OLTC mechanical fault monitoring method based on vibration signal multi-index evaluation

Abstract

The invention relates to the technical field of on-load tap-changer state monitoring, and particularly discloses an OLTC mechanical fault monitoring method based on vibration signal multi-index evaluation, which comprises the following steps: collecting a vibration signal sample when an OLTC performs gear shifting operation, and determining a single gear shifting analysis window; preprocessing the collected vibration signal sample; performing process segmentation on the preprocessed vibration signal sample in the single gear shifting analysis window; multiple types of statistical indexes are extracted from the vibration subsequences of all the stages, and a multi-dimensional feature vector of single gear shifting is constructed; constructing a normal operation multi-index baseline and a typical mechanical fault feature fingerprint database; and comparing and analyzing the multi-dimensional feature vector obtained by the current gear shifting operation with the normal operation multi-index baseline and the typical mechanical fault feature fingerprint database, and carrying out online evaluation and grading diagnosis on the mechanical state of the OLTC. The method is simple in implementation mode, and online monitoring, state evaluation and early fault early warning of the OLTC mechanical state can be achieved.

Description

Technical Field

[0001] This invention relates to the field of on-load tap changer status monitoring technology, and more specifically to an OLTC mechanical fault monitoring method based on multi-index evaluation of vibration signals. Background Technology

[0002] On-load tap changers (OLTCs) are the only load-regulating components in transformers that frequently operate under load current. They maintain a relatively constant DC or AC voltage and compensate for grid voltage fluctuations by switching taps, making them crucial for ensuring the safe and stable operation of power transmission and transformation equipment. Due to the high switching load, frequent operation, and complex transmission links, OLTCs are prone to mechanical failures such as transmission mechanism jamming, contact wear, spring fatigue in the fast-acting mechanism, and loosening of the arc plate. Without effective online monitoring, these faults can lead to prolonged operation, potentially causing transformer shutdowns or even system accidents.

[0003] Existing OLTC (Online Continuous Tracing) condition monitoring methods for vibration signals primarily focus on analyzing impact vibrations during a single gear shift. They utilize time-frequency features such as wavelet packet energy, energy entropy, singular value decomposition (SVD), empirical mode decomposition (EMD), and Hilbert-Huang transform (HHT), combined with pattern recognition algorithms like support vector machines and random forests, as well as methods such as convolutional neural networks and time-frequency graph deep learning, to classify and discriminate vibration signals. These methods have made some progress in improving feature resolution and fault type differentiation. However, these methods typically rely on large amounts of labeled samples for offline training and parameter tuning, placing high demands on computational resources and software environments, making engineering implementation and long-term online operation challenging.

[0004] To reduce implementation complexity, some studies have used a few high-amplitude indicators such as peak value, root mean square, and energy entropy to construct simplified criteria. However, these methods mostly focus on the high-amplitude response of the vibration signal at the moment of contact switching, lacking segmented modeling and targeted feature extraction of the vibration signal throughout the entire process. This results in the slight deterioration of the transmission link and early fault signs being submerged by strong impact components, making it difficult to identify specific mechanical fault types in a timely manner and limiting the ability to locate fault locations and determine their evolution.

[0005] At the same time, some engineering sites have limited conditions and it is difficult to access multi-channel signals. Therefore, how to select a small number of statistical features with clear physical meanings in a targeted manner, and realize online monitoring, condition assessment and early fault warning of OLTC mechanical status without the need for complex classifier training and large-scale computing resources, while taking into account engineering practicality and scalability, is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] In view of the above problems, the present invention proposes an OLTC mechanical fault monitoring method based on multi-index evaluation of vibration signals, so as to overcome the above problems or at least partially solve the above problems.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: The OLTC mechanical fault monitoring method based on multi-index evaluation of vibration signals includes the following steps: S1. Collect vibration signal samples when the on-load tap changer of the transformer is changing gears during operation, and determine the analysis window for a single gear change. S2. Preprocess the collected vibration signal samples; S3. Based on the operating mechanism and vibration timing characteristics of the on-load tap changer, the vibration signal samples in the preprocessed single shift analysis window are segmented to obtain vibration sub-sequences at different stages. S4. Extract multiple statistical indicators that can characterize the stage characteristics from the vibration subsequences of each stage, and construct a multi-dimensional feature vector for a single gear shift. S5. Perform statistical analysis on the multi-dimensional feature vectors obtained from multiple gear shifting operations to construct a multi-index baseline for normal operation; introduce multiple typical mechanical fault conditions under experimental conditions to construct a typical mechanical fault feature fingerprint database. S6. Compare and analyze the multi-dimensional feature vector obtained from the current gear shifting operation with the normal operation multi-index baseline and the typical mechanical fault feature fingerprint database to conduct online evaluation and graded diagnosis of the mechanical state of the on-load tap changer.

[0008] Furthermore, S1 includes: S11. When a shift command is detected or the start of the on-load tap changer shift action is detected by threshold triggering, the vibration signal acquisition of the current shift action is initiated. S12. Expand the acquisition time range before and after the actual switching moment of the contact, so that the acquisition time includes the transmission chain response range before the shift operation, the entire shift operation process, and the vibration attenuation range after the shift operation. Use the expanded acquisition time range as the single shift analysis window.

[0009] Furthermore, S2 includes: S21. Analyze the vibration signal samples collected within the single gear shift analysis window according to the sampling period. Discretization is performed to obtain discrete vibration signals. , ,in, , The number of sampling points within a single gear shift analysis window; S22. The least squares method is used to analyze discrete vibration signals. Perform univariate linear least squares fitting to obtain the trend function. And calculate the trend term corresponding to each discrete vibration signal; S23. Remove the trend term point by point from the discrete vibration signal to obtain the detrended vibration sequence. ; S24. Based on the on-site electromagnetic environment and the noise level of the vibration signal, adopt appropriate filtering methods to process the de-stressed vibration sequence. Filtering is performed to obtain the preprocessed vibration sequence. .

[0010] Furthermore, S3 includes: S31. Collect normal shift vibration signal samples under multiple normal shift operations; S32. Divide the single shift analysis window under each normal shift operation into the transmission energy storage section, the main impact section and the tail section; based on the average or specified quantile of the start and end times of the three stages of multiple normal shift vibration signal samples, obtain the unified reference start and end time interval of the three stages on the time axis. S33. Taking the starting point of the single shift analysis window of the current shift operation as the time origin, convert the reference start and end time intervals of each stage into the corresponding sampling point numbers, and extract the corresponding vibration subsequences of each stage from the preprocessed vibration signal samples.

[0011] Furthermore, S32 includes: S321. Perform preprocessing and sliding window analysis on the vibration signal samples within the single gear shift analysis window for each normal gear shift operation to obtain the short-time root mean square sequence and plot the short-time root mean square curve. S322. Calculate the mean and standard deviation of the short-time root mean square sequence for each normal gear shift. Based on the mean and standard deviation, adaptively set the entry threshold of the main impact segment. and exit threshold ; S323. Perform connectivity analysis on the short-time root mean square curve of each normal gear shift on the time axis. When there exists a certain integer... This ensures that the number of sliding windows is not less than the preset number. All of the following conditions are satisfied on the continuous sliding window At that time, the first The time corresponding to the left end of each sliding window Defined as the start time of the main impact phase of this gear shift, where Let f be the short-time root mean square within the nth sliding window, H be the sliding window step size, and f be the root mean square within the nth sliding window. s The sampling frequency; When there exists a certain integer This ensures that the number of sliding windows is not less than the preset number. All of the following conditions are satisfied on the continuous sliding window At that time, the first The time corresponding to the right end of each sliding window Defined as the end time of the main impact phase of this gear shift; S324. Based on the duration of the single gear shift analysis window and the start and end times of the main impact segment, divide the start and end times of the three stages in this normal gear shift operation. S325. For all normal gear shifting operations, calculate the mean or specified quantile of the start and end times of the three stages to obtain a unified reference start and end time interval for the three stages on the time axis. , , .

[0012] Furthermore, the entry threshold of the main impact phase and exit threshold The calculation method is as follows:

[0013] in, This represents the root mean square value of the sequence under normal shifting conditions. This represents the standard deviation of the root mean square sequence under normal gear shifting operation. This is an empirical coefficient.

[0014] Furthermore, in S4, the extraction process of the multi-dimensional feature vector for a single gear shift includes: In the transmission and energy storage section, amplitude-based statistics, higher-order statistics, and entropy-based indices are selected to describe the vibration amplitude level, waveform sharpness, and the complexity of the spectral energy distribution. In the main impact phase, energy or amplitude statistics are extracted as features to describe the impact intensity and the stability of the contact action process. In the tail section, the proportion of energy in the tail section to the total energy of the three sections is calculated to assess whether the vibration decay process is normal. During the entire gear shifting process, the energy ratio between gear shifts is calculated to characterize the relative distribution pattern of energy at each stage.

[0015] Furthermore, S5 includes: S51. Calculate the mean and standard deviation of each feature component in the multiple sets of multidimensional feature vectors obtained under multiple normal gear shifting operations, and establish a fluctuation range for each feature component based on the mean and standard deviation as the baseline range for multiple indicators during normal operation. S52. Introduce multiple typical mechanical fault conditions under experimental conditions, including at least transmission jamming, contact wear, spring failure and arc plate loosening. Collect vibration signals of gear shifting under the corresponding fault conditions, and obtain multi-dimensional feature vectors of various fault types according to the same processing procedure as normal conditions. S53. For any fault type, calculate the mean and standard deviation of each characteristic component under multiple experiments for that fault type.

[0016] S54. Compare the mean and standard deviation of each characteristic component in each fault type with the mean, standard deviation and baseline interval of the corresponding characteristic component under normal shifting operation to characterize the offset direction and offset magnitude of each characteristic component under fault state relative to normal state. S55. Solidify the characteristic vector offsets that are representative of the baseline interval of the normal state under the fault state into the characteristic fingerprints of the corresponding mechanical faults, and form a mechanical fault characteristic fingerprint library.

[0017] Furthermore, S6 includes: S61. Using the mean and standard deviation of each characteristic component under normal shifting operation, calculate the standardized deviation of each characteristic component under the current shifting operation, and take the maximum value of the absolute value of the standardized deviation of each characteristic component as the multi-index fusion evaluation metric. ; S62, Set two thresholds Based on the evaluation results of multiple consecutive gear shifts, a classification is made: when If this condition is met in the most recent gear shifts, the current gear shift is determined to be within the normal fluctuation range. when Or, if only a slight over-limit occurs during a single gear shift, the shift will be marked as a deteriorating condition or a warning event; when If a stable over-limit occurs during multiple consecutive gear shifts, a mechanical fault or abnormality is determined, and an alarm is triggered. S63. When a mechanical fault or abnormality is determined, the mechanical fault feature fingerprint database is called to perform fault type matching on the multi-dimensional feature vector of this gear shift and output a diagnosis.

[0018] Furthermore, S63 includes: S631. Analyze the components of the multidimensional feature vector obtained from the current gear shift operation. Calculate the standardized deviation The standardized deviations of each component are combined into a standardized deviation vector. ; S632, For each candidate fault type and its set of key indicators Check each gear shift's standardized deviation across all indicators to see if it falls within the pre-defined criterion range for that fault type. ; like fall into Internally, it is believed that the current gear shift is in line with the indicator. Above and fault type The offset pattern must match the typical offset pattern; otherwise, it is considered a mismatch. S633, For each fault type The number of key indicators that meet the offset criterion interval is counted, and the normalized matching degree is calculated. Matching degree This is used to reflect the number of gear shifts currently occurring, and which key indicators are related to the type of fault. The typical offset combination pattern is consistent with this; S634. After obtaining the matching degree set of all fault types... Then, set the type recognition threshold. The fault type results will be output according to the following principles: If a certain type of fault exists Its matching degree The largest among all candidate fault types and not lower than the threshold Then the fault type This is confirmed as the diagnostic output result for this gear shift. If the matching degree of all fault types is lower than If the difference between the matching degrees of multiple fault types is less than the preset threshold, the current shift will be marked as "abnormal and unclassified" or "awaiting manual review".

[0019] As can be seen from the above technical solution, compared with the prior art, the present invention has the following beneficial effects: Based on the OLTC mechanism's action mechanism and vibration timing characteristics, the fixed time window of a single gear shift is divided into a transmission energy storage segment, a main impact segment, and a tail segment. A small number of key statistical indicators are extracted from each stage to form a lightweight feature vector suitable for online computation. On this basis, a multi-index baseline of the OLTC vibration signal under normal operating conditions is constructed, and a feature fingerprint library covering typical mechanical fault modes is established. A multi-index fusion fault occurrence discrimination quantity based on standardized deviation and a two-level threshold classification rule are introduced to quantitatively determine whether the current gear shift deviates from the normal baseline and its severity. Combined with the offset combination criterion of the fault feature fingerprint in the standardized space, the automatic identification of typical mechanical fault types such as transmission jamming, contact wear, spring failure, and arc plate loosening is realized, achieving online monitoring and early fault warning of the mechanical state of the OLTC gear shift in online operation.

[0020] Because the features used in this invention have low dimensionality and the calculation and criterion forms are simple and clear, they are easy to be solidified and deployed in online monitoring devices using software algorithms and parameter configurations. This enables long-term stable operation and maintenance in engineering sites, thereby significantly improving the reliability and sensitivity of OLTC online mechanical condition monitoring in practical applications, enabling graded diagnosis and early warning of faults, and providing strong support for operation and maintenance decisions. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0022] Figure 1 This is a flowchart of the OLTC mechanical fault monitoring method based on multi-index evaluation of vibration signals provided in an embodiment of the present invention; Figure 2 The short-time root mean square curve is provided for embodiments of the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] like Figure 1 As shown in the figure, this invention discloses a method for monitoring OLTC mechanical faults based on multi-index evaluation of vibration signals, including the following steps: S1. Collect vibration signal samples when the on-load tap changer of the transformer is changing gears during operation, and determine the analysis window for a single gear change. S2. Preprocess the collected vibration signal samples; S3. Based on the operating mechanism and vibration timing characteristics of the on-load tap changer, the vibration signal samples in the preprocessed single shift analysis window are segmented to obtain vibration sub-sequences at different stages. S4. Extract multiple statistical indicators that can characterize the stage characteristics from the vibration subsequences of each stage, and construct a multi-dimensional feature vector for a single gear shift. S5. Perform statistical analysis on the multi-dimensional feature vectors obtained from multiple gear shifting operations to construct a multi-index baseline for normal operation; introduce multiple typical mechanical fault conditions under experimental conditions to construct a typical mechanical fault feature fingerprint database. S6. Compare and analyze the multi-dimensional feature vector obtained from the current gear shifting operation with the normal operation multi-index baseline and the typical mechanical fault feature fingerprint database to conduct online evaluation and graded diagnosis of the mechanical state of the on-load tap changer.

[0025] The specific implementation process of each step of the present invention will be further explained below.

[0026] S1. Vibration Signal Acquisition and Determination of Single Gear Shift Analysis Window: One or more accelerometers are placed near the outer surface of the OLTC enclosure to collect vibration samples of the transformer OLTC gear shifting action during operation. When a gear shift command is received or the start of a gear shift action is detected through threshold triggering, edge detection, or other methods, vibration data acquisition for the current action is initiated. The acquisition time range is appropriately extended before and after the actual contact switching moment, so that the acquisition duration includes the transmission chain response interval before the gear shift action, the entire gear shifting process, and the vibration decay interval after the action ends. The aforementioned acquisition time range is defined as the single gear shift analysis window; vibration signals continuously acquired by the accelerometers within this analysis window are used as vibration samples. Specifically, this includes: S11. An acceleration sensor is installed on the outer surface of the housing near the OLTC drive train and contact switching area, avoiding the reinforcing ribs, to collect the vibration response of the housing during gear shifting. When a shift command is detected or the start of the on-load tap changer shifting action is detected through threshold triggering, vibration signal acquisition for the current shifting action is initiated.

[0027] When the on-load tap changer is not shifting, vibration acceleration is pre-acquired to obtain a raw time-domain sequence lasting several seconds. . Use it at sampling frequency Sampling is performed to obtain discrete vibration sequences. ,in, .

[0028] The short-time root mean square is calculated using a sliding window method under static conditions. The root mean square of the moving window is defined as shown in equation (1): (1) in, This represents the m-th sample within the n-th frame; m is the index of the sampling point within the window; L is the length of the sliding window; H is the window step size; is the sliding window number; N is the total number of sampling points in the entire vibration sequence.

[0029] Moving root mean square sequence for all background windows The mean and standard deviation are calculated as shown in equation (2) below: (2) in, This represents the total number of windows used in the background pre-capture stage. (The above...) and Used as a time-domain background baseline under static operating conditions.

[0030] The action initiation threshold is adaptively set based on the background baseline, for example, by selecting... As shown in equation (3): (3) in, These are empirical coefficients used to adjust the trigger sensitivity. Since the above parameters are derived entirely from pre-acquired time-domain background data, no frequency-domain transformation is required, making them easy to implement in field devices.

[0031] During the online monitoring phase, the same sliding window parameters were continuously applied to the real-time acquired vibration time-domain signals. Sampling is performed and the moving root mean square is calculated. Let the real-time sampling sequence be denoted as . Then the first The root mean square of the movement of each window is As shown in equation (4): (4) in, This represents the m-th sample within the n-th frame; m is the index of the sampling point within the window; L is the sliding window length; H is the window step size; is the sliding window sequence number; N is the total number of sampling points in the entire vibration sequence. The total number of sliding windows; the first The time center corresponding to each window can be represented as: .

[0032] When there exists a certain integer This ensures that the number of sliding windows is not less than the preset number. All of the following conditions must be met on the continuous sliding window: From the first All sliding windows starting from the first sliding window have a moving root mean square value greater than the trigger threshold. A significant increase in the time-domain energy level of the vibration signal relative to the static background is considered the initiation of a gear shift, thus triggering vibration data acquisition for the current gear shift. The sampling frequency can be set from 20kHz to 100kHz, denoted as _____. It is used to cover the frequency bands of typical mechanical vibrations such as gear meshing, spring energy storage, and contact impact.

[0033] S12. During the data acquisition process, to ensure a complete acquisition of the vibration evolution process of a single gear shift, the acquisition time range is extended before and after the actual contact switching moment. This ensures that the acquisition duration includes the transmission chain response range before the gear shift, the entire gear shift process, and the vibration attenuation range after the gear shift. The extended acquisition time range is used as the analysis window for a single gear shift. For example, the time can be extended to 50–200 ms before and 300–500 ms after the expected contact switching moment, so that the acquisition window includes the weak vibration before the gear shift, the impact vibration at the moment of switching, and the attenuation response after the action.

[0034] The extended time range described above is defined as the single gear shift analysis window, denoted as the interval. Within this analysis window, the accelerometer is sampled at a fixed frequency. The continuously acquired vibration signals are denoted as: .

[0035] S2. Preprocessing of the acquired vibration signal samples: Based on a time axis strictly aligned with the original vibration signal, the least squares method is used to linearly fit the original vibration signal, generating a linear trend sequence corresponding to the original signal. Following the strict time axis alignment principle, the amplitude of the trend sequence at each sampling point is successively subtracted from the amplitude of the original signal at that time point to obtain a detrended and de-DC processed vibration signal sequence. Depending on the noise level of the vibration signal sequence, target frequency band enhancement processing can be selectively applied to the vibration signal, implementing filtering operations to suppress power frequency and low-frequency structural oscillation components, while simultaneously reducing high-frequency random noise. Specifically, this includes: S21. The vibration signal samples collected within the single gear shift analysis window... According to the sampling period Discretization is performed to obtain discrete vibration signals. , ,in, , This represents the number of sampling points within a single gear shift analysis window.

[0036] S22, to Linear detrending processing is performed to reduce low-frequency interference such as sensor bias, temperature drift, and slow changes in mechanical structure. Specifically, the least squares method is used for discrete vibration signals. Perform univariate linear least squares fitting to obtain the trend function. And calculate the trend term corresponding to each discrete vibration signal; trend function As shown in the following formula (5): (5) Among them, the trend coefficient , As shown in the following formula (6): (6) S23. Remove the trend term point by point from the discrete vibration signal to obtain the detrended vibration sequence. As shown in the following formula (7): - (7) in, The slow linear drift component characterizing the vibration signal within the analysis window is mainly caused by low-frequency factors such as sensor zero drift and slow structural deformation. This trend term is then... Removing the middle part is equivalent to performing linear detrending processing on the original vibration signal, resulting in... It primarily retains the effective vibration components related to mechanical motion.

[0037] S24. Based on the on-site electromagnetic environment and the noise level of the vibration signal, adopt appropriate filtering methods to process the de-stressed vibration sequence. Filtering is performed to obtain the preprocessed vibration sequence. .

[0038] For example, bandpass or high-pass digital filtering methods can be used to suppress power frequency and low-frequency structural oscillations while reducing random high-frequency noise, thus highlighting the effective vibration components related to transmission chain meshing, spring energy storage, and contact impact. The specific type, order, and passband range of the filter can be selected according to site requirements. For instance, when the power frequency magnetic field near the transformer body is strong and there are significant low-frequency oscillations in the support or bushing (mainly concentrated in the 0–100 Hz range), a bandpass filter with a passband of approximately 500–5000 Hz and an order of 4–6 can be used to effectively suppress power frequency and its harmonics, as well as low-frequency structural oscillations, highlighting the mid-to-high frequency components of gear meshing and contact impact. In situations where power frequency interference is weak but the installation location introduces a large amount of low-frequency structural resonance, a high-pass filter with a cutoff frequency of approximately 200 Hz and an order of 2–4 can be used to suppress low-frequency resonance and retain high-frequency impact components. For sites with significant high-frequency electromagnetic noise, the upper limit of the passband can be narrowed accordingly, or a low-pass stage can be added to the high-pass filter to reduce random noise above 10 kHz.

[0039] S3. Segmentation of a Single Gear Shift Process Based on Action Timing: After preprocessing the vibration signal, a sliding window can be used to calculate the short-time root mean square curve to characterize the changes in vibration intensity and spectral complexity over time. Based on this, and combining the typical action sequence of OLTC transmission chain drive, spring energy storage, and contact switching, the vibration signal is analyzed on the time axis to identify three key stages with clear physical meaning during the gear shift process: the transmission energy storage stage, the main impact stage, and the tail stage. The transmission energy storage stage corresponds to the weak vibration response caused by the force change in the transmission chain and spring mechanism before contact switching; the main impact stage corresponds to the high-amplitude vibration range caused by contact switching and concentrated impact force; and the tail stage corresponds to the mechanism return and system vibration decay process. Through time annotation and statistical analysis of a large number of normal gear shift samples in the above three stages, reference start and end intervals of the transmission energy storage stage, main impact stage, and tail stage on the time axis are obtained. These are then solidified into unified segmentation boundaries, allowing different gear shift samples to be divided into consistent stages within the same analysis window, enabling cross-sample comparison. This includes the following steps: S31. During the offline phase, collect normal shift vibration signal samples from multiple (e.g., at least 30) normal shift operations.

[0040] S32. Plot the preprocessed vibration waveforms of the vibration samples within the single gear shift analysis window for each normal gear shift operation. Short-time root mean square curve The operator, based on the OLTC drive chain, spring energy storage, and contact switching mechanism, performs stage marking for each sample, recording the start and end times of the main impact phase. , ,in For sample number, , They represent the first The start and end times of the main impact segment in the subsample (timed relative to the start of the analysis window).

[0041] The interval before the main impact phase, where the vibration energy is relatively stable, is marked as the transmission energy storage phase, and its start and end times are recorded as follows: , The section where the vibration gradually decays after the main impact phase is marked as the tail segment, and its start and end times are recorded as follows: , .

[0042] Statistical analysis was performed on the time-marked results of all normal samples. For example, the mean or quantile of the start and end times for each stage were calculated to obtain the reference start and end times of the three stages on the timeline. .

[0043] Among them, three segments satisfy , , , , This represents the total duration of the analysis window for a single gear shift.

[0044] Subsequently, based on the average or specified quantile of the start and end times of the three stages from multiple normal shift vibration signal samples, a unified reference start and end time interval for the three stages on the time axis was obtained.

[0045] Specifically, for each gear shift operation within the single gear shift analysis window, the algorithm automatically identifies the reference time. The specific steps include: S321. To characterize the change of vibration intensity over time, the vibration signal samples within the single shift analysis window under each normal shift operation are preprocessed and subjected to sliding window analysis to obtain a short-time root mean square sequence, and a short-time root mean square curve is plotted.

[0046] In the Within each window, the short-time root mean square As shown in the following formula (8): (8) in: This represents the m-th sample within the n-th frame; m is the index of the sampling point within the window; L is the sliding window length; H is the window step size; is the sliding window sequence number; N is the total number of sampling points in the entire vibration sequence. The total number of sliding windows; the first The time center corresponding to each window can be represented as: ; The above calculations yield a short-time root mean square curve that varies with time, which serves as the segmentation criterion.

[0047] S322. Calculate the mean and standard deviation of the short-time root mean square sequence for each normal gear shift. Based on the mean and standard deviation, adaptively set the entry threshold of the main impact segment. and exit threshold ; The moving root mean square sequence of the entire gear shift process The mean and standard deviation are calculated as shown in equation (9): (9) in, This represents the total number of windows within the current gear shift analysis window. and Based on this, adaptively set the entry and exit thresholds for action segments, for example, selecting... (10) in, This is an empirical coefficient used to adjust the sensitivity and hysteresis range of motion segment start and end detection.

[0048] S323. Perform connectivity analysis on the short-time root mean square curve of each normal gear shift on the time axis to find continuous intervals that satisfy the entry condition: when there exists a certain integer... This ensures that the number of sliding windows is not less than the preset number. All of the following conditions are satisfied on the continuous sliding window At that time, the first The time corresponding to the left end of each sliding window Defined as the start time of the main impact phase of this gear shift, where Let f be the short-time root mean square within the nth sliding window, H be the sliding window step size, and f be the root mean square within the nth sliding window. s The sampling frequency; Continue to track When there exists a certain integer This ensures that the number of sliding windows is not less than the preset number. All of the following conditions are satisfied on the continuous sliding window At that time, the first The time corresponding to the right end of each sliding window Defined as the end of the main impact phase of the gear shift; in typical samples, this moment usually falls near the first obvious trough after the short-time root mean square main peak.

[0049] This allows for the automatic delineation of the start and end time intervals of the main impact segment on the preprocessed vibration waveform and the short-time root mean square curve, with the leftmost time point, i.e., the start time, denoted as . The rightmost time point, i.e., the end time, is recorded as... , and respectively serve as the start and end times of the main impact segment of the k-th sample. .

[0050] S324. Based on the duration of the single gear shift analysis window and the start and end times of the main impact segment, divide the start and end times of the three stages in this normal gear shift operation.

[0051] On the left side of the main impact segment, [0, Within the given interval, the short-time root mean square remains relatively stable, and the entire segment can be considered as a transmission and energy storage section. , .

[0052] On the right side of the main impact segment ( Within the interval [T], the short-time root mean square gradually decreases from a high value and falls back to the background level, which can be regarded as the tail segment. , , where T is the total duration of a single gear shift analysis window.

[0053] S325. For all normal gear shifting operations, calculate the mean or specified quantile of the start and end times of the three stages to obtain a unified reference start and end time interval for the three stages on the time axis. , , .

[0054] In practical applications, a small number of samples can be manually labeled to obtain an initial reference time. Then, an automatic algorithm can be used to batch correct and supplement subsequent samples, eventually forming a stable set of reference start and end time parameters.

[0055] S33. Taking the starting point of the single shift analysis window of the current shift operation as the time origin, convert the reference start and end time intervals of each stage into corresponding sampling point numbers, and extract the corresponding vibration sub-sequences of each stage from the preprocessed vibration signal samples. Specifically, this includes: During online monitoring, the starting point of the single gear shift analysis window is taken as the time origin, and the reference start and end times of each stage are set. , , Converted into corresponding sampling point numbers, used in the preprocessed vibration sequence Extract the corresponding subsequence from the sample. Preferably, the sampling point numbering is as shown in the following formula (11): (11) in, Sampling frequency, For rounding operators, and Each represents a stage The starting and ending sampling point numbers in the discrete vibration sequence.

[0056] By converting the time to the sampling point number as described above, the vibration sequences of the transmission energy storage section, the main impact section, and the tail section after preprocessing can be determined. The range of each corresponding sampling point.

[0057] Specifically, when the sampling point number satisfy At that time, the corresponding pre-processed vibration samples This is denoted as the transmission energy storage segment sequence. ,Right now:

[0058] When sampling point number satisfy At that time, the corresponding sample is recorded as the main impact segment sequence. ,Right now:

[0059] When sampling point number satisfy At that time, the corresponding sample is denoted as the tail subsequence. ,Right now:

[0060] Therefore, in the same preprocessed vibration sequence The system is divided into three sections: transmission energy storage section, main impact section, and tail section.

[0061] S4. Multi-index extraction and feature vector construction of vibration signals at each stage: The transmission energy storage stage mainly reflects the weak amplitude disturbances generated during transmission chain meshing and spring energy storage. Therefore, amplitude-based statistics (such as root mean square RMS), higher-order statistics (such as kurtosis), and entropy-based indicators (such as spectral entropy) are selected in this stage to describe the vibration amplitude level, waveform sharpness, and complexity of spectral energy distribution. The main impact stage corresponds to the concentrated impact generated by contact switching, and its vibration amplitude changes drastically. Energy-based or amplitude-based statistics (such as root mean square) can be extracted as features in this stage to characterize the impact intensity and stability of the contact action process. The tail stage mainly reflects the mechanism's return and the system's free decay characteristics. By calculating the energy ratio of the tail stage (such as the proportion of tail stage energy to the total energy of the three stages), the normality of the vibration decay process can be evaluated. In addition, to characterize the energy distribution relationship between different stages throughout the shifting process, inter-stage energy ratio indicators (such as the ratio of energy in the transmission energy storage stage to energy in the main impact stage) can be constructed to characterize the relative energy distribution pattern in each stage. The above-mentioned statistical indicators are combined in a preset order to form a multi-dimensional feature vector describing a single gear shift action.

[0062] Specifically, to characterize the vibration intensity and energy level of each stage, the mean square energy and root mean square value (RMS) are uniformly defined for the transmission energy storage section A, the main impact section B, and the tail section C. Let the stages be... The range of sampling point numbers is The corresponding number of sampling points is Then the stage mean square energy With root mean square value The calculation is shown in equation (12): (12) in, The average vibration energy levels of the transmission energy storage section, the main impact section, and the tail section are respectively characterized. Primarily used to characterize the overall vibration intensity of the transmission energy storage section and the main impact section; in tail section scenarios, energy indicators can be prioritized. Its proportion and characteristics are used in subsequent evaluations.

[0063] To characterize the sharpness and impact of the vibration waveform in the transmission energy storage section, a kurtosis index is calculated on transmission energy storage section A. First, based on the subsequence... Calculate the mean with standard deviation As shown in equation (13): (13) The kurtosis of the transmission energy storage section As shown in equation (14): (14) in, It reflects the degree of peak in the vibration waveform of the transmission energy storage section relative to the normal distribution, and is used to identify abnormal meshing of the transmission chain or local impact characteristics.

[0064] To describe the complexity and randomness of the spectral energy distribution in the transmission energy storage section, the spectral entropy index is calculated within transmission energy storage section A. Specifically: Preprocessed vibration sample sequences within the transmission energy storage section Apply a window function (e.g., a Hanning window) and perform a process of length [length missing]. The Fast Fourier Transform (FFT) is used to obtain the single-sided spectrum. The corresponding normalized power spectrum is defined as shown in equation (15): (15) Among them, Frequency point index; The number of effective frequency points in a single-sided spectrum; For summation index.

[0065] Then the spectral entropy of the transmission energy storage section The definition is shown in equation (16): (16) Through the above normalization The value of is constrained to Within the range, a higher value indicates a more dispersed and complex distribution of spectral energy; a lower value indicates that the energy is concentrated in a few frequency bands, which is convenient for reflecting the complexity change of the transmission energy storage section from normal engagement to abnormal state.

[0066] To reflect the energy distribution and attenuation characteristics at the end of the gear shift, an index is constructed to represent the energy percentage at the end of the shift and the energy ratio between shift stages. (Energy percentage at the end of the shift) The definition is shown in equation (17): (17) This indicator is used to characterize the proportion of the mechanism's return and system attenuation processes in the overall shifting energy. When the return is abnormal or there is mechanical loosening, it indicates an issue. It often appears to be abnormally high or low.

[0067] Inter-segment energy ratio The definition is shown in equation (18): (18) in, Used to characterize the energy distribution pattern between the transmission energy storage section and the main impact section, when the transmission chain resistance increases, the spring energy storage process is abnormal, or the contact impact weakens. and The relative relationships will change, thus being reflected in On the offset.

[0068] Based on the statistical indicators of the above stages, after each gear shift, the features corresponding to a single gear shift are combined in a preset order to form a multi-dimensional feature vector. Preferably, the following six core indicators can be selected to form the feature vector.

[0069] in, For the transmission energy storage section RMS; The kurtosis of the transmission energy storage section; The spectral entropy of the transmission energy storage section; RMS in the main impact segment; The proportion of energy in the tail section; This refers to the energy ratio between the transmission energy storage section and the main impact section.

[0070] S5. Establishment of a multi-indicator baseline for normal operation and a fingerprint database of typical mechanical fault features: Statistical analysis is performed on the multi-dimensional feature vectors obtained for each gear shift to obtain the mean, dispersion range, and typical variation intervals of various indicators under normal operating conditions. This data describes the reasonable fluctuation range of normal gear shifting behavior, thus forming a multi-indicator baseline for normal operation. Based on this, combined with the analysis of the OLTC mechanism's operating mechanism, and using typical fault simulation samples constructed under experimental conditions, such as transmission jamming, contact wear, spring failure, and arc plate loosening, as feature offset references, the multi-indicator features of each stage under abnormal operating conditions are compared with the normal baseline. This identifies the regular deviation patterns of different fault types in terms of amplitude, higher-order statistics, entropy indicators, tail-segment energy ratio, and inter-segment energy ratio. The representative feature vector offsets relative to the normal baseline are solidified as feature fingerprints of the corresponding mechanical faults, forming a mechanical fault feature fingerprint database. During subsequent online monitoring operations, the fingerprint entries are continuously corrected and improved by incorporating new samples. This includes the following steps: S51. For each feature component in the multiple sets of multi-dimensional feature vectors obtained from multiple normal gear shifting operations, calculate the mean and standard deviation. Based on the mean and standard deviation, establish a fluctuation range for each feature component as the baseline range for multiple indicators during normal operation. Specifically, this includes: Statistical analysis is performed on the multi-dimensional feature vectors obtained from each gear shift to establish a multi-indicator baseline for normal operation and to construct a feature fingerprint database of typical mechanical faults. Specifically, for the first... The next gear shift action is denoted by its segmented multi-index feature vector as follows: .

[0071] To obtain the total under normal working conditions Group feature vectors ( Each component in the dataset is statistically analyzed separately. For any characteristic component... For example, its sample mean under normal operating conditions and standard deviation The calculation is shown in equation (19): (19) in, Indicates the first under normal operating conditions Feature vector corresponding to secondary gear shift The value of this component in the equation.

[0072] Based on the above statistical results, for each feature component... A reasonable fluctuation range is established as the baseline range for multiple indicators during normal operation, as shown in equation (20): (20) in, This is a tolerance factor set based on engineering experience or desired confidence levels (e.g., 2.0 corresponds to approximately 95% confidence interval).

[0073] For sets All components are sequentially established within the aforementioned intervals, thus forming a multi-index baseline for the single gear shift feature vector under normal operating conditions, used to describe the statistical characteristics and reasonable fluctuation range of normal gear shifting behavior.

[0074] S52. Introduce multiple typical mechanical fault conditions under experimental conditions, including at least transmission jamming, contact wear, spring failure, and arc plate loosening. Collect vibration signals of gear shifting under the corresponding fault conditions, and obtain multi-dimensional feature vectors of various fault types according to the same processing procedure as normal conditions.

[0075] S53. For any fault type, calculate the mean and standard deviation of each characteristic component under multiple experiments for that fault type.

[0076] S54. Compare the mean and standard deviation of each characteristic component in each fault type with the mean, standard deviation and baseline interval of the corresponding characteristic component under normal shifting operation to characterize the offset direction and offset magnitude of each characteristic component under fault state relative to normal state.

[0077] Record No. Mechanical failures (m=1,…, (representing various types of faults) in the first... The feature vector in this experiment is:

[0078] To standardize the notation, let the value of any characteristic component in the j-th test of the m-th type of fault be denoted as . ,in Indicates the feature name. This can be calculated. sample mean and standard deviation This is compared with the mean obtained from a sample of multiple gear shifts under normal operating conditions. Standard deviation and baseline interval By making a comparison, we can depict the direction and magnitude of the deviation of this indicator under fault conditions relative to normal conditions.

[0079] S55. The representative feature vector offsets relative to the baseline interval of the normal state under fault conditions are solidified into feature fingerprints corresponding to the mechanical faults, forming a mechanical fault feature fingerprint database. In practical applications, statistical tests and discrimination assessments can be further performed on the differences in feature components between the normal group and each fault group. For example, for the same component... A two-sample hypothesis test was performed on sample sequences from the normal group and a certain fault group to obtain the significance level. Simultaneously calculate the effect size. The component is used to measure its ability to distinguish between normal and fault types, and statistical indicators such as the area under the binary ROC curve (AUC) are constructed based on this component. By combining the feature vector offset patterns of each fault condition relative to the normal operating samples and related statistical distinguishing indicators, a typical mechanical fault feature fingerprint database for different fault modes is constructed.

[0080] S6. Implementation of Online Monitoring and Two-Level Diagnostic Criteria for Mechanical Faults Based on Multi-Indicator Fusion Evaluation: For each gear shift, the extracted multi-dimensional feature vector is compared and analyzed with the normal operation multi-indicator baseline and the mechanical fault feature fingerprint database to achieve real-time evaluation and hierarchical diagnosis of the OLTC's mechanical status. First, at the fault occurrence discrimination level, the current feature vector undergoes a baseline consistency check. By comparing the deviation of key indicators at each stage from the normal operation baseline, it is assessed whether they exceed the normal fluctuation range or exhibit abnormal trends. When multiple key indicators show significant deviations in the same or several stages, or show stable deviations in multiple consecutive gear shifts, a mechanical abnormality is determined in the current gear shift process; if the deviation is small and only occurs in individual gear shifts, it is marked as a deteriorating condition or a warning event to indicate potential early fault symptoms. Second, at the fault type identification level, the current feature vector is matched with patterns in the fault feature fingerprint database. The degree of conformity with each fault fingerprint can be calculated using methods such as rule matching, similarity scoring, or comprehensive offset measurement. When the typical offset pattern of a certain fault type has a high consistency with the multidimensional offset features of the current gear shift, the corresponding fault type identification result is output; if the matching degree of each fault fingerprint is not significant, it is marked as "abnormal and unclassified" or "awaits manual review", and the corresponding feature vector is output for subsequent manual judgment and further correction and supplementation of the fault fingerprint database. This includes the following steps: S61. Using the mean and standard deviation of each characteristic component under normal shifting operation, calculate the standardized deviation of each characteristic component under the current shifting operation, and take the maximum value of the absolute value of the standardized deviation of each characteristic component as the multi-index fusion evaluation metric. Specifically, this includes: For the feature vector of a certain gear shift any component Using normal sample statistics , The standardized deviation is calculated as shown in equation (21): (twenty one) in, To prevent the division by small positive numbers with a denominator of zero, the standardized deviation vector is constructed as shown in equation (22): (twenty two) Used to characterize the normalized deviation of the current shift from the normal operating baseline across various metrics.

[0081] Based on the above-mentioned standardized bias, a multi-index fusion evaluation metric for the fault occurrence layer is defined. As shown in equation (23): (twenty three) in, Indicates the first The largest value among the absolute values ​​of the standardized deviations of each component of the shift is the shift with the higher degree of overall abnormality of the shift relative to the normal operating baseline.

[0082] S62. In a dataset containing only normal shifting conditions, first calculate the fault occurrence level evaluation quantity for each shift. A set of samples was obtained. Based on this, the mean and standard deviation of this evaluation quantity under normal operating conditions are calculated as shown in equation (24): (twenty four) by , As a baseline, two grading thresholds are adaptively set, as shown in equation (25): (25) in This is an empirical coefficient used to adjust the sensitivity of "early warning" and "fault" signals. In engineering, for example... (Corresponding to the 95% confidence interval of normal fluctuations) (This corresponds to the 99.7% confidence interval of normal fluctuations), and can also be adjusted through experiments based on historical data from the field.

[0083] Therefore, two thresholds are set. Based on the evaluation results of multiple consecutive gear shifts, a classification is made: when If this condition is met in the most recent gear shifts, the current gear shift is determined to be within the normal fluctuation range. when Or, if a slight over-limit occurs only in a few gear shifts, the shift will be marked as a deteriorating condition or a warning event, indicating that there may be an early mechanical abnormality; when If a stable over-limit occurs during multiple consecutive gear shifts, a mechanical fault or abnormality is determined, triggering an alarm in the fault occurrence discrimination layer. S63. When the fault occurrence layer determines that there is a mechanical fault or abnormality in the current gear shifting operation, it calls the mechanical fault feature fingerprint database to perform fault type matching on the multi-dimensional feature vector of this gear shift and outputs a diagnosis.

[0084] To maintain symbol consistency and avoid introducing new statistical indicators, the fault type identification layer directly utilizes the aforementioned standardized deviation vector. Perform combination criterion matching within a standardized space.

[0085] During the offline phase, based on the statistical analysis results of samples from normal operating conditions and various fault conditions, each typical mechanical fault type is analyzed. Determine a set of key indicators And the typical offset range of each indicator in the standardized deviation space. Specifically, for fault type and indicators The criterion interval for this index is pre-defined in the Z-space as shown in equation (26): (26) in Determined by the mean, standard deviation, and their deviation patterns obtained from offline statistics: If the indicator under this failure mode If the typical manifestation is significantly elevated, then choose... , Or take the upper bound as .

[0086] If the typical manifestation is significantly lower, then select , Or take the lower bound as .

[0087] If the typical presentation is close to normal, then let And the interval is relatively narrow, used for constraints It should not be too large.

[0088] Thus, each type of fault Each of these corresponds to a set of combined constraints in Z-space. This constitutes a standardized feature fingerprint for this type of fault.

[0089] When running online, when the After the fault occurrence layer has determined that "a mechanical abnormality exists" during the next gear shift, the fault type identification layer operates according to the following steps: S631. The multidimensional feature vector obtained from the current gear shift operation Each component in Calculate the standardized deviation The standardized deviations of each component are combined into a standardized deviation vector. ; S632, For each candidate fault type and its set of key indicators Check each gear shift's standardized deviation across all indicators to see if it falls within the pre-defined criterion range for this fault type: .

[0090] like fall into Internally, it is believed that the current gear shift is in line with the indicator. Above and fault type The offset pattern must match the typical offset pattern; otherwise, it is considered a mismatch.

[0091] S633, For each fault type Count the number of key indicators that meet the offset criterion interval: And calculate the normalized matching degree. As shown in equation (27): (27) Match This is used to reflect the number of gear shifts currently occurring, and which key indicators are related to the type of fault. This is consistent with the typical offset combination pattern, while preserving the degree of deviation of each indicator. The quantitative information depicted.

[0092] S634. After obtaining the matching degree set of all fault types... Then, set the type recognition threshold. The fault type results will be output according to the following principles: If a certain type of fault exists The judgment conditions are as shown in equation (28): (28) This type of fault Matching degree The largest among all candidate fault types and not lower than the threshold Then the fault type This is the diagnostic output result for this gear shift.

[0093] If the matching degree of all fault types is lower than If multiple (or more) fault types have similar matching degrees and none are significantly dominant, the current shift will be marked as "abnormal and unclassified" or "awaiting manual review," and the corresponding feature vector will be... The relevant vibration waveforms and segmentation results are uploaded to the upper-level monitoring or operation and maintenance system for manual interpretation and subsequent correction and expansion of the fault fingerprint database. In this embodiment, if the matching degree of all fault types is less than the preset threshold, they are considered to be close, and there is no obviously dominant fault type. The preset threshold can be adjusted according to the discrimination accuracy requirements, for example, a value between 0.2 and 0.3.

[0094] Through the above two-level diagnostic process, the corresponding status assessment results can be output after the gear shifting action is completed, and the corresponding fault type or unclassified abnormality mark can be given when a fault occurs, realizing online monitoring and hierarchical diagnosis of OLTC mechanical condition based on multi-index fusion of vibration signals.

[0095] In a specific example, the method of the present invention includes the following steps: S1. The vibration sensor is magnetically attached to the surface of the transformer OLTC housing, away from the reinforcing ribs. The sensor's sampling frequency is 81920Hz. After receiving a shift command from the control terminal or triggering a threshold, vibration data acquisition for the current shift action is initiated. (Threshold triggering is used here, and the detection location is the start of the impact phase of the action.) Considering the OLTC's action characteristics, the acquisition duration is set to extend forward by at least 70ms and backward by at least 400ms from the shift trigger moment, covering the transmission chain response range before the actual contact switching, the entire process of contact switching and impact, as well as the mechanism return and vibration free decay stages; thus, a single shift analysis window and the acquired vibration signal are obtained.

[0096] S2. The collected vibration samples are first subjected to linear detrending processing. Then, on the time axis that is strictly aligned with the original vibration signal, the least squares method is used to linearly fit the entire vibration sequence to obtain a linear trend curve that changes with time. Subsequently, according to the principle of aligning the time axis point by point, the amplitude of the original signal at each sampling moment is successively subtracted from the trend value corresponding to that moment, thereby simultaneously eliminating DC bias and slow linear drift, and obtaining the vibration signal sequence after detrending and DC removal processing.

[0097] Considering the presence of low-frequency structural oscillations and high-frequency random noise at the operating site, further target frequency band enhancement processing is performed based on the above detrended sequence: a 4th-order Butterworth digital bandpass filter is selected, with the passband roughly set between 100Hz and 10000Hz. This can be fine-tuned according to the specific device's frequency band characteristics. Zero-phase bidirectional filtering is used to process the signal, effectively suppressing power frequency and low-frequency structural components, while weakening random high-frequency noise and highlighting the effective frequency band of typical mechanical vibrations such as gear meshing, spring energy storage, and contact impact. This yields the pre-processed vibration signal sequence.

[0098] S3. Process segmentation based on action timing is achieved using offline statistics and online fixed segmentation. First, 68 normal gear switching samples are selected in the offline stage. The preprocessed vibration waveform and short-time root mean square (mRMS) curve within the time range of 0-480 ms for each sample are plotted and statistically compared. The stages are manually divided in combination with the sequence of transmission chain drive, spring energy storage, and contact switching. Alternatively, automatic segmentation can be performed using threshold detection.

[0099] Statistical results show that the contact switching and impact response are mainly concentrated in the range of approximately 70–265 ms. Before this, the vibration energy gradually increases from low to high within approximately 0–70 ms, which can be regarded as the transmission energy storage segment of the transmission chain meshing and spring energy storage. After approximately 265 ms, the vibration amplitude generally shows a decaying trend, corresponding to the mechanism return and the tail segment of the system's free decay. Based on the above statistical laws, the single gear shift analysis window [0, 480] is fixedly divided into three stages: the transmission energy storage segment corresponds to 0–70 ms, the main impact segment corresponds to 70–265 ms, and the tail segment corresponds to 265–480 ms. During online monitoring, the starting point of the analysis window is taken as the time origin, and the reference time range of the above three stages is converted into the corresponding sampling point number interval. The corresponding subsequence is directly extracted from the preprocessed vibration signal sequence to achieve consistency of stage division and cross-sample comparability in all gear shift samples. Among them, the stage division and root mean square curve of a single sample are shown in the figure. Figure 2 As shown.

[0100] S4. Based on the vibration response mechanisms of the three stages mentioned above, extract multiple statistical indicators that characterize the stage-specific features from each subsequence. First, calculate the mean square energy of each of the three subsequences. and root mean square value This is used to characterize the average energy level and overall vibration intensity at each stage; the vibration of the transmission energy storage section mainly reflects the weak amplitude disturbances generated by the meshing of the transmission chain and the energy storage of the springs, and the kurtosis of this section is further calculated. Spectral Entropy Kurtosis reflects the sharpness and impact of the vibration waveform, while spectral entropy reflects the complexity of the spectral energy distribution. The main impact segment corresponds to the concentrated impact caused by contact switching, and the root mean square value is used. The core strength indicator for this section is used to characterize the impact force of the contact and the stability of the action process; the tail section mainly reflects the mechanism's return and the system's free decay characteristics, and the energy proportion of the tail section is calculated. Used to assess the proportion of vibration decay process in the total energy.

[0101] Meanwhile, to characterize the energy distribution relationship between different stages throughout the gear shifting process, the energy ratio between the transmission section and the main impact section is constructed. This ratio will exhibit a stable deviation when transmission resistance increases, spring energy storage is abnormal, or contact impact weakens. Finally, after each gear shift, the above six core indicators are combined into a feature vector in a preset order. As a multidimensional feature representation describing the vibration behavior of a single gear shift.

[0102] S6. Perform statistical analysis on multiple gear shifting data, establish a multi-index baseline for normal operation, and construct a fingerprint database of typical mechanical fault characteristics. For several gear shifting tests under 5 typical operating conditions (normal, transmission jamming, contact wear, spring failure, and loose arc plate), extract the 6-dimensional feature vector of each gear shift according to a unified process. .

[0103] Taking 68 gear shifts under normal operating conditions as an example, the mean and standard deviation of each component were statistically analyzed to obtain the baseline of multiple indicators for normal operation, among which the root mean square of the transmission energy storage section was calculated. Approximately , cliff Approximately Spectral entropy Approximately The root mean square of the main impact segment Approximately Tail-end energy percentage Approximately Inter-segment energy ratio Approximately .

[0104] Based on this, the tolerance coefficient is taken. This establishes a reasonable fluctuation range under normal operating conditions. , , , , , This serves as the baseline range for normal operation.

[0105] Subsequently, feature vector sets were also extracted from various fault condition samples. And compared statistically with the normal baseline.

[0106] Taking the case of loose arc plate as an example, it can be observed in 68 samples that the root mean square of the transmission energy storage section is: Approximately The value was significantly higher than normal, and the root mean square value of the main impact segment was [missing information]. Approximately It also increased significantly, and the energy ratio between segments increased. Similarly, the p-value in the two-sample test was much smaller than 0.01, and the effect size... A value significantly greater than 0.8 indicates that the curved plate is in a loosening condition. The above-mentioned combined characteristics exhibit stable overall high levels. Based on the above statistical results, combined with discrimination indices such as p-value, effect size, and AUC, sensitive components with stable offset direction, significant offset amplitude, and good discrimination ability under a certain fault type are selected. These components, along with stage information, are solidified into a combined feature description of the fault type. For example, the loosening of the arc plate is defined as the transmission energy storage section. Significantly increased, main impact segment Significantly increased and An elevated fingerprint pattern. Maintaining a 6-dimensional feature vector. Under fixed conditions, the multi-index baseline obtained from the statistical analysis of normal operating samples is used as the benchmark, and the representative offset combination under each fault condition is used as the fingerprint entry to construct a typical mechanical fault feature fingerprint database as shown in Table 1 below.

[0107] Table 1: Fingerprint Database of Typical Mechanical Fault Features

[0108] Symbol explanation: 1) "Baseline" refers to the normal operating statistics and their tolerance range; 2) "↑" indicates an upward trend relative to the normal baseline mean, and "↑↑" indicates a significant increase; 3) "↓" indicates a decreasing trend relative to the normal baseline mean, and "≈" indicates a small or insignificant difference from the normal.

[0109] S6. The baseline range of multiple indicators for normal operation and the fingerprint database of typical mechanical fault characteristics are solidified and written into the parameter configuration file of the online monitoring device for real-time identification and diagnosis during subsequent operation.

[0110] After the transformer is put into operation, each time the OLTC receives a shift command and completes a shift action, it automatically extracts the 6-dimensional feature vector of this shift. .

[0111] First, using the mean and standard deviation of the normal operating sample, the standardized deviation of each component in this gear shift is calculated. For example: the normal mean is about 0.039, and the standard deviation is about 0.006. This time it is 0.053, which is about 0.014 higher than the mean, equivalent to about One standard deviation, i.e. Similarly, we can conclude that: , ,and The absolute values ​​of the standardized deviations all do not exceed 1. During this gear shift... , , The standardization bias is most significant, according to Calculate the overall deviation of this gear shift. In this embodiment, T1=1.0 and T2=2.0 are selected to determine the overall deviation. Exceeding the preset fault occurrence threshold Therefore, the fault detection layer determines that this gear shift is due to a mechanical abnormality.

[0112] After the fault occurrence layer confirms a mechanical abnormality in the current gear shift, the fault type identification layer calls the mechanical fault feature fingerprint database in the standardized deviation space. Pattern matching is performed on the feature vector of this gear shift. The fingerprint database records "loose curved plate" as the fault type. Its key indicator set Based on offline statistics and mechanism analysis, the loosening of the curved plate... The typical offset pattern in space can be described as follows: the standardized deviations of both the root mean square (RMS) of the transmission energy storage section and the root mean square (RMS) of the main impact section are significantly positive, i.e. and Generally greater than approximately 1.5; inter-segment energy ratio It is positive and not less than approximately 0.5; other indicators The standardized deviations are all less than 0.5. For this gear shift, the calculated values ​​are... , , All three fall within the pre-defined offset range for the loosening type of the curved plate, and The deviations are all near zero. Therefore, it can be concluded that, in terms of fault type... Key Indicator Set Among them, three indicators meet its typical offset pattern, namely Normalized matching degree When applying the same judgment to other fault types in the fingerprint database (such as transmission jamming, contact wear, spring failure, etc.), the standardized deviation of this gear shift on the corresponding key indicators either does not meet its typical offset direction, or the number of matching indicators is significantly less, making the matching degree of these fault types less accurate. All are below the preset type recognition threshold. And significantly smaller than Therefore, in the fault type identification layer, the fault type of the arc plate loosening is identified. This is the diagnostic output result for this gear shift.

[0113] Through the above two-level discrimination process, this embodiment can determine the current OLTC status based on multi-index fusion evaluation of vibration signals after the gear shifting action is completed. On the one hand, by comprehensively considering the deviation... The system enables online identification of faults and early deterioration relative to changes in the normal baseline. On the other hand, by matching the system with a mechanical fault feature fingerprint database within the standardized deviation space, it enables classification and diagnosis of typical mechanical fault types, thereby meeting the needs of online monitoring, evaluation, and fault early warning of transformer OLTC operating status.

[0114] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0115] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for monitoring mechanical faults in OLTC based on multi-index evaluation of vibration signals, characterized in that, Includes the following steps: S1. Collect vibration signal samples when the on-load tap changer of the transformer is changing gears during operation, and determine the analysis window for a single gear change. S2. Preprocess the collected vibration signal samples; S3. Based on the operating mechanism and vibration timing characteristics of the on-load tap changer, the vibration signal samples in the preprocessed single shift analysis window are segmented to obtain vibration sub-sequences at different stages. S4. Extract multiple statistical indicators that can characterize the stage characteristics from the vibration subsequences of each stage, and construct a multi-dimensional feature vector for a single gear shift. S5. Perform statistical analysis on the multi-dimensional feature vectors obtained from multiple gear shifting operations to construct a multi-indicator baseline for normal operation. Under experimental conditions, multiple typical mechanical fault conditions were introduced to construct a fingerprint database of typical mechanical fault features. S6. Compare and analyze the multi-dimensional feature vector obtained from the current gear shifting operation with the normal operation multi-index baseline and the typical mechanical fault feature fingerprint database to conduct online evaluation and graded diagnosis of the mechanical state of the on-load tap changer.

2. The OLTC mechanical fault monitoring method based on multi-index evaluation of vibration signals as described in claim 1, characterized in that, S1 includes: S11. When a shift command is detected or the start of the on-load tap changer shift action is detected by threshold triggering, the vibration signal acquisition of the current shift action is initiated. S12. Expand the acquisition time range before and after the actual switching moment of the contact, so that the acquisition time includes the transmission chain response range before the shift operation, the entire shift operation process, and the vibration attenuation range after the shift operation. Use the expanded acquisition time range as the single shift analysis window.

3. The OLTC mechanical fault monitoring method based on multi-index evaluation of vibration signals as described in claim 1, characterized in that, S2 include: S21. Analyze the vibration signal samples collected within the single gear shift analysis window according to the sampling period. Discretization is performed to obtain discrete vibration signals. , ,in, , The number of sampling points within a single gear shift analysis window; S22. The least squares method is used to analyze discrete vibration signals. Perform univariate linear least squares fitting to obtain the trend function. And calculate the trend term corresponding to each discrete vibration signal; S23. Remove the trend term point by point from the discrete vibration signal to obtain the detrended vibration sequence. ; S24. Based on the on-site electromagnetic environment and the noise level of the vibration signal, adopt appropriate filtering methods to process the de-stressed vibration sequence. Filtering is performed to obtain the preprocessed vibration sequence. .

4. The OLTC mechanical fault monitoring method based on multi-index evaluation of vibration signals as described in claim 1, characterized in that, S3 include: S31. Collect normal shift vibration signal samples under multiple normal shift operations; S32. Divide the single shift analysis window under each normal shift operation into the transmission energy storage section, the main impact section and the tail section; based on the average or specified quantile of the start and end times of the three stages of multiple normal shift vibration signal samples, obtain the unified reference start and end time interval of the three stages on the time axis. S33. Taking the starting point of the single shift analysis window of the current shift operation as the time origin, convert the reference start and end time intervals of each stage into the corresponding sampling point numbers, and extract the corresponding vibration subsequences of each stage from the preprocessed vibration signal samples.

5. The OLTC mechanical fault monitoring method based on multi-index evaluation of vibration signals as described in claim 4, characterized in that, S32 includes: S321. Perform preprocessing and sliding window analysis on the vibration signal samples within the single gear shift analysis window for each normal gear shift operation to obtain the short-time root mean square sequence and plot the short-time root mean square curve. S322. Calculate the mean and standard deviation of the short-time root mean square sequence for each normal gear shift. Based on the mean and standard deviation, adaptively set the entry threshold of the main impact segment. and exit threshold ; S323. Perform connectivity analysis on the short-time root mean square curve of each normal gear shift on the time axis. When there exists a certain integer... This ensures that the number of sliding windows is not less than the preset number. All of the following conditions are satisfied on the continuous sliding window At that time, the first The time corresponding to the left end of each sliding window Defined as the start time of the main impact phase of this gear shift, where Let f be the short-time root mean square within the nth sliding window, H be the sliding window step size, and f be the root mean square within the nth sliding window. s The sampling frequency; When there exists a certain integer This ensures that the number of sliding windows is not less than the preset number. All of the following conditions are satisfied on the continuous sliding window At that time, the first The time corresponding to the right end of each sliding window Defined as the end time of the main impact phase of this gear shift; S324. Based on the duration of the single gear shift analysis window and the start and end times of the main impact segment, divide the start and end times of the three stages in this normal gear shift operation. S325. For all normal gear shifting operations, calculate the mean or specified quantile of the start and end times of the three stages to obtain a unified reference start and end time interval for the three stages on the time axis. , , .

6. The OLTC mechanical fault monitoring method based on multi-index evaluation of vibration signals as described in claim 5, characterized in that, Entry threshold of the main impact segment and exit threshold The calculation method is as follows: in, This represents the root mean square value of the sequence under normal shifting conditions. This represents the root mean square standard deviation of the sequence under normal shifting conditions. This is an empirical coefficient.

7. The OLTC mechanical fault monitoring method based on multi-index evaluation of vibration signals as described in claim 1, characterized in that, In S4, the process of extracting the multidimensional feature vector of a single gear shift includes: In the transmission and energy storage section, amplitude-based statistics, higher-order statistics, and entropy-based indices are selected to describe the vibration amplitude level, waveform sharpness, and the complexity of the spectral energy distribution. In the main impact phase, energy or amplitude statistics are extracted as features to describe the impact intensity and the stability of the contact action process. In the tail section, the proportion of energy in the tail section to the total energy of the three sections is calculated to assess whether the vibration decay process is normal. During the entire gear shifting process, the energy ratio between gear shifts is calculated to characterize the relative distribution pattern of energy at each stage.

8. The OLTC mechanical fault monitoring method based on multi-index evaluation of vibration signals as described in claim 1, characterized in that, S5 include: S51. Calculate the mean and standard deviation of each feature component in the multiple sets of multidimensional feature vectors obtained under multiple normal gear shifting operations, and establish a fluctuation range for each feature component based on the mean and standard deviation as the baseline range for multiple indicators during normal operation. S52. Introduce multiple typical mechanical fault conditions under experimental conditions, including at least transmission jamming, contact wear, spring failure and arc plate loosening. Collect vibration signals of gear shifting under the corresponding fault conditions, and obtain multi-dimensional feature vectors of various fault types according to the same processing procedure as normal conditions. S53. For any fault type, calculate the mean and standard deviation of each characteristic component under multiple experiments for that fault type. S54. Compare the mean and standard deviation of each characteristic component in each fault type with the mean, standard deviation and baseline interval of the corresponding characteristic component under normal shifting operation to characterize the offset direction and offset magnitude of each characteristic component under fault state relative to normal state. S55. Solidify the characteristic vector offsets that are representative of the baseline interval of the normal state under the fault state into the characteristic fingerprints of the corresponding mechanical faults, and form a mechanical fault characteristic fingerprint library.

9. The OLTC mechanical fault monitoring method based on multi-index evaluation of vibration signals as described in claim 8, characterized in that, S6 include: S61. Using the mean and standard deviation of each characteristic component under normal shifting operation, calculate the standardized deviation of each characteristic component under the current shifting operation, and take the maximum value of the absolute value of the standardized deviation of each characteristic component as the multi-index fusion evaluation metric. ; S62, Set two thresholds Based on the evaluation results of multiple consecutive gear shifts, a classification is made: when If this condition is met in the most recent gear shifts, the current gear shift is determined to be within the normal fluctuation range. when Or, if only a slight over-limit occurs during a single gear shift, the shift will be marked as a deteriorating condition or a warning event; when If a stable over-limit occurs during multiple consecutive gear shifts, a mechanical fault or abnormality is determined, and an alarm is triggered. S63. When a mechanical fault or abnormality is determined, the mechanical fault feature fingerprint database is called to perform fault type matching on the multi-dimensional feature vector of this gear shift and output a diagnosis.

10. The OLTC mechanical fault monitoring method based on multi-index evaluation of vibration signals as described in claim 9, characterized in that, S63 includes: S631. Analyze the components of the multidimensional feature vector obtained from the current gear shift operation. Calculate the standardized deviation The standardized deviations of each component are combined into a standardized deviation vector. ; S632, For each candidate fault type and its set of key indicators Check each gear shift's standardized deviation across all indicators to see if it falls within the pre-defined criterion range for that fault type. ; like fall into Internally, it is believed that the current gear shift is in line with the indicator. Above and fault type The offset pattern must match the typical offset pattern; otherwise, it is considered a mismatch. S633, For each fault type The number of key indicators that meet the offset criterion interval is counted, and the normalized matching degree is calculated. Matching degree This is used to reflect the number of gear shifts currently occurring, and which key indicators are related to the type of fault. The typical offset combination pattern is consistent with this; S634. After obtaining the matching degree set of all fault types... Then, set the type recognition threshold. The fault type results will be output according to the following principles: If a certain type of fault exists Its matching degree The largest among all candidate fault types and not lower than the threshold Then the fault type This is confirmed as the diagnostic output result for this gear shift. If the matching degree of all fault types is lower than If the difference between the matching degrees of multiple fault types is less than the preset threshold, the current shift will be marked as "abnormal and unclassified" or "awaiting manual review".

Images