A power plant air supply fan fault diagnosis method, system and electronic equipment

Abstract

The present application relates to a kind of power plant air supply fan fault diagnosis method, system and electronic equipment.The present application is first to the original vibration signal of nonlinear non-stationary initial noise reduction, remove part of random particle noise;Second, by latent factor model (Latent Factor Model, LFM), signal modeling is obtained to hidden feature, and signal decomposition is carried out to the noise reduction signal, realizes secondary noise reduction;Finally, based on the extracted product function (Product Function, PF) component (i.e. second PF component), using the efficient scalable multi-fault classifier obtains the diagnosis result, and then efficiently and accurately realizes power plant air supply fan fault diagnosis.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fault diagnosis, in particular to a method and system for diagnosing faults of a forced draft fan in a thermal power plant, and an electronic device. BACKGROUND

[0002] The forced draft fan in a thermal power plant is an important auxiliary device of a device management system in a thermal power plant, and plays an important role in improving the core competitiveness of the thermal power plant. The failure of auxiliary equipment will not only directly affect the quality and output of power generation in the power plant, but also may cause power generation failure and even unnecessary personal and property losses. Therefore, it is of great significance to conduct fault diagnosis research on auxiliary equipment in the power plant. Most of the water pumps, fans and other auxiliary equipment in the power plant are motor-driven rotating devices. Abnormal vibration of such devices is one of the main factors affecting the safe and stable operation of the unit. Therefore, how to accurately and efficiently diagnose abnormal vibration faults of rotating devices has become a technical problem to be solved in the field. SUMMARY

[0003] The present application relates to the technical field of fault diagnosis, in particular to a method and system for diagnosing faults of a forced draft fan in a thermal power plant, and an electronic device.

[0004] To achieve the above-mentioned purpose, the present application provides the following solutions:

[0005] A method for diagnosing faults of a forced draft fan in a thermal power plant, comprising:

[0006] obtaining an original vibration signal of the forced draft fan in the thermal power plant, and performing noise reduction processing on the original vibration signal to obtain a noise reduction signal;

[0007] performing signal decomposition on the noise reduction signal to obtain a first PF component;

[0008] eliminating the first PF component from the original vibration signal to obtain a residual signal, and taking the residual signal as a new noise reduction signal, returning to perform "performing signal decomposition on the noise reduction signal to obtain a first PF component" until iteration k times, to obtain a second PF component;

[0009] inputting the second PF component into a multi-fault classifier to obtain a diagnosis result.

[0010] Preferably, the performing signal decomposition on the noise reduction signal to obtain a first PF component specifically comprises:

[0011] performing LFM decomposition on the noise reduction signal to obtain an implicit feature; the implicit feature comprises an implicit feature P and an implicit feature Q;

[0012] searching for the implicit feature to obtain an implicit feature extreme point;

[0013] determine a mean value of adjacent implicit feature extreme points and an envelope estimation value of the adjacent implicit feature extreme points based on the implicit feature extreme points;

[0014] determine a sliding average value of the mean value and a sliding average value of the envelope estimation value;

[0015] determine a first local mean signal and a second local mean signal; the first local mean signal is a local mean signal of the sliding average value of the mean value; the second local mean signal is a local mean signal of the sliding average value of the envelope estimation value;

[0016] separate the first local mean signal from the original vibration signal to obtain a separated signal;

[0017] demodulate the separated signal based on the second local mean signal to obtain a demodulated signal;

[0018] determine an envelope signal based on the second local mean signal, and determine the first PF component based on the envelope signal and the demodulated signal.

[0019] Preferably, the first PF component is a product of the envelope signal and the demodulated signal.

[0020] Preferably, the original vibration signal is denoised by a mean filtering method to obtain the denoised signal.

[0021] According to the specific embodiments of the present application, the following technical effects are provided:

[0022] The power plant air supply fan fault diagnosis method provided by the present application first performs initial denoising on the original non-linear and non-stationary vibration signal to remove part of the random particle noise; secondly, through a latent factor model (Latent Factor Model, LFM), the signal is modeled to obtain implicit features, and the denoised signal is decomposed to realize secondary denoising; finally, based on the extracted product function (Product Function, PF) component (i.e. the second PF component), a high-efficiency and scalable multi-fault classifier is used to obtain a diagnosis result, thereby efficiently and accurately realizing power plant air supply fan fault diagnosis.

[0023] In addition, the present application also provides the following implementation system:

[0024] The power plant air supply fan fault diagnosis system comprises:

[0025] a signal denoising module, configured to acquire an original vibration signal of a power plant air supply fan, and to perform denoising on the original vibration signal to obtain a denoised signal;

[0026] a signal decomposition module, configured to perform signal decomposition on the noise-reduced signal to obtain a first PF component;

[0027] a cycle separation module, configured to remove the first PF component from the original vibration signal to obtain a residual signal, and return the residual signal as a new noise-reduced signal to perform the processing flow in the signal decomposition module until the iteration k times, to obtain a second PF component;

[0028] a fault diagnosis module, configured to input the second PF component into a multi-fault classifier to obtain a diagnosis result.

[0029] An electronic device, comprising:

[0030] a memory, configured to store a logic control instruction; the logic control instruction is used to implement the power plant air blower fault diagnosis method provided in the above;

[0031] a processor, connected with the memory, configured to call and execute the logic control instruction.

[0032] Preferably, the memory is a computer readable storage medium.

[0033] The technical effects achieved by the above system and electronic device provided by the present application are the same as those achieved by the power plant air blower fault diagnosis method provided by the present application, and therefore will not be described here. BRIEF DESCRIPTION OF DRAWINGS

[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0035] Figure 1 a flow chart of the power plant air blower fault diagnosis method provided by the present application;

[0036] Figure 2 an implementation framework diagram of the power plant air blower fault diagnosis method provided by the present application;

[0037] Figure 3 a partial time domain waveform diagram after mean filtering provided by the embodiment of the present application; wherein, Figure 3 part (a) is a mean filtered transverse vibration time domain waveform diagram; Figure 3 part (b) is a mean filtered longitudinal vibration time domain waveform diagram;

[0038] Figure 4 an LFM decomposition loss function curve diagram provided by the embodiment of the present application;

[0039] Figure 5 The LMD decomposition result diagram provided by the embodiment of the present application is shown in the following figure: Figure 5 The loss function curve of the PF1 component in the (a) part; Figure 5 The loss function curve of the PF2 component in the (b) part; Figure 5 The loss function curve of the PF3 component in the (c) part; Figure 5 The loss function curve of the PF4 component in the (d) part; Figure 5 The loss function curve of the PF5 component in the (e) part; Figure 5 The loss function curve of the PF6 component in the (f) part; Figure 5 The loss function curve of the PF7 component in the (g) part; Figure 5 The loss function curve of the PF8 component in the (h) part; Figure 5 The loss function curve of the PER component in the (i) part;

[0040] Figure 6 The PF component correlation coefficient analysis diagram provided by the embodiment of the present application is shown in the following figure:

[0041] Figure 7 The ROC curve diagram of different fault types provided by the embodiment of the present application is shown in the following figure:

[0042] Figure 8 The structural schematic diagram of the fault diagnosis system of the induced draft fan of the thermal power plant provided by the present application is shown in the following figure: DETAILED DESCRIPTION

[0043] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0044] The purpose of the present application is to provide a fault diagnosis method, system and electronic equipment for the induced draft fan of a thermal power plant, which can accurately and efficiently diagnose abnormal vibration faults of the induced draft fan of the thermal power plant.

[0045] In order to make the above-mentioned purposes, characteristics and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0046] As shown in the following figure, the fault diagnosis method for the induced draft fan of a thermal power plant provided by the present application comprises: Figure 1

[0047] ​Step 100: Obtain the original vibration signal of the air supply fan of the thermal power plant, and perform noise reduction processing on the original vibration signal to obtain a noise reduction signal. The original vibration signal is mainly processed by the mean filtering method to eliminate part of the random particle noise in the original vibration signal.

[0048] Step 101: Perform signal decomposition on the noise reduction signal to obtain a first PF component. The implementation process of this step can be:

[0049] Step 1010: Perform LFM decomposition on the noise reduction signal to obtain an implicit feature. The implicit feature includes an implicit feature P and an implicit feature Q.

[0050] Step 1011: Search for an implicit feature extreme point. Wherein, the mean value b i and the envelope estimation value d i of adjacent extreme points calculated are respectively:

[0051]

[0052]

[0053] Wherein, k i is an implicit feature extreme point.

[0054] Step 1012: Determine the mean value and the envelope estimation value of adjacent implicit feature extreme points based on the implicit feature extreme points.

[0055] Step 1013: Determine the sliding average value of the mean value and the sliding average value of the envelope estimation value.

[0056] Step 1014: Determine a first local mean signal and a second local mean signal. The first local mean signal is a local mean signal of the sliding average value of the mean value. The second local mean signal is a local mean signal of the sliding average value of the envelope estimation value.

[0057] Further, based on the sliding average value of the mean value, a local mean equation of the first local mean signal is obtained, and based on the sliding average value of the envelope estimation value, a local mean equation of the second local mean signal is obtained.

[0058] Step 1015: Separate the first local mean signal from the original vibration signal to obtain a separation signal. For example:

[0059] h 11 (t)=x(t)-m 11 (t) (3)

[0060] Wherein, m 11 (t) is the first local mean signal, x(t) is the original vibration signal, and h 11 (t) is the separation signal.

[0061] Step 1016: demodulating the separated signal based on the second local mean value signal to obtain a demodulated signal. The demodulated signal is s 11 (t):

[0062]

[0063] wherein a 11 (t) is the second local mean value signal.

[0064] Step 1017: determining an envelope signal based on the second local mean value signal, and determining a first PF component based on the envelope signal and the demodulated signal. The envelope signal is a1(t):

[0065] a1(t) = a 11 (t) a 12 (t)...a 1n (t) (5)

[0066] wherein a 11 (t), a 12 (t),...,a 1n (t) are all obtained through the local mean value equation of the first local mean value signal.

[0067] The first PF component is PF1(t):

[0068] PF1(t) = a1(t)s 11 (t) (6)

[0069] Step 102: removing the first PF component from the original vibration signal to obtain a residual signal, and taking the residual signal as a new denoised signal, returning to execute step 101 until iteration k times, to obtain a second PF component. This second PF component is a true PF component. Wherein the residual signal is marked as u1(t).

[0070] Step 103: inputting the second PF component into a multi-fault classifier to obtain a diagnosis result. Wherein the multi-fault classifier selects an XGBoost multi-fault classifier, and the training optimization is to train the fault features of the removed false components, and input the test set into the trained classifier for classification.

[0071] A specific embodiment is provided below to illustrate the specific implementation process of the above-mentioned power plant forced draft fan fault diagnosis method.

[0072] For example Figure 2As shown, first, mean filtering is used to initially denoise the nonlinear and non-stationary original vibration signal, removing part of the random particle noise. Second, through the latent factor model, the signal is modeled to obtain the hidden feature vector, and the local mean decomposition method is used for correlation coefficient analysis to remove false components, realizing secondary denoising. Finally, based on the extracted product function components, the XGBoost is used to classify the bearing fault state, so as to obtain the diagnosis result.

[0073] In this embodiment, the fault diagnosis method specifically includes the following steps:

[0074] S1, the inner ring fault, outer ring fault, rolling element fault and normal state are simulated respectively, and the fault bearing signal is selected: light, moderate and severe health states, a total of 10 kinds of signals. Obtain the vibration signal of each state, a total of 80 groups. Each group contains 1075 data points, and a total of 800 samples of 10 states. In each state, 60 samples are randomly selected as training objects, and 20 samples are used as test objects. The specific data description is shown in Table 1.

[0075] Table 1 Data Description Table

[0076]

[0077]

[0078] S2, the original vibration signal is subjected to mean filtering denoising processing. The longitudinal and transverse data under the normal state are selected to illustrate the method provided in this embodiment. The output of the original vibration signal after mean filtering is the simple average value of the data contained in the filtering template neighborhood, as shown in Figure 3 The part (a) and (b) of the LFM decomposition loss function curve are shown in the normal state signal mean filtering part time domain waveform diagram.

[0079] S3, the initial denoising signal is subjected to LFM decomposition to obtain the hidden feature P and the hidden feature Q, so as to obtain Figure 4 The LFM decomposition loss function curve is shown. After LFM decomposition, the LMD analysis is performed on each hidden feature vector to obtain a group of PF components and a residual component. Figure 5 Taking the normal state signal as an example, the LMD algorithm is used to decompose it into nine components, eight PF components (PF1-PF8) and a residual component (RES). Each component represents different regional information, and also contains the change of the vibration signal from high frequency band to low frequency band.

[0080] S4, the nine components contain pseudo components and noise interference signals irrelevant to the original vibration signal. In order to further accurately extract the PF component with high sensitivity and obtain a better feature vector, the correlation of each PF component with the initial signal needs to be calculated respectively, and the result is shown in Figure 6 Figure 6 As can be seen from Figure 6 , PF1-PF5 components all have high contribution rates, indicating strong correlation with the original vibration signal. Although the contribution rate of PF6 component is small, it is still higher than that of other components, so the correlation is also high. Therefore, the first six PF components can represent the fault information of bearing vibration, and secondary noise reduction is realized.

[0081] S5, finally, based on the extracted product function components as the input of the integrated learning model, the XGBoost multi-fault classifier is used to classify the bearing fault state. In order to verify the effectiveness of the XGBoost multi-fault classifier, the precision, recall, F1 value and other indicators are used to evaluate the sample classification results. After several tests, the fault pattern recognition results of the XGBoost multi-fault classifier provided by the present application are shown in Table 2, and from Table 2 it can be seen that the accuracy is about 96.80%, the recall is about 96.53%, the precision is about 96.79%, and the F1 rate is about 96.90%.

[0082] Table 2 Model evaluation index table

[0083]

[0084]

[0085] At the same time, the XGBoost multi-fault classifier diagnostic model is evaluated by the receiver operating characteristic (ROC) curve, which can also be understood as probability. As shown in Figure 7 , the area under the curve (AUC) of the ROC curve is close to 1, and generally the closer the value of AUC is to 1, the better the diagnostic effect is.

[0086] In addition, compared with the prior art, the present application has the following advantages based on the above-mentioned technical solutions:

[0087] (1) The mean filter noise reduction algorithm can effectively eliminate part of the random noise of the original vibration signal and improve the accuracy of the feature vector.

[0088] (2) The latent factor model is used for hidden vector decomposition of the initial noise-reduced signal, so as to fully mine the rich correlation and high correlation in the data. The hidden vector is further decomposed by the LMD algorithm, and the PF component with a higher contribution rate is selected to realize secondary noise reduction. This method effectively reduces the iteration number and greatly improves the calculation efficiency.

[0089] (3) The parallel XGBoost algorithm is introduced to construct a bearing fault diagnosis model and perform fault pattern recognition. The diagnosis accuracy can reach 96.8%, which is higher than that of LightGBM, K-NN and other fault models.

[0090] In addition, the present application also provides the following implementation system:

[0091] Among them, one is a power plant air supply fan fault diagnosis system, as shown in the figure, the system comprises: Figure 8

[0092] The signal denoising module 800 is used for acquiring the original vibration signal of the power plant air supply fan and performing denoising processing on the original vibration signal to obtain a denoised signal.

[0093] The signal decomposition module 801 is used for signal decomposition of the denoised signal to obtain a first PF component.

[0094] The cycle separation module 802 is used for removing the first PF component from the original vibration signal to obtain a residual signal, and taking the residual signal as a new denoised signal, returning to execute the processing flow in the signal decomposition module 801 until the iteration k times, and obtaining a second PF component.

[0095] The fault diagnosis module 803 is used for inputting the second PF component into a multi-fault classifier to obtain a diagnosis result.

[0096] Another is an electronic device, which comprises:

[0097] The memory is used for storing the logic control instruction. The logic control instruction is used for implementing the power plant air supply fan fault diagnosis method provided in the above. The memory used in the present application can be a computer readable storage medium.

[0098] The processor is connected with the memory, and is used for calling and executing the logic control instruction.

[0099] In the present specification, each embodiment is described in a progressive manner, and each embodiment focuses on the difference from other embodiments. The same or similar parts of each embodiment can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the related parts can be referred to the method part.

[0100] ​The principles and implementation manners of the present application are described by using specific examples in the present application, and the above examples are only used to help understand the method of the present application and its core idea; meanwhile, for the general technical personnel in the art, the specific implementation manners and application ranges will be changed according to the idea of the present application. In conclusion, the content of the present specification should not be understood as the limitation of the present application.

Claims

1. A method for diagnosing faults in a thermal power plant's forced draft fan, characterized in that, include: The original vibration signal of the blower of the thermal power plant is acquired, and the original vibration signal is subjected to noise reduction processing to obtain a noise-reduced signal; The signal decomposition of the denoised signal to obtain the first PF component includes: performing LFM decomposition on the denoised signal to obtain hidden features; the hidden features include hidden features P and hidden features Q; Searching for the hidden features yields the extreme points of the hidden features; based on the extreme points of the hidden features, the mean and envelope estimates of adjacent extreme points of the hidden features are determined; the moving average of the mean and the moving average of the envelope estimates are determined; a first local mean signal and a second local mean signal are determined; the first local mean signal is a local mean signal of the moving average of the mean; the second local mean signal is a local mean signal of the moving average of the envelope estimates; the first local mean signal is separated from the original vibration signal to obtain a separated signal; the separated signal is demodulated based on the second local mean signal to obtain a demodulated signal; the envelope signal is determined based on the second local mean signal, and the first PF component is determined based on the envelope signal and the demodulated signal; The first PF component is removed from the original vibration signal to obtain the remaining signal, and the remaining signal is used as the new noise-reduced signal. The process of "decomposing the noise-reduced signal to obtain the first PF component" is repeated until k iterations to obtain the second PF component. The second PF component is input into the multi-fault classifier to obtain the diagnostic results.

2. The method for diagnosing faults in thermal power plant blowers according to claim 1, characterized in that, The first PF component is the product of the envelope signal and the demodulated signal.

3. The method for diagnosing faults in thermal power plant blowers according to claim 1, characterized in that, The original vibration signal is denoised using a mean filtering method to obtain the denoised signal.

4. A fault diagnosis system for forced draft fans in thermal power plants, characterized in that, A system for implementing the fault diagnosis method for thermal power plant blowers as described in any one of claims 1-3; the system comprises: The signal noise reduction module is used to acquire the original vibration signal of the blower of the thermal power plant and to perform noise reduction processing on the original vibration signal to obtain a noise-reduced signal. A signal decomposition module is used to decompose the noise-reduced signal to obtain a first PF component; The loop separation module is used to remove the first PF component from the original vibration signal to obtain the remaining signal, and use the remaining signal as a new noise reduction signal to return to the processing flow in the signal decomposition module until it is iterated k times to obtain the second PF component. The fault diagnosis module is used to input the second PF component into the multi-fault classifier to obtain the diagnosis result.

5. An electronic device, characterized in that, include: A memory for storing logic control instructions; the logic control instructions are used to implement the fault diagnosis method for thermal power plant blowers as described in any one of claims 1-3. A processor, connected to the memory, is used to retrieve and execute the logic control instructions.

6. The electronic device according to claim 5, characterized in that, The memory is a computer-readable storage medium.

Images