A data amplification method, a rotating machinery fault diagnosis method and a device

Abstract

The present invention discloses a data amplification method, a rotating machinery fault diagnosis method and a device. In the case of limited fault state data, small sample data is effectively amplified while maintaining sufficient similarity and diversity. When the amplified data is applied to rotating machinery fault diagnosis, the accuracy of the diagnosis result is greatly improved. The data amplification method includes the following steps: collecting the fault signal of the rotating machinery; arbitrarily intercepting continuous data points of length N from the fault signal as the original finite fault signal x; constructing a random measurement matrix, projecting the original finite fault signal from a high-dimensional space to a low-dimensional space to obtain a variety of compressed signals y; reconstructing the various compressed signals y to obtain sparse coefficients; performing inverse Fourier transform on the sparse coefficients to obtain a variety of reconstructed signals, that is, the amplified samples are composed of the various reconstructed signals.

Description

Technical Field

[0001] The present invention relates to the technical field of mechanical fault diagnosis, and in particular to a data amplification method, a rotating machinery fault diagnosis method and a device. Background Art

[0002] In modern industry, rotating machinery is becoming increasingly complex and intelligent. Monitoring its operating status can effectively avoid significant losses caused by mechanical failures. Therefore, it is of great significance to study effective rotating machinery fault diagnosis methods.

[0003] However, in actual applications, machines are usually operated under healthy working conditions. Once a fault occurs, the machine will usually be shut down immediately to protect life and property, reduce maintenance costs, and avoid more serious accidents. Coupled with the limitations of sensor installation or operating conditions, the collected fault status data is limited and cannot be compensated by fault reproduction. In addition, due to the limitations of transmission channel bandwidth, transmission rate and computing power, it is also possible that only a very small number of collected signals are in a fault state. In the case of limited fault status data, there will be further problems such as omission of fault status data or loss during transmission, which ultimately leads to very limited fault status data being obtained. Traditional rotating machinery fault diagnosis methods are difficult to solve the problem of limited fault status data, which seriously affects the accuracy of mechanical equipment fault diagnosis.

[0004] In order to solve the problem of limited fault status data, data enhancement (or data amplification) strategies are widely used. Among them, Generative Adversarial Network (GAN) is a widely used data amplification method. However, the disadvantage of GAN is that it is difficult to train, and its ability to generate new data is limited when the number of fault samples is limited. Specifically, GAN requires a long training time and rich training data to make the generated data distribution closer to the real data distribution. Therefore, it is necessary to study more effective data amplification methods to overcome the shortcomings of existing methods. Furthermore, data amplification methods are applied to rotating machinery fault diagnosis to solve the problem of low fault diagnosis accuracy. Summary of the Invention

[0005] The purpose of the present invention is to provide a data amplification method, a rotating machinery fault diagnosis method and a device, which can effectively amplify small sample data while maintaining sufficient similarity and diversity under limited fault status data; when the amplified data is applied to rotating machinery fault diagnosis, the accuracy of the diagnosis results is greatly improved.

[0006] In a first aspect, the present invention provides a data augmentation method, comprising the following steps:

[0007] S10. Collecting fault signals of rotating machinery;

[0008] S11. arbitrarily extracting continuous data points of length N from the fault signal as the original finite fault signal x;

[0009] S12. construct a random measurement matrix, and use the random measurement matrix to project the original finite fault signal x from the high-dimensional space to the low-dimensional space to obtain a variety of compressed signals y;

[0010] S13. Reconstruct the various compressed signals y to obtain sparse coefficients

[0011] S14. Sparse coefficients Perform inverse Fourier transform to obtain various reconstructed signals That is, the various reconstructed signals Constitute the amplified sample.

[0012] Compared with the prior art, the present invention obtains diverse amplified samples based on the compressed sampling strategy and data reconstruction strategy under the compressed sensing framework. Specifically, because the measurement matrix of each compressed sampling is randomly constructed, the randomly constructed measurement matrices are different from each other, that is, they have diversity, so that the compressed signal y also has diversity. It needs to be further explained that the above-mentioned random measurement matrix is ​​not completely random. The random measurement matrix of the compressed sampling process can retain the main characteristic information of the original fault signal to effectively avoid the reconstruction of the signal. Distortion. In addition, due to the diversity of the compressed signal y, the reconstructed signal obtained by the reconstruction algorithm will also be diverse. Moreover, since the accuracy of each signal reconstruction is also different, the diversity of signal reconstruction is further ensured. The data amplification method provided by the present invention can achieve diverse data amplification based on the sampling strategy and the reconstruction strategy, while also ensuring that the obtained amplified data has a high similarity with the original fault signal in the frequency domain, and can retain the main characteristic information of the original fault signal. In other words, by adopting the data amplification method provided by the present invention, small sample data can be effectively amplified while maintaining sufficient similarity and diversity. Furthermore, the data amplification method provided by the present invention requires less training data and lower computational complexity to achieve data amplification. Further, when the diverse amplified samples are applied to fault diagnosis of rotating machinery, while ensuring the accuracy of fault diagnosis, it also has the advantages of less required training data and low computational complexity.

[0013] In a second aspect, the present invention further provides a method for diagnosing a fault in a rotating machine, comprising the following steps:

[0014] S20. Using the data amplification method provided in the first aspect and / or any one of the implementations of the first aspect to obtain an amplified sample of the limited monitoring data of the faulty component included in the rotating machinery i is the number of amplified samples;

[0015] S21. All amplified samples All inputs are used for model training in the convolutional neural network, and the trained model is tested using real fault signal samples to obtain diagnostic results.

[0016] Compared with the prior art, the beneficial effects of the rotating machinery fault diagnosis method provided by the present invention are the same as the beneficial effects of the data amplification method provided by the first aspect and / or any implementation of the first aspect, and are not elaborated here.

[0017] In a third aspect, the present invention further provides a rotating machinery fault diagnosis device, characterized in that it comprises:

[0018] A signal acquisition device for acquiring fault signals of rotating machinery;

[0019] A signal processing device includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the following steps can be performed:

[0020] S20. Using the data amplification method provided in the first aspect and / or any one of the implementations of the first aspect to obtain an amplified sample of the limited monitoring data of the faulty component included in the rotating machinery i is the number of amplified samples;

[0021] S21. All amplified samples All inputs are used for model training in the convolutional neural network, and the trained model is tested using real fault signal samples to obtain diagnostic results.

[0022] Compared with the prior art, the beneficial effects of the rotating machinery fault diagnosis device provided by the present invention are the same as the beneficial effects of the data amplification method provided by the first aspect and / or any implementation of the first aspect, and are not elaborated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0024] Figure 1 is a flow chart of a data augmentation method provided by an embodiment of the present invention;

[0025] Figure 2Schematic diagram of the deep convolutional neural network structure provided by an embodiment of the present invention;

[0026] Figure 3 This is a comparison result of the classification accuracy of the data augmentation method provided by the embodiment of the present invention and the traditional data augmentation method;

[0027] Figure 4 This is a comparison result of the visualization confusion matrix of the classification results of the data augmentation method provided by the embodiment of the present invention and the traditional data augmentation method;

[0028] Figure 5 is the probability distribution comparison result of the original data and the generated data provided by the embodiment of the present invention;

[0029] Figure 6 is a comparison result of the original sample envelope spectrum and the generated sample envelope spectrum provided by an embodiment of the present invention;

[0030] Figure 7 is the Pearson correlation coefficient between the original sample envelope spectrum and all generated sample envelope spectra provided by the embodiment of the present invention;

[0031] Figure 8 This is a flow chart of a rotating machinery fault diagnosis method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0032] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0033] It should be noted that when an element is referred to as being “fixed on” or “disposed on” another element, it may be directly on the other element or indirectly on the other element. When an element is referred to as being “connected to” another element, it may be directly connected to the other element or indirectly connected to the other element.

[0034] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Thus, a feature specified as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, "plurality" means two or more, unless otherwise specifically defined. "Several" means one or more, unless otherwise specifically defined.

[0035] In the description of the present invention, it should be understood that the terms "up", "down", "front", "back", "left", "right", etc., indicating directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as a limitation on the present invention.

[0036] In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "mounted," "connected," and "connected" should be understood in a broad sense. For example, they may refer to fixed, detachable, or integral connections; mechanical or electrical connections; direct or indirect connections through an intermediate medium; and may encompass internal communication between two components or interaction between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0037] The data augmentation method provided in an embodiment of the present invention is aimed at fault diagnosis tasks where rotating machinery fault sample data is scarce. The method uses compressed sensing technology to amplify the original small sample data, thereby increasing the scale and diversity of fault samples while retaining the main feature information of the original samples, and improving the fault diagnosis accuracy of deep learning algorithms using data-driven modeling.

[0038] In practical applications, taking a rotating machine like a bearing as an example, faulty components of a bearing include the outer ring, inner ring, rolling element, and cage. In this case, the same accelerometer can be used to collect fault signals from the outer ring, inner ring, and rolling element at different time intervals. Specifically, an accelerometer and an NI data acquisition card can be used to acquire the acceleration of these different bearing components in real time at different time intervals. When the acceleration changes, a vibration signal with a certain vibration frequency is obtained as the fault signal. After signal acquisition is completed, vibration signals of the outer ring, inner ring, and rolling element can be obtained. Based on this, the data amplification method provided in an embodiment of the present invention can be further used to perform data amplification on the outer ring vibration signals, inner ring vibration signals, and rolling element vibration signals to obtain diverse and large-scale amplified samples. In this case, a convolutional neural network can be used as input to perform bearing fault diagnosis and obtain diagnostic results.

[0039] Of course, in addition to the vibration signal mentioned above, bearing fault sample data can also include other signals such as sound signals, bearing temperature signals, oil temperature signals, and metal chip particle counts. Specifically, a microphone sensor can be used to collect sound signals, a non-contact sensor such as an infrared temperature sensor can be used to collect bearing temperature signals, a contact sensor such as a thermocouple sensor can be used to collect oil temperature signals, and a metal chip particle count signal can be collected using a metal chip particle signal sensor. The fault signal described below is any one of a vibration signal, a sound signal, a bearing temperature signal, an oil temperature signal, or a metal chip particle count signal. Accordingly, the original finite fault signal is any one of an original finite vibration signal, an original finite sound signal, an original finite bearing temperature signal, an original finite oil temperature signal, or an original finite metal chip particle count signal.

[0040] See also Figure 1 The data augmentation method provided by the embodiment of the present invention may specifically include the following steps:

[0041] S10. Collecting fault signals of the rotating machinery. As an example, an acceleration sensor and an NI data acquisition card can be used to collect vibration signals of the rotating machinery in a fault state.

[0042] S11. Randomly intercept continuous data points of length N from the fault signal as the original finite fault signal x. That is, the original finite fault signal x in the embodiment of the present invention is [x1, x2, x3, ..., x N ] T , usually the number of data points N intercepted is not less than the number of points contained in two rotations of the faulty component to retain sufficient operating status information.

[0043] S12. Construct a random measurement matrix to project the original finite fault signal from high-dimensional space to low-dimensional space to obtain a variety of compressed signals y. As previously mentioned, during the signal compression sampling phase, the data augmentation provided by the embodiments of the present invention is based on the sampling strategy of compressed sensing technology and can be achieved by constructing a variety of random measurement matrices, such as Gaussian random measurement matrices, Bernoulli random measurement matrices, and sparse random measurement matrices.

[0044] Because the Gaussian random measurement matrix can better satisfy the restricted isometry property (Restricted IsometryProperty, referred to as RIP), in view of this, as a preferred method, the Gaussian random measurement matrix can be used as the measurement matrix for compressed sampling. At this time, various Gaussian random measurement matrices Φ∈R can be constructed M×N , the original finite fault signal is projected from the high-dimensional space to the low-dimensional space through the projection operation y = Φx to obtain a variety of compressed signals y, y = [y1, y2, y3, ..., y M ]T , the compressed signal y retains the main characteristic information of the original fault signal, and the obtained compressed signals y are not consistent and have diversity. Specifically, because the Gaussian measurement matrix of each compressed sampling is randomly constructed, the Gaussian measurement matrices are different from each other and have diversity, thereby making the compressed observation value y diverse. It should be further explained that the above randomness is not completely random. This random measurement matrix in the compressed sampling process retains the main characteristic information of the original signal. Through this operation, the present invention realizes diversity data amplification based on the sampling strategy.

[0045] The compression sampling rate in the process of projecting the original finite fault signal x from the high-dimensional space to the low-dimensional space to obtain the diverse compressed signals y is CSR. Where M is the dimension of the compressed signal, and N is the dimension of the original finite fault signal. The quality of the amplified data is closely related to the size of the CSR. To ensure that the amplified sample retains the main characteristic information of the original sample, it is necessary to select an appropriate CSR. Specifically, based on the fidelity of the amplified sample to the original sample and the diagnostic accuracy, a reasonable CSR value can be determined within a range of 0.1 to 1, for example, CSR = 0.1, CSR = 0.4, CSR = 0.5, CSR = 0.6, CSR = 0.7, CSR = 0.8, CSR = 0.9, and CSR = 1.

[0046] S13. Reconstruct the various compressed signals y to obtain sparse coefficients Methods for reconstructing compressed signals mainly include optimization algorithms and greedy algorithms. The Orthogonal Matching Pursuit (OMP) algorithm is a classic iterative algorithm among greedy algorithms and can quickly achieve signal reconstruction.

[0047] As a preferred embodiment, the present invention may adopt OMP as a signal reconstruction method, which may specifically include the following steps:

[0048] Input: M×1 dimensional compressed measurement value y, M×N dimensional sensor matrix A = ΦΨ, signal sparsity k;

[0049] S130. Initialization: r0 = y, t = 1;

[0050] S131. Find index λ t , such that:

[0051] S132. Let Λ t =Λ t-1 ∪{λ t},

[0052] S133. Find y = A t θ t The least squares solution of :

[0053] S134. Update residual

[0054] S135.t=t+1, if t≤2k, return to S131, otherwise stop iteration and go to S136;

[0055] S136. Reconstruction In Λ t There are non-zero entries at the last iteration.

[0056] Output: sparse coefficients N×1 dimensional residual

[0057] Among them, r t represents the residual, t represents the number of iterations, represents the empty set, Λ t represents the index set of t iterations, λ t represents the index of the tth iteration, a j The jth column of matrix A, A t Indicates by index Λ t The selected set of columns of matrix A.

[0058] S14. Perform inverse Fourier transform on the sparse coefficients to obtain various reconstructed signals That is, the various reconstructed signals The amplification sample is formed, which specifically includes the following steps:

[0059]

[0060] where Ψ is the Fourier dictionary,

[0061] As mentioned above, in the signal reconstruction stage, the data amplification provided by the embodiment of the present invention is a data amplification based on the reconstruction strategy under the compressed sensing technology, which can be achieved by reconstructing a variety of compressed signals. Specifically, due to the diversity of the compressed signal y, the reconstructed signal obtained by the reconstruction algorithm will also be diverse. Moreover, there will be certain errors in the signal reconstruction itself. This difference in signal reconstruction accuracy further ensures the diversity of signal reconstruction. Through this operation, the present invention realizes diverse data amplification based on the reconstruction strategy. That is, by adopting the data amplification method provided by the present invention, small sample data can be effectively amplified and maintain sufficient similarity and diversity.

[0062] Continuing with the bearing as an example, the above steps S10 to S14 can be used to first obtain an amplified sample of outer race fault data, and then S10 to S14 can be repeated to obtain an amplified sample of inner race fault data and an amplified sample of rolling element fault data in sequence.

[0063] See also Figure 8 Based on the data amplification method provided in the first embodiment, in a second aspect, the present invention further provides a rotating machinery fault diagnosis method, comprising the following steps:

[0064] S20. Using the data amplification method provided in the first aspect and / or any one of the implementations of the first aspect to obtain an amplified sample of the limited monitoring data of the faulty component included in the rotating machinery i is the number of amplified samples;

[0065] S21. All amplified samples All inputs to the convolutional neural network are used for model training. The trained model is then tested using real fault signal samples to obtain diagnostic results. The constructed deep convolutional neural network primarily consists of an input layer, a convolutional layer, a pooling layer, a fully connected layer, and a softmax output layer. In practical applications, the constructed deep convolutional neural network is trained using all the augmented operating status data generated in the previous steps. The trained model is then verified and tested using actual operating data collected in S10 to obtain diagnostic results.

[0066] (1) Analysis of diagnostic results

[0067] In order to verify the effectiveness of the rotating machinery fault diagnosis method provided by the embodiment of the present invention relative to the prior art, three comparative examples corresponding to the embodiment provided by the present invention are set. The amplified samples obtained in the embodiment and the comparative example are input into the same convolutional neural network to obtain the diagnosis results respectively. The diagnosis results are then compared and analyzed from the aspects of the average value and standard deviation of the classification accuracy. Among them, the three comparative examples (Comparative Example 1, Comparative Example 2 and Comparative Example 3) respectively adopt GAN, Variational Auto-Encoder (VAE), and Long Short-Term Memory-Variational Auto-Encoder (LSTM-VAE).

[0068] Table 1 Drive end bearing parameters

[0069]

[0070] Table 2 Parameters of the constructed deep convolutional neural network

[0071]

[0072] The above-mentioned embodiment, comparative example 1, comparative example 2 and comparative example 3 are all comparisons and analyses of the Case Western Reserve University bearing data set. The above-mentioned bearing data (public data) comes from the Case Western Reserve University bearing test bench, which is mainly composed of a motor, a torque sensor, a power tester, a drive-end bearing, an acceleration sensor, etc. The above-mentioned bearing data set comes from the acceleration data of the drive-end bearing. Among them, the acceleration data of the drive-end bearing is acquired by the acceleration sensor. The drive-end bearing model is SKF6205 deep groove ball bearing, and the bearing parameters are shown in Table 1. The bearing fault is a single-point damage implant fault using electrospark machining. The fault diameters are 0.007 inches, 0.014 inches and 0.021 inches respectively. The motor speed is 1797 rpm. The vibration acceleration signal of the bearing is collected by an acceleration sensor. The sampling frequency is 12kHz. The motor load is 0 horsepower.

[0073] Combine Figure 2 ,The parameter settings of the constructed deep convolutional neural network are shown in Table 2.

[0074] Example:

[0075] For each type of fault data (for example, the inner ring fault data, outer ring fault data, and rolling element fault data of the drive-end bearing), a signal segment of 3000 continuous data points is taken as a finite vibration signal, the compression sampling rate CSR = 0.4, and the data amplification method provided by the present invention is used to generate 700 amplified samples as the training set of the deep convolutional neural network.

[0076] Comparative Example 1: GAN

[0077] For each fault type, a segment of 3,000 continuous data points was used as a finite vibration signal. A GAN was used to generate 700 amplified samples as a training set for the deep convolutional neural network. During GAN training, to improve the quality of the generated data, the 3,000 data points were divided into 15 training runs, with 200 data points per run. After training, the model was saved and used to generate 700 samples. This also reflects the difficulty of GAN training.

[0078] Comparative Example 2: VAE

[0079] For each fault type data, a signal with a length of 3000 continuous data points is taken as the finite vibration signal, and 700 amplified samples are generated by VAE as the training set of the deep convolutional neural network.

[0080] Comparative Example 3: LSTM-VAE

[0081] For each fault type data, a finite vibration signal with a length of 3000 continuous data points was taken. LSTM-VAE was constructed by combining the LSTM unit with the advantage of processing time series data. LSTM-VAE was used to generate 700 amplified samples as the training set of the deep convolutional neural network.

[0082] It should be further explained that in order to further increase the diversity among the above training set samples, in each training set sample, a point is randomly selected from [0,500] as the starting point, and 2048 consecutive data points are taken to form the actual training sample. The validation set samples and test set samples are derived from real measured data, and each sample also contains 2048 data points. The ratio of the training set, validation set, and test set is [0.7:0.2:0.1]. That is, the total number of samples in the training set, validation set, and test set are 7000, 2000, and 1000 respectively. In order to avoid the randomness of the experimental results, a total of 15 repeated experiments were carried out, see Figure 3 .

[0083] Table 3 compares the average and standard deviation of the classification accuracy of the four methods in 15 experiments. The comparison results are shown in Table 3.

[0084] from Figure 3 As shown in Table 3, among the four data augmentation methods, the proposed method performs best, with an average accuracy of 95.74%. The GAN-based data augmentation method performs second best, with an average classification accuracy of 90.74%. The LSTM-VAE-based data augmentation method has an average diagnostic accuracy of 69.81%. The VAE-based data augmentation method has a diagnostic accuracy of only 67.20%.

[0085] Table 3 Comparison of diagnostic performance of four data augmentation methods

[0086]

[0087] Figure 4 The visualization confusion matrix of one of the experiments of the four data augmentation methods in Table 3 is shown respectively. Figure 4 In (a), only 22% of the test samples in B21_222 were misdiagnosed as B14_185, and the diagnostic accuracy of the remaining categories was above 96%. It is worth noting that B14_185 and B21_222 are both bearing rolling element faults, but the fault diameter is different. Figure 4 In (b), B14_185, OR21_234, and B21_222 were all seriously misclassified, and the remaining categories were also slightly misclassified. This shows that the quality of the data generated by the GAN data augmentation method is significantly lower than that of the data generated by the method proposed in this invention. Figure 4In (c) and (d), as many as five categories are seriously misclassified, indicating that the quality of data generated by LSTM-VAE and VAE is poor.

[0088] The above examples clearly demonstrate that the rotating machinery fault diagnosis method and apparatus, based on the data amplification method provided in the first aspect of the present invention, can effectively address the limited data available for bearing fault diagnosis. Using the generated high-quality data as a training set for the fault recognition model helps achieve excellent fault diagnosis accuracy.

[0089] (2) Feature similarity analysis of amplified samples

[0090] To further verify the advantages of the data amplification method proposed in the present invention, the feature similarity of the amplified samples will be analyzed. To avoid redundant analysis, this embodiment only selects the running status data numbered 1, 2, 4, 8, and 10 in Table 1 for analysis.

[0091] In terms of data augmentation, the generated data should conform to the distribution of the original data. Figure 5 As shown in the figure, taking the Normal and IR7_105 operating states as examples, the data generated using the proposed method conforms to the probability distribution of the original data. This reflects the effectiveness of the proposed method from the perspective of data distribution.

[0092] For data amplification for the purpose of fault identification, the original data does not need to be completely reconstructed. In other words, the reconstructed signal does not need to be completely consistent with the original signal. It is only necessary for the generated sample to retain the main characteristic information of the original sample. It is worth noting that the envelope spectrum of the signal contains rich information about the fault characteristic frequency. In order to further evaluate the quality of the generated data, the present invention analyzes the envelope spectrum information of the original data and the generated data, and proposes an indicator of the mean Pearson correlation coefficient (MPCC), which can quickly evaluate the overall correlation between the envelope spectrum of all generated samples and the envelope spectrum of the original sample. The derivation process of MPCC is as follows:

[0093]

[0094]

[0095] PCC is the Pearson correlation coefficient, x f 、 Represent the original sample x and the generated sample respectively The envelope spectrum of Represents x f The average value of The average value, Represents the envelope spectrum of the jth generated sample. The range of MPCC is from -1 to 1. The closer the MPCC is to -1 or 1, the more f and The stronger the overall correlation is, the more the generated sample contains the main feature information of the original sample; the closer the MPCC is to 0, that is, x f and The worse the overall correlation is, the more the generated sample loses the main feature information contained in the original sample.

[0096] Table 4 Average Pearson correlation coefficient (MPCC) between original samples and generated samples

[0097]

[0098] Figure 6 The envelope spectra of the original samples of five bearing states are compared with the envelope spectra of one of the generated samples. It can be found that the envelope spectra of the generated samples are very similar to those of the original samples. The generated samples successfully capture the main characteristic frequencies of the original samples. Figure 6 、 Figure 7 As can be seen from the information in Table 4, both PCC and MPCC are around 0.9, indicating that the generated samples have a high correlation with the original samples in terms of envelope spectrum. This high correlation further demonstrates that the generated samples retain the main characteristic information of the original samples, verifying the effectiveness of the proposed method.

[0099] In a third aspect, an embodiment of the present invention further provides a rotating machinery fault diagnosis device, characterized by comprising:

[0100] A signal acquisition device for acquiring fault signals of rotating machinery;

[0101] A signal processing device includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the following steps can be performed:

[0102] S20. Using the data amplification method provided in the first aspect and / or any one of the implementations of the first aspect to obtain an amplified sample of the limited monitoring data of the faulty component included in the rotating machinery i is the number of amplified samples;

[0103] S21. All amplified samples All inputs are used for model training in the convolutional neural network, and the trained model is tested using real fault signal samples to obtain diagnostic results.

[0104] Compared with the prior art, the beneficial effects of the rotating machinery fault diagnosis device provided by the present invention are the same as the beneficial effects of the data amplification method provided by the first aspect and / or any implementation of the first aspect, and are not elaborated here.

[0105] In a fourth aspect, an embodiment of the present invention further provides an electronic device, comprising a processor and a communication interface, wherein the communication interface and the processor are coupled, and the processor is used to run a computer program or instruction to implement the rotating machinery fault diagnosis method described in the second aspect.

[0106] In a fifth aspect, an embodiment of the present invention further provides a computer-readable storage medium, in which instructions are stored. When the instructions are executed, the rotating machinery fault diagnosis method described in the second aspect is implemented.

[0107] In the description of the above embodiments, specific features, structures, materials or characteristics may be combined in an appropriate manner in any one or more embodiments or examples.

[0108] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A data augmentation method, characterized in that: The following steps are involved: S10. Collecting fault signals of rotating machinery; S11. Randomly extracting N consecutive data points from the fault signal as the original finite fault signal ; S12. Construct a random measurement matrix and transform the original finite fault signal Projecting from high-dimensional space to low-dimensional space to obtain diverse compressed signals ; S13. The various compressed signals Perform signal reconstruction to obtain sparse coefficients ; The orthogonal matching pursuit algorithm is used to reconstruct the various compressed signals to obtain sparse coefficients. The following steps are involved: enter: dimensional compression measurement , dimensional sensor matrix , the sparsity of the signal ; S130. Initialization: , , , ; S131. Find the index , such that: ; S132. Order ; S133. Ask The least squares solution of : ; S134. Update residual ; S135. ,if Then return to S131, otherwise stop the iteration and go to S136; S136. Reconstruction exist There are non-zero entries at the last iteration. ; Output: sparse coefficients , dimensional residual ; in, represents the residual, represents the number of iterations, represents the empty set, represents the index set of t iterations, represents the index of the t-th iteration, matrix The jth column of Indicates by index Selected matrix The column set of For the sparse coefficient Perform inverse Fourier transform , obtain diverse reconstructed signals , , the reconstructed signal It is to amplify the sample, including: ； in, is the Fourier dictionary, ; S14. For the sparse coefficient Perform inverse Fourier transform to obtain various reconstructed signals , that is, the diverse reconstructed signals Constitute the amplified sample.

2. The data augmentation method according to claim 1, wherein: The original finite fault signal is , , N is not less than the number of points included in two rotations of the faulty component included in the rotating machinery.

3. The data augmentation method according to claim 1, wherein: The random measurement matrix is ​​any one of a Gaussian random measurement matrix, a Bernoulli random measurement matrix or a sparse random measurement matrix.

4. The data augmentation method according to claim 3, wherein: In the case where the random measurement matrix is ​​a Gaussian random measurement matrix, the original finite fault signal Projecting from high-dimensional space to low-dimensional space to obtain diverse compressed signals ,include: Constructing diverse Gaussian random measurement matrices ; Through projection operation The original finite fault signal Projecting from high-dimensional space to low-dimensional space to obtain diverse compressed signals , , where M is the dimension of the compressed signal and N is the dimension of the original finite fault signal.

5. The data augmentation method according to claim 4, characterized in that: The original finite fault signal Projecting from high-dimensional space to low-dimensional space to obtain diverse compressed signals The compression sampling rate of the process is CSR, , where M is the dimension of the compressed signal and N is the dimension of the original finite fault signal.

6. The data augmentation method according to claim 5, characterized in that: 0.1≤CSR≤1.

7. A method for diagnosing a rotating machinery fault, characterized in that: The steps include: S20. Obtaining an amplified sample of limited monitoring data of a faulty component included in a rotating machine using the data amplification method according to any one of claims 1 to 6. , is the number of amplified samples; S21. Input all amplified samples into the convolutional neural network for model training. Use real fault signal samples to test the trained model and obtain diagnostic results.

8. A rotating machinery fault diagnosis device, characterized in that: include: A signal acquisition device for acquiring fault signals of rotating machinery; A signal processing device includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the following steps can be performed: S20. Obtaining an amplified sample of limited monitoring data of a faulty component included in a rotating machine using the data amplification method according to any one of claims 1 to 6. , is the number of amplified samples; S21. Input all amplified samples into the convolutional neural network for model training. Use real fault signal samples to test the trained model and obtain diagnostic results.

Images