Rotary mechanical equipment fault diagnosis method and device based on sample amplification

Abstract

The invention provides a rotating mechanical equipment fault diagnosis method and device based on sample amplification, and the method comprises the steps: carrying out the phase-space reconstruction of a one-dimensional vibration signal, converting the one-dimensional vibration signal into a high-dimensional vibration signal, and enhancing the expression capability of fault features; according to the method, a false neighbor ratio is determined for each space sample, and a corresponding weight is determined according to the false neighbor ratio, so that a low weight can be given to a low-stability sample located in an edge noise area, a high weight can be given to a high-stability sample, and a space sample (reference sample) with a high weight is used as a data enhancement reference; when a new fault sample is generated subsequently according to the reference sample and the target adjacent sample, it can be ensured that the newly generated fault sample is close to the reference sample, the newly generated fault sample is prevented from deviating from real distribution, and therefore the reliability of the newly generated fault sample is improved; as the number of the fault samples is increased, unbalance between the fault diagnosis samples and normal samples is avoided, and the fault diagnosis accuracy of the rotating mechanical equipment is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis of rotating mechanical equipment, and in particular to a method and device for fault diagnosis of rotating mechanical equipment based on sample amplification. Background Art

[0002] In the rotating machinery power system, rotating machinery serves as the core power output equipment, and its operating status is directly related to the safety, stability and production efficiency of the entire system. Such equipment usually includes steam turbines, gas turbines, compressors, generators and other equipment. Such equipment usually operates for a long time under extreme working conditions such as high temperature, high pressure, and high speed, and is subjected to the coupling of multiple physical fields such as mechanical stress, thermal stress, and medium corrosion. The key components of rotating machinery equipment (such as bearings, gears, rotors, etc.) are prone to typical faults such as wear, cracks, imbalance, and misalignment. What is more serious is that in modern industrial production, rotating machinery equipment often needs to operate continuously for 24 hours, which makes the equipment in a state of fatigue accumulation for a long time, further increasing the risk of failure.

[0003] Among related technologies, the core challenge facing fault diagnosis is sample imbalance. During fault diagnosis, samples typically represent over 99% of the time when equipment is operating normally. This extreme data imbalance significantly impacts the performance of fault diagnosis models, primarily in the following ways: First, classification algorithms tend to classify the majority of samples as normal, significantly reducing the recognition rate for minority faults such as bearing spalling and gear tooth breakage. Second, rare fault modes often lack sufficient training samples, making it difficult for models to learn effective fault feature representations.

[0004] Based on this, a new type of rotating mechanical equipment fault diagnosis method is currently needed to solve the above problems in the existing technology. Summary of the Invention

[0005] In response to the problems existing in the prior art, an embodiment of the present invention provides a method and device for fault diagnosis of rotating mechanical equipment based on sample amplification, so as to solve or partially solve the technical problem in the prior art that when fault diagnosis of rotating mechanical equipment is performed, the accuracy of fault diagnosis of rotating mechanical equipment is affected due to imbalance of fault diagnosis samples.

[0006] A first aspect of the present invention provides a method for diagnosing faults of rotating machinery based on sample amplification, the method comprising:

[0007] collecting a plurality of first vibration signals of the rotating mechanical equipment in different fault states, performing standardization processing on the plurality of first vibration signals to obtain a first standard signal; collecting a plurality of second vibration signals of the rotating mechanical equipment in a normal state, performing standardization processing on the plurality of second vibration signals to obtain a second standard signal;

[0008] Performing phase space reconstruction on the first standard signal to obtain a first phase space trajectory matrix; performing spatial reconstruction on the second standard signal to obtain a second phase space trajectory matrix;

[0009] Determining a false neighbor ratio of each spatial sample in the first phase space trajectory matrix; determining a weight corresponding to each spatial sample according to the false neighbor ratio of each spatial sample, and determining a reference sample and a target neighbor sample according to the weight of each spatial sample;

[0010] generating a new fault sample phase space trajectory matrix according to the reference sample and the target neighbor sample, and merging the new fault sample phase space trajectory matrix with the second phase space trajectory to obtain a fault data set;

[0011] The fault data set is used as a pre-built fault diagnosis model for training, and the trained fault diagnosis model is used to perform fault diagnosis on rotating mechanical equipment.

[0012] In the above solution, the first standard signal includes a plurality of signals, and the phase space reconstruction of the first standard signal to obtain a first phase space trajectory matrix includes:

[0013] At different delay times, determining an autocorrelation function value of the first standard signal using an autocorrelation function to obtain a plurality of autocorrelation function values;

[0014] Comparing each of the autocorrelation function values ​​with a preset first threshold, determining a target autocorrelation function value for the autocorrelation function value consistent with the first preset threshold, and taking the delay time corresponding to the target autocorrelation function value as the optimal delay time;

[0015] determining an optimal embedding dimension of the first phase space trajectory matrix;

[0016] The first standard signal is reconstructed in a high-dimensional phase space according to the optimal embedding dimension and the optimal delay time to obtain the first phase space trajectory matrix.

[0017] In the above solution, the determining the autocorrelation function value of the first standard signal by using the autocorrelation function includes:

[0018] Using the formula Determine the autocorrelation function value R(τ) of the first standard signal; wherein,

[0019] N is the total length of the first standard signal, τ is the delay time, t is the sampling point of the first standard signal, x(t) is the tth first standard signal, is the mean of all first standard signals, and σ is the variance of all first standard signals.

[0020] In the above solution, reconstructing the first standard signal in a high-dimensional phase space according to the optimal embedding dimension and the optimal delay time to obtain the first phase space trajectory matrix includes:

[0021] According to the formula Reconstructing the first standard signal in a high-dimensional phase space to obtain the first phase space trajectory matrix T;

[0022] The M is the total number of spatial samples in the first phase space trajectory matrix, and the X M is the Mth spatial sample in the first phase space trajectory matrix, the x(M) is the value of the Mth sampling point of the first standard signal in the time series, the τ′ is the optimal delay time, and the d * is the optimal embedding dimension.

[0023] In the above solution, determining the false neighbor ratio of each spatial sample in the first phase space trajectory matrix includes:

[0024] According to the formula Determine the false neighbor ratio R(X i );

[0025] The X i is the i-th spatial sample in the first phase space trajectory matrix, i=1,2...M; the d * is the optimal embedding dimension, is the i-th spatial sample in the d-dimensional first phase space trajectory matrix, is the d-dimensional first phase space trajectory matrix X i The jth nearest neighbor sample of the i-th spatial sample, k is the number of neighbors of the i-th spatial sample, is the i-th spatial sample in the d+1-dimensional first phase space trajectory matrix, is the d+1-dimensional first phase space trajectory matrix X i+1 The jth nearest neighbor sample, the R th is the preset distance threshold, and II() is an indicator function for indicating when , the output is 1.

[0026] In the above solution, determining the weight corresponding to each spatial sample according to the false neighbor ratio of each spatial sample includes:

[0027] According to the formula Determine the i-th spatial sample X i The weight w(X i );in,

[0028] The R(X i ) is the i-th spatial sample X i The false neighbor ratio of θ is a preset second threshold, and the i-th spatial sample is any spatial sample in the first phase space trajectory matrix.

[0029] In the above solution, determining the reference sample according to the weight of each spatial sample includes:

[0030] According to the formula Determine the i-th spatial sample X i is the first probability P(X i );

[0031] The spatial sample whose first probability is greater than a third threshold is determined as the reference sample; the third threshold is the average value of the first probabilities of all spatial samples; wherein,

[0032] The w(X i ) is the i-th spatial sample X i The weight of M is the total number of spatial samples in the first phase space trajectory matrix.

[0033] In the above solution, determining the target neighbor sample according to the weight of each spatial sample includes:

[0034] Determining a neighbor set of each spatial sample;

[0035] For each neighbor sample in the neighbor set, according to the formula Determine the second probability P(X) of each of the neighbor samples being the target neighbor sample j |X i );

[0036] The neighboring samples whose second probability is greater than a fourth threshold are determined as the target neighboring samples; the fourth threshold is the average value of the second probabilities of all neighboring samples; wherein,

[0037] The w(X j ) is the jth nearest neighbor sample X j The weight of M is the total number of spatial samples in the first phase space trajectory matrix, and N k (Xi ) is the set of neighboring samples of the i-th spatial sample, and k is the total number of the neighboring samples.

[0038] In the above solution, generating a new fault sample phase space trajectory matrix based on the reference sample and the target neighbor sample includes:

[0039] According to the formula X new =X i′ +λ×w(X i′ )×(X j′ -X i′ ) Generate a new fault space sample X in the corresponding position in the first phase space trajectory matrix new , obtain the phase space trajectory matrix of the new fault sample;

[0040] The X i′ is the i′th reference sample, the w(X i′ ) is the weight of the i′th reference sample, λ is the interpolation position correction coefficient, and the value range of λ is [0,1]. j′ is the i′th target neighbor sample.

[0041] A second aspect of the present invention provides a rotating machinery fault diagnosis device based on sample amplification, characterized in that the device comprises:

[0042] a processing unit configured to collect a plurality of first vibration signals of the rotating mechanical device in different fault states, perform standardization processing on the plurality of first vibration signals, and obtain a first standard signal; and collect a plurality of second vibration signals of the rotating mechanical device in a normal state, perform standardization processing on the plurality of second vibration signals, and obtain a second standard signal;

[0043] a reconstruction unit, configured to perform phase space reconstruction on the first standard signal to obtain a first phase space trajectory matrix; and perform spatial reconstruction on the second standard signal to obtain a second phase space trajectory matrix;

[0044] a determination unit, configured to determine a false neighbor ratio of each spatial sample in the first phase space trajectory matrix; determine a weight corresponding to each spatial sample according to the false neighbor ratio of each spatial sample; and determine a reference sample and a target neighbor sample according to the weight of each spatial sample;

[0045] a generating unit, configured to generate a new fault sample phase space trajectory matrix according to the reference sample and the target neighbor sample, and merge the new fault sample phase space trajectory matrix with the second phase space trajectory to obtain a fault data set;

[0046] The training unit is used to train the fault data set as a pre-built fault diagnosis model, and perform fault diagnosis on the rotating mechanical equipment using the trained fault diagnosis model.

[0047] The present invention provides a method and device for diagnosing faults of rotating mechanical equipment based on sample amplification, the method comprising: collecting multiple groups of first vibration signals of the rotating mechanical equipment in different fault states, performing standardization processing on the multiple groups of first vibration signals to obtain first standard signals; collecting multiple groups of second vibration signals of the rotating mechanical equipment in a normal state, performing standardization processing on the multiple groups of second vibration signals to obtain second standard signals; performing phase space reconstruction on the first standard signals to obtain a first phase space trajectory matrix; performing spatial reconstruction on the second standard signals to obtain a second phase space trajectory matrix; determining the false neighbor ratio of each spatial sample in the first phase space trajectory matrix; determining the weight corresponding to each spatial sample according to the false neighbor ratio of each spatial sample, and determining a reference sample and a target neighbor sample according to the weight of each spatial sample; generating a new fault sample phase space trajectory matrix according to the reference sample and the target neighbor sample, and comparing the new fault sample phase space trajectory matrix with the second phase space trajectory matrix The fault data set is used as a pre-built fault diagnosis model for training, and the trained fault diagnosis model is used to diagnose the fault of the rotating mechanical equipment. In this way, by reconstructing the phase space of the first vibration signal, the one-dimensional vibration signal is converted into a high-dimensional vibration signal, thereby enhancing the expression ability of the fault feature. The false neighbor ratio is determined for each spatial sample, and the corresponding weight is determined according to the false neighbor ratio, so that the low-stability samples located in the edge noise area can be given a lower weight, and the high-stability samples can be given a higher weight. The spatial samples with higher weights (reference samples) are used as the reference for data enhancement. When new fault samples are subsequently generated according to the reference samples and the target neighboring samples, it can be ensured that the newly generated fault samples are close to the reference samples, so that the newly generated fault samples are prevented from deviating from the true distribution, thereby improving the reliability of the newly generated fault samples. Due to the increase in the number of fault samples, the imbalance between the fault diagnosis samples and the normal samples can be avoided, thereby improving the accuracy of fault diagnosis of rotating mechanical equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present invention. The same reference symbols are used throughout the drawings to represent the same components. In the drawings:

[0049] Figure 1A schematic flow chart of a method for diagnosing faults of rotating machinery equipment based on sample amplification according to an embodiment of the present invention is shown;

[0050] Figure 2 A schematic diagram showing the results of predicting a test set using a conventional method according to an embodiment of the present invention is shown;

[0051] Figure 3 A schematic diagram showing the results of predicting a test set using a rotating machinery fault diagnosis method based on sample amplification according to an embodiment of the present invention is shown;

[0052] Figure 4 A schematic structural diagram of a rotating mechanical equipment fault diagnosis device based on sample amplification according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0053] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0054] The present invention provides a method for fault diagnosis of rotating mechanical equipment based on sample amplification, such as Figure 1 As shown, the method mainly includes the following steps:

[0055] S110, collecting multiple groups of first vibration signals of the rotating mechanical equipment in different fault states, standardizing the multiple groups of first vibration signals to obtain first standard signals; collecting multiple groups of second vibration signals of the rotating mechanical equipment in a normal state, standardizing the multiple groups of second vibration signals to obtain second standard signals.

[0056] In this step, a vibration sensor or acceleration sensor installed on the rotating mechanical equipment can be used to collect multiple sets of first vibration signals of the rotating mechanical equipment in different fault states and multiple sets of second vibration signals of the rotating mechanical equipment in a normal state.

[0057] The signal lengths of multiple groups of first vibration signals and multiple groups of second vibration signals can be unified by using a zero-padding method or an interpolation method, and the multiple groups of first vibration signals with unified lengths are standardized to obtain first standard signals; and the multiple groups of second vibration signals are standardized to obtain second standard signals; so that the first standard signals and the second standard signals satisfy a standard normal distribution with a mean of 0 and a standard deviation of 1.

[0058] There are a plurality of first standard signals and a plurality of second standard signals.

[0059] S111 , performing phase space reconstruction on the first standard signal to obtain a first phase space trajectory matrix; performing spatial reconstruction on the second standard signal to obtain a second phase space trajectory matrix.

[0060] Since the first standard signal and the second standard signal are one-dimensional vibration signals, in order to improve the expression of fault characteristics, the present invention needs to perform phase space reconstruction on the first standard signal to obtain a first phase space trajectory matrix.

[0061] Similarly, in order to facilitate the subsequent fusion of the amplified fault signal and the normal vibration signal, the second standard signal is also required to perform spatial reconstruction to obtain the second phase space trajectory matrix.

[0062] In the present invention, the method for reconstructing the phase space of the first standard signal and the second standard signal is the same. Here, the phase space reconstruction of the first standard signal is taken as an example for explanation:

[0063] In one embodiment, performing phase space reconstruction on the first standard signal to obtain a first phase space trajectory matrix includes:

[0064] Under different delay times, determining the autocorrelation function value of the first standard signal using the autocorrelation function to obtain a plurality of autocorrelation function values;

[0065] Comparing each autocorrelation function value with a preset first threshold, determining a target autocorrelation function value for an autocorrelation function value that is consistent with the first preset threshold, and taking a delay time corresponding to the target autocorrelation function value as an optimal delay time;

[0066] Determine the optimal embedding dimension of the first phase space trajectory matrix;

[0067] The first standard signal is reconstructed in a high-dimensional phase space according to an optimal embedding dimension and an optimal delay time to obtain a first phase space trajectory matrix.

[0068] In one embodiment, determining the autocorrelation function value of the first standard signal using the autocorrelation function includes:

[0069] Determine the autocorrelation function value R(τ) of the first standard signal using formula (1);

[0070]

[0071] In formula (1), N is the total length of the first standard signal, τ is the delay time, t is the sampling point of the first standard signal, x(t) is the tth first standard signal, is the mean of all first standard signals, and σ is the variance of all first standard signals.

[0072] In one embodiment, reconstructing the first standard signal in a high-dimensional phase space according to the optimal embedding dimension and the optimal delay time to obtain a first phase space trajectory matrix includes:

[0073] According to formula (2), the first standard signal is reconstructed in the high-dimensional phase space to obtain the first phase space trajectory matrix T;

[0074]

[0075] In formula (2), M is the total number of spatial samples in the first phase space trajectory matrix, X M is the Mth spatial sample in the first phase space trajectory matrix, x(M) is the value of the Mth sampling point of the first standard signal in the time series, x(M+τ′) is the value of the M+τ′th sampling point of the first standard signal in the time series, x(M+(d * -1)×τ′) is the M+(d * -1)×τ′ is the value of the sampling points, τ′ is the optimal delay time, d * is the optimal embedding dimension.

[0076] Specifically, since the phase space reconstruction of the first standard signal requires two important parameters, one is the optimal delay time, and the other is the optimal embedding dimension of the matrix.

[0077] The process of determining the optimal delay time is as follows:

[0078] The present invention can pre-set multiple delay times τ, and use formula (1) to calculate the autocorrelation function value of the first standard signal at each delay time. When it is determined that a certain autocorrelation function value is consistent with the first threshold, the autocorrelation function value is determined as the target autocorrelation function value, and the time delay corresponding to the target autocorrelation function value is determined as the optimal delay time τ′. The first threshold can be set according to the specific situation, for example, it can be 1 / e.

[0079] The process of determining the optimal embedding dimension is as follows:

[0080] When initially reconstructing the first phase space trajectory matrix, the optimal dimension is unknown. Therefore, it is necessary to construct the initial first phase space trajectory matrices of different dimensions for the one-dimensional vibration signal based on the initial embedding dimension. This facilitates subsequent calculations of the initial first phase space trajectory matrices of different dimensions to determine the optimal embedding dimension. The initial embedding dimension is generally 3.

[0081] For example, when the embedding dimension is 3, the corresponding initial first phase space trajectory matrix is

[0082]

[0083] When the embedding dimension is 4, the corresponding initial first phase space trajectory matrix is

[0084] By analogy, we can finally obtain multiple high-dimensional first phase space trajectory matrices.

[0085] For the initial first phase space trajectory matrix under different dimensions, the false neighbor ratio of each spatial sample in the initial first phase space matrix is ​​determined, and the mean of the false neighbor ratio of all spatial samples in the initial first phase space matrix is ​​taken. If the mean is lower than the preset false neighbor ratio threshold (which can be 0.01), it means that the dimension is the optimal embedding dimension.

[0086] For example, assuming that when the embedding dimension is 5, the mean of the false neighbor ratio is 0.02; when the embedding dimension is 6, the mean of the false neighbor ratio is 0.005, then the optimal embedding dimension is 6.

[0087] The method for determining the false neighbor ratio may refer to the method for determining the false neighbor ratio in step S112, which will not be described in detail here.

[0088] S112, determining the false neighbor ratio of each spatial sample in the first phase space trajectory matrix, determining the weight corresponding to each spatial sample according to the false neighbor ratio of each spatial sample, and determining the reference sample and the target neighbor sample according to the weight of each spatial sample.

[0089] In order to make the subsequent fault sample amplification more consistent with the actual fault distribution, the present invention needs to determine the false neighbor ratio of each spatial sample in the first phase space trajectory matrix, determine the weight corresponding to each spatial sample based on the false neighbor ratio of each spatial sample, and determine the benchmark sample and target neighbor sample based on the weight of each spatial sample. Since the benchmark sample and target neighbor sample are screened based on weights, the samples with larger weights are determined as the benchmark sample and target neighbor sample, which can improve the reliability of the fault sample. Then, when the fault sample is subsequently amplified based on the benchmark sample and the target neighbor sample, it can also prevent the amplified fault sample from deviating from the actual distribution.

[0090] In one embodiment, determining the false neighbor ratio of each spatial sample in the first phase space trajectory matrix includes:

[0091] According to formula (3), the false neighbor ratio R(X i ):

[0092]

[0093] In formula (3), X iis the i-th spatial sample in the first phase space trajectory matrix, i = 1, 2...M; d * is the optimal embedding dimension, is the i-th spatial sample in the d-dimensional first phase space trajectory matrix, is the d-dimensional first phase space trajectory matrix X i The jth nearest neighbor sample of , k is the number of neighbors of the i-th spatial sample, is the i-th spatial sample in the d+1-dimensional first phase space trajectory matrix, is the d+1-dimensional first phase space trajectory matrix X i+1 The jth nearest neighbor sample, R th is the preset distance threshold, II() is the indicator function, which is used to indicate when When the output is 1, it means X j It's X i False neighbor; otherwise, the output is 0, representing X j It's X i 's real neighbors.

[0094] When selecting the nearest neighbors for the i-th spatial sample, we need to calculate the Euclidean distance between each remaining spatial sample and the i-th spatial sample. We then sort all the Euclidean distances from smallest to largest and select the spatial samples corresponding to the first k Euclidean distances as the nearest neighbors of the i-th spatial sample. The value k can be set based on the actual situation and is not limited here.

[0095] According to formula (3), the false neighbor ratio of each spatial sample in the first phase space trajectory matrix can be determined, and then the weight corresponding to each spatial sample can be determined according to the false neighbor ratio of each spatial sample, including:

[0096] According to formula (4), the i-th spatial sample X is determined i The weight w(X i ):

[0097]

[0098] In formula (4), R(X i ) is the i-th spatial sample X i , θ is a preset second threshold, and the i-th spatial sample is any spatial sample in the first phase space trajectory matrix.

[0099] For example, assuming the second threshold is 0.6, if the i-th spatial sample X i If the false neighbor ratio is greater than 0.6, it means that the i-th spatial sample X i Unreliable, low stability sample, X iThe weight is reduced to 1-0.6=0.4. If the i-th spatial sample X i If the false neighbor ratio is less than or equal to 0.6, it means that the i-th spatial sample X i Reliable, for high stability samples, then X i The weight is directly set to 1.

[0100] After the weight of each spatial sample is determined, the baseline sample and the target neighbor sample must be determined based on the weight of each spatial sample, so that when new fault samples are subsequently interpolated based on the baseline sample and the target neighbor sample, the newly generated fault samples can be more consistent with the actual fault distribution.

[0101] In one embodiment, determining a reference sample according to the weight of each spatial sample includes:

[0102] According to formula (5), the i-th spatial sample X is determined i is the first probability P(X i ):

[0103]

[0104] The spatial samples whose first probability is greater than the third threshold are determined as reference samples; the third threshold is the average value of the first probabilities of all spatial samples; wherein,

[0105] w(X i ) is the i-th spatial sample X i The weight of , M is the total number of spatial samples in the first phase space trajectory matrix.

[0106] Specifically, after calculating the first probability of each spatial sample, the average of the first probabilities of all spatial samples can be determined and used as the third threshold. Then, spatial samples with a first probability greater than the third threshold can be determined as reference samples.

[0107] Similarly, the target neighbor samples are determined according to the weight of each spatial sample, including:

[0108] Determine the neighbor set of each spatial sample;

[0109] For each neighbor sample in the neighbor set, the second probability P(X) of each neighbor sample being the target neighbor sample is determined according to formula (6). j |X i ):

[0110]

[0111] In formula (6), the neighbor samples whose second probability is greater than the fourth threshold are determined as target neighbor samples; the fourth threshold is the average value of the second probabilities of all neighbor samples; where,

[0112] w(X j ) is the jth nearest neighbor sample X j The weight of M is the total number of spatial samples in the first phase space trajectory matrix, N k (X i ) is the set of neighboring samples of the i-th spatial sample, and k is the total number of neighboring samples.

[0113] In this way, the benchmark sample and the target neighbor sample are determined. In this embodiment, the benchmark sample is recorded as S113. A new fault sample phase space trajectory matrix is ​​generated based on the benchmark sample and the target neighbor sample. The new fault sample phase space trajectory matrix and the second phase space trajectory are merged to obtain a fault data set.

[0114] Then, a new fault sample phase space trajectory matrix can be generated according to the benchmark samples and the target neighbor samples, and the new fault sample phase space trajectory matrix and the second phase space trajectory can be merged to obtain the fault data set.

[0115] In one embodiment, generating a new fault sample phase space trajectory matrix based on the reference sample and the target neighbor sample includes:

[0116] According to formula (7), a new fault space sample X is generated in the corresponding position in the first phase space trajectory matrix new , and obtain the phase space trajectory matrix of the new fault sample:

[0117] X new =X i′ +λ×w(X i′ )×(X j′ -X i′ ) (7)

[0118] In formula (7), X i′ is the i′th benchmark sample, w(X i′ ) is the weight of the i′th benchmark sample, λ is the interpolation position correction coefficient, which is used to control the interpolation position. The value range of λ is [0,1]. j′ is the i′th target neighbor sample.

[0119] Weight w(X i′ ) is used to dynamically adjust the interpolation amplitude. The larger the weight, the closer the generated new fault sample is to the benchmark sample, avoiding deviation from the actual fault distribution. j′ -X i′ Represents the difference vector between the target neighbor sample and the reference sample, which is used to determine the generation direction of the new fault sample.

[0120] In order to detect whether the new fault sample is located at the edge of the distribution or in the noise region, after generating the new fault sample, the method further includes:

[0121] According to formula (8), the false neighbor ratio R(X new ):

[0122]

[0123] In formula (8), X new New fault sample;d * is the optimal embedding dimension, is the new fault sample in the d-dimensional first phase space trajectory matrix, is the j′th nearest neighbor sample of the new fault sample in the d-dimensional first phase space trajectory matrix, k′ is the number of neighbors of the new fault sample, is the new fault sample in the d+1-dimensional first phase space trajectory matrix, is the j′th neighbor sample of the new fault sample in the d+1-dimensional first phase space trajectory matrix, R th is the preset distance threshold, II() is the indicator function, which is used to indicate when , the output is 1.

[0124] Then, the weight of the new fault sample is updated according to the false neighbor ratio of the new fault sample, including:

[0125] According to formula (9), the weight of the new fault sample is updated to obtain the weight w of the new fault sample updated (X new ):

[0126]

[0127] In formula (9), w(X i ) is the weight of the i′th benchmark sample, w(X′ j ) is the weight of the j′th target neighbor sample, Represents the weight of the parent sample inherited by the new fault sample, (1-R(X new )) represents the modification of the corresponding weight according to the reliability of the new fault sample. Then the above steps of generating new fault samples are repeated until the number of fault samples reaches the expected number (for example, the same number as the normal samples). Then the present invention calculates the weight by (1-R(X new)) generation is used to correct the weight of the new fault sample. In the subsequent process of repeatedly generating new fault samples, the sample generation process can be adaptively adjusted according to the weight of the new fault sample (for example, new fault samples with small weights will not be determined as benchmark samples), so that the generated samples are more consistent with the actual fault distribution and the reliability of the entire data set is improved.

[0128] After the new fault sample phase space trajectory matrix is ​​determined, it is merged with the second phase space trajectory to obtain a fault dataset. Because the fault dataset contains the same number of faulty and normal samples, sample balance is ensured. Furthermore, after training the pre-built fault diagnosis model using the fault dataset, even with a small sample size, the model can still accurately identify fault characteristics, improving fault diagnosis accuracy.

[0129] S114 , training the fault data set as a pre-built fault diagnosis model, and performing fault diagnosis on the rotating mechanical equipment using the trained fault diagnosis model.

[0130] After the fault data set is determined, the fault data set is used as a pre-built fault diagnosis model for training, and the trained fault diagnosis model is used to perform fault diagnosis on the rotating mechanical equipment.

[0131] The pre-built fault diagnosis model may be a Gaussian Process Regression (GRP) model, or another model, such as a neural network.

[0132] The Gaussian process regression model primarily involves selecting an appropriate kernel function and setting initial values ​​for hyperparameters, thereby determining the prior model in the form of a probability distribution. The covariance function in the Gaussian regression model is the central moment of the random output variable corresponding to two random input points in space. It measures the degree of similarity or correlation between different samples and is a key factor influencing the prediction performance of the GPR model.

[0133] Assume that the fault data set is D, each fault space sample in the fault data set has a corresponding fault label, and assume that the GRP model obeys Gaussian distribution, then the improved weighted covariance function K w (X a ,X b ) can be defined as:

[0134]

[0135] In formula (10), w(X a ) is the weight of any sample in the fault data set, w(X b ) Sample X in the fault data set b The weight of Xb For X a The nearest neighbor sample of ; σ f is the signal variance hyperparameter, which is used to control the overall amplitude of the covariance function. l is the length scale parameter, which is used to determine the smoothness of feature changes. a -X b || represents X b and X a The Euclidean distance is used to measure the similarity between two samples; Represents the noise variance, which is used to measure sensor measurement error or signal disturbance through the Kronecker delta function δ ab (1 when a=b, 0 otherwise) Ensure that the diagonal elements of the covariance matrix contain noise corrections to avoid matrix singularity problems.

[0136] It can be seen that the model suppresses the interference of edge or noise samples by adjusting the weights of samples, and uses the probability of Gaussian process to output quantitative prediction results, thereby improving the accuracy of fault diagnosis.

[0137] In order to further prove that the accuracy of the fault diagnosis method provided by the present invention is reliable, the present invention compares and illustrates a common fault diagnosis method and the fault diagnosis method of the present invention:

[0138] Vibration signals of different states are collected by the acceleration sensor. Since the normal state time of the equipment is often much longer than the abnormal state time, the number of abnormal state samples is much less than the number of normal samples, resulting in the problem of data sample imbalance between different state samples. The experimental data includes 100 normal state samples and 10 fault state samples. The length of each sample is 256. The normal state samples remain unchanged at 100, and the fault state samples are expanded from 10 to 100 to balance the prediction results of the normal state sample classifier. Taking the confusion matrix after model convergence as an example, the number of normal state and fault state samples in the test set remains the same, and the training set and test set are divided according to the ratio of 7:3. Figure 2 As shown, the accuracy on the test set is 93.3%.

[0139] When using the method of the present invention to generate fault samples, the fault samples and normal state samples are balanced, and then input into the Gaussian process regression model for classification. Figure 3 As shown in the figure, the model can accurately predict the fault status samples without misclassification, and the accuracy of the test set can reach 100%.

[0140] The present invention reconstructs the phase space of the first vibration signal to convert the one-dimensional vibration signal into a high-dimensional vibration signal, thereby enhancing the expressive ability of the fault characteristics; and determines the false neighbor ratio for each spatial sample, and then determines the corresponding weight according to the false neighbor ratio, so that low-stability samples located in the edge noise area can be given lower weights, and high-stability samples can be given higher weights. The spatial samples with higher weights (reference samples) are used as the benchmark for data enhancement. When new fault samples are subsequently generated according to the benchmark samples and the target neighboring samples, it can be ensured that the newly generated fault samples are close to the benchmark samples, and the newly generated fault samples are prevented from deviating from the true distribution, thereby improving the reliability of the newly generated fault samples; due to the increase in the number of fault samples, the imbalance between fault diagnosis samples and normal samples can be avoided, thereby improving the accuracy of fault diagnosis of rotating mechanical equipment.

[0141] Based on the same inventive concept as in the above embodiment, this embodiment also provides a rotating mechanical equipment fault diagnosis device based on sample amplification, such as Figure 4 As shown, the device includes:

[0142] The processing unit 41 is configured to collect multiple sets of first vibration signals of the rotating mechanical device in different fault states, perform standardization processing on the multiple sets of first vibration signals, and obtain first standard signals; collect multiple sets of second vibration signals of the rotating mechanical device in a normal state, perform standardization processing on the multiple sets of second vibration signals, and obtain second standard signals;

[0143] The reconstruction unit 42 is configured to perform phase space reconstruction on the first standard signal to obtain a first phase space trajectory matrix; and perform spatial reconstruction on the second standard signal to obtain a second phase space trajectory matrix;

[0144] a determination unit 43 configured to determine a false neighbor ratio of each spatial sample in the first phase space trajectory matrix; determine a weight corresponding to each spatial sample based on the false neighbor ratio of each spatial sample; and determine a reference sample and a target neighbor sample based on the weight of each spatial sample;

[0145] A generating unit 44 is configured to generate a new fault sample phase space trajectory matrix based on the reference sample and the target neighbor sample, and merge the new fault sample phase space trajectory matrix with the second phase space trajectory to obtain a fault data set;

[0146] The training unit 45 is configured to train the fault data set as a pre-built fault diagnosis model, and perform fault diagnosis on the rotating mechanical equipment using the trained fault diagnosis model.

[0147] Since the device described in the embodiments of the present invention is used to implement the method for diagnosing rotating mechanical equipment faults based on sample amplification according to the embodiments of the present invention, the specific structure and variations of the device are readily understood by those skilled in the art based on the methods described in the embodiments of the present invention, and therefore, no further details are given here. All devices used in the methods according to the embodiments of the present invention fall within the scope of protection of the present invention.

[0148] Through one or more embodiments of the present invention, the present invention has the following beneficial effects or advantages:

[0149] The present invention provides a method and device for diagnosing faults of rotating mechanical equipment based on sample augmentation, the method comprising: collecting multiple groups of first vibration signals of the rotating mechanical equipment in different fault states, performing standardization processing on the multiple groups of first vibration signals to obtain first standard signals; collecting multiple groups of second vibration signals of the rotating mechanical equipment in a normal state, performing standardization processing on the multiple groups of second vibration signals to obtain second standard signals; performing phase space reconstruction on the first standard signal to obtain a first phase space trajectory matrix; performing spatial reconstruction on the second standard signal to obtain a second phase space trajectory matrix; determining the false neighbor ratio of each spatial sample in the first phase space trajectory matrix; determining the weight corresponding to each spatial sample according to the false neighbor ratio of each spatial sample, and determining a reference sample and a target neighbor sample according to the weight of each spatial sample; generating a new fault sample phase space trajectory matrix according to the reference sample and the target neighbor sample, and comparing the new fault sample phase space trajectory matrix with the second phase space trajectory matrix. The trajectories are merged to obtain a fault data set; the fault data set is used as a pre-built fault diagnosis model for training, and the trained fault diagnosis model is used to diagnose the fault of the rotating mechanical equipment; in this way, by reconstructing the phase space of the first vibration signal, the one-dimensional vibration signal is converted into a high-dimensional vibration signal, thereby enhancing the expression ability of the fault feature; and the false neighbor ratio is determined for each spatial sample, and then the corresponding weight is determined according to the false neighbor ratio, so that the low-stability samples located in the edge noise area can be given a lower weight, and the high-stability samples can be given a higher weight, and the spatial samples with higher weights (reference samples) are used as the reference for data enhancement. When new fault samples are subsequently generated according to the reference samples and the target neighboring samples, it can be ensured that the newly generated fault samples are close to the reference samples, so that the newly generated fault samples are prevented from deviating from the true distribution, thereby improving the reliability of the newly generated fault samples; due to the increase in the number of fault samples, the imbalance between the fault diagnosis samples and the normal samples can be avoided, thereby improving the accuracy of fault diagnosis of rotating mechanical equipment.

[0150] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0151] The above description is only a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for fault diagnosis of rotating mechanical equipment based on sample amplification, characterized in that: The method comprises: collecting a plurality of first vibration signals of the rotating mechanical equipment in different fault states, performing standardization processing on the plurality of first vibration signals to obtain a first standard signal; collecting a plurality of second vibration signals of the rotating mechanical equipment in a normal state, performing standardization processing on the plurality of second vibration signals to obtain a second standard signal; Performing phase space reconstruction on the first standard signal to obtain a first phase space trajectory matrix; performing spatial reconstruction on the second standard signal to obtain a second phase space trajectory matrix; Determining a false neighbor ratio of each spatial sample in the first phase space trajectory matrix; determining a weight corresponding to each spatial sample according to the false neighbor ratio of each spatial sample, and determining a reference sample and a target neighbor sample according to the weight of each spatial sample; generating a new fault sample phase space trajectory matrix according to the reference sample and the target neighbor sample, and merging the new fault sample phase space trajectory matrix with the second phase space trajectory to obtain a fault data set; The fault data set is used as a pre-built fault diagnosis model for training, and the trained fault diagnosis model is used to perform fault diagnosis on rotating mechanical equipment.

2. The method according to claim 1, wherein The first standard signals include a plurality of signals, and the phase space reconstruction of the first standard signals to obtain a first phase space trajectory matrix includes: At different delay times, determining an autocorrelation function value of the first standard signal using an autocorrelation function to obtain a plurality of autocorrelation function values; Comparing each of the autocorrelation function values ​​with a preset first threshold, determining a target autocorrelation function value for the autocorrelation function value consistent with the first preset threshold, and taking the delay time corresponding to the target autocorrelation function value as the optimal delay time; determining an optimal embedding dimension of the first phase space trajectory matrix; The first standard signal is reconstructed in a high-dimensional phase space according to the optimal embedding dimension and the optimal delay time to obtain the first phase space trajectory matrix.

3. The method according to claim 2, wherein The determining the autocorrelation function value of the first standard signal by using the autocorrelation function includes: Using the formula Determine the autocorrelation function value R(τ) of the first standard signal; wherein, N is the total length of the first standard signal, τ is the delay time, t is the sampling point of the first standard signal, x(t) is the tth first standard signal, is the mean of all first standard signals, and σ is the variance of all first standard signals.

4. The method according to claim 2, wherein The reconstructing the first standard signal in a high-dimensional phase space according to the optimal embedding dimension and the optimal delay time to obtain the first phase space trajectory matrix includes: According to the formula Reconstructing the first standard signal in a high-dimensional phase space to obtain the first phase space trajectory matrix T; The M is the total number of spatial samples in the first phase space trajectory matrix, and the X M is the Mth spatial sample in the first phase space trajectory matrix, the x(M) is the value of the Mth sampling point of the first standard signal in the time series, the τ′ is the optimal delay time, and the d * is the optimal embedding dimension.

5. The method according to claim 1, wherein Determining the false neighbor ratio of each spatial sample in the first phase space trajectory matrix includes: According to the formula Determine the false neighbor ratio R(X i ); The X i is the i-th spatial sample in the first phase space trajectory matrix, i=1,2...M; the d * is the optimal embedding dimension, is the i-th spatial sample in the d-dimensional first phase space trajectory matrix, is the d-dimensional first phase space trajectory matrix X i The jth nearest neighbor sample of the i-th spatial sample, k is the number of neighbors of the i-th spatial sample, is the i-th spatial sample in the d+1-dimensional first phase space trajectory matrix, is the d+1-dimensional first phase space trajectory matrix X i+1 The jth nearest neighbor sample, the R th is the preset distance threshold, and II() is an indicator function for indicating when , the output is 1.

6. The method according to claim 1, wherein The determining the weight corresponding to each spatial sample according to the false neighbor ratio of each spatial sample includes: According to the formula Determine the i-th spatial sample X i The weight w(X i );in, The R(X i ) is the i-th spatial sample X i The false neighbor ratio of θ is a preset second threshold, and the i-th spatial sample is any spatial sample in the first phase space trajectory matrix.

7. The method according to claim 1, wherein The determining of the reference sample according to the weight of each spatial sample includes: According to the formula Determine the i-th spatial sample X i is the first probability P(X i ); The spatial sample whose first probability is greater than a third threshold is determined as the reference sample; the third threshold is the average value of the first probabilities of all spatial samples; wherein, The w(X i ) is the i-th spatial sample X i The weight of M is the total number of spatial samples in the first phase space trajectory matrix.

8. The method according to claim 1, wherein The determining of the target neighbor sample according to the weight of each spatial sample includes: Determining a neighbor set of each spatial sample; For each neighbor sample in the neighbor set, according to the formula Determine the second probability P(X) of each of the neighbor samples being the target neighbor sample j |X i ); The neighboring samples whose second probability is greater than a fourth threshold are determined as the target neighboring samples; the fourth threshold is the average value of the second probabilities of all neighboring samples; wherein, The w(X j ) is the jth nearest neighbor sample X j The weight of M is the total number of spatial samples in the first phase space trajectory matrix, and N k (X i ) is the set of neighboring samples of the i-th spatial sample, and k is the total number of the neighboring samples.

9. The method according to claim 1, wherein Generating a new fault sample phase space trajectory matrix according to the reference sample and the target neighbor sample includes: According to the formula X new =X i′ +λ×w(X i′ )×(X j′ -X i′ ) Generate a new fault space sample X in the corresponding position in the first phase space trajectory matrix new , obtain the phase space trajectory matrix of the new fault sample; The X i′ is the i′th reference sample, the w(X i′ ) is the weight of the i′th reference sample, λ is the interpolation position correction coefficient, and the value range of λ is [0,1]. j′ is the i′th target neighbor sample.

10. A rotating mechanical equipment fault diagnosis device based on sample amplification, characterized in that: The device comprises: a processing unit configured to collect a plurality of first vibration signals of the rotating mechanical device in different fault states, perform standardization processing on the plurality of first vibration signals, and obtain a first standard signal; and collect a plurality of second vibration signals of the rotating mechanical device in a normal state, perform standardization processing on the plurality of second vibration signals, and obtain a second standard signal; a reconstruction unit, configured to perform phase space reconstruction on the first standard signal to obtain a first phase space trajectory matrix; and perform spatial reconstruction on the second standard signal to obtain a second phase space trajectory matrix; a determination unit, configured to determine a false neighbor ratio of each spatial sample in the first phase space trajectory matrix; determine a weight corresponding to each spatial sample according to the false neighbor ratio of each spatial sample; and determine a reference sample and a target neighbor sample according to the weight of each spatial sample; a generating unit, configured to generate a new fault sample phase space trajectory matrix according to the reference sample and the target neighbor sample, and merge the new fault sample phase space trajectory matrix with the second phase space trajectory to obtain a fault data set; The training unit is used to train the fault data set as a pre-built fault diagnosis model, and perform fault diagnosis on the rotating mechanical equipment using the trained fault diagnosis model.

Images