Converter transformer mechanical instability diagnosis method and device based on vibration voiceprint correlation

Abstract

This application provides a method and apparatus for diagnosing mechanical instability of a converter transformer based on vibration acoustic signature correlation. The method includes: acquiring multiple vibration signals from the converter transformer using multiple vibration acoustic signature sensors installed on the outer surface of the converter transformer; processing the multiple vibration signals using a pre-trained vibration analysis network to obtain signal feature quantities corresponding to each vibration signal; analyzing the correlation between the multiple signal feature quantities to obtain a correlation matrix; wherein the correlation matrix includes a correlation index between every two signal feature quantities; and determining the mechanical instability fault diagnosis result of the converter transformer based on whether each correlation index in the correlation matrix meets preset correlation requirements. This scheme simultaneously acquires and analyzes multiple vibration signals fed back by multiple vibration acoustic signature sensors on the converter transformer, thus accurately analyzing whether the converter transformer as a whole has a mechanical instability fault, and has higher accuracy.

Description

Technical Field

[0001] This invention relates to the field of converter transformer fault diagnosis technology, and in particular to a method and apparatus for diagnosing mechanical instability of converter transformers based on vibration acoustic waveform correlation. Background Technology

[0002] Converter transformers (CERTs), especially ultra-high voltage (UHV) converter transformers, are the key to voltage fluctuations in the power grid. Their safe and stable operation is crucial for maintaining the normal operation of the power grid. To prevent sudden converter transformer failures from causing adverse effects on the power grid, how to promptly identify potential problems in converter transformers through online monitoring and assessment has become an important research topic that urgently needs to be addressed.

[0003] Faults in converter transformers are mainly classified into two categories: electrical performance faults and mechanical instability faults. Statistics show that mechanical instability faults account for as much as 55.6% of converter transformer accidents. In the initial stage of a mechanical instability fault, the electrical performance of the converter transformer remains normal, making it difficult to detect through electrical performance testing. As the fault accumulates and worsens, it will eventually lead to equipment damage or even unexpected power outages. Therefore, in fault monitoring, it is necessary to detect mechanical instability faults in their early stages to prevent them from escalating and causing related accidents.

[0004] Currently, there are two main methods for detecting mechanical instability faults in converter transformers in power grids: power outage detection and online detection (i.e., detection without power outage, while the converter transformer is running). Power outage detection requires frequent starting and stopping of the converter transformer, which is obviously detrimental to power grid operation. Existing online detection methods mainly detect the presence of mechanical instability faults by collecting and analyzing local vibration signals of the converter transformer. Obviously, local vibration signals cannot accurately reflect whether there are mechanical instability faults in the converter transformer as a whole, therefore, the accuracy of existing online detection methods is low. Summary of the Invention

[0005] To address the shortcomings of the prior art, this invention provides a method and apparatus for diagnosing mechanical instability of converter transformers based on vibration acoustic signature correlation, thereby providing a more accurate solution for detecting mechanical instability faults in converter transformers.

[0006] The first aspect of this application provides a method for diagnosing mechanical instability of commutator transformers based on vibration acoustic signature correlation, including:

[0007] Multiple vibration signals of the converter transformer are collected using multiple vibration and acoustic sensors installed on the outer surface of the converter transformer.

[0008] The multiple vibration signals are processed using a pre-trained vibration analysis network to obtain the signal feature quantity corresponding to each vibration signal;

[0009] A correlation matrix is ​​obtained by analyzing the correlation between multiple signal features; wherein the correlation matrix includes a correlation index between every two signal features.

[0010] Based on whether each correlation index in the correlation matrix meets the preset correlation requirements, the mechanical instability fault diagnosis result of the converter transformer is determined.

[0011] A second aspect of this application provides a commutator mechanical instability diagnostic device based on vibration acoustic signature correlation, comprising:

[0012] The acquisition unit is used to acquire multiple vibration signals of the converter transformer using multiple vibration and acoustic sensors installed on the outer surface of the converter transformer.

[0013] The processing unit is used to process the multiple vibration signals using a pre-trained vibration analysis network to obtain the signal feature quantity corresponding to each vibration signal;

[0014] An analysis unit is configured to analyze and obtain a correlation matrix based on the correlation between multiple signal features; wherein the correlation matrix includes a correlation index between every two signal features.

[0015] The diagnostic unit is used to determine the mechanical instability fault diagnosis result of the converter transformer based on whether each correlation index in the correlation matrix meets the preset correlation requirements.

[0016] This application provides a method and apparatus for diagnosing mechanical instability of a converter transformer based on vibration acoustic signature correlation. The method includes: acquiring multiple vibration signals from the converter transformer using multiple vibration acoustic signature sensors installed on the outer surface of the converter transformer; processing the multiple vibration signals using a pre-trained vibration analysis network to obtain signal feature quantities corresponding to each vibration signal; analyzing the correlation between the multiple signal feature quantities to obtain a correlation matrix; wherein the correlation matrix includes a correlation index between every two signal feature quantities; and determining the mechanical instability fault diagnosis result of the converter transformer based on whether each correlation index in the correlation matrix meets preset correlation requirements. This scheme simultaneously acquires and analyzes multiple vibration signals fed back by multiple vibration acoustic signature sensors on the converter transformer, thus accurately analyzing whether the converter transformer as a whole has a mechanical instability fault, and has higher accuracy. Attached Figure Description

[0017] 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.

[0018] Figure 1 A schematic diagram illustrating the training process of a vibration analysis network provided in an embodiment of this application;

[0019] Figure 2 This is a schematic diagram of the structure of a converter transformer testing system provided in an embodiment of this application;

[0020] Figure 3 A schematic diagram of the installation position of a vibration acoustic sensor on a converter transformer provided in an embodiment of this application;

[0021] Figure 4 This is a schematic diagram of the structure of a vibration analysis network provided in an embodiment of this application;

[0022] Figure 5 A flowchart illustrating a commutator mechanical instability diagnosis method based on vibration acoustic signature correlation, provided in an embodiment of this application;

[0023] Figure 6 A schematic diagram of the correlation matrix calculated by a commutator mechanical instability diagnosis method based on vibration acoustic signature correlation provided in an embodiment of this application;

[0024] Figure 7 This is a schematic diagram of the structure of a converter mechanical instability diagnostic device based on vibration acoustic signature correlation, provided in an embodiment of this application. Detailed Implementation

[0025] 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.

[0026] The method for diagnosing mechanical instability of converter transformers based on vibration acoustic signature correlation provided in this application first trains a vibration analysis network. Then, this network is used to process multiple vibration signals from the converter transformer, obtaining signal characteristic quantities corresponding to each vibration signal. Finally, the mechanical instability fault diagnosis result of the converter transformer is determined by analyzing the correlation index between pairs of signal characteristic quantities, that is, determining whether the converter transformer has a mechanical instability fault. The multiple vibration signals of the converter transformer are acquired by multiple vibration acoustic signature sensors pre-installed on the outer surface of the converter transformer.

[0027] The training process of the vibration analysis network described above will be explained below. Please refer to [link / reference]. Figure 1 This is a schematic diagram of the training process of a vibration analysis network provided in an embodiment of this application. The process may include the following steps.

[0028] S101, obtain multiple training samples of the converter transformer.

[0029] Each training sample includes multiple sample vibration signals collected by multiple vibration acoustic sensors under preset operating conditions, as well as the load of the converter transformer under the preset operating conditions.

[0030] The training samples in step S101 can be obtained using methods such as... Figure 2 The data was collected by the test system of the converter transformer shown.

[0031] As can be seen, the testing system includes a converter transformer, a test power supply, current measurement and vibration measurement modules, as well as a current recording and test recording module.

[0032] The test power supply can input different loads to the converter transformer according to the input control commands. The current measurement module can measure the load input to the converter transformer and record the measurement results in the current recording module. The vibration measurement module can measure the vibration signals at multiple locations on the converter transformer under the current load and store these vibration signals as sample vibration signals in the vibration recording module.

[0033] Therefore, for any specific operating condition, the load of the converter transformer recorded in the current recording module under that operating condition, and the multiple sample vibration signals at multiple points of the converter transformer recorded in the vibration recording module under that operating condition, constitute a training sample corresponding to that specific operating condition.

[0034] The test power supply adjusts the load parameters of the input converter transformer according to the control command, which enables the converter transformer to switch under different operating conditions, thereby obtaining multiple training samples corresponding to multiple different operating conditions.

[0035] Figure 2The vibration measurement module shown can specifically consist of multiple vibration and acoustic sensors respectively installed on the surface of the converter transformer. The number and installation location of these vibration and acoustic sensors can be determined according to the actual situation and are not limited.

[0036] In some alternative embodiments, these vibration and acoustic sensors can be installed at multiple points in a matrix configuration on the long axial side of the converter transformer. These vibration and acoustic sensors can be installed on the casing of the converter transformer by magnetic adsorption. Therefore, the converter transformer does not need to be powered off during installation, and the safe operation of the converter transformer is not affected.

[0037] As an example, a device such as a converter transformer can be mounted on its surface. Figure 3 The eight vibration acoustic sensors shown are shown. Figure 3 This is a schematic diagram of the installation position of a vibration acoustic sensor on a converter transformer provided in an embodiment of this application. The eight black dots with numerical numbers represent vibration acoustic sensors installed on the converter transformer, and the numerical numbers are equivalent to the corresponding vibration acoustic sensor numbers.

[0038] The vibration acoustic sensor used in this embodiment can be a high-sensitivity vibration acceleration micro-electro-mechanical system (MEMS) sensor. This type of sensor integrates a piezoelectric element and an analog-to-digital converter chip, enabling the sensing and electrical conversion of vibration acoustic signals. The sensor quantizes the mechanical wave of the vibration acoustic signal sensed by the front end into an acceleration value and converts it into a current signal output. The sensor's output signal line is connected to the high-speed acquisition channel of the monitoring device to realize the transmission and acquisition of the vibration acoustic signal from the converter transformer.

[0039] For each vibration acoustic sensor, the signal composed of several sampled values ​​obtained by the sensor through multiple consecutive samplings within a preset sampling period is equivalent to the vibration signal output by the vibration acoustic sensor. Furthermore, the vibration signal output by the vibration acoustic sensor in the testing system is the sample vibration signal of this embodiment. For example, if the sampling duration is 30 seconds and the sampling period is 0.1 seconds, then a vibration acoustic sensor samples once every 0.1 seconds within 30 seconds, continuously sampling for 30 seconds. The signal composed of multiple sampled values ​​obtained according to the sampling time is the vibration signal output by the vibration acoustic sensor.

[0040] In some optional embodiments, after obtaining multiple training samples in S101, these training samples can be preprocessed first, and then S102 can be executed.

[0041] The data preprocessing process includes:

[0042] Identify and remove missing and isolated values ​​from multiple sample vibration signals;

[0043] Noise data in multiple sample vibration signals is smoothed.

[0044] A missing value is a value that occurs when a vibration sensor performs a sampling but fails to obtain the corresponding sample value. In this case, the sample value will be set to 0 or set to a specific symbol to indicate a missing value.

[0045] Isolated values ​​refer to sampled values ​​in a vibration signal that clearly do not conform to the signal pattern. Generally, sampled values ​​that differ significantly from other sampled values ​​can be considered isolated values. During data preprocessing, for each sampled value in the vibration signal, the difference between the sampled value and the sampled values ​​obtained from previous and subsequent samplings can be calculated. If the difference is too large, the sampled value can be considered an isolated value and deleted.

[0046] Noise data refers to the sampled values ​​of the vibration signal that are affected by noise during sampling. In the data preprocessing process, any existing noise data identification method can be used to identify noise data from the vibration signal. Then, for each noise data, it is smoothed according to the following process:

[0047] According to the formula: The smoothing value x corresponding to the noise data is calculated, and then the noise data is replaced with the smoothing value x to complete the smoothing process of the noise data. Here, m is a preset positive integer representing the window size, and m satisfies the condition that 1 / (mTs) is not higher than twice the highest frequency in the range of vibration frequencies of interest corresponding to the currently processed sample vibration signal, and Ts is the sampling period of the vibration acoustic sensor. k The k-th sample value is defined by m within a smoothing window, which contains m sample values, and the first sample value is the next sample value of the noisy data.

[0048] The vibration frequency range of interest, also referred to as the target frequency range in this embodiment, is generally set based on experience. Furthermore, the target frequency range differs depending on the vibration signal collected by the vibration acoustic sensor at different locations under different loads. The target frequency range means that, for a vibration acoustic sensor installed at a specific location, when the converter transformer experiences a mechanical instability fault and operates under a specific load, the probability of finding noise signals caused by the mechanical instability fault within the target frequency range of the vibration signal collected by the sensor is highest. In other words, under these circumstances, the noise signal caused by the mechanical instability fault is most significant within the target frequency range.

[0049] In this embodiment, each training sample contains a vibration signal for each sample, and each sample vibration signal is pre-labeled with a corresponding target frequency range. The target frequency ranges of different sample vibration signals can be the same or different.

[0050] S102, for each training sample, calculate the sample signal feature quantity corresponding to each sample vibration signal in the training sample based on multiple sample vibration signals of the training sample and the target frequency range of each sample vibration signal pre-labeled.

[0051] In step S102, for each sample vibration signal, the sample signal characteristic quantity of the sample vibration signal can be calculated by performing the following process:

[0052] A1 divides the sample vibration signal into multiple sample signal segments with a duration equal to the preset segmentation duration.

[0053] The segmentation duration is a preset parameter, and it is shorter than the sampling duration, which is shorter than the duration of the sample vibration signal. For example, if the sample vibration signal is a signal lasting 30 seconds, the segmentation duration can be set to 3 seconds. Thus, in step A1, a sample vibration signal can be divided into 10 sample signal segments of 3 seconds each. The first 3 seconds of the sample vibration signal are the first sample signal segment, the 4th to 6th seconds are the second sample signal segment, and so on.

[0054] A2 performs a Fourier transform on each sample signal segment of the sample vibration signal to obtain the spectrum of each sample signal segment.

[0055] The specific process of Fourier transform can be found in relevant existing technologies, and will not be repeated here.

[0056] For each sample signal segment, the spectrum of the sample signal segment reflects which frequencies of sine wave signals make up the sample signal segment, and the signal amplitude of each frequency of sine wave signal.

[0057] A3. Based on the spectrum of each sample signal segment, the signal amplitudes of the signals in each sample signal segment whose frequencies are within the pre-marked target frequency range are accumulated, and the result is used as the feature value of each sample signal segment.

[0058] As shown in step A2, the spectrum of the sample signal segment reflects the frequency distribution of the sinusoidal signal that makes up the sample signal segment and the signal amplitude corresponding to each frequency. Therefore, in A3, for a sample signal segment, signals whose frequencies are within the target frequency range of the sample vibration signal being processed can be identified in the spectrum of the sample signal segment. The sum of the signal amplitudes of these signals can be used as the feature value of the sample signal segment.

[0059] A4. The sequence of characteristic values ​​composed of the characteristic values ​​of each sample signal segment in the sample vibration signal is determined as the sample signal characteristic quantity of the sample vibration signal.

[0060] Continuing from the example in step A1, suppose a sample vibration signal is divided into 10 sample signal segments, which are denoted as segment 1 to segment 10 in chronological order. The feature values ​​corresponding to these sample signal segments are denoted as z1 to z10, respectively. Then, the sample signal feature quantity corresponding to this sample vibration signal can be denoted as Z1(z1, z2, z3, ..., z10). Here, Z1 indicates that the sample signal feature quantity is the sample signal feature quantity corresponding to the sample vibration signal collected by vibration acoustic sensor No. 1.

[0061] In addition to the above methods, other methods can also be used in step S102 to calculate the sample signal feature quantity corresponding to the vibration signal of each sample in the training sample. This embodiment does not limit the specific calculation process.

[0062] S103, using the vibration analysis network to be trained to process multiple sample vibration signals of each training sample, to obtain multiple output signal feature quantities corresponding to each training sample.

[0063] The vibration analysis network to be trained used in this embodiment can be as follows: Figure 4 The neural network shown has the following structure. Of course, other neural network structures can be used as needed, without limitation.

[0064] The following is based on Figure 4 Let's take step S103 as an example to illustrate.

[0065] The vibration analysis network to be trained consists of an input layer, a hidden layer, and an output layer.

[0066] The input layer can include two nodes, and the hidden layer has one node. The activation function of the node in the hidden layer is set to the logsig function, and the expression of the function is shown in formula (1):

[0067] f(x) = 1 ÷ (1 + e -x )----(1)

[0068] The number of nodes in the output layer can be set as needed, for example, it can be set to 3. The activation function of the nodes in the output layer can be set to the purelin function, the expression of which is shown in formula (2):

[0069] f(x)=x----(2)

[0070] Set the performance function to the mean squared error function Ep and its index to the default value of 0.

[0071] Before executing step S103 for the first time, the parameters in the vibration analysis network to be trained, that is, the weights and thresholds between the above layers, can be randomly initialized, that is, the initial parameter values ​​are randomly assigned to these parameters.

[0072] Alternatively, before executing S103 for the first time, the loss function of the vibration analysis network to be trained can be set to the mean square error function Ep, and the convergence index can be set to 0, indicating that the best effort is made to build the optimal vibration analysis network.

[0073] After constructing the vibration analysis network to be trained according to the above configuration, the sample vibration signal of each training sample obtained in S101 can be input into the vibration analysis network to be trained one by one. After the vibration analysis network to be trained processes each sample vibration signal, it will output the output signal feature quantity of the sample vibration signal.

[0074] S104. The network loss of the vibration analysis network to be trained is calculated based on the deviation between the sample signal feature quantities corresponding to multiple sample vibration signals in the training samples and the multiple output signal feature quantities corresponding to the training samples.

[0075] As described in S103, the loss function of the vibration analysis network to be trained is the mean square error function Ep. In S104, for each training sample, the sample signal feature quantity corresponding to each sample vibration signal in the training sample and the output signal feature quantity corresponding to each sample vibration signal can be substituted into the mean square error function. The sample loss of the training sample is calculated through this function. Finally, the network loss of the vibration analysis network to be trained is calculated based on the sample losses of all training samples.

[0076] Specifically, for a training sample, the sample loss Loss can be calculated using the following formula (3):

[0077]

[0078] In formula (3), L represents the number of vibration signals in the training samples, and t k o represents the sample signal feature quantity corresponding to the k-th sample vibration signal. k This represents the output signal feature quantity of the k-th sample vibration signal after it has been processed by the vibration analysis network to be trained.

[0079] The network loss of the vibration analysis network to be trained can be calculated from the sample loss of all training samples as follows:

[0080] The sum of the sample losses of all training samples is used as the network loss of the vibration analysis network to be trained. Alternatively, the average of the sample losses of all training samples is calculated and used as the network loss of the vibration analysis network to be trained.

[0081] S105, determine whether the network loss meets the preset convergence conditions.

[0082] If the network loss does not meet the convergence condition, proceed to step S106. If the network loss meets the convergence condition, the training ends. The vibration analysis network to be trained at this time is the vibration analysis network required by the diagnostic method of this embodiment.

[0083] In step S105, the convergence condition can be set as follows: the network loss is less than or equal to a preset convergence metric, or the number of iterations is greater than or equal to a preset maximum number of iterations. The maximum number of iterations can be set as needed; for example, it can be set to 1000.

[0084] The number of iterations can be defined as the number of times step S103 is executed.

[0085] S106, Update the parameters of the vibration analysis network to be trained based on the network loss.

[0086] As mentioned earlier, the parameters of the vibration analysis network include the weights and thresholds between the layers in the vibration analysis network.

[0087] In step S105, the weights between the hidden layer and the output layer can be updated according to the following formulas (4) and (5):

[0088] w ki+ =w ki -ηδ k o i =w ki -ηo k (1-o k )(t k -o k )o i ----(4)

[0089]

[0090] In formulas (4) and (5), η represents the preset learning rate, o i For hidden layer output, o k For output layer output, w ki w represents the weights between the hidden layer and the output layer before the update. ki+ This represents the updated weights between the hidden layer and the output layer.

[0091] The weights between the input layer and the hidden layer can be updated according to the following formula (6):

[0092]

[0093] In formula (6), w ij w represents the weights between the input layer and the hidden layer before the update. ij+ This represents the weights between the input layer and the hidden layer after the update.

[0094] Optionally, the weights of each layer can be combined into a weight matrix W(m), and the correction error δ can be recorded simultaneously. k From (m) to the error matrix E(m), the formulas for updating the weights shown in formulas (4) to (6) above can be integrated into the matrix form shown in formula (7) below:

[0095] W(m+1)=W(m)-A -1 (m)g(m)----(7)

[0096] In formula (7), m represents the number of iterations. For example, W(m) represents the weight matrix composed of the weights of each layer in the vibration analysis network to be trained when the m-th iteration is executed, that is, when step S103 is executed for the m-th time. A(m) represents the Hessian matrix of the loss function under the current weights when the m-th iteration is executed. This matrix can be approximately solved using the following formula (8):

[0097] A(m)=P T P----(8)

[0098] Where P is the Jacobian matrix, its expression is shown in the following formula (9):

[0099]

[0100] In formula (9) f ij (w) represents the product of the activation function of the i-th hidden layer and the activation function of the j-th output layer. w1 to w n This represents the n weights in the vibration analysis network to be trained.

[0101] The g(m) in formula (7) can be calculated using the following formula (10):

[0102]

[0103] g(m) is the gradient matrix of the network output error with respect to each weight or threshold in the m-th iteration. The negative sign indicates the steepest descent direction of the gradient, indicating system convergence.

[0104] In formulas (4) to (10) above, i, j, and k correspond to the node numbers of the input layer, hidden layer, and output layer, respectively.

[0105] After the update in step S106 is completed, return to step S103 until the network loss meets the convergence condition.

[0106] Based on the vibration analysis network obtained through the above training, this application provides a method for diagnosing mechanical instability of commutator transformers based on vibration acoustic signature correlation. Please refer to [link to relevant documentation]. Figure 5 Here is a flowchart of the method, which may include the following steps.

[0107] S501 uses multiple vibration and acoustic sensors installed on the outer surface of the converter transformer to collect multiple vibration signals of the converter transformer.

[0108] In some optional embodiments, after obtaining multiple vibration signals, and before processing these vibration signals using a vibration analysis network, data preprocessing can be performed on these vibration signals. That is, before executing S502, the following steps can also be performed:

[0109] Identify and remove missing and isolated values ​​from multiple vibration signals;

[0110] Smoothing is performed on noise data from multiple vibration signals.

[0111] For details on identifying and removing missing and outlier values, as well as the specific process for smoothing noisy data, please refer to [link to documentation]. Figure 1 The data preprocessing process in the corresponding embodiments will not be described in detail.

[0112] The process of acquiring multiple vibration signals using a vibration acoustic sensor in S501 can be found in [reference needed]. Figure 1 Step S101 of the corresponding embodiment will not be described again.

[0113] S502 uses a pre-trained vibration analysis network to process multiple vibration signals and obtain the signal characteristic quantities corresponding to each vibration signal.

[0114] In step S502, each vibration signal is input into the vibration analysis network trained through the aforementioned training process, thereby obtaining the signal characteristic quantity corresponding to each vibration signal.

[0115] For example, suppose n vibration and acoustic sensors are installed on the casing of a converter transformer, denoted as sensor 1 to sensor n. The n vibration signals collected by these n sensors are denoted as vibration signal 1 to vibration signal n. After these n vibration signals are processed by a vibration analysis network, n signal feature quantities can be obtained, denoted as X1 to Xn, where X1 represents the signal feature quantity obtained after processing the vibration signal 1 collected by sensor 1, and Xn represents the signal feature quantity obtained after processing the vibration signal n collected by sensor n.

[0116] It should be noted that the signal features have the same representation as the sample signal features in the aforementioned embodiments. For example, the signal feature X1 can be represented as X1(x11, x12, x13, x1m), where x11 to x1m are the first to m components of the signal feature X1, and all signal features have the same number of components, for example, all have m components.

[0117] In this embodiment, the advantage of using a vibration analysis network to process vibration signals and obtain corresponding signal characteristic quantities is that:

[0118] Referring to step S102, it can be seen that to calculate the signal characteristic quantity of each vibration signal in the manner described in step S102, it is necessary to first divide the vibration signal into multiple segments, then perform a Fourier transform on each segment, and analyze the spectrum obtained after the transform to obtain the corresponding signal characteristic quantity. Obviously, this process is time-consuming. However, in this embodiment, the above process can be ignored by using a trained vibration analysis network, and the signal characteristic quantity of each vibration signal can be obtained directly through the vibration analysis network.

[0119] Therefore, by obtaining signal characteristic quantities through vibration analysis network processing, the execution efficiency of the diagnostic method in this embodiment can be effectively improved, and the diagnostic results of mechanical instability faults of converter transformers can be obtained more quickly.

[0120] S503, based on the correlation between multiple signal features, obtains the correlation matrix.

[0121] The correlation matrix includes the correlation index between every two signal features.

[0122] In step S503, the correlation index between every two signal features can be calculated first, and then these correlation indices can be used to construct a correlation matrix according to the corresponding signal feature numbers.

[0123] Continuing from the example in step S502, suppose there are n signal features from X1 to Xn. For any two signal features Xi and Xk, the correlation index between the two signal features can be denoted as ρ(Xi, Xk). When constructing the correlation matrix, the correlation index ρ(Xi, Xk) can be determined as the element in the i-th row and k-th column of the correlation matrix.

[0124] The correlation index ρ(Xi, Xk) between any two signal features Xi and Xk can be calculated using the following formulas (11) and (12):

[0125]

[0126]

[0127] Formula (11) is used to calculate the covariance Cov(Xi, Xk) between two signal feature quantities Xi and Xk, and Formula (12) is used to calculate the correlation index between the two signal feature quantities Xi and Xk based on the covariance between them.

[0128] In formula (11), xit represents the t-th component of signal feature Xi, xkt represents the t-th component of signal feature Xk, and m is the total number of components in a signal feature.

[0129] In formula (11), E(Xi) and E(Xk) represent the expected values ​​of the two signal characteristic quantities Xi and Xk. In this embodiment, the expected value of the signal characteristic quantity can be represented by the average value of all components of the signal characteristic quantity. That is, the expected value E(Xi) of the signal characteristic quantity Xi can be calculated by the following formula (13):

[0130]

[0131] S504. Based on whether each correlation index in the correlation matrix meets the preset correlation requirements, determine the mechanical instability fault diagnosis result of the converter transformer.

[0132] Optionally, a specific implementation of step S504 may include:

[0133] If every correlation index in the correlation matrix is ​​greater than the preset first threshold, the mechanical instability fault diagnosis result of the converter transformer is determined to be no mechanical instability fault.

[0134] If every correlation index in the correlation matrix is ​​greater than the preset second threshold, and at least one correlation index is less than the first threshold, the mechanical instability fault diagnosis result of the converter transformer is determined to be that there is a risk of mechanical instability; wherein, the second threshold is less than the first threshold.

[0135] If at least one correlation index in the correlation matrix is ​​less than the second threshold, the mechanical instability fault diagnosis result of the converter transformer is determined to be that a mechanical instability fault exists.

[0136] Optionally, after S504, corresponding prompt information can also be output based on the mechanical instability fault diagnosis results.

[0137] Specifically, if the result indicates no mechanical instability fault, no prompt message needs to be output; if the result indicates a risk of mechanical instability, a risk prompt message will be output to remind relevant maintenance personnel to observe the operation of the converter transformer to further determine whether there is a mechanical instability fault; if the result indicates a mechanical instability fault exists, a fault alarm message will be output to remind relevant maintenance personnel that the converter transformer has experienced a mechanical instability fault and should be repaired in a timely manner.

[0138] The first and second thresholds mentioned above can be set according to actual conditions and are not limited. For example, the first threshold can be set to 0.8 and the second threshold can be set to 0.7.

[0139] The principle of the diagnostic method in this embodiment is that when the converter transformer is running in a normal state (referring to a state without mechanical instability faults), all the internal components operate according to the set operating rules. Therefore, the operation of each component is highly correlated. Thus, the signal characteristic quantities of each vibration signal generated by the operation of each component should also be highly correlated. In terms of correlation index, the correlation index between any two signal characteristic quantities is greater than the first threshold.

[0140] Conversely, when a mechanical instability fault occurs in the converter transformer, the components affected by the fault will no longer operate according to the operating rules set under normal conditions. This leads to a weakening of the correlation between the components affected by the fault, as well as between the affected and unaffected components. Consequently, the correlation between the signal characteristics of some vibration signals generated by the operation of these components will also weaken, which is reflected in the correlation index as some correlation indicators being less than the second threshold.

[0141] Therefore, this scheme can determine whether or not a mechanical instability fault exists in the converter transformer by the magnitude of the correlation index between various signal characteristic quantities.

[0142] This application provides a method for diagnosing mechanical instability of a converter transformer based on vibration acoustic signature correlation. The method includes: acquiring multiple vibration signals from the converter transformer using multiple vibration acoustic signature sensors mounted on its outer surface; processing these signals using a pre-trained vibration analysis network to obtain signal characteristic quantities corresponding to each vibration signal; analyzing the correlation between these signal characteristic quantities to obtain a correlation matrix; wherein the correlation matrix includes a correlation index between every two signal characteristic quantities; and determining the mechanical instability fault diagnosis result of the converter transformer based on whether each correlation index in the correlation matrix meets preset correlation requirements. This scheme simultaneously acquires and analyzes multiple vibration signals fed back by multiple vibration acoustic signature sensors on the converter transformer, thus accurately analyzing whether the converter transformer as a whole has a mechanical instability fault, resulting in higher accuracy.

[0143] Furthermore, in the early stages of mechanical instability faults, the amplitude of the noise caused by the fault is weak and therefore difficult to identify. The method of this embodiment determines whether the converter transformer has mechanical instability faults by analyzing the correlation between the signal characteristics of vibration signals at different locations. Compared with the scheme that diagnoses based on noise amplitude, the scheme of this embodiment can detect mechanical instability faults in their early stages in a timely manner, which helps to carry out timely maintenance in the early stages of the fault and avoid more serious accidents caused by the continued development and deterioration of mechanical instability faults.

[0144] To verify the monitoring capability of the method provided in this embodiment, short-circuit tests and mechanical vibration tests were conducted on a 1000kVA converter transformer model. When a short-circuit current of 560A was applied, mechanical instability occurred in the left winding of the transformer, exhibiting abnormal mechanical vibration at 0.25Hz. At this time, the amplitude of the sensor output changed by less than 10% compared to the rated operating condition, making it difficult to identify in the noise signal. However, the correlation matrix obtained by the above method at this time is as follows: Figure 6 As shown.

[0145] As can be seen, in Figure 6 In the correlation matrix shown, there are multiple elements that are less than the second threshold of 0.7 in the previous example. This indicates that there is a mechanical instability fault in the converter transformer at this time. This shows that the method provided in this embodiment can detect mechanical instability faults when the noise caused by the initial stage of mechanical instability is weak, and has higher accuracy and timeliness.

[0146] According to the method for diagnosing commutator mechanical instability based on vibration acoustic signature correlation provided in the embodiments of this application, the embodiments of this application also provide a device for diagnosing commutator mechanical instability based on vibration acoustic signature correlation. Please refer to [link to relevant documentation]. Figure 7 This is a schematic diagram of the structure of the device, which may include the following units.

[0147] The acquisition unit 701 is used to acquire multiple vibration signals of the converter transformer using multiple vibration and acoustic sensors installed on the outer surface of the converter transformer.

[0148] The processing unit 702 is used to process multiple vibration signals using a pre-trained vibration analysis network to obtain the signal feature quantity corresponding to each vibration signal.

[0149] Analysis unit 703 is used to analyze and obtain a correlation matrix based on the correlation between multiple signal features; wherein, the correlation matrix includes a correlation index between every two signal features.

[0150] The diagnostic unit 704 is used to determine the diagnostic result of mechanical instability fault of the converter transformer based on whether each correlation index in the correlation matrix meets the preset correlation requirements.

[0151] Optionally, the device also includes a training unit 705 for training the vibration analysis network;

[0152] Training unit 705 is specifically used for training vibration analysis networks when training them:

[0153] Multiple training samples of the converter transformer are obtained; each training sample includes multiple sample vibration signals collected by multiple vibration and acoustic sensors under preset operating conditions, as well as the load of the converter transformer under preset operating conditions.

[0154] For each training sample, based on multiple sample vibration signals of the training sample and the target frequency range of each sample vibration signal pre-labeled, the sample signal feature quantity corresponding to each sample vibration signal in the training sample is calculated.

[0155] The vibration analysis network to be trained is used to process multiple sample vibration signals of each training sample to obtain multiple output signal feature quantities corresponding to each training sample.

[0156] The network loss of the vibration analysis network to be trained is calculated based on the deviation between the sample signal feature quantities corresponding to multiple sample vibration signals in the training samples and the multiple output signal feature quantities corresponding to the training samples.

[0157] If the network loss does not meet the preset convergence condition, update the parameters of the vibration analysis network to be trained according to the network loss, and return to the step of using the vibration analysis network to be trained to process multiple sample vibration signals of each training sample to obtain multiple output signal feature quantities corresponding to each training sample, until the network loss meets the convergence condition.

[0158] Optionally, when the training unit 705 calculates the sample signal feature quantity corresponding to each sample vibration signal in the training sample based on multiple sample vibration signals of the training sample and the pre-labeled target frequency range of each sample vibration signal, it is specifically used for:

[0159] For each sample vibration signal, the following process is performed:

[0160] The sample vibration signal is divided into multiple sample signal segments with a duration equal to the preset segmentation duration;

[0161] Perform Fourier transform on each sample signal segment of the sample vibration signal to obtain the spectrum of each sample signal segment;

[0162] Based on the spectrum of each sample signal segment, the signal amplitudes of the signals in each sample signal segment whose frequencies are within the pre-marked target frequency range are accumulated, and the result is used as the feature value of each sample signal segment.

[0163] The sequence of feature values ​​composed of the feature values ​​of each segment of the sample vibration signal is determined as the sample signal feature quantity of the sample vibration signal.

[0164] Optionally, the acquisition unit 701 is also used for:

[0165] Identify and remove missing and isolated values ​​from multiple vibration signals;

[0166] Smoothing is performed on noise data from multiple vibration signals.

[0167] Optionally, when determining the mechanical instability fault diagnosis result of the converter transformer based on whether each correlation index in the correlation matrix meets the preset correlation requirements, the diagnostic unit 704 is specifically used for:

[0168] If every correlation index in the correlation matrix is ​​greater than the preset first threshold, the mechanical instability fault diagnosis result of the converter transformer is determined to be no mechanical instability fault.

[0169] If every correlation index in the correlation matrix is ​​greater than the preset second threshold, and at least one correlation index is less than the first threshold, the mechanical instability fault diagnosis result of the converter transformer is determined to be that there is a risk of mechanical instability; wherein, the second threshold is less than the first threshold.

[0170] If at least one correlation index in the correlation matrix is ​​less than the second threshold, the mechanical instability fault diagnosis result of the converter transformer is determined to be that a mechanical instability fault exists.

[0171] The specific working principle and beneficial effects of the commutator mechanical instability diagnosis device based on vibration acoustic signature correlation provided in this application embodiment can be found in the relevant steps and beneficial effects of the commutator mechanical instability diagnosis method based on vibration acoustic signature correlation provided in this application embodiment, and will not be repeated here.

[0172] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0173] It should be noted that the concepts of "first" and "second" mentioned in this invention are only used to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0174] Those skilled in the art will be able to implement or use this application. 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 this application. Therefore, this application 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 diagnosing mechanical instability of commutator transformers based on vibration acoustic signature correlation, characterized in that, include: Multiple vibration signals of the converter transformer are collected using multiple vibration and acoustic sensors installed on the outer surface of the converter transformer. The multiple vibration signals are processed using a pre-trained vibration analysis network to obtain the signal feature quantity corresponding to each vibration signal; Calculate the correlation index between every two signal features, and then construct a correlation matrix according to the corresponding signal feature number based on each correlation index; wherein, the correlation matrix includes the correlation index between every two signal features. If each correlation index in the correlation matrix is ​​greater than a preset first threshold, the mechanical instability fault diagnosis result of the converter transformer is determined to be no mechanical instability fault. If every correlation index in the correlation matrix is ​​greater than a preset second threshold, and at least one correlation index is less than the first threshold, the mechanical instability fault diagnosis result of the converter transformer is determined to be at risk of mechanical instability; wherein, the second threshold is less than the first threshold. If at least one of the correlation indicators in the correlation matrix is ​​less than the second threshold, the mechanical instability fault diagnosis result of the converter transformer is determined to be that a mechanical instability fault exists.

2. The method according to claim 1, characterized in that, The process of training the vibration analysis network includes: Multiple training samples of the converter transformer are obtained; wherein each training sample includes multiple sample vibration signals collected by the multiple vibration and acoustic sensors under a preset operating condition, and the load of the converter transformer under the preset operating condition. For each training sample, based on the multiple sample vibration signals of the training sample and the target frequency range of each sample vibration signal pre-labeled, the sample signal feature quantity corresponding to each sample vibration signal in the training sample is calculated. The vibration analysis network to be trained is used to process multiple sample vibration signals of each training sample to obtain multiple output signal feature quantities corresponding to each training sample. The network loss of the vibration analysis network to be trained is calculated based on the deviation between the sample signal feature quantities corresponding to multiple sample vibration signals in the training sample and the multiple output signal feature quantities corresponding to the training sample. If the network loss does not meet the preset convergence condition, the parameters of the vibration analysis network to be trained are updated according to the network loss, and the process of using the vibration analysis network to be trained to process multiple sample vibration signals of each training sample to obtain multiple output signal feature quantities corresponding to each training sample is returned until the network loss meets the convergence condition.

3. The method according to claim 2, characterized in that, The step of calculating the sample signal feature quantity corresponding to each sample vibration signal in the training sample based on multiple sample vibration signals of the training sample and the pre-labeled target frequency range of each sample vibration signal includes: For each of the aforementioned sample vibration signals, the following process is performed: The sample vibration signal is divided into multiple sample signal segments with a duration equal to a preset segmentation duration; Perform a Fourier transform on each of the sample signal segments of the sample vibration signal to obtain the spectrum of each sample signal segment; Based on the spectrum of each sample signal segment, the signal amplitudes of the signals in each sample signal segment whose frequencies are within the pre-marked target frequency range are accumulated, and the result is used as the feature value of each sample signal segment. The feature value sequence composed of the feature values ​​of each of the sample signal segments in the sample vibration signal is determined as the sample signal feature quantity of the sample vibration signal.

4. The method according to claim 1, characterized in that, Before processing the multiple vibration signals using a pre-trained vibration analysis network to obtain the signal feature quantity corresponding to each vibration signal, the method further includes: Identify and remove missing and isolated values ​​from the plurality of vibration signals; The noise data in the multiple vibration signals are smoothed.

5. A commutator mechanical instability diagnostic device based on vibration acoustic signature correlation, characterized in that, include: The acquisition unit is used to acquire multiple vibration signals of the converter transformer using multiple vibration and acoustic sensors installed on the outer surface of the converter transformer. The processing unit is used to process the multiple vibration signals using a pre-trained vibration analysis network to obtain the signal feature quantity corresponding to each vibration signal; The analysis unit is used to calculate the correlation index between every two signal features, and then construct a correlation matrix according to the corresponding signal feature number based on each correlation index; wherein, the correlation matrix includes the correlation index between every two signal features. The diagnostic unit is used to determine the mechanical instability fault diagnosis result of the converter transformer based on whether each correlation index in the correlation matrix meets the preset correlation requirements. When the diagnostic unit determines the mechanical instability fault diagnosis result of the converter transformer based on whether each correlation index in the correlation matrix meets the preset correlation requirements, it is specifically used for: If each correlation index in the correlation matrix is ​​greater than a preset first threshold, the mechanical instability fault diagnosis result of the converter transformer is determined to be no mechanical instability fault. If every correlation index in the correlation matrix is ​​greater than a preset second threshold, and at least one correlation index is less than the first threshold, the mechanical instability fault diagnosis result of the converter transformer is determined to be at risk of mechanical instability; wherein, the second threshold is less than the first threshold. If at least one of the correlation indicators in the correlation matrix is ​​less than the second threshold, the mechanical instability fault diagnosis result of the converter transformer is determined to be that a mechanical instability fault exists.

6. The apparatus according to claim 5, characterized in that, The device also includes a training unit for training the vibration analysis network; When the training unit trains the vibration analysis network, it is specifically used for: Multiple training samples of the converter transformer are obtained; wherein each training sample includes multiple sample vibration signals collected by the multiple vibration and acoustic sensors under a preset operating condition, and the load of the converter transformer under the preset operating condition. For each training sample, based on the multiple sample vibration signals of the training sample and the target frequency range of each sample vibration signal pre-labeled, the sample signal feature quantity corresponding to each sample vibration signal in the training sample is calculated. The vibration analysis network to be trained is used to process multiple sample vibration signals of each training sample to obtain multiple output signal feature quantities corresponding to each training sample. The network loss of the vibration analysis network to be trained is calculated based on the deviation between the sample signal feature quantities corresponding to multiple sample vibration signals in the training sample and the multiple output signal feature quantities corresponding to the training sample. If the network loss does not meet the preset convergence condition, the parameters of the vibration analysis network to be trained are updated according to the network loss, and the process of using the vibration analysis network to be trained to process multiple sample vibration signals of each training sample to obtain multiple output signal feature quantities corresponding to each training sample is returned until the network loss meets the convergence condition.

7. The apparatus according to claim 6, characterized in that, When the training unit calculates the sample signal feature quantity corresponding to each sample vibration signal in the training sample based on multiple sample vibration signals of the training sample and the pre-labeled target frequency range of each sample vibration signal, it is specifically used for: For each of the aforementioned sample vibration signals, the following process is performed: The sample vibration signal is divided into multiple sample signal segments with a duration equal to a preset segmentation duration; Perform a Fourier transform on each of the sample signal segments of the sample vibration signal to obtain the spectrum of each sample signal segment; Based on the spectrum of each sample signal segment, the signal amplitudes of the signals in each sample signal segment whose frequencies are within the pre-marked target frequency range are accumulated, and the result is used as the feature value of each sample signal segment. The feature value sequence composed of the feature values ​​of each of the sample signal segments in the sample vibration signal is determined as the sample signal feature quantity of the sample vibration signal.

8. The apparatus according to claim 5, characterized in that, The acquisition unit is also used for: Identify and remove missing and isolated values ​​from the plurality of vibration signals; The noise data in the multiple vibration signals are smoothed.

Images