A method for diagnosing faults of a converter transformer based on phase difference of multi-frequency vibration

Abstract

This invention discloses a fault diagnosis method for converter transformers based on multi-frequency vibration phase difference, comprising: arranging vibration sensors at corresponding winding positions on the surface of the converter transformer, collecting vibration signals and calculating the spectrum, extracting the amplitude and phase of the fundamental frequency and its harmonics; determining the dominant vibration frequency and high-load section, constructing a multi-frequency phase matrix; calculating the multi-frequency phase difference stability index and the multi-frequency phase difference fluctuation mode consistency index within the high-load section, and performing fault diagnosis accordingly. This invention extracts the fundamental frequency and its harmonic phase information, covering a wide frequency range, adaptable to different equipment models and operating conditions; constructing a frequency phase difference matrix for the vibration sensors is unaffected by propagation path differences and requires no current reference signal; the dual indicators characterize the mechanical state from two complementary dimensions of fluctuation amplitude and fluctuation mode, avoiding misjudgment from a single indicator, and can locate deteriorated frequency pairs; it is insensitive to the location of the measuring points, facilitating engineering installation and horizontal comparison of multiple devices.

Description

Technical Field

[0001] This invention belongs to the field of power equipment fault detection technology, specifically relating to a fault diagnosis method for converter transformers based on multi-frequency vibration phase difference. Background Technology

[0002] With the continuous growth of new energy grid connection scale and DC transmission capacity, the operational safety of power equipment faces increasingly severe challenges. As a core component of the DC transmission system, the reliability of the converter transformer directly affects the safety and stability of the entire system. Actual operation has revealed that the failure rate of high-voltage DC converter transformers is approximately twice that of transformers in AC power systems, with converter transformer and converter valve equipment failures being the main factors causing DC system outages. Because converter transformers are constantly subjected to the combined effects of harmonics, DC bias, and complex electromagnetic forces, their internal mechanical structures are prone to problems such as increased vibration, localized loosening, fatigue damage, and winding deformation. In severe cases, this can even lead to short-circuit faults and system outages. Therefore, accurate assessment of the mechanical vibration state of converter transformers is of great significance for ensuring the safe and stable operation of DC transmission systems.

[0003] Currently, converter transformers are a relatively new type of transformer used in high-voltage direct current (HVDC) transmission in recent years. Existing fault diagnosis methods for converter transformers mainly borrow from and transfer traditional AC transformer diagnosis methods, focusing primarily on the vibration amplitude information of each frequency component in the vibration spectrum. For example, the literature [S. Wu, S. Ji, Y. Zhang, S. Wang and H. Liu. A novel vibration frequency response analysis method for detecting the mechanical condition of converter transformer windings, IEEE Transactions on Industrial Electronics, Vol. 71, No. 7, pp. 8176-8180, July 2024] proposes an online vibration frequency response analysis (VFRA) method based on high-amplitude, wide-band electromagnetic forces generated by harmonic currents. VFRA is defined as the ratio of vibration acceleration to the square of the current. By sweeping the vibration frequency, the VFRA curve as a function of vibration frequency is plotted, and the correlation coefficient of the curve is calculated to determine the degree of change of the curve, thereby reflecting the mechanical faults of the converter transformer. However, these methods still have many shortcomings: the vibration spectrum of converter transformers under no-load and load conditions differs significantly, and the dominant frequency and energy distribution of each frequency band are strongly correlated with the load level. The unpredictability of vibration changes introduced by faults means that there is currently no effective means to determine whether the changes in vibration energy structure originate from the fault or the harmonic current itself. Furthermore, most existing fault diagnosis methods are developed and tested in laboratory simulation environments, and their effectiveness and versatility in actual operating environments remain unknown. Therefore, the advancement of fault diagnosis methods based on vibration frequency amplitude spectrum analysis is fraught with difficulties, and the design and development of new fault diagnosis methods are urgently needed.

[0004] On the other hand, some progress has been made in fault diagnosis research targeting phase. The literature [J. Zheng, H. Huang, P. Jie, Y. Hu and X. Jiang. Mechanical Fault Feature Extraction Based on Transformer Vibration Phase Changes, Applied Acoustics, 2022, Vol. 185, No. 108440] systematically elucidated for the first time the intrinsic relationship between transformer vibration phase information and its mechanical stability. Compared with traditional analysis methods based on amplitude or spectrum, phase change characteristics are more sensitive and direct to mechanical faults, and are less affected by changes in operating current, voltage, and temperature. It points out that winding loosening leads to a significant increase in phase angle, while winding deformation causes more drastic phase changes, and deformation has a greater impact on system stability than loosening. The literature [J. Zhao, J. Zheng, Y. Yuan, and N. Wang. Winding Deformation Detection Based on Multi-Sensor Vibration Phase Difference of Transformers, 6th International Conference on Electronic Engineering and Informatics, Chongqing, China, 2024, pp. 169-175] points out that the deformation of transformer windings changes their structural stiffness and vibration transmission path, resulting in a time delay in the propagation of vibration signals from the source (windings and core) to different positions on the tank surface, i.e., a phase difference exists. This phenomenon is experimentally verified. Therefore, the index based on the change of vibration phase difference between multiple sensors can effectively and reliably detect the deformation of power transformer windings. However, current research on fault diagnosis targeting the vibration phase characteristics of converter transformers, especially the multi-frequency phase difference characteristics, is still relatively lacking, making it difficult to meet the actual needs of online assessment and fault diagnosis of the mechanical condition of converter transformers under complex operating conditions.

[0005] Chinese patent application CN111273100A discloses a method for assessing the condition of power transformer windings based on vibration phase. This method leverages the mechanism of winding vibration generation, utilizing the fact that vibration phase is more sensitive to mechanical structural changes caused by faults and less susceptible to environmental factors. It directly correlates winding faults with vibration phase information, providing a basis for the effectiveness and scientific rigor of fault diagnosis. Chinese patent application CN118643448A discloses a method for locating faults in three-phase transformer windings. It obtains the phase of the vibration components of each phase winding of the three-phase transformer by extracting vibration information; calculates the corresponding initial vibration parameters based on the vibration phases; analyzes the operating state of the three-phase transformer based on the initial vibration parameters to obtain the current operating state, which includes normal operation and operating fault; in the case of an operating fault, it calculates both two-phase and single-phase parameters based on the vibration phases to obtain the maximum values ​​of the two-phase and single-phase parameters; calculates the abnormal contribution value of each phase based on the maximum values ​​of the two-phase and single-phase parameters, and locates the faulty phase based on the abnormal contribution value.

[0006] The aforementioned methods primarily characterize the state of AC transformers based on the vibration phase difference between multiple measuring points. However, under the combined influence of harmonic currents and DC bias, the vibration sources of converter transformers are more complex, spatially unevenly distributed, and their intensity dynamically changes with the load. This means that the phase relationship between measuring points is no longer uniquely dependent on the mechanical structure, affecting the accuracy of the diagnostic results. Existing methods only extract the 100Hz fundamental frequency vibration phase for diagnosis. However, under the influence of harmonic currents, the vibration frequency components of converter transformers are complex, covering a wide frequency range up to 2000Hz, and the dominant frequency shifts significantly under different load conditions. Existing methods have failed to effectively utilize harmonic information, especially the high-frequency vibration phase information, which is more sensitive to fault changes.

[0007] Furthermore, existing methods require simultaneous acquisition of current signals as a reference for correction when calculating the phase, but measuring the converter transformer current is difficult and its field feasibility is poor. Moreover, the waveform is severely distorted, and the 100Hz vibration component may be generated by the interaction between harmonics, rendering the mathematical premise of using the power frequency current phase as a reference invalid. On the other hand, calculating pairwise differences between multiple measurement points means that when the signal at a certain measurement point is abnormal (such as sensor detachment, low local signal-to-noise ratio, or environmental interference), all related differences are distorted, affecting the final measurement result and resulting in poor system robustness. Summary of the Invention

[0008] In view of the above, the present invention provides a fault diagnosis method for converter transformers based on multi-frequency vibration phase difference. By extracting the phase information of fundamental frequency and harmonic vibration to construct a phase difference matrix between frequencies, a fault diagnosis method that is sensitive and reliable, without the need for current reference and is insensitive to measurement points and operating conditions is achieved.

[0009] A fault diagnosis method for converter transformers based on multi-frequency vibration phase difference includes: Vibration sensors are arranged on the surface of the converter transformer at the corresponding winding positions to collect vibration signals, calculate the spectrum of the vibration signals, and extract the amplitude and phase of the fundamental frequency and its harmonics in the spectrum. The dominant vibration frequency and high-load zone of the vibration sensor are calculated and determined, and the multi-frequency phase matrix of the vibration sensor is constructed. In high-load sections, the multi-frequency phase difference stability index and the multi-frequency phase difference fluctuation mode consistency index of the vibration sensor are calculated using a multi-frequency phase matrix. Fault diagnosis of converter transformers is performed based on multi-frequency phase difference stability index and multi-frequency phase difference fluctuation mode consistency index.

[0010] Furthermore, the method for calculating and determining the dominant vibration frequency is as follows: For any vibration signal collected by the vibration sensor m The proportion of energy of each frequency component in the total energy of all components in its spectrum is calculated according to the following expression:

[0011] in: Vibration signal m In the spectrum n The amplitude of the first frequency, Vibration signal m In the spectrum n The proportion of the energy of the first frequency component in the total energy of all components. m The sample number of the vibration signal and 1 ≤ m ≤ M , M The number of vibration signal samples collected by the vibration sensor; For 2≤ n ≤20 n The average normalized energy of the subharmonic across all samples is calculated using the following expression:

[0012] in: R ( n )express n The average normalized energy of the subharmonic across all samples; Finally, the dominant vibration frequency of the vibration sensor was determined to be... k i f 0, i It is a natural number and 1≤ i ≤ K , f 0 is the fundamental frequency. K The number of dominant vibration frequencies. k 1, k 2,…, k K For all R ( n The largest front in ) K The harmonic order corresponding to each energy value.

[0013] Furthermore, the high-load segment is determined as follows: First, the activity of each vibration signal collected by the vibration sensor is calculated; then, the Otsu method is used to maximize the inter-class variance on the activity histogram to automatically obtain the segmentation threshold between the high-load and low-load classes; vibration signals with activity greater than or equal to the segmentation threshold are selected, and the samples of these vibration signals are numbered to form the high-load segment.

[0014] Furthermore, the degree of activity is calculated as follows: for any vibration signal m Its dominant vibration frequency k i fThe amplitude at 0 is First, calculate the normalized amplitude. Thus, vibration signals are obtained. m Activity ,in This indicates that the median amplitude of each dominant vibration frequency is taken. This indicates that all vibration signals collected by the vibration sensor are at the dominant vibration frequency. k i f The amplitude value obtained by taking the median of the amplitude at point 0.

[0015] Furthermore, the expression for the multi-frequency phase matrix is ​​as follows:

[0016] Where: the first in the multi-frequency phase matrix S i Line 1 m Column element values For the corrected vibration signal m Vibration frequency k i f Phase of 0 , Equal to the vibration signal before correction m Vibration frequency k i f Phase of 0 minus k i The fundamental frequency phase is times that of the fundamental frequency.

[0017] Furthermore, the calculation process for the multi-frequency phase difference stability index is as follows: First, define any two dominant vibration frequencies. k i f 0 and k j f 0 phase difference :

[0018] in: The first frequency phase matrix is ​​the first frequency phase matrix. i Line 1 m Column element values, The first frequency phase matrix is ​​the first frequency phase matrix. j Line 1 m Column element values; Then, the phase difference is statistically analyzed in the high-load section. Circular standard deviation:

[0019] in: Phase difference Circular standard deviation within the high-load section Ω This indicates the number of samples within the high-load section. As an intermediate variable; Finally, all the cyclic standard deviations Constitutes sequence S σ =[ σ 12 , σ 13 ,…, σ K-1,K ], take the sequence S σ The mean and maximum values ​​are used as the multi-frequency phase difference stability indicators of the vibration sensor. and .

[0020] Furthermore, the calculation process of the multi-frequency phase difference fluctuation mode consistency index is as follows: First, define any two dominant vibration frequencies. k i f 0 and k j f 0 circular correlation coefficient :

[0021] in: The first frequency phase matrix is ​​the first frequency phase matrix. i Line 1 m Column element values, The first frequency phase matrix is ​​the first frequency phase matrix. j Line 1 m Column element values; for The ring-shaped mean value within the high-load section Ω for The ring-shaped mean value within the high-load section Ω Indicates the number of samples within the high-load section; Then all the ring correlation coefficients Constitutes sequence S r =[ r 12 , r 13 ,…, r K-1,K ], take the sequence S r The mean value is used as a consistency index for the multi-frequency phase difference fluctuation mode of the vibration sensor. .

[0022] Furthermore, the standard for fault diagnosis of converter transformers is: only when and and The converter transformer was determined to be in normal operating condition; otherwise, it was determined to be in abnormal operating condition. , , The threshold value is set.

[0023] Furthermore, based on the vibration data accumulated by the vibration sensors from multiple converter transformers under normal or fault conditions, the multi-frequency phase difference stability index was further developed. and and consistency index of multi-frequency phase difference fluctuation mode The converter transformer's state is used as a label for the constituent feature quantities. Supervised or unsupervised machine learning methods are employed to train the converter transformer state assessment model. For the converter transformer to be diagnosed, a multi-frequency phase difference stability index is calculated using vibration signals collected by vibration sensors. and and consistency index of multi-frequency phase difference fluctuation mode These indicators are combined into characteristic quantities and input into the converter transformer condition assessment model, so that the model can automatically output the condition assessment results of the converter transformer.

[0024] A computer device includes a memory and a processor, wherein the memory stores a computer program and the processor executes the computer program to implement the above-described method for diagnosing converter transformer faults based on multi-frequency vibration phase difference.

[0025] A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-described method for diagnosing converter transformer faults based on multi-frequency vibration phase difference.

[0026] Based on the above technical solution, the present invention has the following beneficial technical effects: 1. This invention is no longer limited to a single point at the fundamental frequency, but extracts the vibration amplitude and phase information of the fundamental frequency and its 2nd to 20th harmonics, covering a wide frequency range of 100 to 2000 Hz. Based on this, the main frequency set is dynamically selected according to the energy level of each frequency component, enabling the diagnostic features to adapt to different types of converter transformers and different operating conditions, thus improving system reliability. Simultaneously, the vibration phase in the high-frequency band is more sensitive to faults (measured phase difference fluctuations in high-frequency bands of faulty equipment are significantly greater than those in low-frequency bands), and incorporating it into the diagnostic indicators can improve the sensitivity of fault diagnosis.

[0027] 2. This invention extracts the phase difference between multiple frequency components and its own fundamental frequency at the same measuring point, constructs a phase difference matrix between frequencies, and eliminates the cross-calculation of differences between measuring points. The characteristics are no longer affected by the uneven spatial distribution of vibration sources and the differences in propagation paths, and only reflect the internal mechanical characteristics of the transformer. Furthermore, through phase correction and adaptive selection of high-load sections, the influence of external operating conditions and load fluctuations is eliminated to the greatest extent, resulting in strong robustness.

[0028] 3. This invention extracts features independently for each measuring point. The failure of a single sensor does not affect the normal operation of other measuring points, and the diagnostic system can still output reliable conclusions based on the remaining valid measuring points, thus improving the engineering reliability of the monitoring system. Furthermore, this invention uses the fundamental frequency component of the vibration itself as a reference, eliminating the need for a current reference signal and not relying on the assumption of "power frequency current-vibration phase synchronization." This makes it suitable for scenarios with distorted current waveforms and 100Hz vibration components generated by harmonic interactions, reducing the measurement complexity of the system.

[0029] 4. The dual indicators extracted in this invention characterize the mechanical state from two complementary dimensions: "fluctuation amplitude" and "fluctuation pattern," forming complementary constraints and avoiding misjudgment by a single indicator. Furthermore, the two indicators are insensitive to the overall shift and sign uncertainty of the phase difference, eliminating the need for absolute phase calibration, and allowing direct horizontal comparison between indicators from different measuring points and different equipment. In addition, the pairwise difference stability matrix can locate frequency pairs where degradation occurs, providing clues for fault type and location identification.

[0030] 5. This invention diagnoses based on phase fluctuation characteristics and is insensitive to the location of the measuring points. The index reflects the fluctuation and consistency of the phase difference between frequencies; the constant phase shift introduced by the propagation path has no effect on either, and amplitude response differences are not included in the index calculation. Therefore, even if the measuring points are arranged at different locations on the tank wall of the same transformer, consistent diagnostic conclusions can still be obtained, facilitating engineering installation and lateral comparison between multiple devices. Attached Figure Description

[0031] Figure 1 This is a schematic diagram of the fault diagnosis method for converter transformers based on multi-frequency vibration phase difference according to the present invention.

[0032] Figure 2 This is a schematic diagram of the spectrum of a typical vibration signal of a converter transformer.

[0033] Figure 3 This is a histogram for analyzing the energy of the dominant vibration frequency.

[0034] Figure 4 This is a schematic diagram of the activity curve and the high-load section.

[0035] Figure 5 This is a schematic diagram of the corrected phase curves of healthy samples collected from a portion of the channels at typical frequencies.

[0036] Figure 6 This is a schematic diagram of the corrected phase curves of fault samples collected from some channels at typical frequencies.

[0037] Figure 7 This is a schematic diagram showing the stability index results of healthy samples collected from eight channels (ch1 to ch8).

[0038] Figure 8 This is a schematic diagram showing the stability index results of fault samples collected from eight channels ch1 to ch8.

[0039] Figure 9 This is a schematic diagram showing the consistency index results of healthy samples collected from eight channels (ch1 to ch8).

[0040] Figure 10 This is a schematic diagram showing the consistency index results of fault samples collected from eight channels ch1 to ch8. Detailed Implementation

[0041] To describe the present invention in more detail, the technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0042] This embodiment provides a fault diagnosis method for converter transformers based on multi-frequency vibration phase difference. The specific implementation process is as follows: Figure 1 As shown: First, vibration sensors are placed on the surface of the converter transformer tank at the position corresponding to the winding. Under the normal energized and loaded operating conditions of the converter transformer, vibration signals of a fixed length (L seconds) are collected at fixed intervals (T minutes). The total collection time should be greater than 12 hours (generally 24 hours).

[0043] For vibration sensors, calculate the spectrum of all vibration sampling samples from the vibration sensor, and extract the fundamental frequency from the spectrum of each sample. and its harmonics ( , n The amplitude of (=2,3,…,20) and phase ,in m Represents the sample number. It is 100Hz.

[0044] Calculate the proportion of the energy of each frequency component in the total energy of all components in each sample:

[0045] For each harmonic order Calculate its in all M Average normalized energy in each sample:

[0046] Sure K The first frequency is the dominant vibration frequency of the converter transformer. K The first-order frequency components satisfy:

[0047] The amplitude of the dominant vibration frequency varies with the load level. During low load periods (such as nighttime shutdowns), the amplitude of each dominant frequency is too low and the signal-to-noise ratio is poor, making the extracted phase difference unreliable.

[0048] Record No. m Each sample at the main frequency The vibration amplitude at that location is First, normalize the amplitude of each frequency according to its median over the entire day, then calculate the median of the frequencies to obtain the first value. m Activity of each sample:

[0049] In the formula: This indicates taking the median of each frequency. This indicates that the median of the entire day's sample is taken.

[0050] The Otsu method is used to maximize the inter-class variance on the activity histogram, automatically determining the segmentation threshold between the high-load and low-load classes. θ :

[0051] The threshold can also be changed as needed to the set quantile of activity (e.g., 45%) or the set proportion of the maximum value of the day. Ω takes all samples that meet the conditions, and continuity is not required; short-term troughs of no more than 3 samples (about 15 minutes) are bridged, and 2 samples are added to each end of Ω to retain samples with a relatively high signal-to-noise ratio during the load switching transition; the selection of Ω only determines the range of the index statistics.

[0052] To correct the phase uncertainty introduced by the randomness of the operating condition phase and to remove the influence of the fundamental frequency electromagnetic force introduced by electromagnetic nonlinearity on the high-frequency vibration phase, each dominant frequency component... k i f 0( f The phase of 0 = 100 Hz (i.e., twice the power frequency) minus the phase of the fundamental frequency vibration component. k i Times:

[0053] For each frequency sequence, a 2π phase expansion is first performed, and then a multi-frequency phase matrix is ​​constructed:

[0054] in:K The number of main frequencies participating in the analysis. M The number of samples is [number] in this embodiment. K Set it to 10.

[0055] Define any two frequencies i , j Phase difference (pairwise difference) sequence:

[0056] To ensure the reliability of stability and consistency indicators, statistics are only performed within the high-load segment Ω. During the analysis period Ω (high-load segment, sample size...),... m The cyclic standard deviation within (∈Ω) is:

[0057] The circular standard deviations of all pairwise differences form a vector S. var =[ σ 12 , σ 13 ,…, σ K-1,K The mean and maximum values ​​are taken as the multi-frequency phase difference stability index.

[0058] When the internal structure of the converter transformer is in good condition, the phase difference of each frequency remains relatively constant and the fluctuations are in the same direction. σ ij The phase difference is close to 0; if the internal structure is in a deteriorated state (such as loosening, cracking, etc.), the phase difference fluctuation increases significantly.

[0059] To compare the patterns of phase difference fluctuations at different frequencies (removing the inherent constant offsets of each frequency and retaining only the shape of the fluctuations over time), for any two frequency sequences... and Define the pairwise circular correlation coefficient:

[0060] in: For sequence The circular mean within the time period Ω, i.e. The mean of all pairwise correlations is taken as the consistency index of multi-frequency phase difference fluctuation patterns.

[0061] When the internal structure of the converter transformer is in good condition, the multi-frequency phase difference fluctuation modes are basically consistent. Approaching 1 (actual measured values ​​for multiple channels on health devices are above 0.8); if the internal structure is in a deteriorated state, Significantly less than 1 (actual measurements are mostly below 0.6).

[0062] A linear classification method was used to establish the relationship between the multi-frequency phase difference fluctuation index of converter transformer vibration, the multi-frequency phase difference fluctuation mode consistency index, and the operating status (normal, fault) of the converter transformer:

[0063] in: , , In this embodiment, a threshold value is set as the threshold. and Values , The value is set to 0.5.

[0064] Furthermore, based on the accumulated vibration data from long-term monitoring of multiple converter transformers under normal and fault conditions, the average multi-frequency phase difference fluctuation of the converter transformer vibration sensor can be calculated daily. Maximum multi-frequency phase difference fluctuation Using the consistency index of multi-frequency phase difference fluctuation patterns as a feature quantity and the state of the converter transformer as a label for the feature quantity, a supervised or unsupervised machine learning method is used for training to obtain a converter transformer state assessment model. Finally, for the converter transformer to be evaluated, the average multi-frequency phase difference fluctuation of the converter transformer vibration sensor over a day is calculated. Maximum multi-frequency phase difference fluctuation In addition, the consistency index of multi-frequency phase difference fluctuation mode is input into the evaluation model to obtain the state evaluation result.

[0065] This embodiment tests a converter transformer under normal operation and analyzes the collected tank vibration signals. The data acquisition system records tank vibration at a frequency of 10kHz. Eight vibration measurement points are arranged on the surface of the converter transformer tank at the corresponding winding positions. The vibration signals from the eight channels are analyzed, and the vibration spectrum of a typical sample is shown below. Figure 2 As shown, the vibration spectrum of the converter transformer is extremely rich in harmonic components.

[0066] The collected vibration signal data is processed to calculate the average normalized energy of each frequency component in the entire day's samples. The data is then sorted by energy percentage from highest to lowest, and the dominant frequency set is automatically selected. Figure 3 As shown. Then, based on the dominant frequency vibration amplitude, a motion curve is synthesized, and the Otsu method is used to automatically calculate the segmentation threshold to obtain the high-load section, as shown. Figure 4 As shown, the gray area represents low-load / low signal-to-noise ratio periods automatically identified by the algorithm, while the white area represents high-load segments retained for subsequent statistical analysis. Figure 4As can be seen, the method of the present invention effectively removes samples with excessively low amplitude during the low-load period at night, while retaining high signal-to-noise ratio data during the high-load period during the day, thus ensuring the reliability of subsequent feature extraction.

[0067] For samples within the high-load section Ω, phase correction is performed, and a multi-frequency phase matrix is ​​constructed. Figure 5 and Figure 6 The corrected phase curves of channels 1-4 at three typical frequencies: 200Hz, 300Hz, and 400Hz are shown. During low-load periods, the phase curves exhibit noticeable spikes; however, in the high-load region (white background), the phase curves at each frequency show a continuous and smooth evolution trend. Further comparison of healthy and faulty samples reveals that: under healthy conditions, the phase curves at different frequencies maintain a highly synchronized fluctuation pattern across frequencies; while under faulty conditions, the synchronized fluctuation pattern across frequencies is disrupted, and the curves show a significant trend divergence.

[0068] Based on the multi-frequency phase matrix, stability and consistency indices are calculated respectively. Figure 7 and Figure 8 Stability metrics were compared between healthy and faulty samples. and Calculation results Figure 9 and Figure 10 Consistency metrics were compared between healthy samples and faulty samples. Calculation results. It can be seen that the phase difference fluctuation of healthy samples is significantly smaller than that of faulty samples. In the healthy state, the phase difference fluctuation of all eight channels is less than 25°, while the phase difference fluctuation of channels ch1 and ch5 in the faulty samples exceeds 25°. The mean and variance of the fluctuations. Furthermore, the consistency of fluctuations across channels was significantly higher in healthy samples than in faulty samples.

[0069] Based on linear classification rules, As a standard for judgment, healthy and faulty samples were individually evaluated. Healthy samples met all three requirements across all eight channels and passed the test. For faulty samples, all eight channels... All exceeded the threshold, except for channel ch3. All higher than And channels 1, 4, 6, and 8 All values ​​were below 0.5, failing to meet the normalization criterion, and were identified as abnormal samples. The results of this case fully validate the effectiveness and fault sensitivity of the method described in this invention for converter transformer fault diagnosis.

[0070] The above description of the embodiments is provided to enable those skilled in the art to understand and apply the present invention. Those skilled in the art can readily make various modifications to the above embodiments and apply the general principles described herein to other embodiments without creative effort. Therefore, the present invention is not limited to the above embodiments, and any improvements and modifications made to the present invention by those skilled in the art based on the disclosure thereof should be within the scope of protection of the present invention.

Claims

1. A fault diagnosis method for converter transformers based on multi-frequency vibration phase difference, characterized in that, include: Vibration sensors are arranged on the surface of the converter transformer at the corresponding winding positions to collect vibration signals, calculate the spectrum of the vibration signals, and extract the amplitude and phase of the fundamental frequency and its harmonics in the spectrum. The dominant vibration frequency and high-load section of the vibration sensor are calculated and determined, and the multi-frequency phase matrix of the vibration sensor is constructed. In high-load sections, the multi-frequency phase difference stability index and the multi-frequency phase difference fluctuation mode consistency index of the vibration sensor are calculated using a multi-frequency phase matrix. Fault diagnosis of converter transformers is performed based on multi-frequency phase difference stability index and multi-frequency phase difference fluctuation mode consistency index.

2. The converter transformer fault diagnosis method based on multi-frequency vibration phase difference according to claim 1, characterized in that, The method for calculating and determining the dominant vibration frequency is as follows: For any vibration signal collected by the vibration sensor m The proportion of energy of each frequency component in the total energy of all components in its spectrum is calculated according to the following expression: in: Vibration signal m In the spectrum n The amplitude of the first frequency, Vibration signal m In the spectrum n The proportion of the energy of the first frequency component in the total energy of all components. m The sample number of the vibration signal and 1 ≤ m ≤ M , M The number of vibration signal samples collected by the vibration sensor; For 2≤ n ≤20 n The average normalized energy of the subharmonic across all samples is calculated using the following expression: in: R ( n )express n The average normalized energy of the subharmonic across all samples; Finally, the dominant vibration frequency of the vibration sensor was determined to be... k i f 0, i It is a natural number and 1≤ i ≤ K , f 0 is the fundamental frequency. K The number of dominant vibration frequencies. k 1, k 2,…, k K For all R ( n The largest front in ) K The harmonic order corresponding to each energy value.

3. The converter transformer fault diagnosis method based on multi-frequency vibration phase difference according to claim 2, characterized in that, The high-load segment is determined as follows: First, the activity of each vibration signal collected by the vibration sensor is calculated; then, the Otsu method is used to maximize the inter-class variance on the activity histogram to automatically obtain the segmentation threshold between the high-load and low-load classes; vibration signals with activity greater than or equal to the segmentation threshold are selected, and the samples of these vibration signals are numbered to form the high-load segment.

4. The converter transformer fault diagnosis method based on multi-frequency vibration phase difference according to claim 3, characterized in that, The activity level is calculated as follows: for any vibration signal m Its dominant vibration frequency k i f The amplitude at 0 is First, calculate the normalized amplitude. Thus, vibration signals are obtained. m Activity ,in This indicates that the median amplitude of each dominant vibration frequency is taken. This indicates that all vibration signals collected by the vibration sensor are at the dominant vibration frequency. k i f The amplitude value obtained by taking the median of the amplitude at point 0.

5. The converter transformer fault diagnosis method based on multi-frequency vibration phase difference according to claim 2, characterized in that, The expression for the multi-frequency phase matrix is ​​as follows: Where: the first in the multi-frequency phase matrix S i Line number m Column element values For the corrected vibration signal m Vibration frequency k i f Phase 0 , Equal to the vibration signal before correction m Vibration frequency k i f Phase 0 minus k i The fundamental frequency phase is times that of the fundamental frequency.

6. The converter transformer fault diagnosis method based on multi-frequency vibration phase difference according to claim 2, characterized in that, The calculation process for the multi-frequency phase difference stability index is as follows: First, define any two dominant vibration frequencies. k i f 0 and k j f 0 phase difference : in: The first frequency phase matrix is ​​the first frequency phase matrix. i Line number m Column element values, The first frequency phase matrix is ​​the first frequency phase matrix. j Line number m Column element values; Then, the phase difference is statistically analyzed in the high-load section. Circular standard deviation: in: Phase difference Circular standard deviation within the high-load section Ω This indicates the number of samples within the high-load section. As an intermediate variable; Finally, all the annular standard deviations Constitutes sequence S σ =[ σ 12 , σ 13 ,…, σ K-1,K ], take the sequence S σ The mean and maximum values ​​are used as the multi-frequency phase difference stability indicators of the vibration sensor. and .

7. The converter transformer fault diagnosis method based on multi-frequency vibration phase difference according to claim 2, characterized in that, The calculation process of the multi-frequency phase difference fluctuation mode consistency index is as follows: First, define any two dominant vibration frequencies. k i f 0 and k j f 0 circular correlation coefficient : in: The first frequency phase matrix is ​​the first frequency phase matrix. i Line number m Column element values, The first frequency phase matrix is ​​the first frequency phase matrix. j Line number m Column element values; for The ring-shaped mean value within the high-load section Ω for The ring-shaped mean value within the high-load section Ω Indicates the number of samples within the high-load section; Then all the ring correlation coefficients Constitutes sequence S r =[ r 12 , r 13 ,…, r K-1,K ], take the sequence S r The mean value is used as a consistency index for the multi-frequency phase difference fluctuation mode of the vibration sensor. .

8. The converter transformer fault diagnosis method based on multi-frequency vibration phase difference according to claim 6, characterized in that, The standard for fault diagnosis of converter transformers is: only when and and The converter transformer was determined to be in normal operating condition; otherwise, it was determined to be in abnormal operating condition. , , For the set threshold, It serves as a consistency index for multi-frequency phase difference fluctuation modes.

9. The converter transformer fault diagnosis method based on multi-frequency vibration phase difference according to claim 1, characterized in that: Based on the vibration data accumulated by vibration sensors from multiple converter transformers under normal or fault conditions, the multi-frequency phase difference stability index was... and and consistency index of multi-frequency phase difference fluctuation mode The converter transformer's state is used as a label for the constituent feature quantities. Supervised or unsupervised machine learning methods are employed to train the converter transformer state assessment model. For the converter transformer to be diagnosed, a multi-frequency phase difference stability index is calculated using vibration signals collected by vibration sensors. and and consistency index of multi-frequency phase difference fluctuation mode These indicators are combined into characteristic quantities and input into the converter transformer condition assessment model, so that the model can automatically output the condition assessment results of the converter transformer.

Images