Method and system for intelligently detecting data line based on inductance characteristics

Abstract

The invention discloses a method for intelligently detecting a data line based on inductance characteristics, and the method comprises the steps: obtaining the inductance characteristic measurement data of a target sample set, carrying out the principal component analysis, extracting a feature vector, constructing a Bayesian probability model, and evaluating the signal integrity. And generating a probability correction coefficient by adopting Monte Carlo simulation, optimizing a dynamic probability distribution model, further judging final defective products by utilizing an adaptive detection algorithm, analyzing historical misjudgment characteristics in combination with a random forest, and dynamically updating a detection threshold. According to the invention, the problem of misjudgment in inductance characteristic detection is effectively solved, and the detection precision and reliability are improved. By fusing a plurality of advanced algorithms and dynamic optimization strategies, intelligence and adaptivity of inductance characteristic detection are realized, and an innovative solution is provided for quality control of electronic components. The invention further discloses a system for intelligently detecting the data line based on the inductance characteristics.

Description

Technical Field

[0001] The present invention relates to the field of intelligent manufacturing technology, and particularly to a method and system for intelligently detecting data cables based on inductance characteristics. Background Art

[0002] As an indispensable connecting component in modern electronic devices, the quality and performance of data cables directly affect communication efficiency, device stability, and user experience. Against the backdrop of the rapid development of informatization, the detection of defective data cables has become a crucial research direction in the manufacturing industry and the field of quality control. An efficient and accurate detection method is not only the key to enhancing product competitiveness but also the core guarantee for reducing production costs and user complaints. However, there are still significant deficiencies in current data cable detection methods, and a technological breakthrough is urgently needed. Traditional detection methods mostly rely on manual sampling inspection or simple physical tests, such as appearance inspection and basic electrical performance tests. Although this method is easy to operate, in the face of complex application scenarios, it often fails due to the lack of comprehensive consideration of parameter uncertainties. For example, fluctuations in key variables such as inductance characteristics, transmission distance, and operating frequency are difficult to accurately capture by static standards, resulting in a high undetected rate of defective products. In addition, existing automated detection systems usually rely on fixed threshold judgments and are difficult to adapt to diverse usage requirements and performance changes under extreme working conditions, limiting the universality and reliability of detection. The core challenge in the research field lies in how to effectively address the uncertainties of data cable performance parameters and how to quantify the performance of these parameters in different scenarios. Inductance characteristics, as the basis of data cable transmission efficiency, directly affect signal integrity; transmission distance and operating frequency further amplify the complexity of performance fluctuations. Since these technical factors have not been fully modeled and analyzed by existing methods, the detection system often cannot accurately distinguish defective products from qualified products. Especially under extreme working conditions, the ambiguity of performance boundaries leads to frequent misjudgments. This not only increases the difficulty of subsequent troubleshooting but also poses higher requirements for high-precision selection. Therefore, how to integrate multi-dimensional parameters such as inductance characteristics, transmission distance, and operating frequency through probability modeling, quantify the probability of data cable performance in specific application scenarios, and build a dynamically optimized defective product detection system based on this has become the key problem that needs to be solved urgently in this research. Summary of the Invention

[0003] To solve the above technical problems, in the first aspect of the present invention, a method for intelligently detecting data cables based on inductance characteristics is provided, including:

[0004] S1, obtaining inductance characteristic measurement data of a target sample set, where the measurement data includes transmission distance, operating frequency, and real-time signal integrity parameters, and storing them in a database to form an original data set;

[0005] S2. Perform principal component analysis on the original dataset, extract the eigenvectors of inductance characteristics, transmission distance, and operating frequency, and generate a quantization representation matrix.

[0006] S3. Construct a Bayesian probability model based on the quantization representation matrix, and calculate the probability density function of signal integrity under a specific combination of transmission distance and operating frequency.

[0007] S4. If the coverage of the probability density function below the preset signal integrity threshold is lower than the set value, mark the corresponding samples as the preliminary defective product set.

[0008] S5. Obtain the inductance test data of the preliminary defective product set under extreme working conditions, and generate a set of probability correction coefficients by Monte Carlo simulation.

[0009] S6. Adjust the parameter weights of the Bayesian model according to the set of probability correction coefficients to generate an optimized dynamic probability distribution model.

[0010] S7. Construct an adaptive detection algorithm using the dynamic probability distribution model. If the signal integrity probability value of a sample is lower than the dynamic threshold, classify it into the final defective product set.

[0011] S8. Obtain the misjudgment sample characteristics in the historical detection data, analyze the correlation between the misjudgment characteristics and physical parameters through the random forest algorithm, and generate a set of dynamically optimized parameters.

[0012] S9. Update the decision threshold of the adaptive detection algorithm according to the set of dynamically optimized parameters, and output the real-time updated detection results.

[0013] Optionally, in step S1, obtain the inductance characteristic measurement data of the target sample set. The measurement data includes transmission distance, operating frequency, and real-time signal integrity parameters, and store them in a database to form the original dataset, including:

[0014] Step S11. Obtain the inductance characteristic measurement data of the target sample set, including transmission distance, operating frequency, and real-time signal integrity parameters, and store them in a database to generate the original dataset.

[0015] Step S12. Collect the inductance characteristic data of the target sample through a sensor, including transmission distance, operating frequency, and real-time signal integrity parameters, to obtain a preliminary measurement result.

[0016] Step S13. Process the preliminary measurement result using a filtering method to remove noise interference and generate an optimized dataset.

[0017] Step S14. If the signal integrity parameter in the optimized dataset exceeds the centralized threshold, re-obtain the data by adjusting the acquisition frequency to obtain a corrected dataset.

[0018] Step S15: Calculate the correlation index between the transmission distance and the operating frequency according to the corrected data set, and determine the parameter change trend.

[0019] Step S16: Process the parameter change trend through linear regression analysis to generate a prediction model for the inductance characteristics.

[0020] Step S17: Compare the prediction model with the original data set, calculate the matching degree using the mean square error, and obtain the final analysis result.

[0021] Optionally, in step S2, perform principal component analysis on the original data set, extract the eigenvectors of the inductance characteristics, transmission distance, and operating frequency, and generate a quantization representation matrix, including:

[0022] Step S21: Process the original data set through principal component analysis, extract the eigenvectors of the inductance characteristics, transmission distance, and operating frequency, and generate a first quantization matrix.

[0023] Step S22: Group the first quantization matrix using the K-means clustering method, and obtain a classification data set based on the distribution characteristics of the eigenvectors.

[0024] Step S23: Calculate the mean and variance of the eigenvectors within each group for the classification data set, and determine the parameter distribution characteristics.

[0025] Step S24: If the parameter distribution characteristics exceed the characteristic threshold, re-group by adjusting the parameters of the K-means clustering to obtain a corrected classification data set.

[0026] Step S25: According to the corrected classification data set, extract the correlation index between the inductance characteristics and the transmission distance within each group, and generate a characteristic correlation table.

[0027] Step S26: Process the characteristic correlation table through linear regression to generate a mapping relationship model between the inductance characteristics and the operating frequency, and obtain a prediction result.

[0028] Step S27: Calculate the deviation value from the first quantization matrix for the prediction result, judge the matching degree of the mapping relationship model, and generate a final data set.

[0029] Optionally, in step S3, construct a Bayesian probability model according to the quantization representation matrix, and calculate the signal integrity probability density function under a specific combination of transmission distance and operating frequency, including:

[0030] Step S31: Obtain a data matrix through the quantization matrix, and process the data matrix using the Bayesian model to obtain the distribution characteristics.

[0031] Step S32: Determine the probability density according to the distribution characteristics, generate a parameter combination through the transmission distance and operating frequency, and obtain the density function;

[0032] Step S33: Calculate the signal integrity for the density function to obtain the probability distribution result of the signal integrity;

[0033] Step S34: If the probability distribution result of the signal integrity is lower than the distribution threshold, recalculate the density function by adjusting the parameter combination to obtain the corrected distribution characteristics;

[0034] Step S35: Update the probability density according to the corrected distribution characteristics, and generate a new density function using the calculation process;

[0035] Step S36: Determine the signal integrity through the new density function to obtain the final probability distribution characteristics;

[0036] Step S37: For the final probability distribution characteristics, use the data matrix to verify the calculation process and determine the consistency result.

[0037] Optionally, in step S4, if the coverage of the probability density function at the preset signal integrity threshold is lower than the set value, mark the corresponding samples as the preliminary defective product set, including:

[0038] Step S41: According to the preset signal integrity threshold, use the statistical distribution function in the scipy library to construct a Gaussian distribution probability density function model;

[0039] Step S42: Use the random sampling function in the numpy library to perform Monte Carlo sampling analysis on the probability density function;

[0040] Step S43: Use the KernelDensity class in the scikit-learn library, and use the Gaussian kernel function to perform kernel density estimation on the sampled data to calculate the coverage near the preset signal integrity threshold;

[0041] Step S44: Use the MinMaxScaler class in the scikit-learn library to perform min-max normalization processing on the coverage values to obtain the standardized coverage index;

[0042] Step S45: If the standardized coverage index is lower than the coverage threshold, mark the corresponding samples as preliminary defective products;

[0043] Step S46: Apply the KMeans clustering algorithm in the scikit-learn library to the preliminary defective product sample set to divide different categories;

[0044] Step S47: Perform secondary screening and grading on the preliminary defective product samples according to the clustering results.

[0045] Optionally, the step S5, obtaining the inductance test data of the preliminary defective product set under extreme working conditions, and using Monte Carlo simulation to generate a probability correction coefficient set, includes:

[0046] Step S51, obtaining raw data of inductance testing under extreme working conditions from a preliminary defective product set;

[0047] Step S52, preprocessing the raw data using a box plot method, removing outliers, and removing noise interference by wavelet transform;

[0048] Step S53, based on the processed data, the inductance parameter model is constructed using the least square method, and the inductance value and frequency are set as key variables;

[0049] Step S54, using the Monte Carlo method to generate a large number of random samples according to the inductance parameter model to form a simulation data set;

[0050] Step S55, performing descriptive statistical analysis on the simulated data set to calculate the mean, variance and distribution characteristics of each parameter;

[0051] Step S56, generating an initial value of the probability correction coefficient using maximum likelihood estimation according to the distribution characteristic information;

[0052] Step S57, iteratively optimizing the correction coefficient by gradient descent method to make it close to the real distribution.

[0053] Optionally, the step S6, adjusting the parameter weights of the Bayesian model according to the probability correction coefficient set to generate an optimized dynamic probability distribution model, includes:

[0054] Step S61, obtaining historical data sets and current data samples, cleaning and standardizing the data using data preprocessing technology, and generating a cleaned data set;

[0055] Step S62, based on the cleaned data set, the initial parameters of the Bayesian model are calculated using the maximum likelihood estimation method to obtain the prior probability distribution of the model;

[0056] Step S63, constructing a probability correction coefficient set according to the characteristics of the cleaned data set, making preliminary adjustments to the model parameters, and generating an adjusted model;

[0057] Step S64, performing multiple random sampling on the adjusted model using the Monte Carlo method to generate multiple sets of candidate parameter sets;

[0058] Step S65, evaluating the prediction performance of each group of candidate parameters by a cross-validation method, and selecting the optimal parameter combination;

[0059] Step S66: Update the posterior probability distribution of the Bayesian model based on the optimal parameter combination to form a dynamic probability distribution model;

[0060] Step S67: Adopt the sliding window technique to continuously obtain new data, update the probability correction coefficient in real time, and maintain the dynamic optimization ability of the model.

[0061] Optionally, in step S7, an adaptive detection algorithm is constructed using the dynamic probability distribution model. If the sample signal integrity probability value is lower than the dynamic threshold, it is classified into the final defective product set, including:

[0062] Step S71: Obtain sample signal data, preprocess the signal to remove noise interference, and obtain denoised signal data;

[0063] Step S72: Smooth the denoised signal data using Gaussian filtering, and extract the mean and variance of the signal as features;

[0064] Step S73: Calculate the sample signal integrity probability value according to the extracted mean and variance features;

[0065] Step S74: Train a support vector machine model through historical data and dynamically adjust the dynamic threshold;

[0066] Step S75: If the sample signal integrity probability value is lower than the dynamic threshold, it is determined as a defective product;

[0067] Step S76: Classify the defective products using the K-means clustering algorithm to identify different types of defects;

[0068] Step S77: Generate a quality analysis report for each type of defective product. The report includes the defect type distribution and improvement suggestions, which are used to optimize the production process.

[0069] Optionally, in step S8, obtain the misjudged sample features in the historical detection data, analyze the correlation between the misjudged features and physical parameters through the random forest algorithm, and generate a dynamic optimization parameter set, including:

[0070] Step S81: Obtain the misjudged sample records in the historical detection data, and extract the sample feature vector and the corresponding physical parameter values;

[0071] Step S82: Perform dimensionality reduction processing on the misjudged sample feature vector to obtain the reduced feature dimension;

[0072] Step S83: Take the reduced feature dimension as the analysis object, and use the random forest algorithm to construct a relationship model between the misjudged sample features and physical parameters;

[0073] Step S84, calculate the feature importance scores, sort the physical parameters according to the scores, and filter out a subset of key parameters with greater influence;

[0074] Step S85, perform multi-dimensional combination optimization on the key parameters through the grid search method to generate a set of candidate parameters;

[0075] Step S86, perform cross-validation on the set of candidate parameters, evaluate the performance of each combination on new samples, and select the optimal parameter combination;

[0076] Step S87, update the detection model with the optimal parameter combination to achieve dynamic optimization and improve the detection accuracy.

[0077] In the second aspect of the present invention, a system for intelligently detecting data lines based on inductance characteristics is provided. The data lines are intelligently detected by using the method described above. The system includes:

[0078] A data acquisition module, configured to acquire inductance characteristic measurement data of a target sample set. The measurement data includes transmission distance, operating frequency, and real-time signal integrity parameters, and stores them in a database to form an original data set;

[0079] A feature extraction module, configured to perform principal component analysis on the original data set, extract feature vectors of inductance characteristics, transmission distance, and operating frequency, and generate a quantization representation matrix;

[0080] A probability modeling module, configured to construct a Bayesian probability model according to the quantization representation matrix, and calculate the signal integrity probability density function under a specific combination of transmission distance and operating frequency;

[0081] A preliminary screening module, configured to mark the corresponding samples as a set of preliminary defective products if the coverage of the probability density function below a preset signal integrity threshold is lower than a set value;

[0082] An extreme test module, configured to acquire inductance test data of the set of preliminary defective products under extreme working conditions, and generate a set of probability correction coefficients by using Monte Carlo simulation;

[0083] A model optimization module, configured to adjust the parameter weights of the Bayesian model according to the set of probability correction coefficients to generate an optimized dynamic probability distribution model;

[0084] An adaptive detection module, configured to construct an adaptive detection algorithm by using the dynamic probability distribution model, and classify the samples into a set of final defective products if the sample signal integrity probability value is lower than the dynamic threshold;

[0085] A misjudgment analysis module, configured to acquire the misjudgment sample characteristics in the historical detection data, analyze the correlation between the misjudgment characteristics and the physical parameters through a random forest algorithm, and generate a set of dynamically optimized parameters;

[0086] A dynamic update module, configured to update the decision threshold of the adaptive detection algorithm according to the dynamic optimization parameter set and output a real-time updated detection result.

[0087] The technical solution provided by the embodiment of the present invention has the following beneficial effects:

[0088] The present invention discloses a method for intelligently detecting a data line based on inductance characteristics. By obtaining the inductance characteristic measurement data of a target sample set, performing principal component analysis to extract feature vectors, constructing a Bayesian probability model to evaluate signal integrity, for a preliminary defective product set, using Monte Carlo simulation to generate a probability correction coefficient, optimizing the dynamic probability distribution model, further using an adaptive detection algorithm to determine the final defective products, and combining random forest to analyze historical misjudgment characteristics and dynamically update the detection threshold. The present invention effectively solves the misjudgment problem in inductance characteristic detection, improves the detection accuracy and reliability, and realizes the intelligence and self-adaptability of inductance characteristic detection by integrating a variety of advanced algorithms and dynamic optimization strategies, providing an innovative solution for the quality control of electronic components. Description of the Drawings

[0089] Figure 1 It is a flowchart of a method for intelligently detecting a data line based on inductance characteristics of the present invention.

[0090] Figure 2 It is a schematic structural diagram of a system for intelligently detecting a data line based on inductance characteristics of the present invention. Detailed Embodiments

[0091] In order to enable those skilled in the art to better understand the technical solutions in this specification, the following will clearly and completely describe the technical solutions in the embodiments of this specification with reference to the accompanying drawings in the embodiments of this specification. Obviously, the described embodiments are only a part of the embodiments of this specification, rather than all the embodiments. Based on the embodiments in this specification, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of this specification.

[0092] As Figure 1 shown, in the first aspect of the present invention, a method for intelligently detecting a data line based on inductance characteristics is provided, including:

[0093] S1. Obtain the inductance characteristic measurement data of the target sample set, where the measurement data includes the transmission distance, operating frequency, and real-time signal integrity parameters, and store them in a database to form an original data set.

[0094] Optionally, this step further includes:

[0095] Step S11: Obtain the inductance characteristic measurement data of the target sample set, including the transmission distance, operating frequency, and real-time signal integrity parameters, and store them in a database to generate an original data set.

[0096] Step S12: Collect the inductance characteristic data of the target sample through a sensor, including the transmission distance, operating frequency, and real-time signal integrity parameters, to obtain a preliminary measurement result.

[0097] Step S13: Process the preliminary measurement result using a filtering method to remove noise interference and generate an optimized data set.

[0098] Optionally, use the following formula to process the preliminary measurement result:

[0099]

[0100] where y(t) represents the filtered output signal, M represents the filter length, h(k) represents the filter coefficient, and x(t - k) represents the input signal sequence.

[0101] Step S14: If the signal integrity parameter in the optimized data set exceeds the set threshold, re-acquire the data by adjusting the acquisition frequency to obtain a corrected data set.

[0102] Step S15: Calculate the correlation index between the transmission distance and the operating frequency based on the corrected data set to determine the parameter change trend.

[0103] Optionally, use the following formula to calculate the parameter change trend:

[0104]

[0105] where D represents the comprehensive change amount of the transmission distance and frequency, n represents the number of measurement points, f represents the operating frequency, and d represents the transmission distance.

[0106] Step S16: Process the parameter change trend using the linear regression analysis method to generate a prediction model of the inductance characteristics.

[0107] Step S17: Compare the prediction model with the original data set, calculate the matching degree using the mean square error, and obtain the final analysis result.

[0108] Specifically, obtaining the inductance characteristic measurement data of the target sample set is the basic link of the entire process, and the purpose is to establish a reliable original data set through actual measurement.

[0109] For example, an inductor device test scenario can be envisioned where technicians use high-precision measurement equipment to collect data with a transmission distance of 5 meters and a working frequency of 10 MHz, while recording real-time signal integrity parameters such as a signal attenuation rate of 2 dB. After storing this data in a database, a structured raw dataset is formed, providing material for subsequent analysis.

[0110] It should be noted that the key to this step is to ensure accurate calibration of the measurement equipment to reduce initial errors. The process of collecting inductor characteristic data through sensors further refines the measurement process.

[0111] In one possible implementation, sensors are deployed in the test environment to monitor inductor characteristics in real-time at a transmission distance of 3 meters and a working frequency of 15 MHz, obtaining preliminary results such as a signal integrity parameter showing 90%.

[0112] Specifically, the high sensitivity of the sensors can capture minute changes, which lays the foundation for subsequent optimization.

[0113] Preferably, the sensors can also record environmental interference factors to ensure the comprehensiveness of the data. The purpose of using filtering methods to process the preliminary measurement results is to remove noise interference.

[0114] For example, assuming that the preliminary measurement results are contaminated with high-frequency noise, through a moving average filtering method, the filter length can be set to 5 to smooth the signal waveform and generate an optimized dataset.

[0115] Exemplarily, the original signal integrity parameter is increased from 85% to 92%, significantly improving the data quality.

[0116] It can be understood that this method is simple and efficient, especially suitable for scenarios with high real-time requirements. Its beneficial effect lies in improving the accuracy of subsequent analysis. If the signal integrity parameter in the optimized dataset exceeds a preset centralized threshold, such as a centralized threshold of 95% while the actual value is 90%, then the acquisition frequency needs to be adjusted to re-acquire the data.

[0117] In one embodiment, the acquisition frequency is adjusted from 10 Hz to 20 Hz, and after re-measurement, the signal integrity is increased to 96%, generating a corrected dataset. This dynamic adjustment method can effectively handle abnormal situations and ensure that the data meets the expected standards. When calculating the correlation index between the transmission distance and the working frequency based on the corrected dataset, trend analysis can be started.

[0118] For example, the number of measurement points is 10, the transmission distance increases from 2 meters to 6 meters, and the working frequency increases from 5 MHz to 20 MHz. The calculation results show a positive correlation trend between the two.

[0119] Specifically, this correlation index reveals the regularity of parameter changes, which helps to understand the performance of inductance characteristics under different conditions. Generating a prediction model through linear regression analysis method is one of the core technologies.

[0120] In one embodiment, a regression model is constructed with transmission distance and operating frequency as input variables and signal integrity as the output variable.

[0121] Exemplarily, the model predicts that when the transmission distance is 4 meters and the operating frequency is 12 MHz, the signal integrity is approximately 93%. The advantage of this method is that it can quantify the parameter relationship and provide a basis for optimizing the design. By comparing the prediction model with the original data set and calculating the matching degree using the mean square error, the effectiveness of the model can be verified.

[0122] For example, the mean square error between the predicted value and the actual value is 1.5, indicating a high model matching degree.

[0123] It should be noted that this comparison not only verifies the model accuracy but also provides data support for further improvement. Its beneficial effect is that through error analysis, the model can be continuously optimized to improve the prediction ability.

[0124] S2. Perform principal component analysis on the original data set, extract the eigenvectors of inductance characteristics, transmission distance, and operating frequency, and generate a quantization representation matrix.

[0125] Optionally, this step further includes:

[0126] Step S21. Process the original data set through principal component analysis method, extract the eigenvectors of inductance characteristics, transmission distance, and operating frequency, and generate a first quantization matrix.

[0127] Step S22. Use the K-means clustering method to group the first quantization matrix, and obtain a classification data set according to the distribution characteristics of the eigenvectors.

[0128] Step S23. For the classification data set, calculate the mean and variance of the eigenvectors within each group to determine the parameter distribution characteristics.

[0129] Step S24. If the parameter distribution characteristics exceed the characteristic threshold, re-group by adjusting the parameters of the K-means clustering to obtain a corrected classification data set.

[0130] Step S25. According to the corrected classification data set, extract the correlation index between the inductance characteristics and the transmission distance within each group, and generate a characteristic correlation table.

[0131] Step S26. Process the characteristic correlation table through linear regression method to generate a mapping relationship model between inductance characteristics and operating frequency, and obtain the prediction result.

[0132] Step S27. For the prediction result, calculate the deviation value from the first quantization matrix, judge the matching degree of the mapping relationship model, and generate the final data set.

[0133] Specifically, by processing the original data set through the principal component analysis method, the complex data of inductance characteristics, transmission distance, and operating frequency can be reduced in dimension, key feature vectors can be extracted, and the first quantization matrix can be formed.

[0134] For example, in an inductance test scenario, a technician can retrieve a data set containing multiple samples from a database and, through principal component analysis, identify several variables that have the greatest impact on signal changes.

[0135] Exemplarily, the transmission distance and operating frequency may be identified as the main components, generating a simplified quantization matrix for subsequent processing. The advantage of this method is that it reduces the data dimension while retaining the core information.

[0136] In a possible implementation, the K-means clustering method is used to group the first quantization matrix. According to the distribution characteristics of the feature vectors, the data is divided into several categories.

[0137] For example, assume that the matrix contains 100 sample points. A technician sets the K value to 3 and divides the data into three groups, each group reflecting a different inductance characteristic pattern.

[0138] Specifically, one group may correspond to short distance and low frequency, and another group may correspond to long distance and high frequency.

[0139] It should be noted that the key to grouping is to select an appropriate K value to reflect the internal law of the data. For the classified data set, calculating the mean and variance of the feature vectors within each group can reveal the parameter distribution characteristics.

[0140] For example, the mean transmission distance of the first group is 4 meters and the variance is small, indicating a high data concentration; the mean operating frequency of the second group is 15 MHz and the variance is large, indicating obvious fluctuations. This analysis helps to understand the stability of each group of data.

[0141] It can be understood that the calculation of the mean and variance provides a basis for subsequent adjustment. If the parameter distribution characteristics exceed the preset characteristic threshold, such as the variance exceeding 2, the parameters of the K-means clustering need to be adjusted to re-group.

[0142] Preferably, the K value can be adjusted from 3 to 4, and the clustering process is re-run to generate a corrected classified data set.

[0143] For example, after adjustment, the data is divided into four groups, and the variance of each group is controlled within 1.5. This dynamic adjustment method ensures the rationality of classification and lays a foundation for the next step of analysis. According to the corrected classification data set, the correlation index between the inductance characteristics and the transmission distance within each group is extracted to generate a characteristic correlation table.

[0144] In one embodiment, the technician analyzes a set of data and finds that when the transmission distance increases from 3 meters to 5 meters, the inductance characteristics show a linear increasing trend, and the correlation index is recorded as 0.8. This tabular data clearly shows the relationship between variables and is convenient for further processing. By processing the characteristic correlation table through the linear regression method, a mapping relationship model between the inductance characteristics and the operating frequency can be generated.

[0145] For example, based on a set of data, when the model predicts that the operating frequency is 10 MHz, the inductance characteristic value is 85%.

[0146] Specifically, linear regression quantifies the mapping relationship between parameters by fitting a trend line. The benefit of this method is that it provides a predictable reference result. For the prediction result, calculate the deviation value from the first quantization matrix, judge the matching degree of the model, and generate the final data set.

[0147] Exemplarily, if the average deviation between the predicted value and the actual value is 1.2, it is considered that the model has a high reliability.

[0148] In one possible implementation, the technician confirms through multiple verifications that the deviation is stable below 1.5, and the final data set can be used for subsequent applications. This deviation analysis ensures the consistency between the data and the model and provides reliable support for the optimization design.

[0149] S3. Construct a Bayesian probability model according to the quantization representation matrix, and calculate the signal integrity probability density function under a specific combination of transmission distance and operating frequency.

[0150] Optionally, this step further includes:

[0151] Step S31. Obtain a data matrix through the quantization matrix, and process the data matrix using the Bayesian model to obtain distribution characteristics.

[0152] Optionally, use the following calculation formula to determine the distribution characteristics:

[0153]

[0154] Among them, P(d,f) represents the probability density distribution characteristics of the signal during transmission, d represents the transmission distance, f represents the operating frequency, α represents the amplitude coefficient, μ represents the distance mean, σ represents the standard deviation, and β represents the frequency attenuation index.

[0155] Step S32: Determine the probability density according to the distribution characteristics, generate a parameter combination through the transmission distance and the operating frequency, and obtain the density function.

[0156] Step S33: Calculate the signal integrity for the density function to obtain the probability distribution result of the signal integrity.

[0157] Step S34: If the probability distribution result of the signal integrity is lower than the distribution threshold, recalculate the density function by adjusting the parameter combination to obtain the corrected distribution characteristics.

[0158] Step S35: Update the probability density according to the corrected distribution characteristics and generate a new density function using the calculation process.

[0159] Step S36: Determine the signal integrity through the new density function to obtain the final probability distribution characteristics.

[0160] Step S37: For the final probability distribution characteristics, use the data matrix to verify the calculation process and determine the consistency result.

[0161] Specifically, after obtaining the data matrix through the quantization matrix, using the Bayesian model for processing is an effective means.

[0162] Exemplarily, a technician extracts a data matrix containing multiple samples from an inductance test scenario, and the matrix records the original data of the transmission distance and the operating frequency. The Bayesian model derives the posterior distribution characteristics of these parameters by introducing the prior distribution and combining the observed data.

[0163] In a possible implementation, assuming that the prior distribution is based on historical test experience, the technician sets the mean of the transmission distance to 5 meters and the standard deviation to 1 meter. After calculation by the model, the obtained distribution characteristics reflect the central tendency and dispersion degree of the data. The advantage of this method is that it can integrate uncertainty factors and provide a more robust analysis basis. When determining the probability density according to the distribution characteristics, the technician needs to generate a parameter combination by combining the transmission distance and the operating frequency.

[0164] For example, a scenario can be imagined where the transmission distance range is set from 2 meters to 6 meters and the operating frequency range is from 5 MHz to 20 MHz. Through these combinations, a density function is generated to describe the distribution law of the signal under different parameters.

[0165] Specifically, the technician may find that when the transmission distance is 3 meters and the operating frequency is 10 MHz, the density function shows that the signal distribution is relatively concentrated. This way is convenient for subsequent analysis of the influence of parameters on the signal. When calculating the signal integrity for the density function, what the technician is concerned about is the probability that the signal remains stable during transmission.

[0166] In one embodiment, assume that integrity is defined by measuring the degree of signal amplitude attenuation. If the amplitude attenuation is less than a certain distribution threshold, the signal is considered intact.

[0167] Exemplarily, the calculation results show that when the transmission distance is 4 meters, the integrity probability is 90%, and it drops to 70% when the distance increases to 6 meters. This probability distribution result intuitively reflects the impact of parameter changes. If the probability of signal integrity is lower than a preset distribution threshold, such as 80%, it is necessary to adjust the parameter combination and recalculate.

[0168] For example, a technician adjusts the operating frequency from 15 MHz to 12 MHz, regenerates the density function, and finds that the integrity probability increases to 85%.

[0169] Preferably, this adjustment is based on an understanding of the distribution characteristics to ensure a reasonable direction for parameter optimization. The adjusted modified distribution characteristics provide a basis for subsequent updates. When updating the probability density, a technician generates a new density function according to the modified distribution characteristics.

[0170] It can be understood that the new density function makes the distribution closer to the actual test requirements through smoothing processing or parameter fine-tuning.

[0171] For example, the adjusted function shows that when the transmission distance is 5 meters, the signal integrity probability is stable at about 88%. This dynamic update method improves the adaptability of the analysis. After judging the signal integrity through the new density function, the final probability distribution characteristics are formed.

[0172] In one possible implementation, a technician verified that when the operating frequency is 8 MHz, the integrity probability reaches 92%, and it drops to 75% when the frequency increases to 18 MHz. This result provides a clear reference for parameter selection. For the final probability distribution characteristics, using a data matrix to verify consistency is a key step.

[0173] Specifically, a technician compares the calculation results with the original data matrix to check whether the distribution characteristics match the actual samples.

[0174] For example, the verification found that the deviation between the mean value of the transmission distance and the value recorded in the matrix is only 0.3 meters, indicating a high degree of consistency. This verification process ensures the reliability of the analysis and provides strong support for the optimized design.

[0175] S4. If the coverage of the probability density function at the preset signal integrity threshold is lower than the set value, mark the corresponding samples as the preliminary defective product set.

[0176] Optionally, this step further includes:

[0177] Step S41: According to the preset signal integrity threshold, use the statistical distribution function in the scipy library to construct a Gaussian distribution probability density function model.

[0178] Step S42: Use the random sampling function in the numpy library to perform Monte Carlo sampling analysis on the probability density function.

[0179] Step S43: Use the KernelDensity class in the scikit-learn library, and adopt the Gaussian kernel function to perform kernel density estimation on the sampled data, and calculate the coverage near the preset signal integrity threshold.

[0180] Step S44: Use the MinMaxScaler class in the scikit-learn library to perform min-max normalization on the coverage values to obtain a standardized coverage index.

[0181] Step S45: If the standardized coverage index is lower than the coverage threshold, mark the corresponding samples as preliminary defective products.

[0182] Step S46: Apply the KMeans clustering algorithm in the scikit-learn library to the set of preliminary defective product samples to divide them into different categories.

[0183] Step S47: Perform secondary screening and grading on the preliminary defective product samples according to the clustering results.

[0184] Specifically, when constructing a Gaussian distribution probability density function model according to the preset signal integrity threshold, it can be assumed that technicians need to analyze the influence of transmission distance on the signal.

[0185] Exemplarily, assume that the signal integrity threshold is set to an amplitude attenuation less than 10%. Technicians set the mean of the Gaussian distribution to 4 meters and the standard deviation to 0.8 meters based on historical data. This method utilizes the characteristics of statistical distribution, incorporates uncertain factors into the model, and facilitates the subsequent analysis of the stability of parameter combinations.

[0186] In a possible implementation, when performing Monte Carlo sampling analysis on the probability density function, technicians simulate the signal transmission scenario by randomly generating a large number of samples.

[0187] For example, set the transmission distance range to be from 3 meters to 5 meters, sample 10,000 times, and observe the distribution law of the signal integrity probability.

[0188] Specifically, the sampling results may show that when the distance is 3.5 meters, the integrity probability is relatively high, while when approaching 5 meters, the probability decreases. This method enhances the comprehensiveness of the analysis through a large number of random samples.

[0189] It should be noted that when using the Gaussian kernel function for kernel density estimation, technicians are concerned with the probability density coverage of the sampled data.

[0190] In one embodiment, assume that the kernel density estimation shows that near a transmission distance of 4 meters, the density curve is relatively steep, indicating that the signal integrity is relatively concentrated.

[0191] Preferably, this estimation method can smooth the noise of the sampled data, highlight the distribution trend, and provide a basis for subsequent threshold judgment.

[0192] It can be understood that when performing maximum-minimum normalization on the coverage, technicians map the original coverage values to the interval from 0 to 1.

[0193] For example, if the original range of the coverage is between 0.2 and 0.9, after normalization, the standardized index corresponding to a transmission distance of 4 meters may be 0.75, while it drops to 0.4 at 5 meters. This kind of standardization facilitates cross-sample comparison and unifies the evaluation criteria. If the standardized coverage index is lower than the set coverage threshold, such as 0.6, it is marked as a preliminary defective product.

[0194] Exemplarily, technicians find that the sample index at a transmission distance of 4.8 meters is 0.5 and is marked as defective. This marking method helps to quickly screen out potential problem samples and improve the analysis efficiency.

[0195] In one possible implementation, when applying a clustering algorithm to the preliminary defective product samples, technicians divide the samples into two categories: one is the attenuation caused by a longer transmission distance, and the other is caused by frequency interference.

[0196] For example, the clustering results show that samples from 4.5 meters to 5 meters are mostly classified into the first category. This division reveals the different causes of defective products and provides a direction for optimizing parameters.

[0197] Specifically, when performing secondary screening and grading according to the clustering results, technicians may preferentially adjust the transmission distance for the first category of samples and adjust the operating frequency for the second category of samples.

[0198] For example, after reducing the first category of samples to 4.2 meters, the integrity probability increases significantly. This grading process improves the pertinence of parameter adjustment and ensures that the analysis results are more in line with actual requirements.

[0199] S5. Obtain the inductance test data of the preliminary defective product set under extreme working conditions, and generate a set of probability correction coefficients by using Monte Carlo simulation.

[0200] Optionally, this step further includes:

[0201] Step S51. Obtain the original inductance test data from the preliminary defective product set under extreme working conditions.

[0202] Step S52: Preprocess the original data using the box - plot method to eliminate outliers and remove noise interference through wavelet transform.

[0203] Step S53: Based on the processed data, construct an inductance parameter model using the least - squares method, setting the inductance value and frequency as key variables.

[0204] Step S54: Use the Monte Carlo method to generate a large number of random samples according to the inductance parameter model to form a simulation data set.

[0205] Step S55: Conduct descriptive statistical analysis on the simulation data set, and calculate the mean, variance, and distribution characteristics of each parameter.

[0206] Step S56: According to the distribution characteristic information, use the maximum - likelihood estimation to generate an initial value of the probability correction coefficient.

[0207] Step S57: Iteratively optimize the correction coefficient through the gradient - descent method to make it approach the true distribution.

[0208] It can be understood that when obtaining the original inductance test data from the preliminary defective product set under extreme working conditions, technicians need to simulate the harsh environment in signal transmission.

[0209] For example, test the inductance value under high - temperature, high - humidity, or strong electromagnetic interference scenarios.

[0210] Exemplarily, assume that under the conditions of a temperature of 80 degrees Celsius and a humidity of 90%, the original inductance data collected may be 100 samples per group, ranging from 5 to 15 microhenries. This way can capture the parameter changes in extreme cases and lay a foundation for subsequent analysis.

[0211] Specifically, when preprocessing the original data using the box - plot method, technicians identify outliers by drawing a box - plot. For example, if in a certain test, the inductance value suddenly jumps to 50 microhenries, far beyond the inter - quartile range, it will be eliminated.

[0212] In a possible implementation, apply wavelet transform to further process the data to remove high - frequency noise interference.

[0213] For example, after the spike noise mixed in the original signal is smoothed, the fluctuation range of the inductance value is reduced to 8 to 12 microhenries. This preprocessing ensures the reliability of the data.

[0214] When constructing an inductance parameter model using the least - squares method based on the processed data, it should be noted that technicians use the inductance value and frequency as key variables.

[0215] Preferably, assuming that the frequency range is set from 1 kHz to 10 kHz, the model fitting results show that the inductance value shows a downward trend as the frequency increases.

[0216] For example, the inductance value is 10 μH at 5 kHz, and drops to 9 μH at 10 kHz. This model reflects the dynamic relationship between parameters.

[0217] In one embodiment, when generating random samples using the Monte Carlo method, the technician sets the frequency and inductance value ranges based on the foregoing model and generates 10,000 simulation samples.

[0218] Exemplarily, the simulation results may show that when the frequency is 3 kHz, the inductance values mostly concentrate in the range of 9.5 to 10.5 μH. This simulation with a large number of samples can reveal potential distribution laws.

[0219] Specifically, when performing descriptive statistical analysis on the simulation data set, the technician calculates the mean and variance.

[0220] For example, the mean inductance corresponding to a frequency of 5 kHz is 9.8 μH, and the variance is 0.3. This analysis quantifies the central tendency and dispersion degree of the data, providing a basis for subsequent steps.

[0221] It can be understood that when generating the initial value of the probability correction coefficient using the maximum likelihood estimation according to the distribution characteristic information, the technician assumes that the data follows a normal distribution.

[0222] In one possible implementation, based on the foregoing mean of 9.8 μH and variance of 0.3, the initial correction coefficient may be set to 1.2. This initial value can better approximate the true distribution characteristics.

[0223] For example, when iteratively optimizing the correction coefficient by the gradient descent method, the technician sets the learning rate to 0.01. After 50 iterations, the correction coefficient is adjusted from 1.2 to 1.15, minimizing the error between the simulated distribution and the actual test data.

[0224] In one embodiment, after applying the optimized coefficient to the model, the prediction accuracy of the inductance value is significantly improved. This iterative process enhances the adaptability of the model, ensuring that the analysis results are more in line with the actual situation.

[0225] S6. Adjust the parameter weights of the Bayesian model according to the set of probability correction coefficients to generate an optimized dynamic probability distribution model.

[0226] Optionally, this step further includes:

[0227] Step S61. Obtain the historical data set and the current data sample, and perform data cleaning and standardization processing on the data through data preprocessing techniques to generate a cleaned data set.

[0228] Step S62: Based on the cleaned dataset, use the maximum likelihood estimation method to calculate the initial parameters of the Bayesian model, and obtain the prior probability distribution of the model.

[0229] Step S63: According to the characteristics of the cleaned dataset, construct a set of probability correction coefficients, preliminarily adjust the model parameters, and generate an adjusted model.

[0230] Step S64: Use the Monte Carlo method to perform multiple random samplings on the adjusted model to generate multiple sets of candidate parameter sets.

[0231] Step S65: Evaluate the prediction performance of each group of candidate parameters through the cross-validation method, and select the optimal parameter combination.

[0232] Step S66: Based on the optimal parameter combination, update the posterior probability distribution of the Bayesian model to form a dynamic probability distribution model.

[0233] Step S67: Adopt the sliding window technique to continuously obtain new data, update the probability correction coefficients in real time, and maintain the dynamic optimization ability of the model.

[0234] Specifically, after obtaining the historical dataset and the current data sample, data preprocessing is the key first step.

[0235] Exemplarily, assume that the historical data contains inductance test records with a sample size of 5,000, and the current data is 1,000 inductance values collected in real time, ranging from 5 to 20 microhenries. During cleaning, technicians may find that some data has missing values due to sensor failures, such as 10 consecutive samples being zero. In one possible implementation, these missing values are filled by interpolation, for example, replacing them with the average of the previous and subsequent data. The normalization process then converts the data into a distribution with a mean of 0 and a variance of 1 to ensure the consistency of subsequent analysis. Based on the cleaned dataset, the maximum likelihood estimation is used to calculate the initial parameters of the Bayesian model.

[0236] It can be understood that technicians assume that the inductance values follow a certain distribution, such as a normal distribution.

[0237] Exemplarily, the calculation results show that the mean of the prior probability distribution is 10 microhenries and the variance is 2. Such initial parameters reflect the basic statistical characteristics of the data and lay the foundation for the model.

[0238] When constructing the set of probability correction coefficients, it should be noted that technicians adjust the model parameters according to the data characteristics.

[0239] Specifically, if it is found that the inductance value is low in a high-temperature environment, a correction coefficient can be introduced.

[0240] In one embodiment, the initial value of the coefficient is 1.1, which represents the compensation for high-temperature scenarios, and the adjusted model is closer to the actual distribution.

[0241] Preferably, when using the Monte Carlo method for random sampling, the technician generates 10,000 simulation samples.

[0242] For example, if the set frequency range is from 1 to 5 kHz, the sampling results may show that the inductance values are mostly concentrated between 9 and 11 μH. This method reveals potential laws through a large number of samples.

[0243] When evaluating candidate parameters through cross-validation, in one possible implementation, the technician divides the dataset into 5 folds.

[0244] Exemplarily, the prediction error of a certain set of parameters on the validation set is 0.5 μH, while that of another set is 0.8 μH, and finally the combination with the smaller error is selected. This method ensures the generalization ability of the parameters.

[0245] When updating the posterior probability of the Bayesian model based on the optimal parameters, specifically, the technician adjusts the distribution by combining new data.

[0246] For example, the posterior mean may change from 10 μH to 9.8 μH, reflecting the dynamic changes in the data. This update improves the adaptability of the model.

[0247] When using the sliding window technique to update the model in real time, it can be understood that the technician sets the window size to 500 data points.

[0248] In one embodiment, every time 100 new data points are collected, the window slides once and the correction coefficient is recalculated, such as adjusted from 1.1 to 1.05. This continuous optimization maintains the sensitivity of the model to the new environment and ensures the real-time and accuracy of the analysis.

[0249] S7. Using the dynamic probability distribution model to construct an adaptive detection algorithm, if the sample signal integrity probability value is lower than the dynamic threshold, it is classified into the final defective product set.

[0250] Optionally, this step further includes:

[0251] Step S71. Obtain the sample signal data, preprocess the signal to remove noise interference, and obtain the denoised signal data.

[0252] Step S72. Use Gaussian filtering to smooth the denoised signal data, and extract the mean and variance of the signal as features.

[0253] Step S73. Calculate the sample signal integrity probability value according to the extracted mean and variance features.

[0254] Step S74: Train a support vector machine model with historical data to dynamically adjust the dynamic threshold.

[0255] Step S75: If the sample signal integrity probability value is lower than the dynamic threshold, it is determined as a defective product.

[0256] Step S76: Use the K-means clustering algorithm to classify defective products and identify different types of defects.

[0257] Step S77: Generate a quality analysis report for each type of defective product. The report includes the distribution of defect types and improvement suggestions, which are used to optimize the production process.

[0258] Specifically, after obtaining the sample signal data, preprocessing is the key to removing noise interference.

[0259] Exemplarily, assume that the collected signal comes from an inductance test device, and the data contains 10,000 sampling points with a frequency range between 1 and 10 kHz.

[0260] It can be understood that noise may be introduced by power fluctuations or environmental interference, resulting in spikes or random jitters in the signal.

[0261] In a possible implementation, technicians use wavelet transform to separate high-frequency noise and retain low-frequency effective signals.

[0262] For example, if continuous 50 sampling points in a certain signal segment show abnormal fluctuations, after wavelet decomposition and extraction of the main component, a smoothed denoised signal data is obtained. Next, Gaussian filtering is performed on the denoised signal data to smooth the curve and extract features.

[0263] Specifically, Gaussian filtering weakens the small fluctuations in the signal through convolution operations.

[0264] In one embodiment, the filtering window size is set to 10 sampling points. After processing, the calculated signal mean is 8 microhenries and the variance is 1.5. This mean and variance reflect the overall trend and fluctuation range of the signal, providing basic data for subsequent analysis.

[0265] When calculating the sample signal integrity probability value based on the extracted features, it should be noted that this step aims to quantify the signal quality.

[0266] In a possible implementation, technicians assume that the signal features conform to a normal distribution and estimate the probability density based on the mean and variance.

[0267] For example, for a certain sample with a mean of 7.8 microhenries and a variance of 1.6, the calculated integrity probability value is 0.85. This probability value intuitively reflects the reliability of the signal.

[0268] Preferably, when training a support vector machine model with historical data to adjust the dynamic threshold, 5000 labeled historical signal data are used, including normal and defective product samples.

[0269] Specifically, the support vector machine differentiates between two types of data through a classification hyperplane, and the dynamic threshold may be adjusted from an initial 0.9 to 0.87.

[0270] For example, after a certain training, it is found that the dynamic threshold of 0.87 can better adapt to the characteristics of the new batch of data. This dynamic adjustment improves the flexibility of detection. If the integrity probability value is lower than the dynamic threshold, it is determined as a defective product.

[0271] For example, the probability value of a certain sample is 0.83, which is lower than 0.87 and is marked as defective. This determination method quickly screens out potential problem products by comparing the probability with the dynamic threshold.

[0272] It can be understood that when using the K-means clustering algorithm to classify defective products, the method groups samples according to features.

[0273] In one embodiment, the number of clusters is set to 3, and the classification results show that the mean value of one type of defective product is low, and the variance of the other type is too large.

[0274] For example, the category with a low mean value may point to material defects, while the category with a large variance may be related to equipment jitter. This classification reveals the diversity of defects.

[0275] When generating a quality analysis report for various types of defective products, specifically, the report includes the distribution of defect types, such as 30% with a low mean value and 20% with abnormal variance.

[0276] Exemplarily, the improvement suggestions may include adjusting the equipment calibration frequency or replacing the supplier's materials. Such a report provides a direct basis for production optimization, helps reduce the defective rate and improve process stability.

[0277] S8. Obtain the misjudgment sample features in the historical detection data, analyze the correlation between the misjudgment features and physical parameters through the random forest algorithm, and generate a set of dynamic optimization parameters.

[0278] Optionally, this step further includes:

[0279] Step S81. Obtain the misjudgment sample records in the historical detection data, and extract the sample feature vectors and the corresponding physical parameter values.

[0280] Step S82. Perform dimensionality reduction processing on the misjudgment sample feature vectors to obtain the reduced feature dimension.

[0281] Step S83. Use the reduced feature dimension as the analysis object, and adopt the random forest algorithm to construct a relationship model between the misjudgment sample features and physical parameters.

[0282] Step S84: Calculate the feature importance scores, sort the physical parameters according to the scores, and filter out a subset of key parameters with greater influence.

[0283] Step S85: Optimize the key parameters through multi-dimensional combination by the grid search method to generate a set of candidate parameters.

[0284] Step S86: Conduct cross-validation on the set of candidate parameters, evaluate the performance of each combination on new samples, and select the optimal parameter combination.

[0285] Step S87: Update the detection model with the optimal parameter combination to achieve dynamic optimization and improve the detection accuracy.

[0286] Specifically, when obtaining the misjudgment sample records in the historical detection data and extracting the sample feature vectors and the corresponding physical parameter values, it can be understood that this step aims to mine the wrong samples from the past data to lay a foundation for subsequent analysis.

[0287] Exemplarily, assume that an inductance test device has recorded 1000 pieces of historical data, and 50 of them are marked as misjudgments. These misjudgment samples may include feature vectors such as signal amplitude and frequency offset, as well as corresponding physical parameters such as voltage and current.

[0288] Specifically, a technician may extract a misjudgment sample from the database. Its feature vector shows an amplitude of 5 volts and a frequency offset of 200 Hz, and the corresponding physical parameters are an input voltage of 12 volts and a current of 0.8 amperes. This data extraction provides the original basis for subsequent modeling. When performing dimensionality reduction on the misjudgment sample feature vector to obtain the reduced feature dimension.

[0289] It should be noted that dimensionality reduction can reduce data redundancy and facilitate analysis.

[0290] In a possible implementation, a technician uses the principal component analysis method to compress the original 10-dimensional feature vector to 3 dimensions.

[0291] Exemplarily, after processing multi-dimensional features such as the amplitude and frequency offset of a sample, the main information is retained, and the dimensionality reduction result may be 3 new feature values, such as 2.5, 1.8, and 0.9. This simplification retains the core characteristics and improves the calculation efficiency. When using the reduced feature dimension as the analysis object and adopting the random forest algorithm to construct a relationship model between the misjudgment sample features and the physical parameters.

[0292] Preferably, the random forest can handle non-linear relationships and is suitable for complex data.

[0293] In one embodiment, a technician inputs 50 misjudgment samples after dimensionality reduction, and the training model finds that the frequency shift is closely related to the current parameter.

[0294] For example, when the frequency shift of a certain sample is 150 Hz, the current fluctuates to 0.9 A, and the model establishes an association based on this. This method helps to reveal the potential laws of misjudgment.

[0295] When calculating the feature importance scores, sorting the physical parameters according to the scores, and screening out the subset of key parameters with greater influence, specifically, the scores reflect the influence degree of each parameter on misjudgment.

[0296] In one embodiment, the random forest output shows that the current score is 0.45, the voltage is 0.25, and the temperature is 0.1. The technician then screens out the current and voltage as the key parameters based on this. This sorting clarifies the optimization focus.

[0297] When performing multi-dimensional combination optimization on the key parameters through the grid search method to generate a candidate parameter set, it can be understood that the grid search can exhaustively combine to find the optimal solution.

[0298] Exemplarily, for a current of 0.8 to 1.0 A and a voltage of 11 to 13 V, 9 groups of candidate combinations are generated, such as {0.8 A, 11 V}, {0.9 A, 12 V}, etc. This comprehensive exploration provides diversity for subsequent selection. Cross-validate the candidate parameter set, evaluate the performance of each combination on new samples, and select the optimal parameter combination.

[0299] For example, using 10-fold cross-validation to test 9 groups of parameters, it is found that {0.9 A, 12 V} has the lowest misjudgment rate among 100 new samples. When updating the detection model with the optimal parameter combination to achieve dynamic optimization and improve the detection accuracy, this verification ensures the adaptability of the parameters.

[0300] In one possible implementation, applying {0.9 A, 12 V} to the model and retesting 500 pieces of data, the accuracy rate is increased from 85% to 92%. This dynamic adjustment optimizes the detection process and reduces the risk of misjudgment.

[0301] S9. Update the decision threshold of the adaptive detection algorithm according to the dynamic optimization parameter set, and output the real-time updated detection result.

[0302] Optionally, this step further includes:

[0303] Step S91. Construct an initial parameter set according to historical data and set the initial threshold of the detection algorithm.

[0304] Step S92: Obtain the latest data samples using the sliding window method. The size of the sliding window is 100 data points, and calculate the mean and standard deviation of the samples as statistical features.

[0305] Step S93: Determine whether there are abnormal situations by comparing the sample mean with the initial threshold of the detection algorithm, and mark the abnormal data points.

[0306] Step S94: Based on the marked abnormal data points, adjust the parameter weights and update the parameter set.

[0307] Step S95: Recalculate the initial threshold of the detection algorithm using the updated parameter set to obtain the optimized detection threshold.

[0308] Step S96: Detect the real-time data stream using the optimized detection threshold and output the detection result.

[0309] Step S97: Evaluate the algorithm performance according to the detection result, adjust the optimization strategy, and enter the next iteration.

[0310] Specifically, when constructing the initial parameter set based on historical data and setting the initial threshold of the detection algorithm, it can be understood that this process aims to provide a benchmark for subsequent detections.

[0311] Exemplarily, technicians extract key parameters from the past 5000 inductance test data, such as a voltage range of 10 to 15 volts and a current range of 0.5 to 1.2 amperes, to generate the initial parameter set.

[0312] Preferably, the initial threshold of the detection algorithm can be set to a voltage mean of 12 volts and a current mean of 0.8 amperes as a preliminary judgment basis. When using the sliding window method to obtain the latest data samples and calculate statistical features, this method utilizes historical rules to ensure a reasonable starting point.

[0313] It should be noted that the sliding window can capture dynamic changes. In one possible implementation, the window size is set to 100 data points, and the output data of the inductance device is collected in real time.

[0314] For example, the sample voltage mean within a certain window is 12.3 volts, the standard deviation is 0.4 volts, the current mean is 0.85 amperes, and the standard deviation is 0.05 amperes. When determining abnormal situations and marking data points by comparing the sample mean with the initial threshold of the detection algorithm, these statistical features reflect the central tendency and dispersion degree of the data, laying a foundation for anomaly detection.

[0315] Specifically, the data exceeding the range of the initial threshold of the detection algorithm is regarded as abnormal.

[0316] In one embodiment, if the average voltage of a certain window is 13.5 volts, exceeding the ±10% range of the initial threshold of 12 volts of the detection algorithm, it is marked as an abnormal point. Similarly, if the average current is 1.3 amperes, exceeding the detection algorithm initial threshold range of 0.8 amperes, it will also be marked. When adjusting the parameter weights based on the marked abnormal data points and updating the parameter set, this comparison method is intuitive and efficient, facilitating the rapid identification of problems.

[0317] Preferably, the abnormal data reveals the deficiencies of the initial parameters.

[0318] For example, if technicians find that the abnormal voltage points are mostly concentrated between 13 and 14 volts, the weight can be increased from 0.3 to 0.5, while the current weight remains unchanged at 0.4. When recalculating the initial threshold of the detection algorithm using the updated parameter set for optimization, the updated parameter set is closer to the actual data distribution, enhancing adaptability.

[0319] It can be understood that this step enhances the dynamic nature of the detection threshold.

[0320] In one embodiment, the adjusted voltage threshold becomes 12.5 volts and the current threshold becomes 0.9 amperes. Using the optimized detection threshold to detect the real-time data stream and output the results, this optimized detection threshold can better reflect the latest trends and reduce the possibility of misjudgment.

[0321] For example, technicians apply the optimized detection threshold to 1000 continuously collected data streams, mark the abnormal points and generate a detection report. In one possible implementation, the system real-time displays that a data point with a voltage of 13.8 volts and a current of 0.95 amperes is abnormal, reminding the operator to check. When evaluating the algorithm performance based on the detection results and adjusting and iterating the optimization strategy, this real-time nature improves the response speed.

[0322] Specifically, the performance evaluation provides a direction for continuous improvement.

[0323] Exemplarily, if the abnormal marking accuracy rate reaches 90% after detecting 1000 data, technicians may further narrow the window to 50 data points to test whether it is more sensitive. This iterative mechanism ensures that the algorithm is gradually improved and more adaptable.

[0324] Please refer to Figure 2 , the second aspect of the present invention provides a system for intelligently detecting data lines based on inductance characteristics, using the method described above to intelligently detect data lines. The system further includes:

[0325] A data acquisition module for acquiring inductance characteristic measurement data of the target sample set. The measurement data includes transmission distance, operating frequency, and real-time signal integrity parameters, and stores them in a database to form an original data set;

[0326] A feature extraction module, which is used to perform principal component analysis on the original data set, extract feature vectors of inductance characteristics, transmission distance, and operating frequency, and generate a quantization representation matrix;

[0327] A probability modeling module, which is used to construct a Bayesian probability model according to the quantization representation matrix and calculate the signal integrity probability density function under a specific combination of transmission distance and operating frequency;

[0328] A preliminary screening module, which is used to mark the corresponding samples as a preliminary defective product set if the coverage of the probability density function below a preset signal integrity threshold is lower than a set value;

[0329] An extreme test module, which is used to obtain the inductance test data of the preliminary defective product set under extreme working conditions and generate a set of probability correction coefficients by using Monte Carlo simulation;

[0330] A model optimization module, which is used to adjust the parameter weights of the Bayesian model according to the set of probability correction coefficients and generate an optimized dynamic probability distribution model;

[0331] An adaptive detection module, which is used to construct an adaptive detection algorithm by using the dynamic probability distribution model. If the sample signal integrity probability value is lower than the dynamic threshold, it is classified into the final defective product set;

[0332] A misjudgment analysis module, which is used to obtain the misjudgment sample characteristics in the historical detection data, analyze the correlation between the misjudgment characteristics and physical parameters by using the random forest algorithm, and generate a set of dynamic optimization parameters;

[0333] A dynamic update module, which is used to update the decision threshold of the adaptive detection algorithm according to the set of dynamic optimization parameters and output the real-time updated detection result.

[0334] The present invention discloses a method for intelligently detecting data lines based on inductance characteristics. By obtaining the inductance characteristic measurement data of a target sample set, performing principal component analysis to extract feature vectors, constructing a Bayesian probability model to evaluate signal integrity, for the preliminary defective product set, using Monte Carlo simulation to generate probability correction coefficients, optimizing the dynamic probability distribution model, further using an adaptive detection algorithm to determine the final defective products, and combining random forest to analyze historical misjudgment characteristics and dynamically update the detection threshold. The present invention effectively solves the misjudgment problem in inductance characteristic detection, improves the detection accuracy and reliability, and realizes the intelligence and self-adaptability of inductance characteristic detection by integrating a variety of advanced algorithms and dynamic optimization strategies, providing an innovative solution for the quality control of electronic components.

[0335] It should be noted that the above are only several specific embodiments of the present invention. Obviously, the present invention is not limited to the above embodiments and there can be many variations. All variations that can be directly derived or associated by those of ordinary skill in the art from the disclosed content of the present invention shall be considered as within the protection scope of the present invention.

Claims

1. A method for intelligently detecting a data line based on inductance characteristics, characterized in that The method includes: S1. Obtain the inductance characteristic measurement data of the target sample set. The measurement data includes transmission distance, operating frequency, and real-time signal integrity parameters, and store them in a database to form an original data set. S2. Perform principal component analysis on the original data set, extract the eigenvectors of inductance characteristics, transmission distance, and operating frequency, and generate a quantization representation matrix. S3. Construct a Bayesian probability model according to the quantization representation matrix, and calculate the signal integrity probability density function under a specific combination of transmission distance and operating frequency. S4. If the coverage of the probability density function below a preset signal integrity threshold is lower than a set value, mark the corresponding samples as a preliminary defective product set. S5. Obtain the inductance test data of the preliminary defective product set under extreme working conditions, and use Monte Carlo simulation to generate a set of probability correction coefficients. S6. Adjust the parameter weights of the Bayesian model according to the set of probability correction coefficients to generate an optimized dynamic probability distribution model. S7. Construct an adaptive detection algorithm using the dynamic probability distribution model. If the sample signal integrity probability value is lower than the dynamic threshold, classify it into the final defective product set. S8. Obtain the misjudgment sample characteristics in the historical detection data, analyze the correlation between the misjudgment characteristics and physical parameters through a random forest algorithm, and generate a set of dynamically optimized parameters. S9. Update the decision threshold of the adaptive detection algorithm according to the set of dynamically optimized parameters, and output the real-time updated detection result.

2. The method according to claim 1, wherein The step S1, obtaining the inductance characteristic measurement data of the target sample set. The measurement data includes transmission distance, operating frequency, and real-time signal integrity parameters, and storing them in a database to form an original data set, includes: Step S11. Obtain the inductance characteristic measurement data of the target sample set, including transmission distance, operating frequency, and real-time signal integrity parameters, and store them in a database to generate an original data set. Step S12. Collect the inductance characteristic data of the target sample through a sensor, including transmission distance, operating frequency, and real-time signal integrity parameters, to obtain a preliminary measurement result. Step S13. Process the preliminary measurement result using a filtering method to remove noise interference and generate an optimized data set. Step S14. If the signal integrity parameter in the optimized data set exceeds the centralized threshold, re-obtain the data by adjusting the acquisition frequency to obtain a corrected data set. Step S15. Calculate the correlation index between the transmission distance and the operating frequency according to the corrected data set, and determine the parameter change trend. Step S16. Process the parameter change trend through a linear regression analysis method to generate a prediction model of inductance characteristics. Step S17. Compare the prediction model with the original data set, calculate the matching degree using the mean square error, and obtain the final analysis result.

3. The method according to claim 1, wherein The step S2, performing principal component analysis on the original data set, extracting the eigenvectors of inductance characteristics, transmission distance, and operating frequency, and generating a quantization representation matrix, includes: Step S21. Process the original data set through a principal component analysis method, extract the eigenvectors of inductance characteristics, transmission distance, and operating frequency, and generate a first quantization matrix. Step S22: Group the first quantization matrix using the K-means clustering method, and obtain a classification data set based on the distribution characteristics of the eigenvectors. Step S23: For the classification data set, calculate the mean and variance of the eigenvectors within each group to determine the parameter distribution characteristics. Step S24: If the parameter distribution characteristics exceed the characteristic threshold, re-group by adjusting the parameters of the K-means clustering to obtain a corrected classification data set. Step S25: According to the corrected classification data set, extract the correlation index between the inductance characteristics and the transmission distance within each group, and generate a characteristic correlation table. Step S26: Process the characteristic correlation table by the linear regression method to generate a mapping relationship model between the inductance characteristics and the operating frequency, and obtain a prediction result. Step S27: For the prediction result, calculate the deviation value from the first quantization matrix, judge the matching degree of the mapping relationship model, and generate a final data set.

4. The method according to claim 1, wherein In step S3, a Bayesian probability model is constructed according to the quantization representation matrix, and the signal integrity probability density function under a specific combination of transmission distance and operating frequency is calculated, including: Step S31: Obtain a data matrix through the quantization matrix, and process the data matrix using the Bayesian model to obtain distribution characteristics. Step S32: Determine the probability density according to the distribution characteristics, generate a parameter combination through the transmission distance and the operating frequency, and obtain a density function. Step S33: Calculate the signal integrity for the density function to obtain the probability distribution result of the signal integrity. Step S34: If the probability distribution result of the signal integrity is lower than the distribution threshold, re-calculate the density function by adjusting the parameter combination to obtain corrected distribution characteristics. Step S35: Update the probability density according to the corrected distribution characteristics, and generate a new density function using the calculation process. Step S36: Judge the signal integrity through the new density function to obtain the final probability distribution characteristics. Step S37: For the final probability distribution characteristics, verify the calculation process using the data matrix to determine the consistency result.

5. The method according to claim 1, wherein In step S4, if the coverage of the probability density function below a preset signal integrity threshold is lower than a set value, mark the corresponding samples as a preliminary defective product set, including: Step S41: According to the preset signal integrity threshold, construct a Gaussian distribution probability density function model using the statistical distribution function in the scipy library. Step S42: Perform Monte Carlo sampling analysis on the probability density function using the random sampling function in the numpy library. Step S43: Use the KernelDensity class in the scikit-learn library, and perform kernel density estimation on the sampled data using the Gaussian kernel function to calculate the coverage of the probability density function near the preset signal integrity threshold. Step S44: Use the MinMaxScaler class in the scikit-learn library to perform min-max normalization processing on the coverage values to obtain a normalized coverage index. Step S45: If the normalized coverage index is lower than the coverage threshold, mark the corresponding samples as preliminary defective products. Step S46: Apply the KMeans clustering algorithm in the scikit-learn library to the preliminary defective product sample set to divide different categories. Step S47, performing secondary screening and grading on the preliminary defective product samples according to the clustering results.

6. The method according to claim 1, wherein The step S5, obtaining the inductance test data of the preliminary defective product set under extreme working conditions, and generating a probability correction coefficient set by Monte Carlo simulation, includes: Step S51, obtaining raw data of inductance testing under extreme working conditions from a preliminary defective product set; Step S52, preprocessing the raw data using a box plot method, removing outliers, and removing noise interference by wavelet transform; Step S53, based on the processed data, the inductance parameter model is constructed using the least square method, and the inductance value and frequency are set as key variables; Step S54, using the Monte Carlo method to generate a large number of random samples according to the inductance parameter model to form a simulation data set; Step S55, performing descriptive statistical analysis on the simulated data set to calculate the mean, variance and distribution characteristics of each parameter; Step S56, generating an initial value of the probability correction coefficient using maximum likelihood estimation according to the distribution characteristic information; Step S57, iteratively optimizing the correction coefficient by gradient descent method to make it close to the real distribution.

7. The method according to claim 1, characterized in that The step S6, adjusting the parameter weights of the Bayesian model according to the probability correction coefficient set to generate an optimized dynamic probability distribution model, includes: Step S61, obtaining historical data sets and current data samples, cleaning and standardizing the data using data preprocessing technology, and generating a cleaned data set; Step S62, based on the cleaned data set, the initial parameters of the Bayesian model are calculated using the maximum likelihood estimation method to obtain the prior probability distribution of the model; Step S63, constructing a probability correction coefficient set according to the characteristics of the cleaned data set, making preliminary adjustments to the model parameters, and generating an adjusted model; Step S64, performing multiple random sampling on the adjusted model using the Monte Carlo method to generate multiple sets of candidate parameter sets; Step S65, evaluating the prediction performance of each group of candidate parameters by a cross-validation method, and selecting the optimal parameter combination; Step S66, based on the optimal parameter combination, updating the posterior probability distribution of the Bayesian model to form a dynamic probability distribution model; Step S67, using sliding window technology, continuously acquires new data, updates the probability correction coefficient in real time, and maintains the dynamic optimization capability of the model.

8. The method according to claim 1, wherein The step S7, using the dynamic probability distribution model to construct an adaptive detection algorithm, if the sample signal integrity probability value is lower than the dynamic threshold, is classified into the final defective product set, including: Step S71, obtaining sample signal data, preprocessing the signal to remove noise interference, and obtaining denoised signal data; Step S72, using Gaussian filtering to smooth the denoised signal data, and extracting the mean and variance of the signal as features; Step S73, calculating the sample signal integrity probability value according to the extracted mean and variance features; Step S74, training the support vector machine model through historical data and dynamically adjusting the dynamic threshold; Step S75, if the sample signal integrity probability value is lower than the dynamic threshold, it is determined to be a defective product; Step S76, using K-means clustering algorithm to classify defective products and identify different types of defects; Step S77: Generate a quality analysis report for various types of defective products. The report includes the distribution of defect types and improvement suggestions, which are used to optimize the production process.

9. The method according to claim 1, wherein In step S8, obtain the misjudgment sample characteristics in the historical detection data, analyze the correlation between the misjudgment characteristics and physical parameters through the random forest algorithm, and generate a set of dynamically optimized parameters, including: Step S81: Obtain the misjudgment sample records in the historical detection data, and extract the sample feature vectors and corresponding physical parameter values. Step S82: Perform dimensionality reduction processing on the misjudgment sample feature vectors to obtain the dimensionality-reduced feature dimensions. Step S83: Use the dimensionality-reduced feature dimensions as the analysis object, and construct a relationship model between the misjudgment sample characteristics and physical parameters using the random forest algorithm. Step S84: Calculate the feature importance scores, sort the physical parameters according to the scores, and screen out the subset of key parameters with greater influence. Step S85: Perform multi-dimensional combination optimization on the key parameters through the grid search method to generate a set of candidate parameters. Step S86: Perform cross-validation on the set of candidate parameters, evaluate the performance of each combination on new samples, and select the optimal parameter combination. Step S87: Update the detection model using the optimal parameter combination to achieve dynamic optimization and improve the detection accuracy.

10. A system for intelligently detecting data lines based on inductance characteristics, characterized in that, Use the method described in any one of claims 1 to 9 to perform intelligent detection on the data cable. The system includes: A data acquisition module for acquiring the inductance characteristic measurement data of the target sample set. The measurement data includes the transmission distance, operating frequency, and real-time signal integrity parameters, and stores them in the database to form an original data set. A feature extraction module for performing principal component analysis on the original data set, extracting the feature vectors of inductance characteristics, transmission distance, and operating frequency, and generating a quantization representation matrix. A probability modeling module for constructing a Bayesian probability model based on the quantization representation matrix, and calculating the signal integrity probability density function under a specific combination of transmission distance and operating frequency. A preliminary screening module for marking the corresponding samples as a preliminary defective product set if the coverage of the probability density function below the preset signal integrity threshold is lower than the set value. An extreme test module for obtaining the inductance test data of the preliminary defective product set under extreme working conditions, and generating a set of probability correction coefficients using Monte Carlo simulation. A model optimization module for adjusting the parameter weights of the Bayesian model according to the set of probability correction coefficients to generate an optimized dynamic probability distribution model. An adaptive detection module for constructing an adaptive detection algorithm using the dynamic probability distribution model, and classifying the samples into the final defective product set if the sample signal integrity probability value is lower than the dynamic threshold. A misjudgment analysis module for obtaining the misjudgment sample characteristics in the historical detection data, analyzing the correlation between the misjudgment characteristics and physical parameters through the random forest algorithm, and generating a set of dynamically optimized parameters. A dynamic update module for updating the decision threshold of the adaptive detection algorithm according to the set of dynamically optimized parameters, and outputting the real-time updated detection results.

Images