A preventive maintenance method and device for transformer winding faults

Abstract

The present application relates to the technical field of transformer winding fault detection, and particularly relates to a preventive maintenance method and device for transformer winding faults. The method comprises: obtaining current signals and vibration acceleration signals in different directions of a three-phase winding of a transformer; obtaining an excitation-response spectrum matrix based on the current signals and the vibration acceleration signals, and constructing a health state vibration mode base; projecting the excitation-response spectrum matrix onto the health state vibration mode base to obtain a vibration mode contribution coefficient vector; constructing a frequency-mode correlation graph structure according to the similar situation of different frequencies in the health state vibration mode base at the initial stage of equipment operation; giving characteristics to the nodes of the graph structure to obtain a winding structure state characteristic vector, and establishing a reference database of each temperature interval; matching the winding structure state characteristic vector and the vibration mode contribution coefficient vector in the current period with the reference database to determine the fault cause. The present application improves the timeliness of transformer winding fault early warning.

Description

Technical Field

[0001] This invention relates to the field of transformer winding fault detection technology, and specifically to a preventive maintenance method and device for transformer winding faults. Background Technology

[0002] As a core hub of the power grid, the safe and stable operation of power transformers is crucial. In new energy power generation scenarios, the widespread application of power electronic devices such as inverters injects wide-spectrum, time-varying harmonic currents into the grid. According to the law of electromagnetic force, these harmonic currents generate complex electromagnetic excitation forces within the transformer windings. Simultaneously, the transformer winding itself is a complex elastic mechanical structure composed of conductors, insulating paper, spacers, and clamps, possessing inherent multiple mechanical vibration modes that slowly change with temperature and its own mechanical health (such as preload and insulation aging). When an electromagnetic excitation force of a certain frequency coincides with or approaches a certain mechanical vibration mode frequency of the winding, mechanical resonance will occur, amplifying the winding vibration and accelerating inter-turn insulation wear, fastener loosening, and structural fatigue, ultimately potentially leading to winding instability and deformation.

[0003] Existing winding deformation diagnosis primarily relies on offline frequency response analysis (FRA) during power outages. This method cannot achieve online monitoring and early warning. Conventional online vibration monitoring methods typically only issue alarms based on fixed vibration amplitudes. This approach suffers from lag in risk identification and cannot effectively distinguish whether increased vibration is caused by external power grid disturbances (enhanced excitation source) or by deterioration of the winding's own mechanical properties (increased structural vulnerability). This lack of diagnostic capability prevents the provision of a clear cause-based decision-making basis for preventative maintenance, leaving maintenance work at a stage of passive response or extensive periodic maintenance. Summary of the Invention

[0004] To address the problem of insufficient timeliness in providing early warnings of transformer winding faults using existing methods, the present invention aims to provide a preventative maintenance method and apparatus for transformer winding faults. The specific technical solution adopted is as follows:

[0005] In a first aspect, the present invention provides a preventive maintenance method for transformer winding faults, the method comprising the following steps:

[0006] Acquire the current signal and vibration acceleration signal in different directions of the three-phase winding of the transformer;

[0007] The excitation-response spectrum matrix of each time window is constructed based on the current signal and vibration acceleration signal of each time window; combined with all excitation-response spectrum matrices in the early stage of equipment commissioning, a healthy state vibration mode basis is constructed; the excitation-response spectrum matrix of each time window is projected onto the healthy state vibration mode basis to obtain the corresponding vibration mode contribution coefficient vector.

[0008] Based on the similarity of different frequencies in the healthy state vibration mode base during the initial operation of the equipment, a frequency-mode correlation graph structure is constructed; features are assigned to the nodes of the graph structure, and the winding structure state feature vector is obtained by combining the graph convolutional network; based on the winding structure state feature vectors, vibration mode contribution coefficient vectors and oil temperature during the initial operation of the equipment, a benchmark database for each temperature range is established.

[0009] The cause of the fault is determined by the deviation between the winding structure state feature vector in the current time period and the winding structure state feature vector in the corresponding reference database, as well as the deviation between the vibration mode contribution coefficient vector and the vibration mode contribution coefficient vector in the corresponding reference database.

[0010] Preferably, the construction of the excitation-response spectrum matrix for each time window based on the current signal and vibration acceleration signal for each time window includes:

[0011] The phase current signal with the largest effective value within each time window is recorded as the target signal; the electromagnetic excitation force spectrum vector is obtained by performing a fast Fourier transform on the square of the target signal.

[0012] The vibration response spectrum vector is obtained by performing fast Fourier transform on each vibration acceleration signal within the time window;

[0013] Based on the electromagnetic excitation force spectrum vector and vibration response spectrum vector corresponding to the same time window, an excitation-response spectrum matrix corresponding to the time window is constructed; the first column of the excitation-response spectrum matrix is ​​composed of electromagnetic excitation force spectrum vectors, and the second to seventh columns are composed of vibration response spectrum vectors.

[0014] Preferably, the construction of a healthy-state vibration mode basis by combining all excitation-response spectrum matrices during the initial operation of the equipment includes:

[0015] The elements in each column of the excitation-response spectrum matrix are standardized to obtain the standardized excitation-response spectrum matrix.

[0016] All standardized excitation-response spectrum matrices were expanded and stitched together column-wise during the initial operation of the equipment to form... A long vector of dimension, where The number of frequency points calculated for the Fast Fourier Transform;

[0017] Obtain the covariance matrix of all long vectors, and perform eigenvalue decomposition on the covariance matrix to obtain the column vectors of the eigenvector matrix;

[0018] A predetermined number of main eigenvectors are selected from the column vectors of the eigenvector matrix to form the basis of the vibration mode of the healthy state.

[0019] Preferably, the step of projecting the excitation-response spectrum matrix of each time window onto the healthy state vibration mode basis to obtain the corresponding vibration mode contribution coefficient vector includes:

[0020] The standardized excitation-response spectrum matrix of each time window is expanded into a long vector. The transpose of the healthy state vibration mode basis is multiplied by the long vector to obtain the vibration mode contribution coefficient vector.

[0021] Preferably, the step of constructing a frequency-mode correlation diagram structure based on the similarity of different frequencies in the healthy state vibration mode base during the initial stage of equipment commissioning includes:

[0022] By using frequencies as nodes in the graph structure and the similarity of the row vectors of each channel in the healthy vibration mode basis corresponding to two frequencies as the edge weights between nodes, a frequency-mode association graph structure is constructed.

[0023] Preferably, the step of assigning features to the nodes of the graph structure and obtaining the winding structure state feature vector by combining the graph convolutional network includes:

[0024] All nodes of the graph structure and their composite feature vectors are input into the graph convolutional network encoder, and the encoder outputs the winding structure state feature vector.

[0025] The composite feature vector is composed of the row vector corresponding to the node in the normalized excitation-response spectrum matrix and the vibration mode contribution coefficient vector.

[0026] Preferably, determining the cause of the fault based on the deviation between the winding structure state feature vector in the current time period and the winding structure state feature vector in the corresponding reference database, and the deviation between the vibration mode contribution coefficient vector and the vibration mode contribution coefficient vector in the corresponding reference database, includes:

[0027] Calculate the first deviation vector between the winding structure state feature vector of each time window and its corresponding first average vector; the first average vector is the average value of all winding structure state feature vectors in the reference database corresponding to the oil temperature of each time window.

[0028] The overall degradation rate of the winding structure is obtained based on the changes in the first deviation vector corresponding to all time windows within the current period.

[0029] If the overall degradation rate of the winding structure is greater than the preset degradation rate threshold, then the second deviation vector between the current vibration mode contribution coefficient vector and its corresponding second average vector is calculated; the second average vector is the average value of all vibration mode contribution coefficient vectors in the reference database corresponding to the oil temperature of all time windows in the current period.

[0030] The cause of the fault is determined based on the second deviation vector.

[0031] Preferably, the step of obtaining the overall degradation rate of the winding structure based on the changes in the first deviation vector corresponding to all time windows within the current time period includes: taking the magnitude of the rate of change vector of the first deviation vector corresponding to all time windows within the current time period as the overall degradation rate of the winding structure.

[0032] Preferably, determining the cause of the fault based on the second deviation vector includes: matching the second deviation vector with a pre-calibrated fault feature library to obtain the cause of the fault.

[0033] Secondly, the present invention provides a preventive maintenance device for transformer winding faults, the device being used to perform the above-described method, the device comprising:

[0034] The signal acquisition module is used to acquire the current signal of the three-phase winding of the transformer and the vibration acceleration signal in different directions;

[0035] The vector construction module is used to construct the excitation-response spectrum matrix of each time window based on the current signal and vibration acceleration signal of each time window; combine all excitation-response spectrum matrices in the early stage of equipment commissioning to construct the healthy state vibration mode basis; project the excitation-response spectrum matrix of each time window onto the healthy state vibration mode basis to obtain the corresponding vibration mode contribution coefficient vector.

[0036] The benchmark database construction module is used to construct a frequency-mode correlation graph structure based on the similarity of different frequencies in the healthy state vibration mode base during the initial operation of the equipment; assign features to the nodes of the graph structure and obtain the winding structure state feature vector by combining the graph convolutional network; and establish a benchmark database for each temperature range based on the winding structure state feature vector, vibration mode contribution coefficient vector and oil temperature during the initial operation of the equipment.

[0037] The fault evaluation module is used to determine the cause of the fault based on the deviation between the winding structure state feature vector in the current time period and the winding structure state feature vector in the corresponding reference database, as well as the deviation between the vibration mode contribution coefficient vector and the corresponding vibration mode contribution coefficient vector.

[0038] The present invention has at least the following beneficial effects:

[0039] This invention first constructs the corresponding excitation-response spectrum matrix using the acquired current signal and vibration acceleration signal. Then, it obtains the healthy-state vibration mode basis by combining all excitation-response spectrum matrices during the initial stage of equipment commissioning. The real-time excitation-response spectrum matrix is ​​projected onto the healthy-state vibration mode basis to obtain the vibration mode contribution coefficient vector. This projection operation is a specific mathematical transformation designed to decouple excitation and response, decomposing a complex high-dimensional spectrum matrix that mixes external operating conditions and internal structural states into a low-dimensional vector. This quantifies how many components of the corresponding basic healthy modes are included in the current overall vibration pattern. The numerical change is mainly dominated by the change in the mechanical transfer function of the winding itself, thus largely independent of the fluctuation of the input excitation force amplitude. This invention is based on the healthy-state vibration modes at different frequencies during the initial stage of equipment commissioning. Based on similar cases in the formula base, a frequency-mode correlation graph structure was constructed. Features were assigned to the nodes of the graph structure to obtain the winding structure state feature vector. A benchmark database indexed by temperature was also constructed. Based on the deviation between the winding structure state feature vector and the vibration mode contribution coefficient vector in the current time period and the vector in the corresponding benchmark database, it was determined whether a fault existed and the cause of the fault was identified. The method provided by this invention solves the technical problem that traditional vibration monitoring cannot avoid false alarms due to increased load harmonic current or missed structural deterioration signals due to decreased load when facing changing operating conditions. It realizes early identification of winding mechanical faults. Its identification time is earlier than conventional vibration amplitude alarm. It can sensitively capture the abnormal distribution of energy in this mode and improve the accuracy and timeliness of transformer winding fault early warning. Attached Figure Description

[0040] To more clearly illustrate the technical solutions and advantages 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 some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 A flowchart illustrating a preventative maintenance method for transformer winding faults provided in an embodiment of the present invention;

[0042] Figure 2 This is a structural block diagram of a preventive maintenance device for transformer winding faults provided in an embodiment of the present invention. Detailed Implementation

[0043] To further illustrate the technical means and effects adopted by the present invention to achieve the intended purpose, the following detailed description of a preventive maintenance method and apparatus for transformer winding faults based on the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0044] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0045] The following description, in conjunction with the accompanying drawings, details a specific scheme for a preventive maintenance method and device for transformer winding faults provided by the present invention.

[0046] An example of a preventive maintenance method for transformer winding faults:

[0047] The specific scenario addressed in this embodiment is as follows: When conducting preventative detection of transformer winding faults, since the vibration of the transformer winding is a direct response to the electromagnetic force it experiences, there is a definite causal and temporal relationship between the two. Any time asynchrony will lead to a mismatch in the causal chain, making it impossible to accurately identify which current component caused which vibration anomaly. Therefore, a multi-channel synchronous acquisition system with a unified clock source is required to ensure that the state between current and vibration is captured at the same instant. Thus, this embodiment will acquire the current signal of the three-phase windings of the transformer and the vibration acceleration signal in different directions. By analyzing these two signals, it will determine whether there is a fault in the transformer winding. If a fault is found, the cause of the fault will be identified, and corresponding maintenance measures will be taken in a timely manner.

[0048] This embodiment proposes a preventive maintenance method for transformer winding faults, such as... Figure 1 As shown, a preventive maintenance method for transformer winding faults in this embodiment includes the following steps:

[0049] Step S1: Obtain the current signal of the three-phase winding of the transformer and the vibration acceleration signal in different directions.

[0050] First, at the high-voltage side bushing of the transformer, the three-phase winding current signal is collected using the existing current transformer. , and Meanwhile, three triaxial accelerometers are securely installed on the outer side of the transformer tank wall, corresponding to the three key positions of the upper, middle, and lower internal windings, to acquire vibration acceleration signals in six directions: upper radial, upper axial, middle radial, middle axial, lower radial, and lower axial.

[0051] A multi-channel synchronous data acquisition card was used to synchronously acquire data from all channels at a sampling frequency of 20 kHz. This sampling rate was chosen to ensure coverage of the harmonic electromagnetic force components generated by high-order harmonic currents (e.g., up to 2.5 kHz). The acquisition system segmented the continuous data stream into time windows of τ = 1 second, generating a time-aligned set of synchronous time-domain signal segments for each analysis period. This set included three current signals and six vibration acceleration signals within the time window.

[0052] Thus, by using the above method, it is possible to obtain the current signal of the three-phase winding of the transformer and the vibration acceleration signal in different directions within each time window.

[0053] Step S2: Construct the excitation-response spectrum matrix for each time window based on the current signal and vibration acceleration signal for each time window; combine all excitation-response spectrum matrices during the initial operation of the equipment to construct a healthy state vibration mode basis; project the excitation-response spectrum matrix of each time window onto the healthy state vibration mode basis to obtain the corresponding vibration mode contribution coefficient vector.

[0054] Because the electromagnetic force borne by the winding and the vibration it generates are crucial information such as the magnitude of the harmonic excitation force and the location of the mechanical resonance peak are more clearly defined in the frequency domain than in the time domain waveform. Furthermore, according to the electromagnetic law, the main electromagnetic force generated by the harmonic current with frequency f has a frequency of 2f. Therefore, it is necessary to convert the time-domain signal to the frequency domain and correctly characterize the electromagnetic excitation force by squaring the current signal, ultimately integrating all signal information into a unified matrix.

[0055] First, from the set of signal segments within each time window, select the phase current signal segment with the largest effective value (RMS) and denote it as the target signal, denoted as: Calculate its squared signal. The relationship with electromagnetic force is represented by a formula; then, a fast Fourier transform is applied to it to obtain the electromagnetic excitation force spectrum that directly represents the magnitude of the electromagnetic excitation force at each frequency, and this spectrum is used as a column vector, thus obtaining the electromagnetic excitation force spectrum vector.

[0056] Next, the vibration acceleration signal is processed by response spectroscopy. Specifically, the six vibration acceleration signal segments in the set are processed one by one by Fast Fourier Transform (FFT) to obtain six vibration response spectra that characterize the actual vibration state of the winding at different spatial positions and in different directions. These are used as column vectors, that is, six vibration response spectrum vectors are obtained.

[0057] Furthermore, a matrix is ​​constructed based on the electromagnetic excitation force spectrum vector and vibration response spectrum vector obtained above. Specifically, the electromagnetic excitation force spectrum vector is used as the first column of the excitation-response spectrum matrix, and the subsequent six columns of the excitation-response spectrum matrix are the six vibration response spectrum vectors, thus obtaining the excitation-response spectrum matrix with dimensions of [missing information]. ,in, The number of frequency points calculated for FFT. Each row of this matrix corresponds to a specific frequency, and fully records the excitation input of the system at that frequency and the response output of all measurement points.

[0058] In this embodiment, the excitation-response spectrum matrix is ​​obtained by combining instantaneous operating conditions and structural health information using the above method. Then, through a progressive decoupling and projection analysis, low-dimensional features that are stable, unique, and reflect the integrity of the internal mechanical connection and support structure of the winding are extracted.

[0059] Considering that the physical units and numerical ranges of the data from different measurement channels (such as electromagnetic excitation force spectrum and vibration response spectrum) in the excitation-response spectrum matrix differ greatly, if processed directly, the electromagnetic excitation force spectrum with larger values ​​will dominate the subsequent statistical analysis results, thereby masking the key information that exists in the vibration response spectrum with smaller values ​​caused by minor structural changes.

[0060] Therefore, the excitation-response spectrum matrix is ​​first standardized to eliminate dimensional and scale differences between measurement channels. Specifically, for each column j of all initially acquired excitation-response spectrum matrices, its mean is calculated. and standard deviation And store it. During online monitoring, for each element of the current excitation-response spectrum matrix, perform the following transformation:

[0061]

[0062] in, This represents the standardized data of the element in the i-th row and j-th column of the excitation-response spectrum matrix. This represents the element in the i-th row and j-th column of the excitation-response spectrum matrix. This indicates the preset zero-prevention parameter.

[0063] When standardizing the excitation-response spectrum matrix, a preset zero-prevention parameter is introduced into the denominator of the calculation formula to prevent the denominator from being 0. In this embodiment, the preset zero-prevention parameter is 0.001. In specific applications, the implementer can set it according to the specific situation.

[0064] Using the above method, the obtained excitation-response spectrum matrix is ​​standardized to obtain a standardized excitation-response spectrum matrix. All elements in the standardized excitation-response spectrum matrix are at similar numerical scales, which lays the foundation for subsequent accurate mode decomposition.

[0065] The windings of a specific transformer are mechanical structures determined by their design, materials, and assembly processes, resulting in a set of inherent, independent vibration modes in a healthy state. Any complex vibration observed at any given time can be considered a linear superposition of these fundamental modes. Early degradation of the windings (such as hardening of spacers and loosening of clamps) manifests itself by altering the morphology of these modes or the energy distribution among them. Therefore, by statistically analyzing a large amount of health data collected during the initial commissioning phase, this set of constant vibration modes can be extracted, thereby constructing a reference coordinate system describing all mechanical behaviors of the equipment. This step is performed once during the initial commissioning phase, the duration of which is set by the implementer based on specific circumstances.

[0066] Specifically, all standardized excitation-response spectrum matrices during the initial operation of the equipment are expanded and spliced ​​column-wise to form a... We calculate the covariance matrix of these long vector sets, which quantifies the stable cooperative vibration relationships between all frequency points and sensor locations during healthy operation of the transformer coil. Eigenvalue decomposition is performed on this covariance matrix to obtain each column vector of the eigenvector matrix, which accurately depicts the distribution of fundamental, independent vibration modes across all frequencies and sensors. The top d principal eigenvectors that explain more than 99% of the total variance are selected to form a covariance matrix. A dimensional matrix, once established, is fixed and becomes the basis of the transformer's life-long, unchanging healthy-state vibration modes; where, The preset quantity can be set by the implementer according to the specific circumstances in actual application.

[0067] Thus, this embodiment has constructed a healthy state vibration mode basis.

[0068] Since structural degradation of the winding typically does not generate new vibrational modes, but rather alters the energy distribution among existing healthy vibrational mode bases—for example, causing an anomalous activation of a higher-order bending mode—it is possible to accurately assess the contribution intensity of each fundamental mode by projecting the real-time monitored vibrational state onto an established fixed mode base. This transforms the fuzzy overall vibrational changes into clear, quantifiable mode energy transfer indices.

[0069] Specifically, the standardized excitation-response spectrum matrix monitored in real time is expanded into a long vector, and the transpose of the healthy state vibration mode basis is multiplied by the long vector to obtain the corresponding vibration mode contribution coefficient vector, which can be expressed as follows:

[0070]

[0071] in, This represents the vector of vibration mode contribution coefficients. This represents the vibrational mode base of a healthy state. Indicates transpose. This represents a long vector.

[0072] The vibration mode contribution coefficient vector is a d-dimensional column vector. The elements in this vector precisely quantify the intensity of healthy mode components in the overall vibration observed at the corresponding time. This vector is a highly condensed and physically meaningful intermediate feature of the current dynamic behavior of the winding.

[0073] Step S3: Based on the similarity of different frequencies in the healthy state vibration mode base during the initial operation of the equipment, construct a frequency-mode correlation graph structure; assign features to the nodes of the graph structure, and obtain the winding structure state feature vector by combining the graph convolutional network; based on the winding structure state feature vector, vibration mode contribution coefficient vector and oil temperature during the initial operation of the equipment, establish a benchmark database for each temperature range.

[0074] Because the fault mechanism of windings often manifests as an abnormal correlation between a specific electromagnetic excitation force (local input) and a specific mechanical vibration mode (global output). For example, an excitation force at a certain frequency may normally primarily excite radial vibration modes, but after the pad has aged, it may abnormally transfer energy to the axial bending mode. The vibration mode contribution coefficient vector reflects the global state, while the normalized excitation-response spectrum matrix reflects local details. Therefore, a model that can effectively integrate the two and capture the abnormal correlation between them is needed.

[0075] First, a static frequency-mode correlation graph structure is constructed based on the healthy-state vibration mode basis. The node set of this graph structure is all There are several frequency points. The edge weight between two nodes is determined by the intensity of their cooperative vibration in all healthy modes. Specifically, the similarity of the row vectors of each channel in the healthy vibration mode basis for the corresponding frequencies of the two nodes is calculated, and this similarity is used as the edge weight between the nodes. There are many methods for calculating the similarity between vectors, such as using cosine similarity, which will not be elaborated on here. This graph structure represents the topology of the inherent mechanical transmission network of a healthy winding, which remains fixed after being constructed once.

[0076] Then, within each monitoring time, i.e., the time window of the current period, a dynamic composite feature vector is assigned to each node of the frequency-mode correlation graph structure constructed above. This composite feature vector is composed of two parts: the first part is the row vector corresponding to the node in the normalized excitation-response spectrum matrix, which contains the local response information of all channels at that frequency; the second part is the complete vibration mode contribution coefficient vector, which contains the global mode state information of the system.

[0077] Next, all nodes and their composite feature vectors from this graph structure, which carries rich dynamic features, are input into a pre-trained Graph Convolutional Network (GCN) encoder. The GCN learns a global representation of the current system state by propagating and aggregating node features across the graph structure. The encoder's final output is a low-dimensional (e.g., 16-dimensional) vector, defined as the final product of this step, i.e., the winding structure state feature vector. Changes in this winding structure state feature vector directly reflect whether the healthy relationship between the local response and global mode participation of the winding has been disrupted, thus providing a solid and reliable input for subsequent trend analysis and fault attribution.

[0078] Because the mechanical properties of transformer windings (such as the elastic modulus of the insulation material) undergo reversible changes with operating temperature, this change also leads to reversible and harmless shifts in the winding structural state characteristic vector. If not distinguished, this can be confused with irreversible shifts caused by structural degradation, resulting in misjudgments. Therefore, it is necessary to establish a benchmark database indexed by temperature for comparison under the same temperature conditions. Specifically, in the initial stage of equipment commissioning, the system stores all accumulated winding structural state characteristic vectors, their corresponding vibration mode contribution coefficient vectors, and oil temperatures in temperature bins at 1°C intervals, meaning each temperature interval is 1°C in length. Using this method, multiple benchmark databases are constructed, with one database corresponding to each temperature interval. Subsequent information matching is performed using temperature as the index for searching.

[0079] Step S4: Determine the cause of the fault based on the deviation between the winding structure state feature vector in the current time period and the winding structure state feature vector in the corresponding reference database, as well as the deviation between the vibration mode contribution coefficient vector and the vibration mode contribution coefficient vector in the corresponding reference database.

[0080] Next, a long-term, multi-dimensional trend analysis will be conducted on the continuously generated winding structure state feature vector and its intermediate characteristic vibration mode contribution coefficient vector with clear physical meaning within the current time period. By quantifying the changes in the winding structure state feature vector and identifying the specific vibration mode anomalies that cause these changes, early warning, accurate attribution, and intelligent decision-making for winding structure degradation can be achieved, and finally, specific preventive maintenance work orders can be output.

[0081] The winding structure state characteristic vector is a low-dimensional representation of the overall mechanical state of the winding, and its deviation from the health baseline directly reflects changes in structural integrity. Therefore, by tracking the long-term rate of change of this vector, the speed of structural degradation can be quantified, providing a basis for judging the urgency of maintenance.

[0082] Specifically, for each newly generated winding structure state feature vector, that is, the winding structure state feature vector of each time window in the current period, the corresponding reference database is first found according to its temperature (oil temperature). For any time window, the average value of all winding structure state feature vectors in the reference database corresponding to the oil temperature of the time window is recorded as the first average vector of the time window. The difference between the winding structure state feature vector of the time window and its corresponding first average vector is calculated, and the difference is used as the first deviation vector corresponding to the time window.

[0083] Obtain the vector formed by the rate of change of the first deviation vector corresponding to all time windows within the current period, and denote this vector as the rate of change vector. The magnitude of the rate of change vector is used as the overall degradation rate of the winding structure. The overall degradation rate of the winding structure is a key indicator for triggering subsequent diagnostic analysis. In this embodiment, the current period is a relatively long cycle, with a quarter as the duration. In specific applications, the implementer can set this according to specific circumstances.

[0084] If the overall degradation rate of the winding structure exceeds a preset degradation rate threshold, it indicates significant and persistent structural degradation, triggering an early warning. The vibration mode contribution coefficient vector is then used to investigate the root cause of this degradation. The preset degradation rate threshold is set according to specific circumstances and will not be elaborated here. Different mechanical faults, such as a decrease in overall clamping force and local aging of pads, can cause energy transfer with specific patterns between healthy vibration mode bases. Therefore, by analyzing the deviation of the vibration mode contribution coefficient vector and matching it with preset fault characteristics, fault mode identification can be achieved.

[0085] Specifically, when an early warning is triggered, the average value of the contribution coefficient vectors of all vibration modes in the baseline database corresponding to the oil temperature of all time windows within the current period is calculated, and this average value is recorded as the second average vector. The difference between the current vibration mode contribution coefficient vector and the second average vector is calculated, and this difference is recorded as the second deviation vector. The elements in the second deviation vector can reflect whether the energy participation of each basic vibration mode is higher or lower than that of the healthy state.

[0086] The second deviation vector is matched with a pre-calibrated fault feature library. This library is established through high-precision simulation or by referencing historical fault data from similar equipment, and stores the second deviation vectors corresponding to various typical winding faults (e.g., fault 1: clamping force decreases by 5%; fault 2: elastic modulus of a certain area of ​​the pad decreases by 20%). The matching process is completed by calculating the cosine similarity between the obtained second deviation vector and each fault feature in the fault feature library. The fault mode with the highest cosine similarity is determined as the current fault cause.

[0087] Thus, using the method provided in this embodiment, preventative maintenance of transformer winding faults has been completed. When a fault is suspected, the cause of the fault is obtained, and the system automatically generates a structured preventative maintenance work order, converting diagnostic information into executable maintenance instructions. This work order includes the following:

[0088] (1) Maintenance warning level: It is set according to the overall degradation rate of the winding structure, for example, it is divided into three levels: "attention", "warning" and "emergency".

[0089] (2) Diagnostic conclusion: Show the cause of the fault, for example: the overall preload of the winding has decreased.

[0090] (3) Diagnostic basis: Provide quantitative data support and directly cite the key changes in the pattern contribution coefficient deviation vector. For example: "The basis is that the radial breathing pattern (pattern 1) contribution coefficient has a significant negative deviation (-0.75), while the higher-order bending pattern (pattern 4, pattern 6) contribution coefficient has a positive deviation (+0.42, +0.58). This feature is highly consistent with the feature library of pretension decrease (cosine similarity 0.92)."

[0091] (4) Recommended maintenance measures: Provide specific maintenance items based on the diagnostic conclusions. For example, for the conclusion of "overall preload reduction", the recommended measures are: "It is recommended to carry out body inspection and torque verification and tightening of all fastening bolts during the next planned maintenance window."

[0092] The maintenance work order provided by the above method in this embodiment can assist staff in repairing faults in a timely and rapid manner.

[0093] This embodiment first constructs the corresponding excitation-response spectrum matrix using the acquired current signal and vibration acceleration signal. Then, it obtains the healthy-state vibration mode basis by combining all excitation-response spectrum matrices during the initial operation of the equipment. The real-time excitation-response spectrum matrix is ​​projected onto the healthy-state vibration mode basis to obtain the vibration mode contribution coefficient vector. This projection operation is a specific mathematical transformation designed to decouple excitation and response, decomposing a complex high-dimensional spectrum matrix that mixes external operating conditions and internal structural states into a low-dimensional vector. This quantifies how many components of the corresponding basic healthy modes are included in the current overall vibration pattern. The numerical change is mainly dominated by the change in the mechanical transfer function of the winding itself, thus being largely independent of the fluctuation of the input excitation force amplitude. This embodiment is based on the healthy-state vibration at different frequencies during the initial operation of the equipment. Based on similar cases in the mode base, a frequency-mode correlation graph structure was constructed. Features were assigned to the nodes of the graph structure to obtain the winding structure state feature vector. A reference database indexed by temperature was also constructed. Based on the deviation between the winding structure state feature vector and the vibration mode contribution coefficient vector in the current time period and the vector in the corresponding reference database, it was determined whether a fault existed and the cause of the fault was identified. The method provided in this embodiment solves the technical problem that traditional vibration monitoring cannot avoid false alarms due to increased load harmonic current or missed structural deterioration signals due to decreased load when facing changing operating conditions. It realizes early identification of winding mechanical faults. Its identification time is earlier than conventional vibration amplitude alarm. It can sensitively capture the abnormal distribution of this mode energy and improve the accuracy and timeliness of transformer winding fault early warning.

[0094] An embodiment of a preventive maintenance device for transformer winding faults:

[0095] See Figure 2 The diagram illustrates a structural block diagram of a preventive maintenance device for transformer winding faults provided in an embodiment of the present invention. The device may include a signal acquisition module, a vector construction module, a reference database construction module, and a fault evaluation module.

[0096] The signal acquisition module is used to acquire the current signal of the three-phase winding of the transformer and the vibration acceleration signal in different directions.

[0097] The vector construction module is used to construct the excitation-response spectrum matrix of each time window based on the current signal and vibration acceleration signal of each time window; combine all excitation-response spectrum matrices in the early stage of equipment commissioning to construct the healthy state vibration mode basis; project the excitation-response spectrum matrix of each time window onto the healthy state vibration mode basis to obtain the corresponding vibration mode contribution coefficient vector.

[0098] The benchmark database construction module is used to construct a frequency-mode correlation graph structure based on the similarity of different frequencies in the healthy state vibration mode base during the initial operation of the equipment; assign features to the nodes of the graph structure and obtain the winding structure state feature vector by combining the graph convolutional network; and establish a benchmark database for each temperature range based on the winding structure state feature vector, vibration mode contribution coefficient vector and oil temperature during the initial operation of the equipment.

[0099] The fault evaluation module is used to determine the cause of the fault based on the deviation between the winding structure state feature vector in the current time period and the winding structure state feature vector in the corresponding reference database, as well as the deviation between the vibration mode contribution coefficient vector and the corresponding vibration mode contribution coefficient vector.

[0100] It should be understood that Figure 2 The structural block diagram and modules of the preventive maintenance device for transformer winding faults shown can be implemented in various ways. For example, in some embodiments, the device and its modules can be implemented by hardware, software, or a combination of both. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated hardware. Those skilled in the art will understand that the above-described methods and devices can be implemented using computer-executable instructions and / or included in processor control code, for example, on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of this specification can be implemented not only by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field-programmable gate arrays, programmable logic devices, etc., but also by software executed by various types of processors, or by a combination of the above-described hardware circuits and software (e.g., firmware).

[0101] For more details about the above modules, please refer to other parts of this manual; they will not be repeated here.

[0102] In other embodiments, a preventive maintenance device for transformer winding faults is also provided, including a memory and a processor. The memory stores executable program code, and the processor calls and runs the executable program code from the memory, causing the device to perform the aforementioned preventive maintenance method for transformer winding faults. Specifically, the device may be a chip, component, or module. The chip may include a connected processor and memory; wherein the memory stores instructions, and when the processor calls and executes the instructions, the chip can perform the preventive maintenance method for transformer winding faults provided in the above embodiments.

[0103] In other embodiments, a computer program product is also provided, which, when run on a computer, causes the computer to perform the aforementioned related steps to implement the preventive maintenance method for transformer winding faults provided in the above embodiments.

[0104] In other embodiments, a computer-readable storage medium is also provided, which stores computer program code. When the computer program code is run on a computer, the computer performs the above-described related method steps to implement the preventive maintenance method for transformer winding faults provided in the above embodiments.

[0105] The provided devices, electronic equipment, computer program products, and computer-readable storage media are all used to execute the corresponding methods provided above. Therefore, the beneficial effects they can achieve can be referred to the beneficial effects in the corresponding methods provided above, and will not be repeated here.

[0106] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A preventive maintenance method for transformer winding faults, characterized in that, The method includes the following steps: Acquire the current signal and vibration acceleration signal in different directions of the three-phase winding of the transformer; The phase current signal with the largest effective value within each time window is recorded as the target signal; the electromagnetic excitation force spectrum vector is obtained by performing a fast Fourier transform on the square of the target signal. The vibration response spectrum vector is obtained by performing fast Fourier transform on each vibration acceleration signal within the time window; Based on the electromagnetic excitation force spectrum vector and vibration response spectrum vector corresponding to the same time window, an excitation-response spectrum matrix corresponding to the time window is constructed; the first column of the excitation-response spectrum matrix is ​​composed of electromagnetic excitation force spectrum vectors, and the second to seventh columns are composed of vibration response spectrum vectors. The elements in each column of the excitation-response spectrum matrix are standardized to obtain the standardized excitation-response spectrum matrix. All standardized excitation-response spectrum matrices were expanded and stitched together column-wise during the initial operation of the equipment to form... A long vector of dimension, where The number of frequency points calculated for the Fast Fourier Transform; Obtain the covariance matrix of all long vectors, and perform eigenvalue decomposition on the covariance matrix to obtain the column vectors of the eigenvector matrix; A predetermined number of main eigenvectors are selected from the column vectors of the eigenvector matrix to form the healthy state vibration mode basis; the standardized excitation-response spectrum matrix of each time window is expanded into a long vector, and the transpose of the healthy state vibration mode basis is multiplied by the long vector to obtain the vibration mode contribution coefficient vector. By using frequencies as nodes in the graph structure and the similarity of the row vectors of each channel in the healthy vibration mode basis corresponding to two frequencies as the edge weights between nodes, a frequency-mode association graph structure is constructed. All nodes of the graph structure and their composite feature vectors are input into the graph convolutional network encoder, and the encoder outputs the winding structure state feature vector. The composite feature vector is composed of the row vector corresponding to the node in the standardized excitation-response spectrum matrix and the vibration mode contribution coefficient vector; a reference database for each temperature range is established based on the feature vectors of the structural state of all windings, the vibration mode contribution coefficient vectors, and the oil temperature during the initial operation of the equipment. The cause of the fault is determined by the deviation between the winding structure state feature vector in the current time period and the winding structure state feature vector in the corresponding reference database, as well as the deviation between the vibration mode contribution coefficient vector and the vibration mode contribution coefficient vector in the corresponding reference database.

2. The preventive maintenance method for transformer winding faults according to claim 1, characterized in that, The determination of the fault cause based on the deviation between the winding structure state feature vector in the current time period and the winding structure state feature vector in the corresponding reference database, and the deviation between the vibration mode contribution coefficient vector and the vibration mode contribution coefficient vector in the corresponding reference database, includes: Calculate the first deviation vector between the winding structure state feature vector of each time window and its corresponding first average vector; the first average vector is the average value of all winding structure state feature vectors in the reference database corresponding to the oil temperature of each time window. The overall degradation rate of the winding structure is obtained based on the changes in the first deviation vector corresponding to all time windows within the current period. If the overall degradation rate of the winding structure is greater than the preset degradation rate threshold, then the second deviation vector between the current vibration mode contribution coefficient vector and its corresponding second average vector is calculated; the second average vector is the average value of all vibration mode contribution coefficient vectors in the reference database corresponding to the oil temperature of all time windows in the current period. The cause of the fault is determined based on the second deviation vector.

3. A preventive maintenance method for transformer winding faults according to claim 2, characterized in that, The step of obtaining the overall degradation rate of the winding structure based on the changes in the first deviation vector corresponding to all time windows within the current time period includes: taking the magnitude of the rate of change vector of the first deviation vector corresponding to all time windows within the current time period as the overall degradation rate of the winding structure.

4. A preventive maintenance method for transformer winding faults according to claim 2, characterized in that, Determining the cause of a fault based on the second deviation vector includes: matching the second deviation vector with a pre-calibrated fault feature library to obtain the cause of the fault.

5. A preventive maintenance device for transformer winding faults, the device being used to perform the method of claim 1, characterized in that, The device includes: The signal acquisition module is used to acquire the current signal of the three-phase winding of the transformer and the vibration acceleration signal in different directions; The vector construction module is used to construct the excitation-response spectrum matrix of each time window based on the current signal and vibration acceleration signal of each time window; combine all excitation-response spectrum matrices in the early stage of equipment commissioning to construct the healthy state vibration mode basis; project the excitation-response spectrum matrix of each time window onto the healthy state vibration mode basis to obtain the corresponding vibration mode contribution coefficient vector. The benchmark database construction module is used to construct a frequency-mode correlation graph structure based on the similarity of different frequencies in the healthy state vibration mode base during the initial operation of the equipment; assign features to the nodes of the graph structure and obtain the winding structure state feature vector by combining the graph convolutional network; and establish a benchmark database for each temperature range based on the winding structure state feature vector, vibration mode contribution coefficient vector and oil temperature during the initial operation of the equipment. The fault evaluation module is used to determine the cause of the fault based on the deviation between the winding structure state feature vector in the current time period and the winding structure state feature vector in the corresponding reference database, as well as the deviation between the vibration mode contribution coefficient vector and the corresponding vibration mode contribution coefficient vector.

Images