Fan gear box fault diagnosis method and system based on spEEMD and double-index screening

Abstract

The application discloses a fan gear box fault diagnosis method and system based on spEEMD and double-index screening, and comprises the following steps: obtaining vibration signals of a healthy or faulty fan gear box; improving an EEMD method by using a single-parameter strategy to obtain a series of intrinsic mode components; screening the intrinsic mode function components by using a double-index screening strategy to obtain target components; taking the target components as input data to establish a convolutional neural network model, and finally obtaining an intelligent diagnosis model of the fan gear box fault combined with a fusion layer; and using the obtained model to perform fault recognition and diagnosis on vibration signals of a wind generator gear box. The method improves the EEMD method by using the single-parameter strategy, improves the adaptability and decomposition precision of the EEMD method, obtains target components according to the double-index screening strategy, and realizes accurate intelligent fault diagnosis.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of wind turbine gearbox fault diagnosis, and particularly relates to a wind turbine gearbox fault diagnosis method and system based on an spEEMD method and a double-index screening strategy. BACKGROUND

[0002] Wind turbines work in extreme environments such as random wind speed, high load and strong impact for a long time, and the gears, bearings and other parts are prone to failure. The downtime caused by gearbox failure accounts for the largest share of all other component downtimes, so it is of great significance to detect the vibration of the wind turbine gearbox. The gearbox vibration signal collected by the vibration sensor is usually a nonlinear and non-stationary time series data, and often interfered by noise and other factors. The traditional signal processing method is difficult to obtain ideal feature extraction effect, thereby reducing the accuracy of fault diagnosis.

[0003] Ensemble Empirical Mode Decomposition (EEMD) is an adaptive signal processing method proposed for the problem that Empirical Mode Decomposition (EMD) performs poorly in the noise scene. EEMD improves the uniform distribution characteristics of the signal in the time-frequency space by adding white noise to the signal, and removes the final noise by ensemble averaging method, and finally decomposes a series of intrinsic mode components (IMF) containing signals of different frequency bands. EEMD has been widely applied to the fault diagnosis of various parts of various gearboxes.

[0004] The EEMD method introduces two very important parameters that need to be determined by the user: noise amplitude and integration times. The inventors found that the EEMD method is very sensitive to the change of noise amplitude. When the amplitude of the added white noise is too small, the noise is difficult to improve the extreme value distribution of the initial signal. When the amplitude of the added white noise is too large, the noise will destroy the extreme value distribution of the low frequency component, resulting in modal aliasing phenomenon. The current improvement methods have their limitations, and they are lacking in signal decomposition accuracy, adaptability to different working conditions, etc.

[0005] A fan gearbox fault diagnosis method based on improved EEMD and spectrogram analysis is disclosed in patent CN113657244A, which includes the following steps: obtaining the vibration signal of the fan gearbox in a healthy or fault state; optimizing the EEMD method by using the tsPSO method to obtain a series of intrinsic mode function components; screening the intrinsic mode function components and performing spectrogram analysis on the selected components to obtain a fault feature spectrogram; inputting the fault feature spectrogram into a convolutional neural network model to obtain an intelligent diagnosis model of the fan gearbox fault; and using the obtained model to identify and diagnose the vibration signal of the wind turbine gearbox; however, this patent is susceptible to noise amplitude parameters, has poor adaptability, the decomposed IMF components are not effectively screened, all decomposed IMF components are used for analysis, the calculation time is long, and the calculation efficiency is low.

[0006] After the signal is decomposed, due to the instability and uncertainty of the number of decomposition components, the randomness of noise, and modal aliasing, there is no unified standard and solution for how to select target components and how many target components to select. For online fault diagnosis and condition monitoring of fan gearboxes under complex working conditions, if the sub-signals containing the health state characteristics are selected less or missed or noise or interference signals are selected by mistake in the target component selection process, the diagnosis result of the system will be strongly affected, and accurate and timely identification of faults cannot be achieved.

[0007] In summary, the inventors found that if the appropriate Gaussian noise amplitude is not set when the traditional EEMD method is used to process the vibration signal of the wind turbine gearbox collected by the vibration sensor, the decomposition precision of EEMD will be affected, and then the accuracy and comprehensiveness of target component selection will be affected. In addition, the selection of target components still largely depends on the experience of the user. If the sub-signals containing the health state characteristics are selected less or missed or noise or interference signals are selected by mistake, the fault diagnosis result will be greatly affected. SUMMARY

[0008] To solve the above problems, the first aspect of the present application provides a fan gearbox fault diagnosis method based on spEEMD and a double-index screening strategy. The method is based on an improved EEMD method, in which the Gaussian noise with a specific amplitude in the original EEMD method is replaced by Gaussian noise without a specific noise amplitude, which can improve the distribution of signal extrema. This can reduce the influence of noise randomness on the decomposition result, reduce the occurrence of modal aliasing to a certain extent, improve the robustness and adaptability of the method, and improve the fault diagnosis accuracy.

[0009] To achieve the above purpose, the first aspect of the present application adopts the following technical solutions:

[0010] A fan gearbox fault diagnosis method based on spEEMD and double-index screening strategy, comprising:

[0011] Collecting vibration signals of fan gearbox faults occurring in actual industrial operation scenarios;

[0012] The spEEMD method is used to decompose the fan gearbox vibration signal into multiple IMF components. Different groups are directly combined by having the same number of IMF components, and groups with different numbers of IMF components are selected for IMF component combination.

[0013] Select IMF components containing sufficient fault feature information as target IMF components through a double-index screening strategy;

[0014] A convolutional neural network model is established using the target IMF components as input data, and an intelligent fault diagnosis model for fan gearbox faults is finally obtained. The obtained model is used for fault recognition and diagnosis of wind turbine gearbox vibration signals.

[0015] As a further technical solution, the steps of the spEEMD method (i.e. single parameter strategy improved EEMD method) are as follows:

[0016] Add a random Gaussian noise sequence to the original signal to obtain a noise signal;

[0017] Evaluate the extreme value distribution characteristics of the noise signal through the extreme value improvement degree index F value. If the noise signal meets the conditions, it will pass the screening and enter the decomposition calculation;

[0018] Decompose the noise signal to obtain a series of IMF components.

[0019] As a further technical solution, the step of selecting IMF components containing sufficient fault feature information as target IMF components through a double-index screening strategy is as follows:

[0020] Calculate the correlation coefficient between each IMF component and the original signal, sort the correlation coefficients and select the IMF component with the highest correlation coefficient as the initial component;

[0021] Calculate the envelope signal of the initial component, and according to the power spectral density, combine the IMF components with larger correlation with the initial component;

[0022] Re-select the initial component, and according to the corresponding power spectral density, combine the final components into a series of new components, and select 1-2 combined components as target IMF components according to the original component sorting.

[0023] As a further technical solution, the process of establishing an intelligent fault diagnosis model is as follows:

[0024] A convolutional neural network model is established, using the selected vibration signals as input. After adaptive feature extraction through fusion, convolution, and pooling operations, fully connected layers are used to classify the features. Finally, the model parameters are initialized, and the established dataset is fed into the model for training. The training accuracy and cross-entropy loss function are used to measure the model's performance. When the model tends to converge and the accuracy reaches above 96%, the model is considered complete.

[0025] Secondly, the present invention also provides a wind turbine gearbox fault diagnosis system, as follows:

[0026] The vibration signal acquisition module is used to acquire vibration signals of a fan gearbox failure in actual industrial operation scenarios.

[0027] The vibration signal decomposition module is used to improve the EEMD method using a single-parameter strategy to obtain a series of final IMF components.

[0028] Target component filtering module: It uses the spEEMD method to decompose the vibration signal of the wind turbine gearbox into multiple IMF components, and performs IMF component merging by directly merging the same number of IMF components and selecting a merging strategy for different numbers of IMF components.

[0029] Fault classification module: It is used to establish a convolutional neural network model with the target component as input, and finally obtain an intelligent fault diagnosis model for wind turbine gearbox faults. The obtained model is used to identify and diagnose the vibration signal of the wind turbine gearbox.

[0030] As a further technical solution, the target component filtering module includes:

[0031] The initial component selection module is used to calculate the correlation coefficient between each IMF component and the original signal, sort the correlation coefficients, and select the IMF component with the highest correlation coefficient as the initial component.

[0032] The component merging module is used to calculate the envelope signal of the initial component and merge the remaining IMF components with the initial component based on the power spectral density.

[0033] The target component determination module reselects the initial components, merges them according to the corresponding power spectral density, and finally merges them into a series of new components. Based on the original component sorting, it selects 1-2 merged components as the target IMF components.

[0034] As a further technical solution, the intelligent fault diagnosis model includes:

[0035] The first submodule is used to build a convolutional neural network model. It takes the selected vibration signal as input, and after adaptive feature extraction through fusion, convolution and pooling operations, it uses a fully connected layer to classify the features.

[0036] The second submodule is used to initialize model parameters, feed the established dataset into the model for training, and use the training accuracy and cross-entropy loss function to measure the model's performance. The model is considered complete when it tends to converge and the model's accuracy reaches 96% or higher.

[0037] Thirdly, the present invention also provides a computer, including a memory and a processor, and computer instructions stored in the memory and running on the processor, wherein the computer instructions, when executed by the processor, complete the steps in the aforementioned wind turbine gearbox fault diagnosis method.

[0038] Fourthly, the present invention also provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the steps in the aforementioned wind turbine planetary gearbox fault diagnosis method.

[0039] The advantages of this invention are:

[0040] (1) In view of the problems that the EEMD method is not easy to select a suitable Gaussian noise amplitude and the poor decomposition effect is easily caused by an inappropriate selection, this embodiment proposes the spEEMD method. This method replaces the Gaussian noise with a specific amplitude in the original EEMD method with Gaussian noise that can improve the distribution of signal extrema but does not have a specific noise amplitude. After the improvement, only one parameter, the number of integrations, needs to be selected, thereby improving the adaptability and decomposition accuracy of the EEMD method.

[0041] (2) More accurate vibration signal decomposition results can be obtained by using the spEEMD method, and the combination of spEEMD and dual index screening strategy can fully extract signals with rich fault feature information.

[0042] (3) A convolutional neural network was used to extract and classify the vibration signals obtained by screening, and a smart fault diagnosis model for wind turbine gearbox with sufficient accuracy and diagnostic efficiency was established. Attached Figure Description

[0043] Figure 1 This is a flowchart of a wind turbine gearbox fault diagnosis method based on spEEMD and a dual-index screening strategy according to an embodiment of the present invention.

[0044] Figure 2 This is a flowchart of the spEEMD method according to an embodiment of the present invention.

[0045] Figure 3This is a time-domain diagram of the simulated signal in an embodiment of the present invention.

[0046] Figure 4 This is the simulation signal decomposition result based on the spEEMD method in an embodiment of the present invention.

[0047] Figure 5 This is the simulation signal decomposition result based on the EWT method in an embodiment of the present invention.

[0048] Figure 6 This is the simulation signal decomposition result based on the SGMD method in this embodiment of the invention.

[0049] Figure 7 This is an example diagram of the intelligent fault diagnosis model according to an embodiment of the present invention.

[0050] Figures 8(a) and 8(b) are curves showing the training process of the intelligent fault diagnosis model according to an embodiment of the present invention. Detailed Implementation

[0051] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0052] As mentioned in the background section, if the Gaussian noise amplitude added to the EEMD method is inappropriate, it will affect the EEMD decomposition accuracy, and consequently the accuracy and comprehensiveness of the target IMF component, making it impossible to obtain ideal vibration signal screening results. This embodiment proposes a method and system for improving EEMD using a single-parameter strategy to solve the above problems. The wind turbine gearbox fault diagnosis method based on spEEMD and dual-index screening proposed in this embodiment includes: collecting vibration signals of wind turbine gearbox faults occurring in actual industrial operation scenarios; decomposing the wind turbine gearbox vibration signal into multiple IMF components using the spEEMD method, and merging IMF components by directly merging the same number of IMF components and by selecting a merging strategy for different numbers of IMF components; selecting the IMF component containing sufficient fault feature information as the target IMF component through a dual-index screening strategy. It should be noted that: sufficient information here means that the fault features are obvious, the interference noise is low, and there is fault information that can clearly determine the fault type; insufficient information means that the fault features are not obvious, there are many interference frequencies and noise, it is impossible to clearly determine the fault type, and multiple interference quantities are merged together. Using the target IMF component as input data, a convolutional neural network model is established, ultimately yielding an intelligent diagnostic model for wind turbine gearbox faults. This model is then used to identify and diagnose faults in the vibration signals of the wind turbine gearbox. The invention will be described in detail below with reference to the accompanying drawings:

[0053] Example 1

[0054] This embodiment discloses a wind turbine gearbox fault diagnosis method based on spEEMD and a dual-index screening strategy, which will be described in detail below:

[0055] First, Empirical Mode Decomposition (EMD) adaptively decomposes a signal into a series of steady-state and linear signals, called IMF components, by fitting the upper and lower envelopes of the signal and removing its local mean. Ensemble Empirical Mode Decomposition (EEMD) is a method that improves mode mixing by adding Gaussian noise to the original input signal.

[0056] The EEMD method flow is as follows:

[0057] 1. First, add an amplitude of A to the original signal x(t). N White noise sequence n i (t), to obtain the signal x after adding noise. i ′(t) can be expressed as:

[0058] x′ i (t)=n i (t)+x(t) (1)

[0059] For ease of discussion, the noise amplitude A will be used here. N Further represented as A N =L N ×σ0, where L N It is a dimensionless noise level coefficient, and σ0 is the standard deviation of the original signal.

[0060] 2. Next, use EMD to extract the noise signal x′ i (t) is decomposed into a series of IMF components C ij (t) and residual component r i (t), where C ij (t) is the j-th IMF component generated by the i-th decomposition. This step can be expressed as:

[0061]

[0062] 3. Repeat steps 1 and 2 N times with different noise sequences, where N is the number of integration attempts set manually;

[0063] 4. Calculate the average value of the corresponding IMF components as the final decomposition result, where C ij (t) represents the j IMF components obtained after EEMD decomposition.

[0064] 5. Finally, the decomposition result of EEMD can be expressed as follows:

[0065]

[0066] Where x(t) is the original signal, C j r(t) is the final IMF component, and r(t) is the final residual component.

[0067] While EEMD suppresses the mode mixing problem in the EMD method, it introduces a new parameter selection problem: noise amplitude and ensemble count, two key parameters, affect the decomposition performance of the EEMD method. Existing methods for enhancing EEMD include empirical selection, principle extension, and parameter optimization. However, these methods are lacking in accuracy, adaptability to different operating conditions, and computational efficiency, making them unsuitable for real-time online condition monitoring under complex operating conditions. To reduce the impact of inappropriate parameter selection on the EEMD method and improve its adaptability and decomposition capability, this invention utilizes a single-parameter strategy to improve the EEMD method.

[0068] The decomposition effect of EEMD is affected by the noise amplitude A N The two key parameters, integration number N, have a significant impact. The single-parameter strategy considers replacing Gaussian noise with a specific amplitude added to the original signal with Gaussian noise without a specific amplitude. Therefore, when using EEMD, it is no longer necessary to set the key noise amplitude parameter A. N Since only one parameter needs to be considered—the number of integrations—the method for improving EEMD using a single-parameter strategy is called spEEMD (single parameter EEMD).

[0069] The single-parameter strategy considers adding Gaussian noise without a specific amplitude. Through random noise generation and filtering mechanisms, it ensures the signal decomposition result without setting a noise amplitude. Random noise generation involves adding Gaussian noise with random amplitudes during the EEMD ensemble computation. Simultaneously, the spEEMD method uses a filtering mechanism to ensure that the added noise improves the signal's extreme value distribution. If the randomly generated noise from spEEMD fails to improve the extreme value distribution, a new noise sequence is generated until it passes the filtering judgment.

[0070] In order to screen for noise that can improve the extreme value distribution of a signal, spEEMD uses formula (4) to evaluate the signal after noise is added. By comparing the evaluation value with that of the initial signal, it can be determined whether the noise can improve the extreme value distribution of the signal.

[0071]

[0072] Where x′(t) is the signal after noise has been added, x′ max(t) and x′ min(t)The maximum and minimum values ​​of x′(t) are respectively, and σ 2 This is the variance operator; t(x) and p(x) are the actual and ideal extreme value sequences of x′(t) on the x-axis, respectively. max (x) is the sequence of actual maxima on the x-axis, t min (x) is the actual minimum sequence on the x-axis, p max (x), p min Similarly, L1 and L2 are the lengths of the x-axis maximum and minimum value sequences, respectively.

[0073] In formula (4), F1 measures the distribution and fluctuation characteristics of signal extrema, and can assess the uniformity of noise on the signal's extrema in the y-axis direction. F2 represents the uniform distribution characteristics of the extrema sequence in the x-axis direction, and can measure the improvement level of noise on the signal's extrema in the x-axis direction. Considering that the uniform distribution characteristics in the x-axis direction are more in line with the purpose of adding noise to EEMD, the two are weighted and combined as the final extrema improvement measure F.

[0074] During the screening process, all F values ​​will be recorded. Only when the added noise causes the F value of the signal to exceed the average value of the previous noise signal F will the noise signal pass the screening and enter the decomposition calculation. This judgment condition is expressed as formula (5).

[0075]

[0076] Where F i Let f be the measure of the degree of improvement of the extreme value when noise is added for the i-th time, and let f be the sequence of F values ​​of the noise signal that meets the screening results. Its initial value is the F value of the original noise-free signal x(t).

[0077] The EEMD method is improved based on the above single-parameter strategy, enabling it to adaptively cope with a variety of complex working conditions and application scenarios.

[0078] Secondly, after decomposition using the spEEMD method, due to the unstable and uncertain number of decomposed components, the randomness of noise, and mode aliasing, there is no unified standard or solution for how to select components and how many components to select. Component selection relies heavily on user experience. However, for online fault diagnosis and condition monitoring of wind turbine gearboxes under complex operating conditions, if too few or nocturnal signals containing health status characteristics are selected, or if noise or interference signals are mistakenly selected, the system's diagnostic results will be severely affected, making accurate and timely fault identification impossible. This invention utilizes a dual-index screening strategy to improve the accuracy of selecting target components after decomposition, avoiding the problems of missed or incorrect selection in traditional selection methods.

[0079] The first indicator in the dual-index screening strategy is the correlation coefficient cc, which is used to evaluate the degree of correlation between the IMFs component and the original signal. The calculation formula is shown in (6).

[0080]

[0081] Here, x(t) and y(t) are the time-domain signals to be processed, and x and y are their means, respectively. If the correlation coefficient cc is closer to 1, then the IMF component obtained after EEMD decomposition is considered to be closer to the original signal components.

[0082] The second indicator in the dual-indicator screening strategy is the power spectral density (PSD). PSD is one of the main means and characteristics for describing the statistical regularity of a random process or signal x(t) from the frequency perspective. PSD can represent the power distribution or energy status of a random process or signal x(t) at different frequencies. By evaluating the correlation of power spectral density between different IMF components, IMF components with similar health status characteristics can be effectively merged.

[0083] The steps of the dual-index screening strategy based on correlation coefficient and power spectral density are as follows:

[0084] 1. For multiple IMFs components, calculate the correlation coefficient cc between each IMF component and the original signal S. i The correlation coefficients are sorted, and the IMF component with the highest correlation coefficient is selected as the initial component, denoted as .

[0085] 2. Calculation envelope signal The calculation formula is shown in (7). power spectral density At the same time, it is normalized to its maximum and minimum values, and the calculation formula is shown in (8);

[0086] E a =|IMF a +jH(IMF a (7)

[0087]

[0088] 3. Evaluation and The degree of correlation between the normalized power spectral density of the other component envelope signals, and all related to Related components and The components are merged to obtain the preliminarily filtered IMF components, denoted as sIMF. i ;

[0089] 4. Select the component with the highest correlation coefficient to the original signal from the remaining IMF components as... Repeat steps 2-3 until all IMF components have been evaluated and screened, ultimately resulting in multiple sIMFs. i (selected IMF) selects 1-2 sIMF components by default, sorted from high to low according to the original component correlation coefficients, to complete the target component selection;

[0090] Figure 1 This is a flowchart of a wind turbine gearbox fault diagnosis method based on spEEMD and a dual-index screening strategy, according to an embodiment of the present invention. Figure 1 As shown, the wind turbine gearbox fault diagnosis method based on spEEMD and a dual-index screening strategy is as follows:

[0091] S101: Collect vibration signals of the wind turbine gearbox for health or fault conditions.

[0092] S102: The vibration signal is decomposed using the EEMD method improved by the single-parameter strategy described above to obtain a series of IMF components. The specific implementation process is as follows: Figure 2 As shown:

[0093] S1021: Set the spEEMD method parameter: number of integrations N.

[0094] S1022: Add a random Gaussian noise sequence to the original signal x(t) to obtain the noise signal x′(t). Use formula (4) to evaluate the extreme value distribution characteristics of the noise signal x′(t) and obtain the extreme value improvement index F. The screening ends when the noise signal x′(t) meets the screening conditions of formula (5).

[0095] S1023: Use EMD to decompose the noise signal x′(t) that meets the screening criteria to obtain a series of IMF components. Decompose N noise signals to obtain N sets of IMF components.

[0096] S1024: Merge N groups of IMFs components. If there are different numbers of IMFs components, merge them separately and then use PSD to select and merge them.

[0097] To verify the effectiveness of this method, the simulated signal shown in Equation (9) was selected for comparison and evaluation of three methods: spEEMD, Empirical Wavelet Transform (EWT), and Symptotic Geometry Mode Decomposition (SGMD). The sampling frequency was 2560Hz, and the sampling time was 1.5 seconds. The time-domain waveform is shown in [Figure 9]. Figure 3Where (a) is the periodic impact component, (b) is the modulated signal component, (c) is the vibration signal component during normal operation, and (d) and (e) are the composite signals before and after the addition of noise, respectively.

[0098]

[0099] Where y(t) is a mixed simulation signal, y1(t) represents the periodic impact component caused by device failure, y2(t) is the vibration signal component generated by normal gear meshing, y3(t) is the vibration signal component generated by the normal operation of components such as rotor and shaft, and n(t) represents the noise component.

[0100] The spEEMD method was used to process and decompose the simulated signal into five IMF components, such as... Figure 4 As shown in the figure, the two IMF components with relatively large correlation coefficients (0.958 and 0.993 respectively), IMF4 and IMF5, completely separated the y2(t) and y3(t) components. The time-domain characteristics of the original signal components can be clearly seen from the decomposition results.

[0101] The EWT method was used to process the simulated signal. Parameters were set empirically based on reference data, and the signal was decomposed into five EMF components, such as... Figure 5 As shown in the figure, the two EMF components EMF2 and EMF4 with relatively large correlation coefficients (0.982 and 0.932 respectively) separated the y3(t) and y2(t) components more completely.

[0102] The SGMD method was used to process the simulated signal. Parameters were set empirically based on reference data, and the signal was decomposed into three SGC components, such as... Figure 6 As shown in the figure, the two components SGC1 and SGC2 with relatively large correlation coefficients (1.000 and 0.942 respectively) effectively extracted the original signal y3(t) and y2(t) components.

[0103] Comparing the decomposition results of the spEEMD, EWT, and SGMD methods on the simulated signal, it can be found that the proposed spEEMD method has better robustness in noisy environments and can more effectively handle fault feature extraction tasks in noisy scenarios. Furthermore, spEEMD only requires manual setting of the relatively unimportant parameter of the number of integration iterations, thus exhibiting stronger adaptability.

[0104] S103: The target component is accurately selected from the decomposition results using the dual-index screening strategy based on correlation coefficient and power spectral density as described above.

[0105] S104: Using the target component as input data, a convolutional neural network is established, and finally a fault intelligent diagnosis model combining the fusion layer is obtained.

[0106] In this embodiment, the accelerometer is mounted on the gearbox housing near the output shaft, while the speed sensor directly acquires the rotational speed of the gearbox output shaft, with sampling frequencies of 25600Hz and 30Hz, respectively. The accelerometer and speed sensor acquire the vibration signals and corresponding speed signals generated when the intermediate speed stage gear operates during the break-tooth fault.

[0107] First, the spEEMD method is used to process the acquired signal, decomposing it into a series of IMF components. A dual-index screening strategy is then used to obtain the target component, establishing an intelligent diagnostic model dataset. A convolutional neural network is then constructed, with the structure as follows: Figure 7 As shown, the target component is used as input, and the input data is fused through a fusion layer. After adaptive feature extraction via convolution and pooling operations, a fully connected layer is used to classify the features. Model parameters are initialized, and the established dataset is fed into the model for training. The training accuracy and cross-entropy loss function are used to measure the model's performance, as shown in Figures 8(a) and 8(b). It can be seen from the figures that when the model tends to converge, the model accuracy can reach 96%, indicating that the model can achieve fault identification and diagnosis of wind turbine gearbox vibration signals under non-stationary operating conditions.

[0108] Example 2

[0109] This embodiment discloses a wind turbine gearbox fault diagnosis system, including:

[0110] Vibration signal acquisition module, which is used to acquire vibration signals of the wind turbine gearbox to indicate health or malfunction;

[0111] The vibration signal decomposition module is used to improve the EEMD method using a single-parameter strategy to obtain a series of IMF components.

[0112] The target component screening module is used to screen target components using a dual-index screening strategy based on correlation coefficient and power spectral density.

[0113] The fault classification module is used to establish a convolutional neural network model with the target component as input, and finally obtain an intelligent fault diagnosis model for wind turbine gearbox faults. The obtained model is used to identify and diagnose faults in the vibration signal of the wind turbine gearbox.

[0114] The functions implemented by the four modules in this embodiment are completely consistent with the steps in Example 1, and will not be repeated here.

[0115] As a further technical solution, the intelligent fault diagnosis system includes:

[0116] The first submodule is used to build a convolutional neural network model. It takes the target component as input, and after adaptive feature extraction through fusion, convolution and pooling operations, it uses a fully connected layer to classify the features.

[0117] The second submodule is used to initialize model parameters, feed the established dataset into the model for training, and use the training accuracy and cross-entropy loss function to measure the model's performance. When the model tends to converge and the model's accuracy reaches 96%, it can be considered that the model has achieved the purpose of fault identification and diagnosis.

[0118] Example 3

[0119] The present invention also provides a computer, including a memory and a processor, and computer instructions stored in the memory and running on the processor, wherein the computer instructions, when executed by the processor, complete the steps in the wind turbine planetary gearbox fault diagnosis method described in Embodiment 1.

[0120] Example 4

[0121] The present invention also provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the steps in the wind turbine gearbox fault diagnosis method described in Embodiment 1.

[0122] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0123] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0124] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0125] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A fault diagnosis method for wind turbine gearboxes based on spEEMD and dual-index screening, characterized in that, include: Collect vibration signals from a fan gearbox failure occurring in an actual industrial operation scenario; The spEEMD method is used to decompose the vibration signal of the wind turbine gearbox into multiple IMF components. The IMF components are then merged by directly merging the IMF components with the same number of IMF components and by selecting a merging strategy for the IMF components with different numbers of IMF components. The IMF component containing sufficient fault characteristic information is selected as the target IMF component through a dual-index screening strategy. Using the target IMF component as input data, a convolutional neural network model is established, and finally an intelligent diagnostic model for wind turbine gearbox faults is obtained. The obtained model is used to identify and diagnose faults in the vibration signal of the wind turbine gearbox. The spEEMD method is as follows: a random Gaussian noise sequence is added to the original signal to obtain a noise signal; the extreme value distribution characteristics of the noise signal are evaluated by the F-value, a measure of the degree of improvement of extreme values; if the noise signal meets the conditions, it is allowed to pass the screening and enter the decomposition calculation; the noise signal is solved to obtain a series of IMFs components. The method of selecting IMF components containing sufficient fault characteristic information as target IMF components through a dual-index screening strategy involves the following steps: Calculate the correlation coefficient between each IMF component and the original signal, sort the correlation coefficients, and select the IMF component with the highest correlation coefficient as the initial component; calculate the envelope signal of the initial component, and merge the remaining IMF components with a high correlation to the initial component based on the power spectral density; reselect the initial component, merge it again based on the corresponding power spectral density, and finally merge it into a series of new components. Then, select 1-2 merged components as target IMF components based on the sorting of the original components.

2. The wind turbine gearbox fault diagnosis method based on spEEMD and dual-index screening as described in claim 1, characterized in that, The process of establishing an intelligent fault diagnosis model is as follows: A convolutional neural network model is established, using the selected vibration signals as input. After adaptive feature extraction through fusion, convolution, and pooling operations, fully connected layers are used to classify the features. Finally, the model parameters are initialized, and the established dataset is fed into the model for training. The training accuracy and cross-entropy loss function are used to measure the model's performance. When the model tends to converge and the accuracy reaches above 96%, the model is considered complete.

3. A wind turbine gearbox fault diagnosis system based on spEEMD and dual-index screening, characterized in that, as follows: The vibration signal acquisition module is used to acquire vibration signals of a fan gearbox failure in actual industrial operation scenarios. The vibration signal decomposition module is used to improve the EEMD method using a single-parameter strategy to obtain a series of final IMF components. Target component filtering module: It uses the spEEMD method to decompose the vibration signal of the wind turbine gearbox into multiple IMF components, and performs IMF component merging by directly merging the same number of IMF components and selecting a merging strategy for different numbers of IMF components. Fault classification module: It is used to establish a convolutional neural network model with the target component as input, and finally obtain an intelligent fault diagnosis model for wind turbine gearbox faults. The obtained model is used to identify and diagnose faults in the vibration signal of wind turbine gearbox. The spEEMD method is as follows: a random Gaussian noise sequence is added to the original signal to obtain a noise signal; the extreme value distribution characteristics of the noise signal are evaluated by the F value, which is a measure of the degree of extreme value improvement; if the noise signal meets the conditions, it is allowed to pass the screening and enter the decomposition calculation. De-noise signals to obtain a series of IMFs components; The target component filtering module includes: The initial component selection module is used to calculate the correlation coefficient between each IMF component and the original signal, sort the correlation coefficients, and select the IMF component with the highest correlation coefficient as the initial component. The component merging module is used to calculate the envelope signal of the initial component and merge the remaining IMF components with the initial component based on the power spectral density. The target component determination module reselects the initial components, merges them according to the corresponding power spectral density, and finally merges them into a series of new components. Based on the original component sorting, it selects 1-2 merged components as the target IMF components.

4. The wind turbine gearbox fault diagnosis system based on spEEMD and dual-index screening as described in claim 3, characterized in that, The intelligent fault diagnosis model includes: The first submodule is used to build a convolutional neural network model. It takes the selected vibration signal as input, and after adaptive feature extraction through fusion, convolution and pooling operations, it uses a fully connected layer to classify the features. The second submodule is used to initialize model parameters, feed the established dataset into the model for training, and use the training accuracy and cross-entropy loss function to measure the model's performance. When the model tends to converge and the model's accuracy reaches more than 96%, the model is considered complete.

5. A computer, comprising a memory and a processor, and computer instructions stored in the memory and running on the processor, characterized in that, When the computer instructions are executed by the processor, they complete the steps in the wind turbine gearbox fault diagnosis method based on spEEMD and dual-index screening as described in any one of claims 1-2.

6. A computer-readable storage medium for storing computer instructions, wherein when the computer instructions are executed by a processor, the instructions are characterized in that... Complete the steps in the wind turbine gearbox fault diagnosis method based on spEEMD and dual-index screening as described in any one of claims 1-2.

Images