A method for generating a magnetic barkhausen noise simulation based on time-frequency characteristic parameters
By using a simulation generation method based on time-frequency characteristic parameters, the problems of high computational load and low matching degree of traditional models are solved, and high matching degree simulation of magnetic Barkhausen noise signals is achieved, thereby improving the accuracy and efficiency of non-destructive testing.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA JILIANG UNIV
- Filing Date
- 2026-04-29
- Publication Date
- 2026-07-21
AI Technical Summary
In the simulation process of generating magnetic Barkhausen noise signals, the existing technology has a large computational load and is time-consuming, while the JA model has a low matching degree, which makes it difficult to meet the needs of batch simulation and rapid optimization in engineering. In addition, the simulation signal and the measured signal have a low matching degree in the time and frequency domain, resulting in a large error.
A simulation generation method based on time-frequency feature parameters is adopted. By constructing feature parameter vectors and combining multi-objective joint optimization and multi-random seed smoothing mechanism, a high-matching magnetic Barkhausen noise signal is generated, including time-domain envelope and frequency-domain colored noise shaping. Iterative optimization is performed using a multi-dimensional feature synthesis cost function.
The generated simulated signal is highly consistent with the measured signal in the time domain envelope and frequency domain power spectrum, with a correlation of over 0.99, which reduces errors and enhances the application value of the simulated signal in nondestructive testing.
Smart Images
Figure CN122436088A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of nondestructive testing and signal processing technology for materials, specifically to the detection technology of microscopic electromagnetic signals during the manufacturing and service of ferromagnetic materials, and particularly to a method for simulating and generating magnetic Barkhausen noise (MBN) based on time-frequency characteristic parameters. Background Technology
[0002] In the field of existing engineering and manufacturing technology, the surface hardness and residual stress of mechanical parts are key indicators determining their fatigue life. Magnetic Barkhausen noise is a discontinuous, jumping signal generated when ferromagnetic materials are dynamically magnetized under an alternating magnetic field, due to the irreversible displacement of magnetic domain walls within the material being hindered by pinning points. This signal is highly sensitive to changes in the microstructure of ferromagnetic materials and has significant application potential in material hardness assessment and non-destructive testing.
[0003] Currently, existing technologies for simulating MBN signals mainly rely on the Ising model or the Jiles-Atherton (JA) model. However, these traditional models have significant limitations in practical engineering applications: First, the micro-spin-based Ising model relies on a large state-space iteration, and its computational load increases dramatically with the increase of the grid parameter N, resulting in extremely high computational costs and making it difficult to meet the needs of large-scale engineering simulations and rapid optimization. Second, although the macroscopic phenomenological Jiles-Atherton model reduces computational complexity to some extent, the simulated signal it generates has poor matching with the measured MBN signal. It often fails to simultaneously match the temporal envelope shape and the frequency domain micro-energy distribution, resulting in significant errors in physical characteristics highly correlated with material hardness, such as the high-frequency band and the proportion of frequency band energy. This limits the simulation signal in specific engineering application scenarios.
[0004] In summary, how to overcome the limitations of the traditional Ising model's massive computation and the JA model's low matching degree, and provide a fast simulation and parameter optimization method for MBN signals that takes into account high matching degree in the time and frequency domains, is a problem that urgently needs to be solved in the engineering field. Summary of the Invention
[0005] To address the problems in existing technologies, this invention provides a simulation generation method for magnetic Barkhausen noise based on time-frequency characteristic parameters. This method overcomes the limitations of traditional models in microscopic evolution calculations and macroscopic waveform fitting. It extracts the core physical features of the MBN signal's external driving modulation and internal random transitions, deconstructs the MBN signal into a parameterized time-domain envelope and frequency-domain colored noise shaping, and utilizes multi-objective joint optimization and multi-random seed smoothing mechanisms to achieve high-matching reconstruction of the simulated signal.
[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:
[0007] A simulation generation method for magnetic Barkhausen noise based on time-frequency characteristic parameters includes the following steps:
[0008] Step S101: Obtain the measured raw signal data of the magnetic Barkhausen noise of the ferromagnetic material under alternating magnetic field excitation, and extract the measured time-domain envelope curve and the measured power spectral density curve of the measured raw signal as a reference.
[0009] Step S102: Construct a feature parameter vector for driving the simulation. The feature parameter vector includes a time-domain envelope parameter for controlling the macroscopic energy distribution in the time domain, a fundamental frequency band parameter for controlling the signal bandwidth, and a spectral fine-tuning parameter for controlling the micromorphology of the power spectrum.
[0010] Step S103: Based on the basic frequency band parameters and spectrum fine-tuning parameters in the feature parameter vector, perform frequency domain shaping on the Gaussian white noise to generate a colored noise carrier; at the same time, generate a time domain modulation envelope based on the time domain envelope parameters, and multiply the two in the time domain to generate the original magnetic Barkhausen noise signal of a single simulation.
[0011] Step S104: Re-extract the simulation time-domain envelope curve and the simulation power spectral density curve from the original simulated magnetic Barkhausen noise signal; wherein, when extracting the simulation power spectral density curve, step S103 is repeatedly executed using multiple different random seeds to generate multiple sets of simulation signals and calculate the power spectral density respectively, and the average value is obtained to obtain the smoothed simulation power spectral density curve.
[0012] Step S105: Construct a multi-dimensional feature synthesis cost function, and calculate the time-domain envelope correlation error between the simulated time-domain envelope curve and the measured time-domain envelope curve, as well as the frequency-domain power spectrum correlation error between the smoothed simulated power spectral density curve and the measured power spectral density curve.
[0013] Step S106: Iteratively update the feature parameter vector using an optimization algorithm to minimize the comprehensive cost function; if the frequency domain power spectrum correlation error does not reach the preset qualified threshold during the iteration process, an adaptive retry mechanism is triggered to dynamically increase the penalty weight of the power spectral density feature in the comprehensive cost function and re-optimize until the optimal feature parameter vector and the final simulation signal are output.
[0014] On the other hand, the present invention 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 program to implement the steps of the above-described simulation generation method.
[0015] As can be seen from the above description, this application innovatively proposes a new time-frequency domain decoupling simulation method based on time-frequency characteristic parameter control. Its beneficial effects are as follows: (1) Innovative frequency domain multi-node anchor point interpolation mechanism. This method abandons the traditional single bandpass filter and introduces a frequency band gain anchor point based on logarithmic frequency axis interpolation to construct a multi-node nonlinear spectrum equalizer, so that the simulated power spectrum can accurately fit the micro-frequency domain noise floor of the real MBN; (2) Excellent robustness against random jitter optimization. This method introduces a multi-random seed averaging mechanism in the cost function, which effectively filters out the random spikes inherent in white noise and greatly improves the stability of the algorithm in finding the global optimal solution; (3) Extremely high physical feature restoration degree. Experiments show that the simulated signal generated by this method not only maintains a correlation of more than 0.99 with the measured reference in the time domain envelope, but its smooth power spectrum correlation also jumps to more than 0.99. Furthermore, the errors in features such as total energy and RMS, which are highly correlated with the Rockwell hardness (HRC) of the material, are reduced to extremely low levels, which greatly enhances the engineering application value of the simulated signal in nondestructive testing. Attached Figure Description
[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments will be briefly introduced below.
[0017] Figure 1 This is a flowchart illustrating the magnetic Barkhausen noise simulation generation method based on feature parameter control in an embodiment of the present invention.
[0018] Figure 2 This is a comparison of the time-domain waveforms of the measured and simulated original magnetic Barkhausen noise signals (low hardness 23.0 HRC) in a specific application example of the present invention.
[0019] Figure 3 This is a comparison of the time-domain waveforms of the measured and simulated original magnetic Barkhausen noise signals (medium hardness 43.4 HRC) in a specific application example of the present invention.
[0020] Figure 4 This is a comparison of the measured and simulated time-domain waveforms of the original magnetic Barkhausen noise signal (high hardness 62.8 HRC) in a specific application example of the present invention.
[0021] Figure 5 This is a comparison chart of time-domain envelope feature matching for three typical hardness materials (low, medium, and high) in a specific application example of the present invention.
[0022] Figure 6 This is a comparison chart of smooth power spectral density (logarithmic scale) matching for three typical hardness materials of low, medium and high hardness in a specific application example of the present invention;
[0023] Figure 7This is a comparison chart of smooth power spectral density (linear scale) matching for three typical hardness materials of low, medium and high hardness in a specific application example of the present invention;
[0024] Figure 8 This is a comparison chart of single-sample detailed power spectral density feature matching in a specific application example of the present invention (low hardness 23.0 HRC).
[0025] Figure 9 This is a comparison chart of single-sample detailed power spectral density features (medium hardness 43.4 HRC) in a specific application example of the present invention.
[0026] Figure 10 This is a comparison chart of single-sample detailed power spectral density features (high hardness 62.8 HRC) in a specific application example of the present invention.
[0027] Figure 11 This is a graph showing the relationship between the temporal envelope fidelity correlation coefficient and the Rockwell hardness (HRC) of the tested material in this invention.
[0028] Figure 12 This is a graph showing the relationship between the frequency domain power spectrum fidelity correlation coefficient and the system comprehensive evaluation score of this invention and Rockwell hardness (HRC).
[0029] Figure 13 This is a bar chart showing the average percentage of simulation error for the core hardness characteristics of this invention. Detailed Implementation
[0030] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.
[0031] Traditional MBN simulations often focus on calculating the evolution of microscopic spin states or simply fitting macroscopic phenomenological features. However, the physical principle of this invention lies in deconstructing the MBN generation process into two independent parts: macroscopic deterministic modulation and microscopic random transitions. Through an engineering implementation scheme combining parameterized bimodal envelope, frequency-domain colored noise shaping, and multi-objective optimization, the simulated signal achieves a high degree of consistency with measured data in both time-domain envelope structure and power spectrum energy distribution.
[0032] Based on the aforementioned physical principles, the specific implementation of the method provided by this invention is as follows, see below. Figure 1 :
[0033] Step S101: Pre-extraction of reference signal features. Acquire the measured MBN signal obtained through a standardized excitation system. The analytic envelope is extracted using the Hilbert transform, and the measured smooth time-domain envelope is obtained by applying a moving average filter. Simultaneously, the Welch periodogram method combined with a frequency-domain Gaussian smoothing algorithm was used to extract the measured smoothed power spectral density of the measured signal within the analysis frequency band (e.g., 5kHz to 1000kHz). This serves as a standard benchmark for subsequent optimization.
[0034] Step S102: Initialization of the multidimensional feature parameter space. Construct a feature parameter vector containing 17 hyperparameters. Specifically, this includes controlling the first peak time of the two main transition burst regions within a magnetization cycle. ,width Amplitude Second peak time ,width Amplitude and quiet zone baseline amplitude High-pass and low-pass cutoff frequencies , and its roll-off order , Center frequency of double resonant peak , High-frequency tilt compensation coefficient and three smooth frequency band gain anchors located in the logarithmic frequency domain. , , To accelerate the convergence of the optimization search, a peak detection algorithm is used. Peak search is performed, and the time and amplitude of the main and secondary peaks are automatically extracted as the initial points for time-domain parameters.
[0035] Step S103: Physical simulation generation for time-frequency decoupling. First, generate a Gaussian white noise sequence. And then transformed to the frequency domain via Fourier transform. Secondly, a composite frequency domain shaping filter is constructed: Specifically, the calculation formulas for each sub-item are as follows: High-pass filter term: Low-pass filter term: First double Gaussian resonance peak term: Second double Gaussian resonance peak term: High-frequency tilt compensation term: in Furthermore, the dual Gaussian time-domain modulation envelope The precise mathematical expression is: For frequency domain interpolation terms Its corresponding three frequency band gain anchor points frequency nodes Reasonable constraints must be imposed within the effective frequency band. In this embodiment, a fixed ratio method is adopted: , , This limits the physical location of the anchor point on the logarithmic frequency axis. In particular, To utilize a nonlinear smooth interpolation algorithm (such as, but not limited to, cubic spline interpolation, Bezier interpolation, etc., the piecewise cubic Hermite conformal interpolation PCHIP algorithm is preferred in this embodiment) to anchor the gain on the logarithmic frequency axis. The envelope deformation curve is generated by smooth interpolation. and After multiplication and inverse transformation back to the time domain, colored noise is obtained. Finally, a double Gaussian time envelope is generated using time-domain parameters. The original simulation signal was synthesized through time-domain modulation. .
[0036] Step S104: Joint feature smoothing using multiple random seeds. To adhere to realistic physical constraints, the simulated envelope and spectrum must be derived from the original signal synthesized in step S103. The extraction process is recalculated to eliminate... The severe random fluctuations in the spectrum caused by the white noise background are substituted into the power spectrum calculation. One (preferred in this embodiment) For example, let's say Multiple pseudo-random signal sequences were repeatedly generated using different pseudo-random seeds. The power spectrum was calculated using the Welch periodogram method, with specific parameters set as follows: Hanning window, 50% overlap, and frequency band analysis covering the main energy transition regions (e.g., 5kHz to 1000kHz). The spectral density is arithmetically averaged in the frequency domain to obtain a stable and smooth simulated spectral density. Furthermore, to eliminate the interference of amplitude dimensions on the optimization process, the colored noise carrier in step S103... and the final simulation signal All values underwent effective value normalization, with an enforced constraint that RMS = 1.
[0037] Step S105: Construction of Multidimensional Joint Cost Function. Construct a comprehensive cost function to evaluate the realism of the simulation. This cost function not only includes the time-domain envelope Pearson correlation coefficient and envelope peak-to-valley ratio error, but also strictly constrains several physical properties at the power spectrum level, including: PSD logarithm and linear correlation coefficient, spectral energy centroid shift, root mean square bandwidth error, and energy proportion error in low / medium / high frequency bands. Furthermore, it incorporates amplitude statistical errors such as the effective value (RMS), peak value, and kurtosis of the original signal.
[0038] Step S106: Adaptive weighted closed-loop iterative optimization. Comprehensive cost function. The error is calculated by multiplying the aforementioned characteristic errors after normalization to their maximum and minimum values, and then summing the results by their corresponding preset penalty weights. Because white noise is extremely sensitive in the frequency domain, the system introduces an adaptive weighted retry mechanism based on power spectrum characteristics. During the iteration process, the correlation coefficient of the smoothed power spectrum is monitored in real time. Set the trigger qualification threshold. (For example ): (1) Triggering condition: After completing one round of local optimization, if This indicates that the algorithm has fallen into a local trap of overfitting in the time domain; (2) Weight update formula: At this time, the control law is triggered, and the penalty weight of the power spectrum correlation error is dynamically amplified. The updated formula is as follows The step size amplification factor (like ), The weight cap is set to prevent gradient explosion; (3) Retry and Stop Conditions: After the weights are updated, the local optimization is forcibly re-executed starting from the current parameters. This retry loop is allowed to trigger a maximum of [number missing] times. Number of times (e.g., maximum number of retries) When satisfied. Or the maximum number of retries has been reached. When the loop stops, the final output makes Furthermore, to achieve engineering applications, the generated target simulation signal is used to construct a benchmark sample library for hardness evaluation or non-destructive testing of ferromagnetic materials, thereby providing highly matched data support for the calibration and feature recognition of actual testing equipment.
[0039] Refer to the instruction manual. Figure 2 To be continued Figure 13 Verification of the effectiveness of this method: (e.g.) Figures 2 to 4 As shown, the original simulated MBN signal synthesized in this invention exhibits a random clustering phenomenon in its macroscopic transition structure that is highly consistent with the measured signal; based on this, as Figure 5 As shown, for samples of different hardness levels, the optimized simulated envelope and the measured baseline envelope exhibit high consistency in the time domain, with a correlation coefficient exceeding 0.99. Figures 6 to 10 As shown, under both logarithmic and linear scales, the smoothed and detailed PSDs faithfully reproduce the complex micro-fluctuation characteristics of the measured benchmark PSD. Figure 11 and Figure 12 Global index verification shows that, under different Rockwell hardness (HRC) ratings, the algorithm can consistently maintain the time-domain envelope correlation coefficient above 0.985. Figure 11 The frequency domain power spectrum correlation coefficient remained stable above the target line of 0.85, and the system comprehensive evaluation score remained stable above 0.95. Figure 12 Furthermore, such as Figure 13 As shown in the feature error statistics chart, this invention enables the average relative simulation error of the core physical statistical features in the time-frequency domain to be strictly controlled at an extremely low level (the error of all extracted features is...). 7%). The above large-scale experimental data fully demonstrate that this method can stably and accurately achieve high-matching simulation of MBN signals in ferromagnetic materials.
[0040] Furthermore, targeting Figure 13The extracted core hardness features can be divided into three categories according to the data dimensions: (1) Original noise features prefixed with "[MBN]", mainly including the effective value (Rms), peak value (Peak), standard deviation (Std), energy, and effective values of frequency bands (Rms B1, Rms B2), etc.; (2) Time-domain envelope features prefixed with "[ENV]", mainly covering macroscopic statistics of the envelope (Max, Mean, Std, Energy), peak-to-valley ratio (PV Ratio), time and amplitude of the first and second main peaks (Pk1 Time, Pk1 Amp, Pk2 Time, Pk2 Amp, etc.) and time-domain detail features (Det Max, Det Mean, Det Std), etc.; (3) Power spectral density features prefixed with "[PSD]", mainly including centroid frequency (Centroid), root mean square bandwidth (RMS BW), and total power (Total) of the smooth spectrum and detail spectrum dimensions. The simulation model includes Pwr), amplitude distribution statistics (Mean, Peak, Median), energy proportions of each frequency band (Mid Ratio, Hi Ratio, Det MidRatio), high-frequency slope (HF Slope), and detailed spectrum statistical characteristics (Det Mean, Det Peak). As shown in the figure, the average relative error of each multidimensional physical property is at most 6.4%, and the overall error is strictly controlled within 7%. Through the comprehensive comparison covering multiple physical properties, the simulation model of this invention fully verifies that it has a very high degree of fidelity to the actual physical properties of materials.
[0041] On the other hand, specific embodiments are used to illustrate the principles and implementation methods of this invention. Those skilled in the art will understand that the embodiments of this invention can be provided as a software product or testing system based on a computer program. The steps in this invention, such as parameter control, frequency domain shaping, Gaussian smoothing, and closed-loop optimization, can be entirely implemented by computer program instructions and stored in a computer-readable storage medium.
[0042] The above description of the embodiments is only for the purpose of helping to understand the method and core idea of the present invention; at the same time, there may be changes in the specific implementation and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.
Claims
1. A method for simulating and generating magnetic Barkhausen noise based on time-frequency characteristic parameters, characterized in that, Includes the following steps: S101: Obtain the measured raw signal data of the magnetic Barkhausen noise of the ferromagnetic material under test under alternating magnetic field excitation, and extract the measured time-domain envelope curve and the measured power spectral density curve of the measured raw signal as a reference. S102: Constructing the feature parameter vector for driving the simulation The feature parameter vector includes a time-domain envelope parameter for controlling the time-domain macroscopic energy distribution, a basic frequency band parameter for controlling the signal bandwidth, and three frequency band gain anchor points located on the logarithmic frequency axis as spectrum fine-tuning parameters. S103: Based on the feature parameter vector The structure explicitly includes a high-pass filter term, a low-pass filter term, a double resonant peak term, a high-frequency tilt compensation term, and a logarithmic frequency domain shape interpolation term. Composite frequency domain shaping function And the It is a curve generated using a nonlinear smooth interpolation algorithm for a given frequency band gain anchor point; the Gaussian white noise spectrum is then compared with the curve. After multiplication and inverse transformation, followed by RMS normalization, the colored noise carrier is obtained. Simultaneously, a dual Gaussian time-domain modulation envelope is generated using the aforementioned time-domain envelope parameters. Multiplying the two in the time domain generates the original magnetic Barkhausen noise signal for a single simulation. ; S104: From the original magnetic Barkhausen noise signal of the simulation The simulation time-domain envelope curve and simulation power spectral density curve were re-extracted; in particular, to eliminate frequency domain random jitter caused by white noise background, the simulation power spectral density curve was extracted using... Repeat step S103 with different pseudo-random seeds to generate multiple sets of simulation signals and calculate the power spectrum of each signal. The smooth simulation power spectral density curve after eliminating random jitter is obtained by taking the arithmetic mean in the frequency domain. S105: Constructing a multidimensional feature synthesis cost function The time-domain envelope correlation error between the simulated time-domain envelope curve and the measured time-domain envelope curve, and the frequency-domain power spectrum correlation error between the smoothed simulated power spectral density curve and the measured power spectral density curve are calculated respectively. S106: Use an optimization algorithm to optimize the feature parameter vector. Perform iterative updates to minimize the comprehensive cost function. ; During the iteration process, the correlation coefficient of the smoothed power spectrum is monitored in real time. If the set qualification threshold is not reached Then the adaptive weight retry mechanism is triggered: according to the update formula. Dynamic amplification involves penalty weights for power spectral density characteristics. ,in The step size amplification factor and , The maximum weight is set as the upper limit, and the optimization iteration is repeated under the constraints of the maximum weight limit and the maximum number of retries to guide the optimization algorithm away from local optima; finally, the optimal feature parameter vector and the target simulation signal are output, and the target simulation signal is used to construct a benchmark sample library for hardness evaluation or non-destructive testing of ferromagnetic materials.
2. The method according to claim 1, characterized in that, The feature parameter vector mentioned in step S102 It contains 17 hyperparameters, including: the center time of the first envelope peak. ,width Amplitude The center time of the second envelope peak ,width Amplitude Quiet zone baseline amplitude ; High-pass cutoff frequency of the spectrum Spectrum low-pass cutoff frequency Roll-off order of Qualcomm and Low-pass filters and The center frequency of the first resonance peak The center frequency of the second resonance peak High-frequency tilt coefficient and three frequency band gain anchor points located at different frequency nodes. , , .
3. The method according to claim 1, characterized in that, Step S103 involves frequency-domain shaping of the Gaussian white noise to generate a colored noise carrier. The specific process is as follows: Generate a white noise sequence The frequency domain signal is obtained by performing a Fourier transform. ; A composite frequency domain shaping function is constructed using the fundamental frequency band parameters and the spectral fine-tuning parameters. ;Will and After multiplication, an inverse Fourier transform is performed, followed by normalization to obtain the colored noise carrier. .
4. The method according to claim 1, characterized in that, The dual Gaussian time-domain modulation envelope The calculation formula includes two Gaussian peak expressions and the quiet zone baseline amplitude. The superposition; the logarithmic frequency domain shape interpolation term The corresponding three frequency band gain anchor points Its node coordinates on the frequency axis According to Qualcomm cutoff frequency With low-pass cutoff frequency The proportional distribution satisfies: lie in Nearby frequency bands, , lie in Nearby frequency bands.
5. The method according to claim 1, characterized in that, The multidimensional feature synthesis cost function mentioned in step S105 includes, in addition to the time-domain envelope correlation error and frequency-domain power spectrum correlation error, the following: envelope normalization shape root mean square error, quiet zone baseline error, smoothed power spectrum linear correlation error, spectrum centroid error, root mean square bandwidth error, energy proportion error of each frequency band, high-frequency slope error, and effective value and kurtosis error of the original signal.
6. The method according to claim 1 or 5, characterized in that, The optimization algorithm described in step S106 adopts a four-stage joint optimization strategy: First, the particle swarm optimization algorithm is used to perform a global pre-optimization search based solely on the temporal envelope error; then, the multi-dimensional feature comprehensive cost function is introduced, and the nonlinear simplex method is used for full-objective fine-tuning; next, a local particle swarm optimization algorithm search is performed near the current optimal point; finally, the nonlinear simplex method is executed again to make it converge to the global optimal solution.
7. The method according to claim 1, characterized in that, The number of pseudo-random seeds mentioned in step S104 The value range is 10 to 200; the power spectral density is calculated using the Welch periodogram method, whose parameters explicitly define the window function and overlap rate; the specific control law of the adaptive weighted retry mechanism described in step S106 is: when the trigger condition... Upon establishment, a weight update is performed. ,in This is the step size amplification factor, and , This represents the upper limit of the weight. The optimal stopping condition is reaching the maximum number of retries. Or satisfy .
8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method according to any one of claims 1 to 7.