HHT-based transformer silicon steel sheet fault diagnosis method, system and device
By processing and filtering magnetic Barkhausen noise signals using a HHT-based method and combining it with a support vector machine model, the accuracy problem of transformer silicon steel sheet condition detection was solved, achieving high-precision defect detection.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINA UNIV OF GEOSCIENCES (WUHAN)
- Filing Date
- 2024-06-07
- Publication Date
- 2026-08-04
AI Technical Summary
Existing non-destructive testing technologies suffer from inaccuracy and large errors in the condition detection of silicon steel sheets in transformer cores, and lack intelligent evaluation models that can clearly characterize the condition of silicon steel sheets in transformer cores.
A method based on HHT is used to process magnetic Barkhausen noise signals, including noise reduction, empirical mode decomposition, and screening of intrinsic mode functions. Defect detection is performed by combining a support vector machine model. The detection accuracy is improved by constructing a threshold screening method for correlation coefficients and feature vector preprocessing.
It achieves high-precision and robust detection of defects in silicon steel sheets of transformers, reduces errors, and improves the accuracy and reliability of detection.
Smart Images
Figure CN118760958B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of nondestructive testing of ferromagnetic materials, and more specifically, to a method, system, and equipment for fault diagnosis of silicon steel sheets in transformers based on HHT. Background Technology
[0002] Transformers are among the most critical pieces of equipment in power systems. Their function extends beyond voltage transformation to include power transmission and distribution, playing a particularly vital role in modern power grids. They ensure that all equipment receives adequate power, thereby guaranteeing the safe, stable, and efficient operation of the power system. Despite their crucial position, transformers may face various problems during operation, such as aging, faults, and losses, which can severely impact the power system. Furthermore, the use of secondary silicon steel sheets in the manufacture of transformer core silicon steel sheets in recent years has introduced risks to transformer quality. Therefore, it is necessary to research methods for inspecting the condition of silicon steel sheets in formed transformers. From the perspective of whether the tested parts are damaged, core inspection methods can be broadly categorized into destructive and non-destructive testing. Non-destructive testing methods suffer from inaccuracies in classification and prediction, with proneness to misclassification and large prediction errors. Destructive testing methods can achieve reliable and accurate detection, but they can damage the product, making it difficult to meet engineering quality requirements. Therefore, non-destructive, non-contact, accurate, and efficient classification and inspection of silicon steel sheets is of great significance.
[0003] Magnetic Barkhausen Noise (MBN) technology, as a novel non-destructive testing technique, can detect and assess the performance degradation and micro-damage of silicon steel sheets. It can determine the surface stress state, fatigue damage status, and microstructure changes of silicon steel sheets in the early stages of their use, thereby enabling early detection of damaged areas and providing a reliable basis for safety evaluation and remaining service assessment of silicon steel sheets.
[0004] Currently, in terms of eigenvalue extraction, scholars both domestically and internationally, such as Wang Ping, Kizkitza, and Dong Haijiang, use root mean square (RMS), peak width, MBN signal peak value, peak-to-peak value, reciprocal of peak amplitude, excitation signal parameters, skewness, slow features, and fusion of magnetic feature parameters to reflect material stress and defect fatigue. Regarding the establishment of quantitative evaluation models, scholars both domestically and internationally, such as Kypris, Ma Xianyao, and Shu Di, employ spectral models, autoregressive models, backpropagation neural networks, Bayesian learning models, and Gaussian mixture models to quantitatively evaluate material stress and defect fatigue. However, current MBN signal extraction methods and evaluation models are mainly used to reflect the stress magnitude of ferromagnetic materials, and research on MBN signal extraction methods and defect evaluation models for characterizing defects in transformer silicon steel sheets is still insufficient.
[0005] After nearly 30 years of development, Barkhausen testing technology has achieved relatively mature applications. However, based on the aforementioned research on Barkhausen testing technology, it can be found that there is still some room for improvement and enhancement:
[0006] (1) The Barkhausen signal is a non-stationary nonlinear signal. Currently, the characteristic values commonly used to reflect the magnitude of the MBN signal include root mean square, peak value, and full width at half maximum (FWHM). However, due to the high randomness of the Barkhausen signal, the detection results of these signal characteristics vary greatly at different times, which can lead to large errors. Existing MBN signal characteristic values are difficult to effectively characterize the state of the object being measured. It is necessary to study time-frequency characteristic values that are more correlated with the state of silicon steel sheets to improve detection accuracy.
[0007] (2) Few studies have been conducted on the relationship between the MBN signal feature values extracted from the side detection of silicon steel sheets in transformer cores and the secondary sheets. It is still unclear which feature values are closely related to the secondary sheets of transformer cores, and there is a lack of intelligent evaluation models that can clearly characterize the state of silicon steel sheets in transformer cores. Summary of the Invention
[0008] The purpose of this invention is to provide a method, system, and equipment for fault diagnosis of transformer silicon steel sheets based on HHT, which can improve the accuracy of defect detection of transformer silicon steel sheets.
[0009] This invention provides a fault diagnosis method for silicon steel sheets in transformers based on HHT, comprising the following steps: S1: acquiring the magnetic Barkhausen noise signal of the silicon steel sheet to be tested, and performing noise reduction processing on the magnetic Barkhausen noise signal of the silicon steel sheet to be tested to obtain the processed magnetic Barkhausen noise signal; S2: performing empirical mode decomposition and screening on the processed magnetic Barkhausen noise signal to obtain the screened intrinsic mode functions; S3: based on the screened intrinsic mode functions, using a support vector machine model, obtaining the defect detection result of the silicon steel sheet to be tested.
[0010] Furthermore, step S2 of the above-mentioned HHT-based transformer silicon steel sheet fault diagnosis method includes the following steps: S21: Perform empirical mode decomposition on the magnetic Barkhausen noise signal to be processed to obtain the intrinsic mode functions and residuals, as shown in the formula:
[0011]
[0012] Where x(t) is the Barkhausen noise signal to be processed, n is the order of the intrinsic mode function, and c i r(t) is the i-th intrinsic mode function, and r(t) is the residual.
[0013] S22: Based on the intrinsic mode function, the correlation coefficient is obtained, as shown in the formula:
[0014]
[0015] Where, r i Let be the correlation coefficient of the i-th order intrinsic mode function sequence, N be the order of the intrinsic mode function sequence, x(u) be the original signal sequence, and C be the correlation coefficient of the i-th order intrinsic mode function sequence. i (u) is the sequence of intrinsic mode functions of order i; S23: Based on the correlation coefficient, the threshold is obtained as shown in the formula:
[0016]
[0017] Where ρ is the threshold and rmax is the maximum correlation coefficient; S24: Based on the threshold, the screening intrinsic mode function is obtained.
[0018] Furthermore, step S3 of the above-mentioned HHT-based transformer silicon steel sheet fault diagnosis method includes the following steps: S31: preprocessing the selected intrinsic mode functions to obtain magnetic Barkhausen noise feature vectors; S32: using the magnetic Barkhausen noise feature vectors to obtain training and test sets; using the training set to train the support vector machine model to obtain a trained support vector machine model; and using the test set and the trained support vector machine model to obtain the defect detection results of the silicon steel sheet to be tested.
[0019] The present invention also provides a system comprising the following modules: a signal acquisition module configured to acquire the magnetic Barkhausen noise signal of the silicon steel sheet under test, and to perform noise reduction processing on the magnetic Barkhausen noise signal of the silicon steel sheet under test to obtain the processed magnetic Barkhausen noise signal; a signal decomposition and filtering module configured to perform empirical mode decomposition and filtering on the processed magnetic Barkhausen noise signal to obtain the filtered intrinsic mode functions; and a defect detection module configured to obtain the defect detection result of the silicon steel sheet under test based on the filtered intrinsic mode functions and using a support vector machine model.
[0020] Furthermore, the signal decomposition and filtering module of the above system is specifically configured as follows: Empirical mode decomposition is performed on the magnetic Barkhausen noise signal to be processed to obtain the intrinsic mode functions and residuals, as shown in the formula:
[0021]
[0022] Where x(t) is the Barkhausen noise signal to be processed, n is the order of the intrinsic mode function, and c i Let r(t) be the i-th intrinsic mode function, and r(t) be the residual. Based on the intrinsic mode function, the correlation coefficient is obtained as shown in the formula:
[0023]
[0024] Where, r iLet be the correlation coefficient of the i-th order intrinsic mode function sequence, N be the order of the intrinsic mode function sequence, x(u) be the original signal sequence, and C be the correlation coefficient of the i-th order intrinsic mode function sequence. i (u) is the sequence of intrinsic mode functions of order i; the threshold is obtained based on the correlation coefficient, as shown in the formula:
[0025]
[0026] Where ρ is the threshold and rmax is the maximum correlation coefficient; based on the threshold, the screening intrinsic mode function is obtained.
[0027] Furthermore, the defect detection module of the above system is specifically configured as follows: preprocessing the selected intrinsic mode functions to obtain magnetic Barkhausen noise feature vectors; using the magnetic Barkhausen noise feature vectors to obtain training and test sets; using the training set to train the support vector machine model to obtain a trained support vector machine model; and using the test set and the trained support vector machine model to obtain the defect detection results of the silicon steel sheet to be tested.
[0028] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described HHT-based transformer silicon steel sheet fault diagnosis method.
[0029] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the above-described HHT-based transformer silicon steel sheet fault diagnosis method.
[0030] This invention also provides an MBN signal acquisition device for the above-mentioned HHT-based transformer silicon steel sheet fault diagnosis method. The MBN signal acquisition device includes: a signal generation system, a power amplification system, a signal acquisition system, and a signal processing system; wherein, the signal generation system is used to generate an excitation signal; the power amplification system is used to receive, amplify, and output the power current of the excitation signal; the signal acquisition system is used to acquire the MBN signal induced by the silicon steel sheet under test; the signal acquisition system includes a magnetic yoke, an excitation coil, and a detection sensor; wherein, the excitation coil is used to receive the signal from the power amplification system... The system outputs an excitation signal to generate an alternating magnetic field, which excites the tested silicon steel sheet to generate MBN (Magnetic Barkhausen Noise). A magnetic yoke is used to transmit the alternating magnetic field generated by the excitation coil. A detection sensor is used to collect the dynamic magnetic field signal on the surface of the silicon steel sheet and convert the dynamic magnetic field signal into a weaker voltage signal to obtain the original magnetic Barkhausen noise signal. A signal processing system is used to condition and preprocess the original magnetic Barkhausen noise signal to obtain the preprocessed magnetic Barkhausen noise signal. A filter is used to perform high-pass filtering on the preprocessed magnetic Barkhausen noise signal to obtain the final magnetic Barkhausen noise signal.
[0031] The HHT-based transformer silicon steel sheet fault diagnosis method, system, and equipment provided by this invention have the following beneficial effects:
[0032] 1. The Hilbert-Huang transform was used to perform time-frequency analysis on the magnetic Barkhausen noise signal, and new eigenvalues were extracted to improve the accuracy of defect detection in transformer silicon steel sheets;
[0033] 2. To address the issues of numerical errors and mode aliasing of Intrinsic Mode Function (IMF) components that may arise during empirical mode decomposition, a threshold for the correlation coefficient is constructed. By using the threshold for the correlation coefficient as a screening method, relevant IMFs are distinguished from IMFs that may be caused by numerical errors, thus obtaining the IMF components that make the main contribution to the signal while filtering out other possible noise interference. This improves the correlation between eigenvalues and signal features, reduces errors, and improves detection accuracy.
[0034] 3. An evaluation method and model were constructed using a classification machine learning algorithm, which enabled highly robust, high-precision, and highly repeatable detection for transformer fault diagnosis and internal defects in ferromagnetic materials. Attached Figure Description
[0035] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings:
[0036] Figure 1 This is a flowchart of the transformer silicon steel sheet fault diagnosis method based on HHT provided by the present invention;
[0037] Figure 2 This is a schematic diagram of the transformer silicon steel sheet fault diagnosis method based on HHT provided by the present invention;
[0038] Figure 3 This is a schematic diagram of the IMF after EMD decomposition of the preprocessed magnetic Barkhausen noise signal provided by the present invention.
[0039] Figure 4 This is a structural block diagram of the computer device provided by the present invention. Detailed Implementation
[0040] To provide a clearer understanding of the technical features, objectives, and effects of the present invention, specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.
[0041] Figure 1 A schematic diagram of the transformer silicon steel sheet fault diagnosis method based on HHT of this embodiment is shown. In this embodiment, the transformer silicon steel sheet fault diagnosis method based on HHT includes the following steps:
[0042] S1: Obtain the magnetic Barkhausen noise signal of the silicon steel sheet under test, and perform noise reduction processing on the magnetic Barkhausen noise signal of the silicon steel sheet under test to obtain the processed magnetic Barkhausen noise signal.
[0043] Specifically, step S1 includes: acquiring the magnetic Barkhausen noise signal of the silicon steel sheet to be tested, and using a signal processing tool to remove the noise and glitches of the magnetic Barkhausen noise signal of the silicon steel sheet to be tested at the baseline, so as to obtain the magnetic Barkhausen noise signal to be processed.
[0044] S2: Perform empirical mode decomposition and screening on the magnetic Barkhausen noise signal to be processed to obtain the selected intrinsic mode functions;
[0045] Specifically, step S2 of the above-mentioned HHT-based transformer silicon steel sheet fault diagnosis method includes the following steps:
[0046] S21: Perform empirical mode decomposition on the magnetic Barkhausen noise signal to be processed to obtain the intrinsic mode functions and residuals, as shown in the formula:
[0047]
[0048] Where x(t) is the Barkhausen noise signal to be processed, n is the order of the intrinsic mode function, and c i r(t) is the i-th intrinsic mode function, and r(t) is the residual.
[0049] S22: Based on the intrinsic mode function, the correlation coefficient is obtained, as shown in the formula:
[0050]
[0051] Where, r i Let be the correlation coefficient of the i-th order intrinsic mode function sequence, N be the order of the intrinsic mode function sequence, x(u) be the original signal sequence, and C be the correlation coefficient of the i-th order intrinsic mode function sequence. i (u) is the sequence of intrinsic mode functions of order i;
[0052] S23: Based on the correlation coefficient, obtain the threshold, as shown in the formula:
[0053]
[0054] Where ρ is the threshold and rmax is the maximum correlation coefficient;
[0055] S24: Based on the threshold, obtain the filtering intrinsic mode function;
[0056] Specifically, when the threshold reaches the preset value, the intrinsic mode function corresponding to the maximum correlation coefficient is used as the screening intrinsic mode function; through the above steps, the feature information extracted from the high contribution rate IMFs obtained through screening includes the time-frequency characteristics and energy distribution of the signal; this feature information is used for subsequent defect detection of silicon steel sheets in transformer cores;
[0057] S3: Based on the selected intrinsic mode functions, the defect detection results of the silicon steel sheet under test are obtained using the support vector machine model;
[0058] Specifically, step S3 of the above-mentioned HHT-based transformer silicon steel sheet fault diagnosis method includes the following steps:
[0059] S31: Preprocess the selected intrinsic mode functions to obtain the magnetic Barkhausen noise feature vector;
[0060] Specifically, the intrinsic mode function is selected as input data, and the extracted feature vectors are preprocessed, including data cleaning, missing value imputation, and outlier handling, to ensure data quality and reliability. Since the data is multidimensional, feature selection and dimensionality reduction are performed to remove irrelevant or redundant features and reduce the dimensionality of the feature space. For missing value imputation, the KNN (K-nearest neighbors interpolation) method is used, where the estimated missing value x is... i Given a missing value, its K nearest neighbors are known values x1, x2, ..., with corresponding distances d1. For each missing value, find the known values of its K nearest neighbors and their corresponding distances. Calculate the weight of each neighbor based on the distance, typically using the reciprocal of the distance as the weight. Therefore, the estimated missing value can be represented as:
[0061]
[0062] in, It is the weight of the j-th neighbor; through the above steps, a more complete characteristic signal of the magnetic Barkhausen signal can be obtained;
[0063] S32: Obtain the training set and test set using the magnetic Barkhausen noise feature vector; train the support vector machine model using the training set to obtain the trained support vector machine model; based on the test set, use the trained support vector machine model to obtain the defect detection results of the silicon steel sheet to be tested.
[0064] This embodiment provides a system, including the following modules:
[0065] The signal acquisition module is configured to: acquire the magnetic Barkhausen noise signal of the silicon steel sheet under test, perform noise reduction processing on the magnetic Barkhausen noise signal of the silicon steel sheet under test, and obtain the processed magnetic Barkhausen noise signal.
[0066] The signal decomposition and filtering module is configured to perform empirical mode decomposition and filtering on the magnetic Barkhausen noise signal to be processed, and obtain the filtered intrinsic mode functions.
[0067] Specifically, the signal decomposition and filtering module of the above system is configured as follows:
[0068] Empirical mode decomposition is performed on the magnetic Barkhausen noise signal to be processed to obtain the intrinsic mode functions and residuals, as shown in the formula:
[0069]
[0070] Where x(t) is the Barkhausen noise signal to be processed, n is the order of the intrinsic mode function, and c i Let r(t) be the i-th intrinsic mode function, and r(t) be the residual. Based on the intrinsic mode function, the correlation coefficient is obtained as shown in the formula:
[0071]
[0072] Where, r i Let be the correlation coefficient of the i-th order intrinsic mode function sequence, N be the order of the intrinsic mode function sequence, x(u) be the original signal sequence, and C be the correlation coefficient of the i-th order intrinsic mode function sequence. i (u) is the sequence of intrinsic mode functions of order i; the threshold is obtained based on the correlation coefficient, as shown in the formula:
[0073]
[0074] Where ρ is the threshold and rmax is the maximum correlation coefficient; based on the threshold, the screening intrinsic mode function is obtained;
[0075] The defect detection module is configured to: obtain the defect detection results of the silicon steel sheet under test by using a support vector machine model based on the selected intrinsic mode functions;
[0076] Specifically, the defect detection module of the above system is configured as follows: preprocessing the selected intrinsic mode functions to obtain magnetic Barkhausen noise feature vectors; using the magnetic Barkhausen noise feature vectors to obtain training and test sets; using the training set to train the support vector machine model to obtain a trained support vector machine model; and using the test set and the trained support vector machine model to obtain the defect detection results of the silicon steel sheet to be tested.
[0077] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of the above-described HHT-based transformer silicon steel sheet fault diagnosis method.
[0078] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of the above-described HHT-based transformer silicon steel sheet fault diagnosis method.
[0079] This embodiment provides an MBN signal acquisition device for the above-mentioned HHT-based transformer silicon steel sheet fault diagnosis method. The MBN signal acquisition device includes: a signal generation system, a power amplification system, a signal acquisition system, and a signal processing system; wherein, the signal generation system is used to generate an excitation signal;
[0080] Specifically, the signal generation system is a DG4202 function generator;
[0081] A power amplifier system is used to receive, amplify, and output power current of excitation signals;
[0082] Specifically, the power amplification system is the LYF-140010 power amplifier;
[0083] The signal acquisition system is used to acquire the MBN signal induced by the silicon steel sheet under test. The signal acquisition system includes a magnetic yoke, an excitation coil, and a detection sensor. The excitation coil is used to receive the excitation signal output from the power amplifier system, generate an alternating magnetic field, and excite the silicon steel sheet under test to generate MBN. The magnetic yoke is used to transmit the alternating magnetic field generated by the excitation coil. The detection sensor is used to collect the dynamic magnetic field signal on the surface of the silicon steel sheet, convert the dynamic magnetic field signal into a weaker voltage signal, and obtain the original magnetic Barkhausen noise signal.
[0084] The signal processing system is used to condition and preprocess the original magnetic Barkhausen noise signal to obtain the preprocessed magnetic Barkhausen noise signal; the preprocessed magnetic Barkhausen noise signal is then high-pass filtered by a filter to obtain the final magnetic Barkhausen noise signal.
[0085] Specifically, the filter is an active filter chip, UAF42.
[0086] In some embodiments, the above-described HHT-based transformer silicon steel sheet fault diagnosis method can also be implemented in the following ways.
[0087] In this embodiment, the transformer silicon steel sheet fault diagnosis method based on HHT is as follows: Figure 2 As shown, the method includes the following steps:
[0088] Step 1
[0089] In the extraction of the MBN test signal used, a detection system for the state of silicon steel sheets based on Barkhausen detection technology was first built. In the signal generation system, a DG4202 function generator was used to generate the excitation signal, and a power amplifier was used to amplify the power of the excitation signal to output sufficient current. The power amplifier used was the LYF-140010 power amplifier manufactured by the same company. The current output from the power amplifier passed through the excitation coil, causing the magnetic yoke to generate an alternating magnetic field, exciting the tested silicon steel sheet to generate MBN. The detection sensor collected the dynamic magnetic field signal on the surface of the silicon steel sheet and converted it into a weaker voltage signal. In the signal acquisition system, a magnetic yoke, excitation coil, and detection sensor were used. The MBN detection probe detects silicon steel sheets. The acquired raw magnetic Barkhausen noise signal undergoes preprocessing, including signal loss and anomalous signal processing, to obtain a preprocessed magnetic Barkhausen noise signal. This preprocessed signal is then high-pass filtered using an active filter chip, UAF42, to remove power frequency interference and other noise. Filtering for various sections is achieved by changing the parameters of the external circuit. After determining the parameters of the signal conditioning circuit, wavelet denoising is performed on the acquired MBN signal in MATLAB to further remove interference noise signals and eliminate baseline noise and glitches in the magnetic Barkhausen noise, making the acquired MBN signal more accurate.
[0090] Step 2
[0091] The Hilbert-Huang Transform (HHT) is performed on the pre-processed data. Since the Hilbert algorithm requires the input signal to be linear and stationary, most signals in real life do not meet this requirement. Therefore, before performing the Hilbert transform, the signal is subjected to Empirical Mode Decomposition (EMD) to generate basis functions called Intrinsic Mode Functions (IMFs). These IMFs can transform any time-domain signal into a "linear and stationary" signal, overcoming the problem of the lack of adaptability of the basis functions. The main purpose of EMD decomposition is to adaptively decompose the signal into a series of IMFs and a residual based on the characteristics of the signal itself. These IMFs must meet the following conditions: (1) The absolute value of the difference between the total number of extreme points and the total number of zero crossings cannot exceed 1; (2) At all times, the mean of the upper envelope and lower envelope obtained from the maximum and minimum points respectively is equal to 0.
[0092] The steps of Hilbert-Huang analysis are as follows:
[0093] Assuming the signal to be decomposed is x(t), find all local extrema of the original signal, and then connect all local maxima to form the upper envelope; repeat the above steps for the local extrema to obtain the lower envelope; define the average of the upper and lower envelopes as m1, and the difference between the data and m1 as component h1, then:
[0094] x(t)-m1=h1
[0095] Where m1 comes from:
[0096]
[0097] Where L is a local maximum and M is a local minimum;
[0098] Next, h1 is treated as the first component, and then the new mean is recalculated; if the new mean is m 11 ,but:
[0099] h1-m 11 =h 11
[0100] After repeating the screening process at most k times, h 1k Becoming an IMF member; that is:
[0101] h 1(k-1) -m 1k =h 1k
[0102] Let h 1k =c1, which becomes the first IMF in the original signal; c1 should contain the small-scale or shortest-period components in the data; the process of generating an IMF can be regarded as an inner loop; separate c1 from the original data:
[0103] x(t)-c1=r1
[0104] Here, r1 is the residual obtained after decomposition, which contains information about the longer periodic components. r1 will re-enter the decomposition process; all subsequent r... n This process must be repeated to obtain all subsequent IMFs;
[0105] After EMD decomposition is complete, the following can be obtained:
[0106]
[0107] After EMD decomposition, the MBN signal will be represented as a superposition of multiple IMFs and a remainder term R; where: c i (t) represents the individual IMF components decomposed, and r(t) represents the remainder; EMD mainly utilizes the characteristics of x(t) itself to adaptively decompose the signal; such as Figure 3The diagram shows the IMF after EMD decomposition of the preprocessed signal. The results show that the MBN signal is adaptively decomposed into 8 IMFs and 1 remainder R. Compared to wavelet transform, the advantage of EMD decomposition is that the decomposition process is adaptive; the frequency band range of the different IMFs obtained is based on the characteristics of the signal itself, while the result of wavelet transform is a fixed frequency band range determined by the decomposition scale. After EMD decomposition, the MBN signal will be represented as a superposition of multiple IMFs and a remainder R. Then, each IMF is subjected to Hilbert transform to obtain its Hilbert spectrum.
[0108] signal c i The Hilbert transform of (t) is defined as follows:
[0109]
[0110] With c i (t) and H[c i [(t)] is the analytic signal z constructed from the conjugate complex pair. i (t) is:
[0111] z i (t)=c i (t)+jH[c i [t]=A i (t)exp[jφ i (t)]
[0112] in:
[0113]
[0114] The instantaneous frequency ω can then be calculated. i (t) is:
[0115]
[0116] Therefore, x(t) can be expressed as:
[0117]
[0118] The Hilbert spectrum gives the relationship between signal time, instantaneous frequency, and amplitude. It can be used to analyze the time-varying patterns of components in a signal containing mixed components to identify local features. This formula omits the remainder r(t) and is called the "Hilbert spectrum," denoted as:
[0119]
[0120] The Hilbert spectrum of the IMF signal obtained after EMD decomposition in the previous step is obtained by performing Hilbert transform on the signal and plotting the corresponding spectrum.
[0121] After obtaining H(ω,t), time integration is performed on the Hilbert spectrum to obtain the marginal spectrum of the original signal, representing the Hilbert energy density corresponding to that frequency:
[0122]
[0123] h(ω) is called the "Hilbert marginal spectrum", where T is the intercept time of the signal; as can be seen from the above formula, H(ω,t) has both time and frequency information, indicating how the signal x(t) is distributed in time and frequency, while h(ω) indicates the amplitude distribution of the signal x(t) at different frequencies, but does not include the distribution of x(t) over the entire time.
[0124] After performing Hilbert-Huang transform and empirical mode decomposition, it is necessary to determine which IMFs are strongly correlated with the signal decomposition process. To obtain IMF components that contribute significantly to signal features and contain necessary MBN characteristic information, all f... IM The selection process is performed to obtain f, which contributes the most to the signal. IM The components are identified, and relevant feature values are extracted from them; the specific steps are as follows:
[0125] f IM The variance contribution rate vi is f IM The ratio of variance to the variance of the original sequence characterizes the degree of influence of different periodic components on the original data.
[0126]
[0127] In the formula: Mi is the i-th f IM The variance of the components; N is the total number of sampling points; u is the number of sampling points; T is the sampling period; the correlation coefficient r i Description of f IM The degree of correlation between the components and the original data; based on this, an effective f can be selected. IM The component is calculated using the following formula:
[0128]
[0129] In the formula: x(u) is the original signal sequence; C i For the i-th order f after decomposition IM Sequence; calculate threshold ρ and select effective components, where rmax is the value of each f IM The maximum correlation coefficient between the component and the original signal data;
[0130]
[0131] Through the above steps, the feature information extracted from the high-contribution IMFs obtained through screening includes the time-frequency characteristics and energy distribution of the signal; this feature information is used for subsequent defect detection of silicon steel sheets in transformer cores.
[0132] Step 3
[0133] A small-sample evaluation model for accurately predicting the defect state of silicon steel sheets based on MBN signals and artificial intelligence modeling is established; the relationship between small-sample MBN signals and silicon steel sheet states is processed using support vector machine (SVM) to test the detection performance of the system; the specific steps of step 3 are as follows:
[0134] Data preprocessing: The signal after HHT feature extraction is used as input data, and the extracted feature vector is preprocessed, including data cleaning, missing value imputation, outlier handling, etc., to ensure data quality and reliability; the data is multidimensional, so feature selection and dimensionality reduction are performed to remove irrelevant or redundant features and reduce the dimensionality of the feature space.
[0135] For missing value imputation in the step, the KNN (K-nearest neighbors interpolation) interpolation method is used; let the estimated missing value be x. i Its K nearest neighbors are known values x1, x2, ..., and their corresponding distances are d1; for each missing value, find the known values of its K nearest neighbors and their corresponding distances;
[0136] The weight of each neighbor is calculated based on distance, typically using the reciprocal of the distance as the weight; therefore, the estimated missing values can be represented as:
[0137]
[0138] in, It is the weight of the j-th neighbor; through the above steps, a more complete characteristic signal of the magnetic Barkhausen signal can be obtained;
[0139] The dataset is divided into training and test sets, and cross-validation is used to ensure that the sample distributions of the training and test sets are consistent.
[0140] SVM maps data to a high-dimensional space through kernel functions, thereby solving nonlinear problems that cannot be solved in low-dimensional space. The radial basis function (RBF) kernel function is chosen to better fit the data.
[0141] The SVM model is trained using the training set data. By adjusting the model parameters, the model can maximize the classification boundary and maintain the model's generalization performance.
[0142] The trained SVM model is evaluated using test set data. Evaluation metrics include accuracy, precision, recall, and F1 score to assess the model's performance and generalization ability.
[0143] Optimizing the model based on the evaluation results may include adjusting kernel function parameters, adjusting regularization parameters, increasing the amount of training data, etc., to improve the model's performance and stability.
[0144] By combining these feature analyses, electrical equipment can be monitored and diagnosed in real time, thereby improving the operating efficiency and reliability of the power grid.
[0145] In some embodiments, the above-described HHT-based transformer silicon steel sheet fault diagnosis method can also be implemented in the following ways.
[0146] In this embodiment, an MBN detection system is used to measure the MBN signal. The excitation signal amplitude is 5V and the frequency is 50Hz. Before the experiment, the surface of each sample is flattened and cleaned to minimize the error caused by the difference in the surface condition of the sample.
[0147] The acquired raw magnetic Barkhausen noise signal is preprocessed to remove signal loss and abnormal signals, resulting in a preprocessed magnetic Barkhausen noise signal. This preprocessed signal is then subjected to high-pass filtering using the UAF42 active filter chip to remove power frequency interference and other noise. Filtering for various segments is achieved by changing the parameters of the external circuit. After determining the parameters of the signal conditioning circuit, wavelet denoising is performed on the acquired MBN signal in MATLAB to further remove interference noise signals and eliminate noise and glitches at the baseline of the magnetic Barkhausen noise, making the acquired MBN signal more accurate.
[0148] For each group of MBN signals collected, traditional feature values (root mean square, peak-to-peak value, mean) and the feature values proposed in this invention are extracted. In order to conduct classification research, this embodiment selects Support Vector Machine (SVM) as the classifier and selects "RBF kernel function" as the kernel function to better fit the data. Fifty samples are randomly selected as the input of a single traditional feature value to SVM and used as training samples to train SVM. The rest are used as test samples.
[0149] Directly using the Hilbert spectrum and traditional single feature values as inputs to train an SVM results in an extremely low average recognition rate. Therefore, the Hilbert spectrum of the MBN signal cannot be directly used as a feature of the MBN signal for sample classification and recognition. The feature extraction method in this embodiment extracts feature vectors composed of corresponding feature values from different sample MBN signals, and then uses these feature vectors as inputs to train the SVM, which effectively improves the accuracy of defect detection in transformer silicon steel sheets.
[0150] This embodiment provides a computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the steps of the above-described HHT-based transformer silicon steel sheet fault diagnosis method. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk drive (HDD), or solid-state drive (SSD), etc.; the storage medium may also include combinations of the above types of memory.
[0151] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of the above-described HHT-based transformer silicon steel sheet fault diagnosis method.
[0152] like Figure 4As shown, the computer device may include: at least one processor 121, such as a CPU (Central Processing Unit), at least one communication interface 123, memory 124, and at least one communication bus 122. The communication bus 122 is used to enable communication between these components. The communication interface 123 may include a display screen and a keyboard; optionally, the communication interface 123 may also include a standard wired interface or a wireless interface. The memory 124 may be high-speed RAM (Random Access Memory) or non-volatile memory, such as at least one disk drive. Optionally, the memory 124 may also be at least one storage device located remotely from the processor 121. The memory 124 stores application programs, and the processor 121 calls the program code stored in the memory 124 to execute any of the aforementioned method steps. The communication bus 122 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The communication bus 122 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 4The term 124 is represented by a single line, but this does not imply a single bus or a single type of bus. The memory 124 may include volatile memory, such as random-access memory (RAM); it may also include non-volatile memory, such as flash memory, hard disk drive (HDD), or solid-state drive (SSD); or a combination of the above types of memory. The processor 121 may be a central processing unit (CPU), a network processor (NP), or a combination of a CPU and an NP. The processor 121 may further include a hardware chip. This hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The aforementioned PLD can be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL), or any combination thereof. Optionally, the memory 124 is also used to store program instructions. The processor 121 can call the program instructions to implement the HHT-based transformer silicon steel sheet fault diagnosis method of this embodiment.
[0153] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.
Claims
1. A fault diagnosis method for transformer silicon steel sheets based on HHT, characterized in that, The method includes the following steps: S1: Obtain the magnetic Barkhausen noise signal of the silicon steel sheet to be tested, and perform noise reduction processing on the magnetic Barkhausen noise signal of the silicon steel sheet to be tested to obtain the processed magnetic Barkhausen noise signal. S2: Perform empirical mode decomposition and screening on the magnetic Barkhausen noise signal to be processed to obtain the selected intrinsic mode functions, including: S21: Perform empirical mode decomposition on the magnetic Barkhausen noise signal to be processed to obtain the intrinsic mode functions and residuals, as shown in the formula: in, The signal to be processed is the Barkhausen magnetic noise signal, where n is the order of the intrinsic mode function. Let i be the i-th order intrinsic mode function. For residuals; S22: Based on the intrinsic mode function, the correlation coefficient is obtained, as shown in the formula: in, Let be the correlation coefficient of the i-th order intrinsic mode function sequence, and N be the order of the intrinsic mode function sequence. The original signal sequence, It is the sequence of intrinsic mode functions of order i; S23: Based on the correlation coefficient, obtain the threshold, as shown in the formula: in, For the threshold, The maximum correlation coefficient; S24: Based on the threshold, obtain the filtering intrinsic mode function; S3: Based on the selected intrinsic mode function, the defect detection results of the silicon steel sheet to be tested are obtained using the support vector machine model.
2. The transformer silicon steel sheet fault diagnosis method based on HHT according to claim 1, characterized in that, Step S3 includes the following steps: S31: Preprocess the selected intrinsic mode function to obtain the magnetic Barkhausen noise feature vector; S32: Obtain a training set and a test set using the magnetic Barkhausen noise feature vector; train the support vector machine model using the training set to obtain a trained support vector machine model; and obtain the defect detection result of the silicon steel sheet to be tested using the trained support vector machine model based on the test set.
3. A transformer silicon steel sheet fault diagnosis system based on HHT, characterized in that, The system includes the following modules: The signal acquisition module is configured to: acquire the magnetic Barkhausen noise signal of the silicon steel sheet under test, and perform noise reduction processing on the magnetic Barkhausen noise signal of the silicon steel sheet under test to obtain the processed magnetic Barkhausen noise signal. The signal decomposition and filtering module is configured to perform empirical mode decomposition and filtering on the magnetic Barkhausen noise signal to be processed to obtain filtered intrinsic mode functions. Specifically, the configuration is as follows: Empirical mode decomposition (EMD) is performed on the magnetic Barkhausen noise signal to be processed to obtain the intrinsic mode functions and residuals, as shown in the formula: in, The signal to be processed is the Barkhausen magnetic noise signal, where n is the order of the intrinsic mode function. Let i be the i-th order intrinsic mode function. For residuals; Based on the intrinsic mode function, the correlation coefficient is obtained, as shown in the formula: in, Let be the correlation coefficient of the i-th order intrinsic mode function sequence, and N be the order of the intrinsic mode function sequence. The original signal sequence, It is the sequence of intrinsic mode functions of order i; Based on the correlation coefficient, the threshold is obtained as shown in the formula: in, For the threshold, The maximum correlation coefficient; Based on the threshold, the filtering intrinsic mode function is obtained; The defect detection module is configured to obtain the defect detection results of the silicon steel sheet under test by using a support vector machine model based on the selected intrinsic mode function.
4. The system according to claim 3, characterized in that, The defect detection module is specifically configured as follows: The selected intrinsic mode function is preprocessed to obtain the magnetic Barkhausen noise feature vector; The training set and test set are obtained using the magnetic Barkhausen noise feature vector; the support vector machine model is trained using the training set to obtain a trained support vector machine model; based on the test set, the defect detection result of the silicon steel sheet to be tested is obtained using the trained support vector machine model.
5. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the HHT-based transformer silicon steel sheet fault diagnosis method as described in any of claims 1-2.
6. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the HHT-based transformer silicon steel sheet fault diagnosis method as described in any of claims 1-2.
7. An MBN signal acquisition device for use in any of the HHT-based transformer silicon steel sheet fault diagnosis methods as described in claims 1-2, characterized in that, The MBN signal acquisition device includes: a signal generation system, a power amplification system, a signal acquisition system, and a signal processing system; The signal generation system is used to generate excitation signals; The power amplifier system is used to receive, amplify, and output the power current of the excitation signal; The signal acquisition system is used to acquire the MBN signal induced by the silicon steel sheet under test; the signal acquisition system includes a magnetic yoke, an excitation coil, and a detection sensor. The excitation coil is used to receive the excitation signal output from the power amplifier system, generate an alternating magnetic field, and excite the silicon steel sheet under test to generate MBN. The magnetic yoke is used to transmit the alternating magnetic field generated by the excitation coil; The detection sensor is used to collect dynamic magnetic field signals on the surface of silicon steel sheets, convert the dynamic magnetic field signals into weaker voltage signals, and obtain the original magnetic Barkhausen noise signal. The signal processing system is used to condition and preprocess the original magnetic Barkhausen noise signal to obtain a preprocessed magnetic Barkhausen noise signal; and to perform high-pass filtering on the preprocessed magnetic Barkhausen noise signal to obtain the final magnetic Barkhausen noise signal.