A generator fault diagnosis method based on FA-BP neural network model

Abstract

The embodiment of the application discloses a generator fault diagnosis method based on a FA-BP neural network model, comprising: collecting operation data and image data of the generator, wherein the operation data comprises electric signals and vibration signals; obtaining an operation feature vector according to the operation data and obtaining image features according to the image data; inputting the operation feature vector into the FA-BP neural network model to obtain an initial fault probability; calculating the correlation strength between the fault type and the image features; correcting the initial fault probability by using the correlation strength to obtain a final fault probability; and outputting the generator fault state according to the final fault probability. The initial fault diagnosis result is corrected by using the image data, so that the accuracy, real-time performance and false alarm rate of the generator fault state judgment can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fault diagnosis methods, in particular to a generator fault diagnosis method based on a FA-BP neural network model. BACKGROUND

[0002] In modern power systems, generators are core equipment, and their operating states are directly related to the stability and reliability of power supply. With the continuous expansion of the scale and increasing complexity of power systems, the importance of generator fault diagnosis technology is increasingly prominent. Accurate and timely diagnosis of generator faults is of great significance to ensure the safe operation of power systems, reduce maintenance costs and reduce downtime.

[0003] Traditional generator fault diagnosis methods mainly rely on sensor data. Although these methods can reflect the operating state of the generator to some extent, they are susceptible to environmental noise, sensor accuracy and other factors due to relying solely on sensor-collected operating data for fault diagnosis, leading to misdiagnosis or missed diagnosis.

[0004] In particular, in the early stages of failure, the changes in operating data can be very subtle, making it difficult to accurately identify faults and resulting in insufficient fault diagnosis accuracy. In particular, for some critical components of the generator, such as bearing seats and insulation damage faults of stator windings, relying solely on raw vibration signals cannot detect early generator faults. SUMMARY

[0005] Therefore, the purpose of the embodiments of the present application is to provide a generator fault diagnosis method based on a FA-BP neural network model, which solves the technical problems of misdiagnosis or missed diagnosis and inability to detect early faults caused by relying solely on sensor data by fusing operating data and image data and fully utilizing image features.

[0006] The embodiments of the present application provide a generator fault diagnosis method based on a FA-BP neural network model, comprising:

[0007] Collecting operating data and image data of the generator, the operating data including electrical signals and vibration signals;

[0008] Obtaining operating feature vectors from the operating data and image features from the image data;

[0009] Inputting the operating feature vectors into a FA-BP neural network model to obtain initial fault probabilities;

[0010] Calculating the correlation strength between the fault type and the image features;

[0011] Correcting the initial fault probabilities using the correlation strength to obtain final fault probabilities;

[0012] outputting a generator fault state according to the final fault probability.

[0013] Further, the operation data further comprises a rotating speed signal, the electric signal comprises an output current signal, and the vibration signal comprises a vibration acceleration signal.

[0014] Further, obtaining an operation feature vector according to the operation data comprises:

[0015] obtaining a current unbalance degree according to the output current signal;

[0016] calculating an amplitude ratio of a one revolution frequency amplitude and a two revolution frequency amplitude according to the vibration signal and the rotating speed signal;

[0017] calculating a high frequency energy proportion according to the vibration signal;

[0018] the operation feature vector at least comprises the current unbalance degree, the amplitude ratio and the high frequency energy proportion.

[0019] Further, the image data comprises a bearing seat image and a stator winding surface image.

[0020] Further, obtaining an image feature according to the image data comprises:

[0021] obtaining a bearing light spot density according to the bearing seat image;

[0022] obtaining a crack proportion according to the stator winding surface image;

[0023] the image feature at least comprises the light spot density and the crack proportion.

[0024] Further, calculating a correlation strength between the fault type and the image feature comprises:

[0025] constructing a correlation matrix of the fault type and the image feature;

[0026] calculating an image feature saliency of the image feature, the image feature saliency representing a degree of the image feature exceeding an image feature threshold;

[0027] calculating an image correction weight according to the correlation matrix and the image feature saliency.

[0028] Further, the correlation matrix of the fault type and the image feature is constructed according to prior knowledge.

[0029] Further, the image correction weight is calculated according to the correlation matrix and the image feature saliency as follows:

[0030] ;

[0031] in, is the image feature The image feature saliency, For the Fault types and The correlation matrix of image features, is the total number of fault types, , is the total number of image features, .

[0032] Furthermore, the initial failure probability is corrected using the correlation strength to obtain the final failure probability as follows:

[0033] ;

[0034] in, For the The final failure probability of class failure, For the The initial failure probability of the class failure, and is the modified weight coefficient, is the positive correction threshold, is the negative correction threshold.

[0035] Furthermore, outputting the generator fault state according to the final fault probability includes:

[0036] When the final failure probability of the kth type of fault is greater than or equal to the failure threshold, the kth type of fault is output; when the final failure probability of the kth type of fault is less than the failure threshold but greater than or equal to the warning threshold, the warning prompt is output; if all failure probabilities are less than the warning threshold, the output is normal.

[0037] The beneficial effects of the embodiments of the present invention are:

[0038] The present invention uses sensor-collected operating data to generate an initial fault probability using a FA-BP neural network model. This initial fault probability reflects the likely fault probability. Image features are then collected synchronously to establish the correlation strength between the generator's image features and the fault type. The initial fault probability is then corrected using information related to the image features. This improves the accuracy of generator fault diagnosis and avoids misjudgments or missed diagnosis. Furthermore, when image features include spot density and crack ratio, the present invention's technical solution can provide early warning capabilities for bearing seat faults and stator winding insulation faults, improving the comprehensiveness, accuracy, and timeliness of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiments will be briefly introduced as follows. It should be understood that the following drawings only show some of the embodiments of the present application, and therefore should not be regarded as a limitation on the scope. Other related drawings can also be obtained by those of ordinary skill in the art without creative labor on the basis of these drawings.

[0040] Fig. 1 The flowchart of the generator fault diagnosis method based on the FA-BP neural network model of the present application;

[0041] Fig. 2 The flowchart of the FA-optimized BP neural network model of the present application. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, but not all the embodiments. The components of the embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations.

[0043] Please refer to Figs. 1-2 The generator fault diagnosis method based on the FA-BP neural network model of the present application comprises the following steps:

[0044] S1, collecting the operation data and image data of the generator.

[0045] Before establishing the fault diagnosis model and performing fault diagnosis, the possible fault types of the generator need to be coded. These codes will be used as the target output of the model, so that the model can correctly identify and classify the fault types according to the input feature parameters. Table 1 shows 10 common generator fault types.

[0046] Table 1: Generator fault types and codes

[0047]

[0048] According to the fault type, the present application collects the operation data corresponding to the fault type. The operation data at least includes electrical signals and vibration signals.

[0049] Among them, the electrical signals include three-phase output current signals The output current is detected by a high-frequency current transformer installed on the three-phase bus of the generator output end. According to the three-phase current signals The three-phase current unbalance degree can be obtained by a conventional current unbalance degree algorithm Also, the zero sequence current transformer can be directly used to collect the three-phase current unbalance degree The electrical signal includes the voltage signal output by the generator, and the overvoltage / undervoltage fault of the generator is detected according to the specific voltage value of the voltage signal; since the output power of the generator will decrease when the generator fails, preferably, the output power of the generator is also collected by the power transmitter, and the output power of the generator is used as an input feature to accurately judge the fault type during fault diagnosis.

[0050] The fault of the generator is usually related to the vibration signal, and the vibration speed sensor is arranged at multiple positions to detect the vibration signal and associate it with the fault type.

[0051] The acoustic vibration sensor on the bearing seat of the generator collects the vibration signal ;

[0052] The one revolution frequency amplitude is calculated according to the vibration signal The one revolution frequency amplitude is the amplitude of the sinusoidal component in the vibration signal that is synchronized with the rotor rotation frequency, The calculation formula is as follows:

[0053] ;

[0054] Wherein, represents the rotation frequency, represents the rotation speed measured by the rotation speed sensor.

[0055] represents the Fourier transform of the vibration signal .

[0056] Then, the two revolution frequency amplitude is calculated according to the vibration signal The two revolution frequency amplitude is the amplitude of the sinusoidal component in the vibration signal with a frequency of twice the rotor rotation frequency, The calculation formula is as follows:

[0057] ;

[0058] Finally, the amplitude ratio of the one revolution frequency amplitude and the two revolution frequency amplitude is calculated, and the formula is as follows:

[0059] ;

[0060] When the amplitude ratio exceeds a certain range, the rotor imbalance fault can be analyzed.

[0061] The high-frequency energy proportion can reflect bearing wear failure, and therefore the vibration signals collected by the acoustic vibration sensor on the generator bearing seat are used to calculate the high-frequency energy proportion Further calculation of the high-frequency energy proportion, the high-frequency frequency range is from frequency to the upper limit of the frequency bandwidth The high-frequency energy proportion , the calculation formula is as follows:

[0062] ;

[0063] Wherein, is the power spectral density, and the upper limit of the analysis bandwidth. The power spectral density is to divide the vibration signal into segments, each segment is windowed (such as Hanning window), the square of the FFT amplitude of each segment is calculated, and then the average value is obtained, the specific calculation process is as follows:

[0064] The vibration signal is segmented and windowed:

[0065] ;

[0066] Wherein, is the number of segments into which the signal is divided, is the time segment, is the Hanning window function, indicates the windowed signal of the vibration signal after the Hanning window function processing in the time segment.

[0067] The segmented and windowed signal is subjected to FFT calculation:

[0068] ;

[0069] Wherein, is the Fourier transform value of the windowed signal at frequency .

[0070] The window function energy normalization factor U is calculated:

[0071] ;

[0072] Wherein, N is the number of sampling points of a single segment signal.

[0073] The power spectral density P(f) calculation formula is as follows:

[0074] :

[0075] wherein, is the sampling frequency, the power spectral density P(f) represents the distribution of vibration energy in the frequency domain.

[0076] In order to detect the insufficient circulation of cooling water or the failure of cooling system, an acoustic sensor such as a condenser microphone is arranged on the wall of the cooling water pipe. After the condenser microphone collects an analog voltage signal V(t), the cavitation sound pressure level is calculated according to the analog voltage signal V(t) to represent the strength of the cavitation phenomenon of the cooling water pipeline. The specific calculation process is to convert the voltage signal into a sound pressure signal , and the sound pressure signal is the change of air pressure caused by the propagation of sound waves, and the calculation formula is as follows:

[0077] ;

[0078] is the microphone sensitivity, unit: V / Pa, typical value 50 mV / Pa.

[0079] The cavitation sound pressure level is calculated from the sound pressure signal , and the calculation process is as follows:

[0080] First, the root mean square value of the sound pressure signal is calculated:

[0081] ;

[0082] wherein, N is the number of sampling points of the sound pressure signal.

[0083] Then, the cavitation sound pressure level is obtained from the root mean square value of the sound pressure signal as follows:

[0084] ;

[0085] wherein, represents the reference sound pressure.

[0086] When it is also necessary to analyze the overheating fault of the generator, temperature sensors can also be arranged in the generator and the cooling water system respectively to detect the temperature inside the generator and the temperature of the cooling water of the generator. The information collected by the temperature sensors is used as the input features of the model for analyzing the overheating fault.

[0087] The application synchronously collects image data of the generator while collecting operation data. Based on some microscopic faults in the generator, such as bearing wear or insulation damage, these fault features will not cause obvious changes in electrical signals or vibration signals in the early stage. By capturing features related to such faults through image data, the faults can be discovered in advance and early diagnosis of the faults can be achieved. For example, the fan fault is a relatively macroscopic and less influential fault type, and it is not necessary to additionally set up an image collection device.

[0088] An ultraviolet imager is arranged near the outer surface of the bearing seat of the generator to collect ultraviolet image data of the bearing. When the generator bearing wears out and causes the oil film to break, micro-discharge, i.e., electric erosion, is triggered. This phenomenon is represented by a light spot on the ultraviolet image data, and the bearing wear state can be reflected through the light spot density. Therefore, in the application, the image data of the bearing seat is collected to achieve early diagnosis of faults and to compensate for the fault information of the bearing seat that cannot be reflected by the vibration signal.

[0089] An industrial visible light high-definition camera is arranged near the stator winding of the generator to collect image data of the stator winding. The insulation of the stator winding of the generator ages and causes cracks on the surface of the winding. However, the cracks of the stator winding are small in the early stage, and other sensor signals are difficult to accurately identify insulation faults. In the application, by analyzing the insulation surface crack features in the image data of the stator winding and fusing them with the sensor data, insulation layer damage faults can be discovered in time and handled to prevent cracks from expanding to cause serious faults such as short circuit.

[0090] S2, respectively, according to the operation data of step S1, an operation feature vector is obtained, and according to the image data of step S1, an image feature is obtained.

[0091] In addition to the calculated current imbalance, amplitude ratio, high-frequency energy proportion, and cavitation sound pressure level, the operation feature vector also includes all collected original sensor data signals to comprehensively reflect the fault data of the generator, such as temperature signals.

[0092] After the ultraviolet image data collected by the ultraviolet imager is filtered by Gaussian filtering and binarized and separated, the number of discharge light spots is obtained by marking the discharge light spots through connected domain analysis, and the light spot density is calculated As follows:

[0093] ;

[0094] Wherein, the light spot density is the number of discharge light spots per unit area (unit: pieces / cm²), representing the bearing wear degree.

[0095] The original image collected by the Basler industrial visible light high-definition camera is denoised through Gaussian filtering, and then the binary edge image is output through edge detection by the Canny operator, and then the crack pixels are obtained by removing the pseudo flaws through morphological processing, and the proportion of the stator winding crack pixels is calculated as follows:

[0096]

[0097] Wherein, the crack proportion represents the degree of damage of the stator winding insulation.

[0098] The crack proportion and the spot density are used as image features to correct the initial fault probability.

[0099] In addition, the electrical short circuit fault will cause the temperature of the generator rotor lead joint to rise, and then cause the hot spot to grow, in order to further optimize the judgment of such a fault, preferably, the rotor lead joint image data is also collected, and the image feature reflecting the hot spot growth rate is obtained.

[0100] S3, input the operation feature vector into the FA-BP neural network model to obtain the initial fault probability.

[0101] The FA-BP neural network model is adopted, that is, the firefly algorithm FA (Firefly Algorithm) is used to optimize the BP neural network (Back Propagation Neural Network, feedforward neural network) model. In the FA-BP neural network model of the present application, the firefly algorithm FA and the BP neural network are integrated with each other, aiming to overcome the problems of local optimal solution trap and slow convergence speed in the training process of the traditional BP neural network. Based on the powerful global search ability of the firefly algorithm, the weights and thresholds of the BP neural network are optimized, so that the model has a more superior parameter initialization setting at the beginning of training. This algorithm strategy not only can significantly improve the convergence efficiency of the model, but also can enhance the global search ability and diagnostic accuracy of the network in complex fault diagnosis tasks to some extent, thereby improving the robustness and stability of the model. This method effectively combines the global exploration characteristics of FA and the strong learning ability of BP, and provides a more reliable and efficient solution for generator fault diagnosis of complex systems. According to the optimization process as shown in Fig. 2 , the firefly algorithm is used to optimize the BP neural network to obtain the best initial weights W1, W2 and biases b1, b2.

[0102] The operation feature vector related to the sensor original data is input into the FA-BP neural network model to obtain the initial fault probability :​

[0103] ;

[0104] in, is the activation function, W1 and W2 are the weights of the BP neural network obtained by FA optimization, and b1 and b2 are the biases of the BP neural network obtained by FA optimization.

[0105] S4. Calculating the correlation strength between the fault type and the image features includes:

[0106] S4-1. Construct a correlation matrix between fault types and image features.

[0107] Correlation Matrix Is a p×q matrix, indicating the kth ( ) types of faults and ( ) image features. Taking 10 fault types and 2 image features as an example, a 10×2 correlation matrix is ​​defined in advance based on physical knowledge or other prior knowledge. .

[0108] S4-2, calculating the image feature saliency of the image feature;

[0109] For each image feature, calculate the image feature The significance is as follows:

[0110] ;

[0111] For example, the proportion of cracks in image features The significance of The calculation formula is as follows:

[0112] ;

[0113] in, is the crack index threshold, which can be set to 0.05%. Indicates the ratio of the actual crack ratio to the threshold.

[0114] For example, the spot density in the image feature The significance of The calculation formula is as follows:

[0115] ;

[0116] in, is the spot density threshold, which can be set to 5 / cm². Indicates the ratio of the actual spot density to the threshold.

[0117] S4-3. Calculate the image correction weight based on the correlation matrix and the image feature saliency.

[0118] For each fault type k (k=1~p), calculate the image correction weight as follows:

[0119] ;

[0120] in, is the image feature The image feature saliency, For the Fault types and image features, is the total number of fault types, , is the total number of image features, .

[0121] Preferably, the acquisition and processing of image data is performed in parallel with the acquisition and processing of sensor data, so as to enable rapid diagnosis of generator faults.

[0122] S5. Use the correlation strength to correct the initial failure probability to obtain the final failure probability .

[0123] The present invention corrects the initial fault probability through the contradictory / supportive dual correction mechanism, wherein the supportive correction means that when the image feature The image feature significance exceeds the positive correction threshold When , it means that the fault type is confirmed by the image features. At this time, increasing the probability of the fault by supporting correction can effectively reduce the missed detection caused by sensor diagnosis and improve the early warning capability. When the positive correction threshold is 0.8, it is judged When the image features The image feature significance is less than the negative correction threshold When , it means that the image feature believes that the fault type does not exist. At this time, reducing the probability of the fault by contradictory correction can suppress the false alarm of the sensor diagnosis result. When the negative correction threshold is 0.5, it is judged that When the contradiction is corrected.

[0124] ;

[0125] in, For the The final failure probability of class failure, For the The initial failure probability of the class failure is, where and is the correction weight coefficient, obtained through experiments The value is 0.25, When the value is 0.30, it can significantly reduce the false alarm rate and improve the early detection rate.

[0126] To avoid the eventual failure probability Exceeding the maximum value of 1, the probability of final failure The constraints are as follows:

[0127] .

[0128] S6. According to the final failure probability Outputs generator fault status.

[0129] After image correction, a more accurate final failure probability is obtained , if the corrected probability of the kth type of failure is If the corrected probability of the kth type of fault is greater than or equal to the fault threshold (for example, 0.85), it is determined to be this type of fault and the kth type of fault is output; if the corrected probability of the kth type of fault is If the probability of the kth type of fault is less than the fault threshold but greater than or equal to the warning threshold (for example, 0.7), it is determined that the kth type of fault is about to occur and an early warning prompt is output; if all failure probabilities are less than the early warning threshold, the generator is in a healthy state and the output is "normal".

[0130] The present invention uses image acquisition equipment to collect image data from some key equipment of the generator, and uses the image data to correct the initial fault diagnosis results, which can improve the accuracy, real-time performance and false alarm rate of generator fault status judgment.

[0131] Finally, it should be noted that the above-described embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the above-described embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily conceive of changes to the technical solutions described in the above-described embodiments within the technical scope disclosed by the present invention, or replace some of the technical features therein with equivalents. Such modifications, changes, or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and 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 generator fault diagnosis method based on FA-BP neural network model, characterized in that: include: Collecting operating data and image data of the generator, wherein the operating data includes electrical signals and vibration signals; and the image data includes an image of the bearing seat and an image of the stator winding surface; Obtaining an operation feature vector according to the operation data, and obtaining an image feature according to the image data; Obtaining image features according to the image data includes: obtaining the bearing spot density according to the bearing seat image; obtaining the crack ratio according to the stator winding surface image; the image features at least include the spot density and the crack ratio; Inputting the operation feature vector into the FA-BP neural network model to obtain an initial failure probability; Calculating the correlation strength between the fault type and the image feature, including: Construct a correlation matrix between fault types and image features; Calculating an image feature saliency of the image feature, where the image feature saliency indicates the degree to which the image feature exceeds an image feature threshold. For each image feature, the image feature saliency is a ratio of an actual value of the image feature to the image feature threshold. Calculating an image correction weight based on the correlation matrix and image feature saliency; Correcting the initial failure probability using the image correction weight to obtain a final failure probability; The generator fault state is output according to the final fault probability.

2. The method according to claim 1, characterized in that The operating data further includes a rotation speed signal, the electrical signal includes an output current signal, and the vibration signal includes a vibration acceleration signal.

3. The method according to claim 2, characterized in that Obtaining an operation feature vector according to the operation data includes: obtaining a current imbalance degree according to the output current signal; Calculate the amplitude ratio of one-time rotation frequency amplitude and two-time rotation frequency amplitude according to the vibration signal and the rotation speed signal; Calculate the proportion of high-frequency energy based on the vibration signal; The operation characteristic vector includes at least the current imbalance, the amplitude ratio and the high-frequency energy proportion.

4. The method according to claim 1, wherein The association matrix between fault types and image features is constructed based on prior knowledge.

5. The method according to claim 4, characterized in that Calculate the image correction weight according to the correlation matrix and the image feature saliency as follows: in, is the image feature The image feature saliency, For the Fault types and The correlation matrix of image features, is the total number of fault types, , is the total number of image features, .

6. The method according to claim 5, characterized in that The calculation formula for correcting the initial failure probability using the image correction weight to obtain the final failure probability is as follows: in, For the The final failure probability of class failure, For the The initial failure probability of the class failure, and is the modified weight coefficient, is the positive correction threshold, is the negative correction threshold.

7. The method according to claim 1, characterized in that Outputting the generator fault state according to the final fault probability includes: When the final failure probability of the kth type of fault is greater than or equal to the failure threshold, the kth type of fault is output; when the final failure probability of the kth type of fault is less than the failure threshold but greater than or equal to the warning threshold, the warning prompt is output; if all failure probabilities are less than the warning threshold, the output is normal.

Images