A method, system and storage medium for early fault detection of a hydropower unit

Abstract

The present invention discloses a method, a system and a storage medium for early fault detection of a hydroelectric generating unit, belonging to the technical field of mechanical fault diagnosis methods. The method includes: collecting vibration data of a water turbine and using the vibration data as source data; performing blind source decomposition on the source data based on a binary decomposition method to decompose it into multiple source signals; respectively performing surrogate data processing based on the fast Fourier transform on the decomposed multiple source signals to obtain surrogate data; and performing non-linear structural damage determination and fault propagation path drawing based on the surrogate data. The present invention is beneficial to the discovery and maintenance of early faults, thereby reducing the maintenance cost of the hydroelectric generating unit.

Description

Technical Field

[0001] The present invention relates to the technical field of mechanical fault diagnosis methods, and more specifically, to a method, a system and a storage medium for early fault detection of a hydro-generator set. Background Art

[0002] At present, as the core equipment of a hydropower station, the fault diagnosis of a hydraulic generator set has always troubled the operation and maintenance personnel of large and small hydropower stations, especially the early fault diagnosis of a hydraulic generator set. With the integration and automation of hydraulic production equipment, a large amount of monitoring data in the production process will show characteristics such as high dimension, strong non-linearity, and susceptibility to interference. How to analyze the massive data to ensure a safe and stable production process is a current research hotspot.

[0003] However, the traditional unit maintenance methods, such as the planned maintenance and the after-failure maintenance modes, have many deficiencies. Planned maintenance means that when no fault is found in the unit, the normal working unit is pre-maintained according to the existing experience of the staff. The maintenance in this way is relatively blind and wastes a lot of manpower and material resources to a large extent. On the contrary, after-failure maintenance is carried out after the unit fails, and the preventive function of unit maintenance cannot be fully exerted, and minor repairs become major repairs, resulting in waste of resources and maintenance equipment and damage to economic benefits. Therefore, solving the fault at an earlier stage can greatly increase the service life of the unit, greatly save the operation and maintenance costs and improve its economic benefits.

[0004] Therefore, how to provide a method, a system and a storage medium for early fault detection of a hydro-generator set is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, the present invention provides a method, a system and a storage medium for early fault detection of a hydro-generator set to solve the problems existing in the above-mentioned prior art.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] The present invention provides a method for early fault detection of a hydro-generator set, where the hydro-generator set is composed of multiple water turbines, and the method includes:

[0008] S100: Collect the vibration data of the water turbine and use the vibration data as source data;

[0009] S200: Blindly decompose the source data into multiple source signals based on the binary decomposition method;

[0010] S300: Perform surrogate data processing on the decomposed multiple source signals respectively based on the fast Fourier transform to obtain surrogate data;

[0011] S400: Determine the non-linear structural damage and draw the fault propagation path based on the alternative data.

[0012] Preferably, the S200 includes the following steps:

[0013] S210: Set an n-dimensional source signal as s(t):

[0014] s(t) = [s1(t), s2(t),..., s n (t)] T ;

[0015] S220: Set an m-dimensional coupled observation signal as x(t):

[0016] x(t) = [x1(t), x2(t), …, x m (t)] T ;

[0017] S230: The relationship between the n-dimensional source signal and the m-dimensional coupled observation signal is:

[0018] x(t) = As(t);

[0019] In the formula, A is the aliasing matrix;

[0020] S240: Use the demixing matrix W to decompose the aliasing matrix A to obtain multiple decomposed signals. The formula is:

[0021]

[0022] In the formula, P is the number of independent components, p + l is the termination constraint condition, and the number of termination constraint conditions is the same as the number of set binary independent component extraction sources; W j is the jth column vector in the demixing matrix W;

[0023] S250: Solve the transfer entropy between W j and the observation signal x(t), and use the transfer entropy as the judgment condition to determine whether the transfer entropy is less than the preset value. If so, stop and obtain multiple source signals;

[0024] S260: If not, return to step S240 for decomposition operation until the transfer entropy between W j and the observation signal x(t) is less than the preset value and stop, obtaining multiple source signals.

[0025] More preferably, in step 240, there is a signal y(t) = Wx(t). Since x(t) is the observation signal and the source signal s(t) needs to be obtained, only let W = A -1, after appropriate processing, the source information is independently and identically distributed. Generally, assume that the distribution function of s(t) is:

[0026]

[0027] The joint distribution function of s(t) is:

[0028]

[0029] According to the theory of the distribution of functions of multi-dimensional random variables in probability theory, the joint distribution of x can be derived:

[0030]

[0031] A method that can estimate the demixing matrix W, the maximum likelihood method MLE. The log-likelihood function of the sample is:

[0032]

[0033] Taking the derivative of the above formula gives:

[0034]

[0035] Using the stochastic gradient ascent method, the update rule of matrix W is obtained as:

[0036]

[0037] The demixing matrix W is the inverse matrix of the aliasing matrix A. Given the observed signal, finding the source signal is equivalent to solving the demixing matrix W. Formula (1) is the iteration of the demixing matrix W.

[0038] More preferably, solve for W j The steps for the transfer entropy with the observed signal x(t) are:

[0039] Let Xn and Yn be two time series with discrete states xn and yn at time n, and Xn and Yn can be approximated as steady-state Markov processes of order k and order l respectively. Then the definition formula of the transfer entropy from Yn to Xn is as follows:

[0040]

[0041] Where: P(X n+1 , X n , Y n ) represents the joint probability of X n+1 , X n , Y n ; P(X n , Y n ) represents the joint probability of X n , Y nJoint probability; P(X n +1|X n ) represents the conditional probability of X n given X n+1 at time n. In the transfer entropy calculation in this paper, the logarithm base is 2 and the unit is bit.

[0042] Preferably, the S300 includes the following steps:

[0043] S310: Set the source data set as and set the Fourier transform amplitude of the source data set to be |X k |;

[0044] S320: Calculate the permutation order of the source data set as (rank order) rank 0 (n), where when is the k-th smallest number in , then rank 0 (n) = k;

[0045] S330: Generate a Gaussian white noise set {g } according to the sorted source data set n and arrange the Gaussian white noise {g n} into an ordered Gaussian white noise set {g′ n}, where the permutation serial number rank n of the ordered Gaussian white noise set {g′ g} is equal to rank 0 (n);

[0046] S340: Perform a fast Fourier transform on the ordered Gaussian white noise set {g′ n} to obtain an alternative data set {s′ n}.

[0047] Preferably, the S300 further includes the following steps:

[0048] S350: Calculate the permutation serial number rank n of the alternative data set {s′ s}, and rearrange the ordered Gaussian white noise set {g′ n} according to rank s (n) into an alternative data set {s n}

[0049] to obtain S′ k , S′ k is the data set set in the operation process;

[0050] S370: Calculate S′ k Perform the inverse Fourier transform of, to obtain {s″ n}, where {s″ n} is the data set set during the operation process, and based on rank 0 (n), obtain {s″′ n}, where {s″′ n} is the desired alternative data;

[0051] S380: Perform iterative operations on S360 and S370 until the alternative data has a power spectral density similar to that of the original data and output it.

[0052] Preferably, it can be understood that {s′ n} is obtained by performing a fast Fourier transform on {g′ n}, and {s n} is rearranged from {g′ n} in the arrangement order of {s′ n}.

[0053] Preferably, the iterative rule of S380 is: the distortion of the time probability distribution caused by each Fourier transform amplitude correction is less than that of the previous iteration, and the distortion of the power spectral density caused by each rearrangement of the time signal is less than that of the previous iteration.

[0054] Preferably, S400 includes the following steps:

[0055] S410: Calculate the mutual transfer entropy of the alternative data to obtain the net transfer entropy between the alternative data;

[0056] S420: Determine whether nonlinear damage has occurred based on the net transfer entropy, and draw the transfer path based on the transfer entropy.

[0057] On the other hand, the present invention provides an early fault detection system for a hydropower unit, including:

[0058] A signal acquisition module, configured to acquire the vibration data of the water turbine and use the vibration data as the source data;

[0059] A decomposition module, connected to the signal acquisition module, configured to perform blind source decomposition of the source data based on the binary decomposition method to decompose it into multiple source signals;

[0060] A data processing module, connected to the decomposition module, configured to perform alternative data processing on the decomposed multiple source signals based on the fast Fourier transform respectively to obtain alternative data;

[0061] An output module, connected to the data processing module, is configured to perform non-linear structural damage determination and fault propagation path drawing based on the alternative data and output the results.

[0062] In another aspect, the present invention provides a computer-readable storage medium storing computer-readable instructions, which when executed by a processor, implement the steps of the detection method as described above.

[0063] As can be seen from the above technical solutions, compared with the prior art, the present invention discloses a method, a system and a storage medium for early fault detection of a hydraulic generator unit. After obtaining the vibration signal of the unit with noise interference removed, source signal separation is first performed based on the blind source separation method of binary decomposition; non-linear structural damage determination and fault propagation path drawing are performed based on alternative data and transfer entropy, and a fault diagnosis model is constructed based on information entropy and transfer entropy to achieve early fault diagnosis of the hydraulic generator unit. The present invention is conducive to the discovery and maintenance of early faults, thereby reducing the maintenance cost of the hydraulic generator unit. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings according to the provided drawings without creative efforts.

[0065] Figure 1 It is a schematic flowchart of the detection method provided by the present invention;

[0066] Figure 2 It is a schematic structural diagram of the detection system provided in Embodiment 1 of the present invention;

[0067] FIG. 3(a) is a time-domain waveform and frequency spectrum diagram of the first group of observed signals;

[0068] FIG. 3(b) is a time-domain waveform and frequency spectrum diagram of the second group of observed signals;

[0069] FIG. 3(c) is a time-domain waveform and frequency spectrum diagram of the third group of observed signals;

[0070] FIG. 4(a) is a time-domain waveform and frequency-domain spectrum diagram of the demixed signal y11;

[0071] FIG. 4(b) is a time-domain waveform and frequency-domain spectrum diagram of the demixed signal y12;

[0072] FIG. 5(a) is a schematic diagram of the transfer entropy between y11 and s1, s2, s3 respectively;

[0073] Figure 5(b) is a schematic diagram of the transfer entropy of y12 with s1, s2, and s3 respectively;

[0074] Figure 6(a) is a time-domain diagram of the surrogate data of the X offset of the water turbine;

[0075] Figure 6(b) is a time-domain diagram of the surrogate data of the Y offset of the water turbine;

[0076] Figure 6(c) is a time-domain diagram of the surrogate data of the X of the guide bearing swing;

[0077] Figure 6(d) is a time-domain diagram of the surrogate data of the Y of the guide bearing swing;

[0078] Figure 6(e) is a time-domain diagram of the surrogate data of the X of the guide bearing vibration;

[0079] Figure 6(f) is a time-domain diagram of the surrogate data of the Y of the guide bearing vibration;

[0080] Figure 7 This is the fault propagation path diagram provided in Embodiment 2 of the present invention. Detailed implementation manners

[0081] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0082] Embodiment 1

[0083] See the appendix Figure 1 As shown, on the one hand, Embodiment 1 of the present invention discloses a method for early fault detection of a hydro-generating unit, including:

[0084] S100: Collect the vibration data of the water turbine and use the vibration data as the source data;

[0085] S200: Perform blind source decomposition of the source data based on the binary decomposition method to decompose it into multiple source signals;

[0086] S300: Perform surrogate data processing on the decomposed multiple source signals respectively based on the fast Fourier transform to obtain surrogate data;

[0087] S400: Perform non-linear structural damage determination and fault propagation path drawing based on the surrogate data.

[0088] In a specific embodiment, S200 includes the following steps:

[0089] S210: Set an n-dimensional source signal as s(t):

[0090] s(t) = [s1(t), s2(t),..., s n (t)] T ;

[0091] S220: Set an m-dimensional coupled observation signal as x(t):

[0092] x(t) = [x1(t), x2(t), …, x m (t)] T ;

[0093] S230: The relationship between the n-dimensional source signal and the m-dimensional coupled observation signal is:

[0094] x(t) = As(t);

[0095] where A is the aliasing matrix;

[0096] S240: Use the demixing matrix W to decompose the aliasing matrix A to obtain multiple decomposed signals, and the formula is:

[0097]

[0098] where P is the number of independent components, p + l is the termination constraint condition, and the number of termination constraint conditions is the same as the number of set binary independent component extraction sources; W j is the j-th column vector in the demixing matrix W;

[0099] S250: Solve the transfer entropy of W j and the observation signal x(t), and use the transfer entropy as the judgment condition. Judge whether the transfer entropy is less than the preset value. If so, stop and obtain multiple source signals;

[0100] S260: If not, return to step S240 for decomposition operations until the transfer entropy of W j and the observation signal x(t) is less than the preset value and stop, obtaining multiple source signals.

[0101] Specifically, in step S240, there is a signal y(t) = Wx(t). Since x(t) is the observation signal and the source signal s(t) needs to be obtained, only let W = A -1 , after appropriate processing of the source information, they are all independently and identically distributed. Generally, assume that the distribution function of s(t) is:

[0102]

[0103] The joint distribution function of s(t) is:

[0104]

[0105] According to the theory of the distribution of multi-dimensional random variable functions in probability theory, the joint distribution of x can be derived:

[0106]

[0107] A method that can estimate the demixing matrix W, the maximum likelihood method MLE. The log-likelihood function of the sample is:

[0108]

[0109] Taking the derivative of the above formula gives:

[0110]

[0111] Using the stochastic gradient ascent method, we obtain the update rule for matrix W as:

[0112]

[0113] The demixing matrix W is the inverse matrix of the aliasing matrix A. Given the observed signal, finding the source signal is equivalent to solving the demixing matrix W. Formula (1) is the iteration of the demixing matrix W.

[0114] In a specific embodiment, solve for W j The steps for the transfer entropy with the observed signal x(t) are as follows:

[0115] Let Xn and Yn be two time series with discrete states xn and yn at time n, and Xn and Yn can be approximated as steady-state Markov processes of order k and order l respectively. Then the definition formula of the transfer entropy from Yn to Xn is as follows:

[0116]

[0117] In the formula: P(X n+1 ,X n ,Y n ) represents the joint probability of X n+1 , X n , Y n ; P(X n ,Y n ) represents the joint probability of X n , Y n ; P(X n +1|X n ) represents the conditional probability of X n given X n+1 at time n. In the calculation of the transfer entropy in this article, the logarithm base is taken as 2, and the unit is bit.

[0118] Preferably, the S300 includes the following steps:

[0119] S310: Set the source data set as and set the Fourier transform amplitude of the source data set as |X k |;

[0120] S320: Calculate the rank order of the source data set as (rank order) rank 0 (n), where when is the k-th smallest number in 0 , then rank

[0121] S330: Generate a Gaussian white noise set {g } according to the sorted source data set n , and arrange the Gaussian white noise {g n} into an ordered Gaussian white noise set {g' n}, where the arrangement serial number rank n of the ordered Gaussian white noise set {g' g} is equal to rank 0 (n);

[0122] S340: Perform a fast Fourier transform on the ordered Gaussian white noise set {g' n} to obtain an alternative data set {s' n}.

[0123] In a specific embodiment, S300 includes the following steps:

[0124] S310: Set the source data set as and set the Fourier transform amplitude of the source data set as |X k |;

[0125] S320: Calculate the rank order of the source data set as (rank order) rank 0 (n), where when is the k-th smallest number in 0 , then rank

[0126] S330: Generate a Gaussian white noise set {g } according to the sorted source data set n and arrange the Gaussian white noise {g n} into an ordered Gaussian white noise set {g' n} such that the arrangement serial number rank n of the ordered Gaussian white noise set {g'g (n) is equal to rank 0 (n);

[0127] S340: Perform a fast Fourier transform on the ordered Gaussian white noise set {g′ n} to obtain an alternative data set {s′ n}.

[0128] In a specific embodiment, S300 further includes the following steps:

[0129] S350: Calculate the permutation number rank n (n) of the alternative data set {s′ s}, and rearrange the ordered Gaussian white noise set {g′ n} according to rank s (n) to form an alternative data set {s n}

[0130] S360: Calculate the Fourier transform S n of the alternative data set {s k . Without changing the phase, replace |S k | with |X k | to obtain S′ k . S′ k is the data set set during the operation process;

[0131] S370: Calculate the inverse Fourier transform of S′ k to obtain {s″ n}. {s″ n} is the data set set during the operation process, and obtain {s″′ 0} according to rank n (n). {s″′ n} is the required alternative data;

[0132] S380: Perform iterative operations on S360 and S370 until the alternative data has a power spectral density similar to that of the original data and output it.

[0133] Specifically, it can be understood that {s′ n} is obtained by performing a fast Fourier transform on {g′ n}, and {s n} is obtained by rearranging {g′ n} in the arrangement order of {s′ n}.

[0134] In a specific embodiment, the iteration rule of S380 is that the distortion of the time probability distribution caused by the Fourier transform amplitude correction each time is less than that of the previous iteration, and the distortion of the power spectral density caused by the time signal rearrangement each time is less than that of the previous iteration.

[0135] In a specific embodiment, S400 includes the following steps:

[0136] S410 calculates the mutual transfer entropy of the surrogate data to obtain the net transfer entropy between pairwise surrogate data.

[0137] S420: determines whether nonlinear damage has occurred based on the net transfer entropy, and draws the transfer path based on the transfer entropy.

[0138] On the other hand, as shown in the attached Figure 2 FIG. 1, Embodiment 1 of the present invention discloses an early fault detection system for a hydropower unit, including:

[0139] A signal acquisition module, configured to acquire vibration data of a water turbine and use the vibration data as source data;

[0140] A decomposition module, connected to the signal acquisition module, configured to perform blind source decomposition of the source data based on a binary decomposition method to decompose it into multiple source signals;

[0141] A data processing module, connected to the decomposition module, configured to perform surrogate data processing on the decomposed multiple source signals respectively based on the fast Fourier transform to obtain surrogate data;

[0142] An output module, connected to the data processing module, configured to perform nonlinear structural damage determination and fault propagation path drawing based on the surrogate data and output the results.

[0143] On the other hand, Embodiment 2 of the present invention also discloses a computer-readable storage medium, which stores computer-readable instructions. When the computer-readable instructions are executed by a processor, the steps of the detection method as described above are implemented.

[0144] Through the above technical solutions, compared with the prior art, the present invention discloses an early fault detection method, system and storage medium for a hydropower unit. The present invention has a wide range of applications and can be used for processing nonlinear and non-stationary signals; at the same time, there is no requirement for the number of sensors, and it is applicable to the blind source separation problem with an indefinite number of signal sources; this method has strong robustness and is not easily affected by noise interference. At the same time, it can realize damage determination and fault propagation path drawing, complete early fault diagnosis work, and achieve the purpose of guiding actual maintenance.

[0145] Embodiment 2

[0146] In this Example 2, three sets of source signals s1, s2, and s3 are obtained, taking the X offset values under the rated condition 1 of the water turbine, normal condition 2, fault condition 1, fault condition 2, fault condition 3, and fault condition 4 as examples.

[0147] Using a 3×3 aliasing matrix A, the source signals are aliased in the form of x = A * [s1, s2, s3], T to obtain three sets of observed signals x1, x2, and x3. Refer to Figures 3(a) to 3(c) shown, which are the time-domain waveforms and frequency spectra of the three sets of observed signals.

[0148] The significant frequency-domain components of the source signals are obtained as shown in Table 1:

[0149] Table 1 Significant Frequency-Domain Components of Source Signals

[0150]

[0151]

[0152] Only use the observed signals x1 and x2 to perform underdetermined binary independent component extraction. The separation results are obtained when the number of source signals is inferred to be 2. Refer to the Figure 4(a)-4(b) shown, which are the time-domain waveforms and frequency spectra of the demixed signal y11 and the demixed signal y12.

[0153] The significant frequency-domain components of the demixed signals are obtained as shown in Table 2:

[0154] Table 2 Significant Frequency-Domain Components of Demixed Signals

[0155]

[0156] It can be found from Table 2 that the characteristic frequency of 21 Hz representing signal s1 may coincide with the characteristic frequency of 1 Hz or 21 Hz representing s3, or the characteristic frequency may appear near 20 Hz after the aliasing of the characteristic frequencies of 20 Hz and 21 Hz. Therefore, it is impossible to judge the proportion of the separation result occupied by s1 after aliasing according to the different characteristic frequencies.

[0157] Use transfer entropy to intuitively describe the connection degree between the source signals and the separated signals. Calculate the transfer entropy of the separation results y11 and y12 from the actual source signals s1, s2, and s3. Refer to the Figure 5(a)-5(b) shown, which is the transfer entropy between the separation results and the actual source signals. The larger the transfer entropy, the stronger the transfer relationship between the demixed signal and the actual source signal, that is, the stronger the connection between the two.

[0158] The separation results y 11 , y 12 both contain the components of the three sets of actual source signals, but the proportions of the separation results occupied by each source signal are different. The separated signal y 11Has a stronger connection with the actual source signals s1 and s2. On the contrary, the separated signal y 12 Has a stronger connection with the actual source signal s3. Therefore, y 11 Is the approximate signal of s1 and s2, while y 12 Is the approximate signal of s3.

[0159] 1000 sample points of misalignment vibration noise reduction data for the X offset, Y offset, X water guide swing, Y water guide swing, X water guide vibration, and Y water guide vibration of the water turbine are measured. The time-domain diagram of the replacement data is shown in the appendix Figure 6(a)-6(f) As shown.

[0160] Calculate the net transfer entropy between the six groups of data, and use |t x→y -t y→x | As the representative value of the transfer entropy of the larger one among the two to obtain the net transfer entropy values between variables as shown in Table 3 below (the transfer direction is simplified to one-way).

[0161] Table 3 Net transfer entropy including transfer direction

[0162]

[0163] According to the transfer relationship between variables in Table 3, construct a transfer relationship diagram between parameters. Since the transfer entropy from the Y offset to the X offset is 0.096, which is significantly greater than the transfer entropy values between other variables, it can be inferred that the two are in a direct transfer relationship. The transfer entropy value from the Y water guide swing to the X water guide vibration is 0.003, and the transfer entropy value from the Y water guide swing to [the relevant variable] is 0.002. Further simplify the transfer relationship and judge that the transfer from the Y water guide swing to the X water guide vibration is an indirect transfer. See the appendix Figure 7 As shown, it is the fault propagation path diagram drawn in Example 2. Similarly, other relationships can be deduced.

[0164] Each embodiment in this specification is described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. The same or similar parts between each embodiment can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple. For the relevant parts, refer to the description in the method part.

[0165] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but will be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for early fault detection of a hydropower unit, the hydropower unit being composed of multiple water turbines, characterized in that, The method includes: S100: Collect the vibration data of the water turbine and use the vibration data as the source data; S200: Blindly decompose the source data into multiple source signals based on the binary decomposition method; S300: Perform surrogate data processing based on the fast Fourier transform on the decomposed multiple source signals respectively to obtain surrogate data; S400: Perform non-linear structural damage determination and fault propagation path drawing based on the surrogate data; Among them, S200 includes the following steps: S210: Set an n-dimensional source signal as s(t): s(t) = [s1(t), s2(t),..., s n (t)] T ; S220: Set an m-dimensional coupled observation signal as x(t): x(t) = [x1(t), x2(t), …, x m (t)] T ; S230: The relationship between the n-dimensional source signal and the m-dimensional coupled observation signal is: x(t) = As(t); In the formula, A is the aliasing matrix; S240: Use the demixing matrix W to decompose the aliasing matrix A to obtain multiple decomposed signals. The formula is: where P is the number of independent components, p + l is the termination constraint condition, and the number of termination constraint conditions is the same as the number of binary independent component extraction sources set; W j is the j-th column vector in the demixing matrix W; S250: Solve for W j Calculate the transfer entropy between the source signal and the observed signal x(t), and use the transfer entropy as a decision condition to determine whether the transfer entropy is less than a preset value. If so, stop and obtain multiple source signals; S260: If not, return to step S240 for the decomposition operation until W is decomposed. j Stop when the transfer entropy with the observed signal x(t) is less than the preset value to obtain multiple source signals. S300 includes the following steps: S310: Set the source data set as and set the magnitude of the Fourier transform of the source data set to |X k |; S320: Calculate the source data set with a permutation order of rank 0 (n), where when is the k-th smallest number in, then rank 0 (n) = k; S330: According to the sorted source data set generate a Gaussian white noise set {g n}, and arrange the Gaussian white noise {g n} into an ordered Gaussian white noise set {g' n}, where the arrangement serial number rank n of the ordered Gaussian white noise set {g' g}(n) is equal to rank 0 (n); S340: Perform a fast Fourier transform on the ordered Gaussian white noise set {g′ n} to obtain an alternative data set {s′ n}; S400 includes the following steps: S410: Calculate the mutual transfer entropy of the surrogate data to obtain the net transfer entropy between the surrogate data; S420: Judge whether non-linear damage occurs according to the net transfer entropy, and draw the transfer path according to the transfer entropy.

2. The method for early fault detection of a hydropower unit according to claim 1, characterized in that, S300 further includes the following steps: S350: Calculate the permutation number rank n of the alternative data set {s′ s}(n), and rearrange the ordered Gaussian white noise set {g′ n} into an alternative data set according to rank s (n), and label it as {s n}; S360: Calculate the rearranged replacement data set {s n The Fourier transform S k , do not change the phase, use |X k |Replace|S k |, get S′ k , S′ k It is the data set set during the operation; S370: Calculate S′ k Take the inverse Fourier transform to obtain {s″ n}, and obtain {s″′ 0} according to rank n (n); {s″′ n} is the required replacement data; S380: Perform iterative operations on S360 and S370 until the surrogate data has a power spectral density similar to the original data and output it.

3. The method for early fault detection of a hydropower unit according to claim 2, characterized in that, The iterative rule of S380 is: the distortion of the time probability distribution caused by each Fourier transform amplitude correction is less than the previous iteration, and the distortion of the power spectral density caused by each rearrangement of the signal over time is less than the previous iteration.

4. An early fault detection system for a hydropower unit using the method for early fault detection of a hydropower unit according to any one of claims 1 - 3, characterized in that, Includes: A signal acquisition module for collecting the vibration data of the water turbine and using the vibration data as the source data; A decomposition module connected to the signal acquisition module for blindly decomposing the source data into multiple source signals based on the binary decomposition method; A data processing module connected to the decomposition module for performing surrogate data processing based on the fast Fourier transform on the decomposed multiple source signals respectively to obtain surrogate data; An output module connected to the data processing module for performing non-linear structural damage determination and fault propagation path drawing based on the surrogate data and outputting.

5. A computer-readable storage medium storing computer-readable instructions, characterized in that, When the computer-readable instructions are executed by a processor, the steps of the detection method according to any one of claims 1 to 3 are implemented.

Images