires.

[0154] Furthermore, the parameter calibration method for objects provided by this invention is not only applicable to metal alloy wires with high resistivity, but can also be extended to the parameter calibration of other nonlinear heating elements such as carbon fiber heating films.

[0155] The object parameter calibration device provided by this invention will be described below. The object parameter calibration device described below can be referred to in correspondence with the object parameter calibration method described above. The object parameter calibration device is used to iteratively optimize the set of parameters to be calibrated until a preset cutoff condition is met, and the set of parameters to be calibrated targeted in the last round of iteration optimization is taken as the parameter set of the object to be tested. See also... Figure 6 and Figure 7 The object parameter calibration device 300 includes the following modules: The data acquisition module 10 is used to acquire the experimental time series data of the test object under the target working condition and the simulation time series data obtained by the simulation model based on the set of calibration parameters of the test object in this round of iterative optimization. The frequency domain feature acquisition module 20 is used to perform frequency domain conversion on experimental time series data and simulation time series data to obtain the first frequency domain feature corresponding to the experimental time series data and the second frequency domain feature corresponding to the simulation time series data. The target loss determination module 30 is used to determine the target loss based on the frequency domain feature difference between the first frequency domain feature and the second frequency domain feature, as well as the amplitude distribution difference between the experimental time series data and the simulation time series data. The parameter adjustment module 40 is used to adjust the set of parameters to be calibrated using the target loss, so as to obtain the set of parameters to be calibrated for the next round of iterative optimization. In this process, after the iterative optimization stops, the set of parameters to be calibrated in the last round of iterative optimization is used as the parameter set of the object to be tested.

[0156] According to the object parameter calibration device 300 provided by the present invention, the target loss determination module 30 includes: Error term determination module 31 is used to determine a first error term based on the difference in the position of the resonant peak in the frequency domain characteristic difference, the gain difference and / or phase difference at several sampling frequencies, and to determine a second error term based on the difference in amplitude distribution between experimental time series data and simulated time series data. Error term combination module 32 is used to combine the first error term and the second error term to obtain the target loss.

[0157] According to the parameter calibration device 300 for an object provided by the present invention, the experimental time series data includes experimental time series sub-data of multiple location points in the object under test, the simulation time series data includes simulation time series sub-data of multiple location points, and the error term determination module 32 includes: The amplitude error term determination module 321 is used to obtain the amplitude error term based on the difference in amplitude between the experimental time series sub-data and the simulation time series sub-data at each location point; The spatial gradient error term determination module 322 is used to obtain first spatial gradient data based on the difference in magnitude between experimental time series sub-data at each adjacent location point, obtain second spatial gradient data based on the difference in magnitude between simulation time series sub-data at each adjacent location point, and determine the spatial gradient error term based on the first spatial gradient data and the second spatial gradient data. The second error term determination module 323 is used to combine the amplitude error term and the spatial gradient error term to obtain the second error term.

[0158] According to the object parameter calibration device 300 provided by the present invention, the target loss is determined based on a first error term determined by frequency domain feature differences and a second error term determined by amplitude distribution differences; the parameter adjustment module 40 includes: The change value determination module 41 is used to obtain the difference between the first error terms in several rounds of iterative optimization to obtain the first change value, and to obtain the difference between the second error terms in several rounds of iterative optimization to obtain the second change value; The adjustment strategy determination module 42 is used to determine the adjustment strategy of the set of parameters to be calibrated based on the relationship between the first change value and the second change value. The parameter optimization module 43 is used to adjust the set of parameters to be calibrated according to the adjustment strategy, so as to obtain the set of parameters to be calibrated for the next round of iterative optimization.

[0159] According to the parameter calibration device 300 for an object provided by the present invention, the adjustment strategy includes the adjustment priority of each parameter to be calibrated in the set of parameters to be calibrated, and the adjustment strategy determination module 42 includes: The target error term determination module 421 is used to take the error term corresponding to the larger of the first change value and the second change value as the target error term; The priority determination module 422 is used to set the adjustment priority of the parameters to be calibrated in the set of parameters to be calibrated that are related to the target error term to the highest priority; Among them, the adjustment range of each parameter to be calibrated in the set of parameters to be calibrated is positively correlated with its adjustment priority.

[0160] According to the parameter calibration device 300 for an object provided by the present invention, the data acquisition module 10 includes: The experimental timing data acquisition module 11 is used to acquire experimental timing sub-data of several locations on the test object under the target working condition of applying dynamic disturbance signal to the test object, and to obtain experimental timing data based on the experimental timing sub-data of several locations. The simulation timing data acquisition module 12 is used to input dynamic disturbance signals to the input terminal of the simulation model and then perform simulation calculations to obtain simulation timing data.

[0161] According to the object parameter calibration device 300 provided by the present invention, the experimental time series data acquisition module 11 includes: The reconstruction loss function determination module 111 is used to construct a reconstruction loss function that includes the physical conservation equation as a regularization term; The physical constraint determination module 112 is used to use the collected experimental time series sub-data as physical constraints. The data reconstruction module 113 is used to solve for the experimental time series data representing the entire domain of the object under test by minimizing the reconstruction loss function.

[0162] Figure 8 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 8As shown, the electronic device may include: a processor 510, a communications interface 520, a memory 530, and a communications bus 540, wherein the processor 510, the communications interface 520, and the memory 530 communicate with each other through the communications bus 540. The processor 510 can call logic instructions in the memory 530 to execute a parameter calibration method for an object. This method includes: iteratively executing the following steps until a preset cutoff condition is met, using the set of parameters to be calibrated in the last round of iterative optimization as the parameter set of the object under test: acquiring experimental time-series data of the object under test under target conditions and simulation time-series data obtained by simulating the object under test based on the set of parameters to be calibrated in the current round of iterative optimization; performing frequency domain transformation on the experimental time-series data and the simulation time-series data to obtain a first frequency domain feature corresponding to the experimental time-series data and a second frequency domain feature corresponding to the simulation time-series data; determining a target loss based on the frequency domain feature difference between the first and second frequency domain features and the amplitude distribution difference between the experimental and simulation time-series data; adjusting the set of parameters to be calibrated using the target loss to obtain the set of parameters to be calibrated for the next round of iterative optimization; wherein, after the iterative optimization stops, the set of parameters to be calibrated in the last round of iterative optimization is used as the parameter set of the object under test.

[0163] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0164] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the object parameter calibration method provided by the above methods. The method includes: iteratively executing the following steps until a preset cutoff condition is met, and using the set of parameters to be calibrated targeted by the last round of iterative optimization as the parameter set of the object under test: acquiring experimental time-series data of the object under test under target working conditions and simulation time-series data obtained by the simulation model based on the set of parameters to be calibrated of the object under test in this round of iterative optimization; performing frequency domain transformation on the experimental time-series data and the simulation time-series data to obtain a first frequency domain feature corresponding to the experimental time-series data and a second frequency domain feature corresponding to the simulation time-series data; determining a target loss based on the frequency domain feature difference between the first frequency domain feature and the second frequency domain feature, and the amplitude distribution difference between the experimental time-series data and the simulation time-series data; adjusting the set of parameters to be calibrated using the target loss to obtain the set of parameters to be calibrated targeted by the next round of iterative optimization; wherein, after the iterative optimization stops, the set of parameters to be calibrated targeted by the last round of iterative optimization is used as the parameter set of the object under test.

[0165] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements a parameter calibration method for an object provided by the above-described methods. The method includes: iteratively executing the following steps until a preset cutoff condition is met, and using the set of parameters to be calibrated targeted by the last round of iterative optimization as the parameter set of the object under test: acquiring experimental time-series data of the object under test under target operating conditions and simulation time-series data obtained by the simulation model in this round of iterative optimization based on the set of parameters to be calibrated of the object under test; performing frequency domain transformation on the experimental time-series data and the simulation time-series data to obtain a first frequency domain feature corresponding to the experimental time-series data and a second frequency domain feature corresponding to the simulation time-series data; determining a target loss based on the frequency domain feature difference between the first frequency domain feature and the second frequency domain feature, and the amplitude distribution difference between the experimental time-series data and the simulation time-series data; adjusting the set of parameters to be calibrated using the target loss to obtain the set of parameters to be calibrated targeted by the next round of iterative optimization; wherein, after the iterative optimization stops, the set of parameters to be calibrated targeted by the last round of iterative optimization is used as the parameter set of the object under test.

[0166] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0167] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0168] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for calibrating the parameters of an object, characterized in that, include: The following steps are executed iteratively until a preset cutoff condition is met, and the set of parameters to be calibrated targeted by the last round of iteration optimization is taken as the parameter set of the object under test: The experimental time series data of the object under test under the target working condition and the simulation time series data of the current round of iterative optimization are obtained. The simulation time series data is obtained by the simulation model based on the set of calibration parameters of the object under test. Obtain the first frequency domain feature corresponding to the experimental time series data and the second frequency domain feature corresponding to the simulation time series data; The target loss is determined based on the frequency domain feature differences between the first frequency domain feature and the second frequency domain feature, as well as the amplitude distribution differences between the experimental time series data and the simulation time series data. The set of parameters to be calibrated is adjusted using the target loss to obtain the set of parameters to be calibrated for the next round of iterative optimization.

2. The object parameter calibration method according to claim 1, characterized in that, Based on the frequency domain feature differences between the first and second frequency domain features, and the amplitude distribution differences between the experimental time series data and the simulated time series data, the target loss is determined, including: The first error term is determined based on the difference in the position of the resonant peak in the frequency domain characteristics, the gain difference and / or phase difference at several sampling frequencies, and the second error term is determined based on the difference in amplitude distribution between the experimental time series data and the simulated time series data. The target loss is obtained by combining the first error term and the second error term.

3. The object parameter calibration method according to claim 2, characterized in that, The experimental time-series data includes experimental time-series sub-data from multiple locations on the object under test, and the simulation time-series data includes simulation time-series sub-data from multiple locations. A second error term is determined based on the amplitude distribution difference between the experimental time-series data and the simulation time-series data, including: Based on the magnitude difference between the experimental time series data and the simulated time series data at each of the aforementioned locations, an amplitude error term is obtained. Based on the magnitude difference between experimental time series sub-data of each adjacent location point, a first spatial gradient data is obtained; based on the magnitude difference between simulation time series sub-data of each adjacent location point, a second spatial gradient data is obtained; and a spatial gradient error term is determined based on the first spatial gradient data and the second spatial gradient data. The second error term is obtained by combining the magnitude error term and the spatial gradient error term.

4. The parameter calibration method for an object according to any one of claims 1 to 3, characterized in that, The target loss is determined based on a first error term based on the frequency domain feature differences and a second error term based on the amplitude distribution differences. The target loss is used to adjust the set of parameters to be calibrated, resulting in the set of parameters to be calibrated for the next round of iterative optimization, including: The difference between the first error terms during several rounds of iterative optimization is obtained to obtain a first change value, and the difference between the second error terms during several rounds of iterative optimization is obtained to obtain a second change value; Based on the relationship between the first change value and the second change value, determine the adjustment strategy for the set of parameters to be calibrated; According to the adjustment strategy, the set of parameters to be calibrated is adjusted to obtain the set of parameters to be calibrated for the next round of iterative optimization.

5. The object parameter calibration method according to claim 4, characterized in that, The adjustment strategy includes the adjustment priority of each parameter in the set of parameters to be calibrated. Based on the relationship between the first change value and the second change value, the adjustment strategy for the set of parameters to be calibrated is determined, including: The error term corresponding to the larger of the first change value and the second change value is taken as the target error term; Set the adjustment priority of the parameters in the set of parameters to be calibrated that are related to the target error term to the highest priority; The adjustment range of each parameter in the set of parameters to be calibrated is positively correlated with its adjustment priority.

6. The parameter calibration method for an object according to any one of claims 1 to 3, characterized in that, Acquire experimental time-series data of the object under test under the target working condition and simulation time-series data of this round of iterative optimization, including: The experimental time series data of several locations on the test object are collected during the experimental test under the target working condition of applying a dynamic disturbance signal to the test object, and the experimental time series data is obtained based on the experimental time series data of several locations. The dynamic disturbance signal is input to the input terminal of the simulation model for simulation calculation to obtain the simulation time series data.

7. The object parameter calibration method according to claim 6, characterized in that, The experimental time series data is obtained based on experimental time series sub-data from several location points, including: Construct a reconstruction loss function that includes the physical conservation equations as regularization terms; The collected experimental time series sub-data are used as physical constraints. By minimizing the reconstruction loss function, experimental time-series data characterizing the entire domain of the object under test are obtained.

8. A parameter calibration device for an object, characterized in that, The parameter calibration device for the object is used to iteratively optimize the set of parameters to be calibrated until a preset cutoff condition is met, and the set of parameters to be calibrated targeted in the last round of iterative optimization is taken as the parameter set of the object to be tested. The parameter calibration device for the object includes: The data acquisition module is used to acquire experimental time-series data of the object under test under target working conditions and simulation time-series data of the current round of iterative optimization. The simulation time-series data is obtained by the simulation model based on the set of calibration parameters of the object under test. A frequency domain feature acquisition module is used to acquire the first frequency domain feature corresponding to the experimental time series data and the second frequency domain feature corresponding to the simulation time series data; The target loss determination module is used to determine the target loss based on the frequency domain feature difference between the first frequency domain feature and the second frequency domain feature, and the amplitude distribution difference between the experimental time series data and the simulation time series data; The parameter adjustment module is used to adjust the set of parameters to be calibrated using the target loss, so as to obtain the set of parameters to be calibrated for the next round of iterative optimization.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the object parameter calibration method as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the parameter calibration method for the object as described in any one of claims 1 to 7.

Images