A Filtering Method for Turbine Displacement Fault Feature Signals Based on Compressed Wavelet Spectral Images
By employing a method based on compressed wavelet spectral images, phase space reconstruction and singular value decomposition, combined with synchronous squeezing wavelet transform, the problem of noise filtering in the characteristic signals of turbine displacement faults was solved. This achieved efficient noise reduction and preservation of useful signals, thereby improving the accuracy of fault diagnosis.
Patent Information
- Application Number
- CN202611043099.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-14
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies cannot effectively filter out random noise and power frequency noise in turbine displacement fault characteristic signals at the same time, and the filtering process is prone to loss of useful signal components, affecting the accuracy of fault diagnosis.
A method based on compressed wavelet spectral images is adopted. By reconstructing the phase space and decomposing the singular value, the optimal singular value reconstruction order is determined. Combined with synchronous squeezing wavelet transform, power frequency noise is located and filtered out to generate a two-dimensional spectral image, and finally a high-quality fault feature signal is obtained.
While filtering out random noise and power frequency noise, it reduces the loss of fault characteristic information, improves the signal-to-noise ratio, and provides a more reliable data foundation for fault analysis and condition assessment.
Smart Images

Figure CN123115591A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of steam turbine condition monitoring and digital signal processing technology, and is specifically applied to noise reduction filtering and fault diagnosis scenarios of steam turbine displacement fault characteristic signals. Background Technology
[0002] In the field of steam turbine operation condition monitoring, displacement fault characteristic signals are an important basis for reflecting the rotor's operating status, and signal quality directly affects the reliability of fault identification and condition assessment. In actual engineering scenarios, due to factors such as the on-site electromagnetic environment, equipment vibration, and power supply system, the steam turbine displacement fault characteristic signals collected by sensors usually contain both random noise and power frequency noise. If noisy signals are directly used for condition analysis, it can easily lead to misjudgment of the actual operating status of the steam turbine.
[0003] Currently, noise reduction for turbine displacement fault characteristic signals typically employs separate techniques to filter out random noise and power frequency noise. Power frequency noise suppression often utilizes notch filters, which rely on a fixed power frequency noise spectrum. However, in actual operating conditions, power frequency noise fluctuates significantly. To ensure effective filtering, the notch filter's spectral range needs to be set relatively wide, which in turn results in the loss of useful signal components and reduces the overall accuracy of signal processing. Existing methods struggle to effectively filter both types of noise while fully preserving the fault characteristic information in the original signal, failing to adequately meet the requirements of subsequent signal analysis and fault diagnosis. Summary of the Invention
[0004] The purpose of this invention is to provide a method for filtering turbine displacement fault feature signals based on compressed wavelet spectral images, which solves the problem that existing technologies are unable to effectively filter out random noise and power frequency noise at the same time, and that the filtering process is prone to causing loss of useful signal components.
[0005] To achieve the above objectives, the present invention provides the following technical solution:
[0006] A method for filtering turbine displacement fault feature signals based on compressed wavelet spectral images includes the following steps:
[0007] Collect characteristic signals of turbine displacement faults;
[0008] The turbine displacement fault characteristic signal is reconstructed in phase space and decomposed into singular values to obtain a singular value sequence and a singular vector sequence.
[0009] Based on the singular value sequence and the singular vector sequence, the order is optimized to obtain the optimal singular value reconstruction order. The components corresponding to the optimal singular value reconstruction order are subjected to inverse singular value decomposition transformation to obtain the noise-reduced fault feature signal.
[0010] The noise-reduced fault feature signal is subjected to synchronous squeezing wavelet transform to generate a two-dimensional spectrum image;
[0011] The optimal frequency range corresponding to the power frequency noise is obtained by locating the frequency range of the two-dimensional spectrum image.
[0012] The wavelet coefficients corresponding to the optimal frequency range in the two-dimensional spectral image are set to zero, and synchronous squeezing wavelet inverse transform is performed to obtain the filtered turbine displacement fault characteristic signal.
[0013] In one possible implementation, the step of performing phase space reconstruction and singular value decomposition on the turbine displacement fault characteristic signal to obtain a singular value sequence and a singular vector sequence includes:
[0014] The turbine displacement fault characteristic signal is reconstructed in phase space to generate a multi-dimensional phase space matrix;
[0015] Singular value decomposition is performed on the multidimensional phase space matrix to obtain two sets of orthogonal singular matrices and diagonal singular value matrices;
[0016] Diagonal elements are extracted from the diagonal singular value matrix to generate a singular value sequence, and column vectors are extracted from the orthogonal singular matrix to generate a singular vector sequence.
[0017] In one possible implementation, the step of performing order optimization based on the singular value sequence and the singular vector sequence to obtain the optimal singular value reconstruction order includes:
[0018] Perform a forward difference operation on the singular value sequence to obtain a difference singular value sequence;
[0019] The initial effective singular value order is determined based on the peak position of the differential singular value sequence;
[0020] The singular vector waveforms within the initial effective singular value order range are characterized by feature discrimination, and the singular vector orders corresponding to noise components are removed to obtain the optimal singular value reconstruction order.
[0021] In one possible implementation, the step of performing synchronous squeezing wavelet transform on the denoised fault feature signal to generate a two-dimensional spectral image includes:
[0022] The noise-reduced fault feature signal is subjected to continuous wavelet transform to obtain a multi-scale wavelet coefficient matrix;
[0023] Perform time-dimensional partial derivative operations on the multi-scale wavelet coefficient matrix to calculate the instantaneous frequency values corresponding to each scale and each time step.
[0024] The frequency axis is divided into intervals, and the wavelet coefficients whose instantaneous frequencies fall into the same frequency interval are superimposed and squeezed to obtain the synchronous squeezed wavelet coefficients corresponding to each frequency interval.
[0025] A two-dimensional spectral image is generated by using time as the horizontal axis and frequency as the vertical axis, combined with the synchronous squeezing wavelet coefficients.
[0026] In one possible implementation, the step of locating the optimal frequency range corresponding to the power frequency noise by analyzing the two-dimensional spectrum image includes:
[0027] The in-row energy of the two-dimensional spectral image is summed along the frequency dimension to convert the two-dimensional coefficient data into a one-dimensional spectral energy sequence.
[0028] Locate the energy peak position corresponding to the power frequency component in the one-dimensional spectral energy sequence;
[0029] Find the energy minimum points on both sides of the energy peak position, and determine the frequency range corresponding to the minimum points on both sides as the optimal spectrum interval for power frequency noise.
[0030] In one possible implementation, the step of performing continuous wavelet transform on the denoised fault feature signal to obtain a multi-scale wavelet coefficient matrix includes:
[0031] A mother wavelet function is selected, and multi-scale continuous wavelet convolution operation is performed on the denoised fault feature signal to obtain wavelet coefficients at different scales and time positions, forming a multi-scale wavelet coefficient matrix.
[0032] In one possible implementation, the step of performing feature discrimination on the singular vector waveform within the initial effective singular value order range and removing the singular vector order corresponding to the noise component includes:
[0033] By analyzing the waveform morphology characteristics of singular vectors of different orders, singular vectors whose waveforms exhibit irregular and random fluctuations are identified as noise components.
[0034] The order corresponding to the noise component is removed from the initial effective singular value order, and the order corresponding to the singular vector containing fault feature information is retained to obtain the optimal singular value reconstruction order.
[0035] In one possible implementation, the step of acquiring the turbine displacement fault characteristic signal includes,
[0036] An eddy current displacement sensor is used to collect displacement signals at the turbine journal. Discrete data points are obtained according to a preset sampling frequency to generate turbine displacement fault characteristic signals.
[0037] Compared with the prior art, the beneficial effects of this invention are as follows: This invention uses a combination of singular value forward difference and singular vector waveform screening to determine the optimal singular value reconstruction order. Compared with the traditional denoising method that selects the reconstruction order based on a single threshold, it can more accurately distinguish between signal components and noise components, reduce the loss of fault feature information while filtering out random noise, and improve the signal-to-noise ratio of the denoised signal.
[0038] This invention generates a two-dimensional spectral image through synchronous squeezing wavelet transform, locates the optimal spectral interval corresponding to power frequency noise based on the spectral energy distribution, and then sets the wavelet coefficients of the corresponding interval to zero and completes the inverse transform. Compared with the fixed bandwidth notch filter method, this invention can match the spectral distribution range of actual power frequency noise, reducing the impact on useful signals in adjacent frequency bands while filtering out power frequency noise, and avoiding excessive loss of useful signal components.
[0039] This invention performs random noise filtering and power frequency noise filtering in stages, successively removing the two types of noise through phase space singular value decomposition and synchronous squeezing wavelet transform. The processed turbine displacement fault characteristic signal can retain more original fault characteristics, providing a more reliable data foundation for subsequent fault analysis and condition assessment. Attached Figure Description
[0040] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0041] Figure 1 This is a schematic diagram of the filtering method for turbine displacement fault feature signals based on compressed wavelet spectral images according to an embodiment of the present invention;
[0042] Figure 2 The original turbine displacement fault characteristic signal time-domain waveform and amplitude spectrum are shown in the embodiment of the present invention.
[0043] Figure 3 This is a schematic diagram of the distribution of singular value sequences according to an embodiment of the present invention;
[0044] Figure 4 This is a comparison diagram of the waveforms of multi-order singular vectors in an embodiment of the present invention;
[0045] Figure 5 This is a schematic diagram of the differential singular value sequence distribution according to an embodiment of the present invention;
[0046] Figure 6 The above are the time-domain waveform and amplitude spectrum of the fault feature signal after singular value decomposition and noise reduction according to an embodiment of the present invention.
[0047] Figure 7 This is a two-dimensional spectrum diagram of the fault characteristic signal synchronous squeezing wavelet in an embodiment of the present invention;
[0048] Figure 8 This is a one-dimensional sequence waveform diagram of spectral energy according to an embodiment of the present invention;
[0049] Figure 9 The images show the time-domain waveform and amplitude spectrum of the final filtered fault characteristic signal in this embodiment of the invention. Detailed Implementation
[0050] 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 this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.
[0051] Example:
[0052] It should be noted that the terms "comprising" and "having" and any variations thereof in the embodiments of the present invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such processes, methods, products, or devices.
[0053] See Figure 1 This embodiment of a method for filtering turbine displacement fault feature signals based on compressed wavelet spectral images is characterized by the following steps:
[0054] Step 101: Collect turbine displacement fault characteristic signals.
[0055] Specifically, the turbine displacement fault characteristic signal can be a fault characteristic signal reflecting the displacement state of the turbine rotor. For example, an eddy current displacement sensor can be used to collect the displacement signal at the turbine journal, with a sampling frequency set to 1024Hz, and 1024 data points collected in a single sampling.
[0056] The step of acquiring the turbine displacement fault characteristic signal includes,
[0057] An eddy current displacement sensor is used to collect displacement signals at the turbine journal. Discrete data points are obtained according to a preset sampling frequency to generate turbine displacement fault characteristic signals.
[0058] Specifically, an eddy current displacement sensor can be a sensor device that measures displacement based on the eddy current effect; the preset sampling frequency can be a pre-set signal sampling rate; the discrete data points can be displacement values collected at fixed time intervals. For example, the eddy current displacement sensor uses the Kaman KD2306S model; the sensor is installed on the journal positions on both sides of the turbine rotor; the preset sampling frequency is 1024Hz, and 1024 discrete data points are acquired in a single acquisition; the generated signal is a one-dimensional discrete time series.
[0059] Step 102: Perform phase space reconstruction and singular value decomposition on the turbine displacement fault characteristic signal to obtain singular value sequence and singular vector sequence.
[0060] Specifically, phase space reconstruction can be a process of converting a one-dimensional time series into a multi-dimensional spatial matrix; singular value decomposition can be a matrix decomposition method that decomposes a matrix into singular values and singular vectors.
[0061] The step of performing phase space reconstruction and singular value decomposition on the turbine displacement fault characteristic signal to obtain a singular value sequence and a singular vector sequence includes:
[0062] The turbine displacement fault characteristic signal is reconstructed in phase space to generate a multi-dimensional phase space matrix;
[0063] Singular value decomposition is performed on the multidimensional phase space matrix to obtain two sets of orthogonal singular matrices and diagonal singular value matrices;
[0064] Diagonal elements are extracted from the diagonal singular value matrix to generate a singular value sequence, and column vectors are extracted from the orthogonal singular matrix to generate a singular vector sequence.
[0065] Specifically, the multidimensional phase space matrix can be a multidimensional matrix obtained by reconstructing a one-dimensional sequence; the orthogonal singular matrix can be a left and right singular matrix that satisfies the orthogonality property; and the diagonal singular value matrix can be a diagonal matrix whose diagonal elements are singular values.
[0066] For example, multidimensional phase space matrices, , This represents an m-row n-dimensional phase space reconstruction matrix. , ,... The original turbine displacement discrete fault feature sampling points are represented by m, which represents the phase space embedding dimension, and n represents the length of a single sequence segment. Two sets of orthogonal singular matrices are the left and right singular matrices, respectively. The singular value sequence is arranged in descending order of value. The singular vector sequence contains waveform features of both signal and noise components, and singular value decomposition is performed. = , , , Describes a left orthogonal singular matrix. Describes a right orthogonal singular matrix. express The transpose of the matrix, This represents a diagonal singular value matrix.
[0067] Step 103: Perform order optimization based on the singular value sequence and the singular vector sequence to obtain the optimal singular value reconstruction order. Perform inverse singular value decomposition transformation on the components corresponding to the optimal singular value reconstruction order to obtain the noise-reduced fault feature signal.
[0068] Specifically, the optimal singular value reconstruction order can be the singular value order used for signal reconstruction after filtering.
[0069] The step of performing order optimization based on the singular value sequence and the singular vector sequence to obtain the optimal singular value reconstruction order includes:
[0070] Perform a forward difference operation on the singular value sequence to obtain a difference singular value sequence;
[0071] The initial effective singular value order is determined based on the peak position of the differential singular value sequence;
[0072] The singular vector waveforms within the initial effective singular value order range are characterized by feature discrimination, and the singular vector orders corresponding to noise components are removed to obtain the optimal singular value reconstruction order.
[0073] Specifically, forward difference operation can be an operation that subtracts adjacent singular values; the difference singular value sequence can be a sequence of differences obtained by difference operation; the initial effective singular value order can be the order initially determined by the difference method; feature discrimination can be a discrimination method that judges signals or noise based on waveform features.
[0074] For example, the forward difference operation calculates the difference between the subsequent term and the preceding term, such as... , Denotes the i-th order difference singular value. Denotes the i-th singular value. Indicates the first Singular values of order, The total number of singular values is denoted by 21; the peak position of the difference sequence corresponds to the 21st order, and the initial effective singular value order is 21; the singular vectors with irregular random fluctuations in the waveform correspond to noise components and are removed; the final optimal singular value reconstruction order is 19.
[0075] Further, the step of performing feature discrimination on the singular vector waveforms within the initial effective singular value order range and removing the singular vector orders corresponding to noise components includes,
[0076] By analyzing the waveform morphology characteristics of singular vectors of different orders, singular vectors whose waveforms exhibit irregular and random fluctuations are identified as noise components.
[0077] The order corresponding to the noise component is removed from the initial effective singular value order, and the order corresponding to the singular vector containing fault feature information is retained to obtain the optimal singular value reconstruction order.
[0078] Specifically, waveform morphology characteristics can be the morphological regularity of singular vector time-domain waveforms; irregular random fluctuations can be noise waveform characteristics without fixed periods or regularities; fault characteristic information can be effective signal components reflecting the fault state of the steam turbine. For example, singular vectors with regular periodic changes correspond to fault characteristic signals; the 20th and 21st order singular vectors exhibit irregular random fluctuations and are identified as noise components; the initial order is 21, and after removing the two order noise components, the optimal singular value reconstruction order is 19.
[0079] Step 104: Perform synchronous squeezing wavelet transform on the noise-reduced fault feature signal to generate a two-dimensional spectrum image.
[0080] Specifically, synchronous squeezing wavelet transform can be a time-frequency analysis method to achieve time-frequency energy aggregation; two-dimensional spectrum image can be two-dimensional image data that reflects the time-frequency distribution of the signal.
[0081] The step of performing synchronous squeezing wavelet transform on the denoised fault feature signal to generate a two-dimensional spectral image includes:
[0082] The noise-reduced fault feature signal is subjected to continuous wavelet transform to obtain a multi-scale wavelet coefficient matrix;
[0083] Perform time-dimensional partial derivative operations on the multi-scale wavelet coefficient matrix to calculate the instantaneous frequency values corresponding to each scale and each time step.
[0084] The frequency axis is divided into intervals, and the wavelet coefficients whose instantaneous frequencies fall into the same frequency interval are superimposed and squeezed to obtain the synchronous squeezed wavelet coefficients corresponding to each frequency interval.
[0085] A two-dimensional spectral image is generated by using time as the horizontal axis and frequency as the vertical axis, combined with the synchronous squeezing wavelet coefficients.
[0086] Specifically, continuous wavelet transform can be a multi-scale time-frequency transform method based on the mother wavelet; the multi-scale wavelet coefficient matrix can be a set of wavelet coefficients at different scales and times; the instantaneous frequency value can be the instantaneous frequency value corresponding to each time-frequency point; frequency interval division can be an operation of dividing the frequency axis into multiple continuous intervals; and synchronously squeezed wavelet coefficients can be the wavelet coefficient values after squeezing and aggregation. For example, using the Morlet mother wavelet to perform continuous wavelet transform, the denoised signal is subjected to multi-scale continuous wavelet convolution to generate a multi-scale wavelet coefficient matrix: In the formula, This represents the wavelet coefficients at scale a and time b. This represents the time-domain signal of the turbine displacement fault after noise reduction, where 'a' represents the wavelet scale and 'b' represents the time shift. ( ) indicates the selection of the mother wavelet function; the time-dimensional partial derivatives of the wavelet coefficients are used to calculate the instantaneous frequencies at each scale and time step: In the formula, The instantaneous frequency is represented by scale a and time b, where i is the imaginary unit; wavelet coefficients in the same frequency range are superimposed and squeezed to obtain the squeezing coefficients, and a two-dimensional spectrum image is constructed: In the formula, Indicates frequency range At time b, the wavelet coefficients are simultaneously squeezed. Indicates the width of the frequency range. Represents the discrete scale. The scale segmentation constant is used; the frequency range is uniformly divided according to a logarithmic scale; the size of the two-dimensional spectrum image is 1280×1024 pixels, with the horizontal axis corresponding to time and the vertical axis corresponding to frequency; after synchronous compression, the spectrum energy is more concentrated and the time-frequency resolution is higher.
[0087] Furthermore, the step of performing continuous wavelet transform on the denoised fault feature signal to obtain a multi-scale wavelet coefficient matrix includes,
[0088] A mother wavelet function is selected, and multi-scale continuous wavelet convolution operation is performed on the denoised fault feature signal to obtain wavelet coefficients at different scales and time positions, forming a multi-scale wavelet coefficient matrix.
[0089] Specifically, the mother wavelet function can be a reference waveform function used for wavelet transform; the multi-scale continuous wavelet convolution operation can be an operation that convolves the signal with the mother wavelet at different scales; the multi-scale wavelet coefficient matrix can be a two-dimensional matrix that records wavelet coefficients at different scales and times. For example, the mother wavelet function is the Morlet mother wavelet; the number of scales is set to 128, covering the entire effective frequency range of the signal; the rows of the multi-scale wavelet coefficient matrix correspond to different scales, and the columns correspond to different time sampling points.
[0090] Step 105: Perform spectrum interval positioning on the two-dimensional spectrum image to obtain the optimal spectrum interval corresponding to the power frequency noise.
[0091] Specifically, the optimal frequency range can be the optimal frequency range corresponding to power frequency noise. For example, power frequency noise corresponds to power interference components at a frequency of 50Hz.
[0092] The step of locating the optimal frequency range corresponding to the power frequency noise by analyzing the two-dimensional spectrum image includes:
[0093] The in-row energy of the two-dimensional spectral image is summed along the frequency dimension to convert the two-dimensional coefficient data into a one-dimensional spectral energy sequence.
[0094] Locate the energy peak position corresponding to the power frequency component in the one-dimensional spectral energy sequence;
[0095] Find the energy minimum points on both sides of the energy peak position, and determine the frequency range corresponding to the minimum points on both sides as the optimal spectrum interval for power frequency noise.
[0096] Specifically, in-row energy summation can be a calculation operation that sums the spectral coefficients of each row; a one-dimensional spectral energy sequence can be a sequence of energy changes with frequency after summation; the energy peak position can be the highest energy point corresponding to the power frequency component; the energy minimum point can be the position point with the lowest energy on both sides of the peak. For example, in-row energy summation is the summation of all element values in each row of a two-dimensional spectral image; the power frequency component corresponds to the energy peak at 50Hz; the minimum value to the left of the peak corresponds to the frequency of 42Hz, and the minimum value to the right corresponds to the frequency of 58Hz, with the optimal spectral range being 42Hz to 58Hz; the optimal range can completely cover the power frequency noise range while minimizing the inclusion of effective signal components.
[0097] Step 106: Set the wavelet coefficients corresponding to the optimal frequency range in the two-dimensional spectrum image to zero, perform synchronous squeezing wavelet inverse transform, and obtain the filtered turbine displacement fault characteristic signal.
[0098] Specifically, the filtered signal removes both random noise and power frequency noise. For example, after setting the wavelet coefficients in the power frequency range to zero, a synchronous squeezing inverse transform is performed to output the filtered signal:
[0099] In the formula, This represents the turbine displacement fault signal after inverse transform output filtering. Denotes the wavelet reconstruction constant. For the mother wavelet conjugate Fourier transform, , These are the scale and frequency discrete intervals, respectively. [] indicates taking the real part of a complex number.
[0100] This embodiment provides a method for filtering turbine displacement fault feature signals based on compressed wavelet spectral images. The overall process of the method is as follows: Figure 1 As shown in the figure. This method first removes random noise through singular value decomposition and then filters out power frequency noise through synchronous squeezing wavelet transform, finally obtaining a high-quality fault characteristic signal that can be used for condition analysis.
[0101] Step 1: Use an eddy current displacement sensor to collect turbine displacement fault characteristic signals.
[0102] Eddy current displacement sensors are installed on the journals on both sides of the turbine rotor, and Kaman KD2306S eddy current displacement sensors are used to measure fault characteristic signals. In this embodiment, displacement fault characteristic signals of several data points can be collected according to a preset sampling frequency. Specifically, when the preset sampling frequency is... Then it can be based on the sampling frequency The fault characteristic signal is composed of data collected from N data points. For example, by collecting data from 1024 points at a preset sampling frequency of 1024Hz, the original fault characteristic signal is obtained. The time-domain waveform is as follows Figure 2 As shown in (a), the corresponding amplitude spectrum is as follows: Figure 2 As shown in (b), power frequency noise and random noise are evident.
[0103] Step 2: Reconstruct the phase space from the fault characteristic signals, perform singular value decomposition on the multidimensional phase space matrix, and generate singular value sequences and singular vector sequences.
[0104] Fault characteristic signals Reconstructed The multidimensional phase space matrix is obtained by performing singular value decomposition and generating a sequence of singular values. ,like Figure 3 As shown in the figure. In the singular value sequence, the first singular value corresponds to the DC component. Compared with the other singular values, this value is very large. To facilitate the display of the overall singular value pattern, it is not shown in the figure. The smaller singular values that follow correspond to the noise component. Only the 2nd to 50th singular values are shown in the figure.
[0105] The generated singular value sequence is as follows Figure 4 As shown, due to space limitations, only a few images are displayed in the diagram. , , , , , Waveform of a singular vector.
[0106] Step 3: Perform forward differencing on the singular value sequence to obtain the difference singular value sequence. Combined with singular vector waveform filtering, optimize to obtain the optimal singular value reconstruction order, and perform inverse singular value decomposition transformation to generate a noise-reduced turbine displacement fault characteristic signal.
[0107] Perform forward differencing on the singular value sequence, i.e. This yields the singular value difference sequence, such as Figure 5 As shown. In Figure 5 In the singular value difference sequence, from left to right, the peak positions are 3, 5, 9, 11, 13, 15, 17, and 20. As the singular value order increases, useful components in the signal are gradually lost. When the peak position is 20, the reconstructed signal can better preserve the main components of the signal. Therefore, an effective singular value order of 21 is chosen. .
[0108] However, further from Figure 4 The singular vector waveforms are visible. Corresponding to The DC component in the equation is therefore a straight line. , and Corresponding to The communication component; , And the subsequent singular vector waveforms clearly show that The random noise component in the data. The effective singular value order is further optimized to... This yields the optimal singular value reconstruction order.
[0109] Singular values in the range of 1 to 19 are retained, and the remaining singular values are set to zero. An inverse singular value decomposition transform is then performed to denoise the fault characteristic signal. The denoised fault characteristic signal is then used... The time-domain waveform and amplitude spectrum are shown as follows: Figure 6 As shown. By Figure 6 As can be seen, random noise was significantly filtered out.
[0110] Step 4: Perform synchronous squeezing wavelet transform on the noise-reduced fault characteristic signal to generate a two-dimensional spectrum image.
[0111] Fault feature signals denoised using Morlet mother wavelet for singular value decomposition Perform continuous wavelet transform to calculate instantaneous frequency. ,
[0112] Fault characteristic signals The length is ,but The sampling time interval is =0.001s, take the scale number =128, let , , .Regulation , The frequency range of fault characteristic signals is divided into frequency intervals. .
[0113] Perform a synchronous squeezing wavelet transform on the fault characteristic signal at the center frequency to generate a two-dimensional spectral image, such as... Figure 7 As shown. The specifications of the two-dimensional spectral image are... The graph shows pixels, with the horizontal axis representing time (number of sampling points) and the vertical axis representing the frequency range of the fault characteristic signal. As can be seen in the graph, the power frequency noise component contained in the fault characteristic signal is extracted. The power frequency noise spectrum is almost a straight line, but it intersects with other adjacent spectral components.
[0114] Step 5: Determine the optimal spectral range corresponding to power frequency noise in the two-dimensional spectral image.
[0115] To determine the optimal spectral interval corresponding to power frequency noise in a two-dimensional spectrogram image, the element values of each row of the two-dimensional spectrogram image are summed, thus converting the two-dimensional array into a one-dimensional array. The values of the one-dimensional array are as follows: Figure 8 As shown in the figure, peak value A represents the region containing the 50Hz power frequency component, and peak value B represents other components of the fault characteristic signal. To further determine the range of this region, first find the location of peak value A (125, peak value 0.0467), then find the location of peak value B (160, peak value 0.087). The location of the minimum value between the two (135, minimum value 0.0105) is the upper limit of the frequency spectrum corresponding to the power frequency noise. Similarly, find the lower limit of the frequency spectrum corresponding to the power frequency noise (88, 1.779e-4).
[0116] Step 5: Determine the optimal spectral range corresponding to power frequency noise in the two-dimensional spectral image.
[0117] To determine the optimal spectral interval corresponding to power frequency noise in a two-dimensional spectrogram image, the element values of each row of the two-dimensional spectrogram image are summed, thus converting the two-dimensional array into a one-dimensional array. The values of the one-dimensional array are as follows: Figure 8As shown in the figure, peak value A represents the region containing the 50Hz power frequency component, and peak value B represents other components of the fault characteristic signal. To further determine the range of this region, first find the location of peak value A (125, peak value 0.0467), then find the location of peak value B (160, peak value 0.087). The location of the minimum value between the two (135, minimum value 0.0105) is the upper limit of the frequency spectrum corresponding to the power frequency noise. Similarly, find the lower limit of the frequency spectrum corresponding to the power frequency noise (88, 1.779e-4).
[0118] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.
[0119] The above embodiments are merely illustrative of the technical concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly. They should not be construed as limiting the scope of protection of the present invention. All equivalent changes or modifications made based on the essence of the content of the present invention should be covered within the scope of protection of the present invention.
Claims
1. A method for filtering displacement fault feature signals of a steam turbine based on compressed wavelet spectrogram images, characterized in that, Includes the following steps: Collect characteristic signals of turbine displacement faults; The turbine displacement fault characteristic signal is reconstructed in phase space and decomposed into singular values to obtain a singular value sequence and a singular vector sequence. Based on the singular value sequence and the singular vector sequence, the order is optimized to obtain the optimal singular value reconstruction order. The components corresponding to the optimal singular value reconstruction order are subjected to inverse singular value decomposition transformation to obtain the noise-reduced fault feature signal. The noise-reduced fault feature signal is subjected to synchronous squeezing wavelet transform to generate a two-dimensional spectrum image; The optimal frequency range corresponding to the power frequency noise is obtained by locating the frequency range of the two-dimensional spectrum image. The wavelet coefficients corresponding to the optimal frequency range in the two-dimensional spectral image are set to zero, and synchronous squeezing wavelet inverse transform is performed to obtain the filtered turbine displacement fault characteristic signal.
2. The method of claim 1, wherein the method is a method of filtering a steam turbine displacement fault signature signal based on a compressed wavelet spectrogram, the method comprising: The step of performing phase space reconstruction and singular value decomposition on the turbine displacement fault characteristic signal to obtain a singular value sequence and a singular vector sequence includes: The turbine displacement fault characteristic signal is reconstructed in phase space to generate a multi-dimensional phase space matrix; Singular value decomposition is performed on the multidimensional phase space matrix to obtain two sets of orthogonal singular matrices and diagonal singular value matrices; Diagonal elements are extracted from the diagonal singular value matrix to generate a singular value sequence, and column vectors are extracted from the orthogonal singular matrix to generate a singular vector sequence.
3. The method of claim 1, wherein the method is a steam turbine displacement fault signature filtering method based on compressed wavelet spectrogram images. The step of performing order optimization based on the singular value sequence and the singular vector sequence to obtain the optimal singular value reconstruction order includes: Perform a forward difference operation on the singular value sequence to obtain a difference singular value sequence; The initial effective singular value order is determined based on the peak position of the differential singular value sequence; The singular vector waveforms within the initial effective singular value order range are characterized by feature discrimination, and the singular vector orders corresponding to noise components are removed to obtain the optimal singular value reconstruction order.
4. The method of claim 1, wherein the method is a steam turbine displacement fault signature filtering method based on compressed wavelet spectrogram images. The step of performing synchronous squeezing wavelet transform on the denoised fault feature signal to generate a two-dimensional spectral image includes: The noise-reduced fault feature signal is subjected to continuous wavelet transform to obtain a multi-scale wavelet coefficient matrix; Perform time-dimensional partial derivative operations on the multi-scale wavelet coefficient matrix to calculate the instantaneous frequency values corresponding to each scale and each time step. The frequency axis is divided into intervals, and the wavelet coefficients whose instantaneous frequencies fall into the same frequency interval are superimposed and squeezed to obtain the synchronous squeezed wavelet coefficients corresponding to each frequency interval. A two-dimensional spectral image is generated by using time as the horizontal axis and frequency as the vertical axis, combined with the synchronous squeezing wavelet coefficients.
5. The method of claim 1, wherein the method is a steam turbine displacement fault signature filtering method based on compressed wavelet spectrogram images. The step of locating the optimal frequency range corresponding to the power frequency noise by performing frequency range localization on the two-dimensional spectrum image includes: The in-row energy of the two-dimensional spectral image is summed along the frequency dimension to convert the two-dimensional coefficient data into a one-dimensional spectral energy sequence. Locate the energy peak position corresponding to the power frequency component in the one-dimensional spectral energy sequence; Find the energy minimum points on both sides of the energy peak position, and determine the frequency range corresponding to the minimum points on both sides as the optimal spectrum interval for power frequency noise.
6. The method of claim 4, wherein the compressed wavelet spectrogram image-based steam turbine displacement fault signature filtering method is characterized by, The step of performing continuous wavelet transform on the denoised fault feature signal to obtain a multi-scale wavelet coefficient matrix includes: A mother wavelet function is selected, and multi-scale continuous wavelet convolution operation is performed on the denoised fault feature signal to obtain wavelet coefficients at different scales and time positions, forming a multi-scale wavelet coefficient matrix.
7. The method of claim 3, wherein the compressed wavelet spectrogram image-based steam turbine displacement fault signature filtering method further comprises: The step of performing feature discrimination on the singular vector waveforms within the initial effective singular value order range and removing the singular vector orders corresponding to noise components includes: By analyzing the waveform morphology characteristics of singular vectors of different orders, singular vectors whose waveforms exhibit irregular and random fluctuations are identified as noise components. The order corresponding to the noise component is removed from the initial effective singular value order, and the order corresponding to the singular vector containing fault feature information is retained to obtain the optimal singular value reconstruction order.
8. The method of filtering turbine displacement fault signature signals based on compressed wavelet spectrogram images of claim 1, wherein, The steps for collecting turbine displacement fault characteristic signals include: An eddy current displacement sensor is used to collect displacement signals at the turbine journal. Discrete data points are obtained according to a preset sampling frequency to generate turbine displacement fault characteristic signals.
