Water turbine guide vane health state monitoring method and system

Abstract

The invention provides a water turbine guide vane health state monitoring method and system, and relates to the technical field of water turbine health monitoring. The method comprises the following steps: respectively acquiring a sound wave signal and a vibration signal generated during the operation of a guide vane through a sound wave sensor and a vibration sensor which are arranged at a volute manhole door of the water turbine; extracting time domain features, frequency domain features and time-frequency combined features of the sound wave signals and the vibration signals; and inputting the time domain feature, the frequency domain feature and the time-frequency combined feature into a pre-trained fault diagnosis model, and monitoring the health state of the guide vane through the fault diagnosis model. The running state of the guide vane can be monitored in real time, the health condition of the guide vane is accurately obtained, abnormal conditions such as water leakage are found in time, and therefore accurate maintenance suggestions are provided.

Description

Technical Field

[0001] The present invention relates to the technical field of hydroturbine health monitoring, and particularly to a method and system for monitoring the health status of a hydroturbine guide vane. Background Art

[0002] As the core equipment of hydropower generation, the health status of the guide vane of a hydroturbine is directly related to power generation efficiency and safety. If the guide vane is damaged, it is easy to cause phenomena such as guide vane leakage, affecting power generation efficiency. During daily work, when the unit is in the working state or standby state, the runner chamber is fully filled with water, and at this time, there are no conditions to check the health status of the guide vane; and when the hydroturbine is running normally, the water flow in the runner chamber is large, and it is impossible to install precise sensors at the guide vane to detect the health condition of the guide vane.

[0003] Traditional methods for judging the health degree of guide vanes mainly rely on manual experience judgment and ray flaw detection after drainage maintenance and entering the guide vane. Manual experience means that production personnel use their experience in power production to judge the health degree of guide vanes. For example, by combining the change of the water consumption rate of the unit, it is inferred that there is a leakage phenomenon in the guide vane, and it is judged that the guide vane is in an unhealthy state. Ray flaw detection after drainage maintenance and entering the guide vane means that after the unit is overhauled and the water filled in the unit is drained, at this time, personnel can enter the volute and carry relevant equipment to detect the guide vane.

[0004] The traditional methods for judging the health degree of guide vanes have the following problems: 1. It is difficult to detect the decline of the guide vane sealing performance and structural damage in a timely manner. Using manual experience judgment is too dependent on the sense of responsibility and personal ability of production personnel, and ray flaw detection after drainage maintenance and entering the guide vane can only be carried out during the overhaul of the unit.

[0005] 2. It is difficult to accurately identify abnormal water flow and abnormal water pressure changes, and abnormal situations such as guide vane leakage cannot be detected in a timely manner, resulting in an increase in water energy loss.

[0006] 3. It is difficult to repair the abnormal situation of the guide vane in a timely manner. Due to the lack of effective means for monitoring the health of the guide vane, abnormal situations of the guide vane are often difficult to detect and accurately judge in a timely manner, and it is necessary to rely on regular complete drainage and maintenance (the maintenance interval is as long as half a year to one year), resulting in the inability to detect the hidden danger of guide vane leakage and repair it in a timely manner, affecting the operation efficiency and safety of the unit.

[0007] 4. The performance of the guide vane health monitoring system is insufficient. Existing means for monitoring the health of guide vanes often cannot accurately analyze the overall health status of guide vanes, nor can they accurately predict potential faults and maintenance requirements of guide vanes, and it is difficult to achieve preventive maintenance.

[0008] Therefore, traditional methods for monitoring the health of turbine guide vanes have many deficiencies and are difficult to meet the requirements of modern hydropower for guide vane health monitoring. Based on this, there is an urgent need for a safe, reliable, and accurate method for monitoring the health status of turbine guide vanes to determine the health of the guide vanes. Summary of the Invention

[0009] The purpose of the present invention is to provide a method and system for monitoring the health status of turbine guide vanes, which are used to solve the problems that traditional methods for monitoring the health of turbine guide vanes have many deficiencies and are difficult to meet the requirements of modern hydropower for guide vane health monitoring. It can monitor the operating status of the guide vanes in real time, accurately analyze the health status of the guide vanes, and timely detect abnormal situations such as water leakage, so as to provide accurate maintenance suggestions.

[0010] To achieve the above purpose, in the first aspect, the present invention provides a method for monitoring the health status of turbine guide vanes, including: Collecting acoustic signals and vibration signals generated during the operation of the guide vanes respectively through an acoustic wave sensor and a vibration sensor installed at the manhole door of the scroll case of the water turbine; Extracting the time-domain features, frequency-domain features, and time-frequency joint features of the acoustic signals and vibration signals; Inputting the time-domain features, frequency-domain features, and time-frequency joint features into a pre-trained fault diagnosis model, and monitoring the health status of the guide vanes through the fault diagnosis model.

[0011] According to the method for monitoring the health status of turbine guide vanes provided by the present invention, the time-domain features include kurtosis, waveform factor, and zero-crossing rate, the frequency-domain features include frequency band energy ratio, and the time-frequency joint features include the cross-correlation peak time delay between the acoustic signal and the vibration signal.

[0012] According to the method for monitoring the health status of turbine guide vanes provided by the present invention, the calculation formula for kurtosis is:

[0013]

[0014] In the formula, is kurtosis, represents the fourth-order central moment of the acoustic signal; represents the standard deviation of the acoustic signal; represents the mean value of the acoustic signal; N represents the number of sampling points within a single frame of the acoustic signal; represents the acoustic signal at the i-th sampling point.

[0015] According to the method for monitoring the health status of turbine guide vanes provided by the present invention, the calculation formula for the waveform factor is:

[0016] Wherein, C is the waveform factor; represents the vibration signal at the i -th sampling point; M represents the number of sampling points within a single frame of the vibration signal. According to a method for monitoring the health status of a water turbine guide vane provided by the present invention, the calculation formula for the zero-crossing rate is:

[0017] Wherein, is the zero-crossing rate; represents the vibration signal at the i -th sampling point; represents the vibration signal at the i +1-th sampling point; M represents the number of sampling points within a single frame of the vibration signal; when the signs of the vibration signals at the i -th sampling point and the i +1-th sampling point are opposite, ; when the signs of the vibration signals at the i -th sampling point and the i +1-th sampling point are the same, .

[0018] According to a method for monitoring the health status of a water turbine guide vane provided by the present invention, the calculation formula for the frequency band energy ratio is:

[0019] Wherein, is the frequency band energy ratio; is the -th frequency component after the acoustic signal undergoes fast Fourier transform, represents the lowest frequency, represents the highest frequency; N represents the number of sampling points within a single frame of the acoustic signal.

[0020] According to a method for monitoring the health status of a water turbine guide vane provided by the present invention, the cross-correlation peak time delay between the acoustic signal and the vibration signal is:

[0021]

[0022] Wherein, represents the cross-correlation peak time delay between the acoustic signal and the vibration signal; represents the acoustic signal, represents the vibration signal, represents the time delay, represents at the time delay The correlation between the lower acoustic wave signal and the vibration signal.

[0023] According to a method for monitoring the health status of a water turbine guide vane provided by the present invention, the training process of the pre-trained fault diagnosis model includes: Collecting multiple groups of test data of the guide vane operating in a healthy state and a fault state; wherein, each group of test data includes an acoustic wave signal and a vibration signal; Respectively extracting historical time-domain features, historical frequency-domain features, and historical time-frequency joint features from the acoustic wave signal and the vibration signal of each group of test data; Taking the historical time-domain features, historical frequency-domain features, and historical time-frequency joint features corresponding to each group of test data as a sample, and taking the state of the guide vane corresponding to this group of test data as the label of this sample, thereby constructing a training sample; Training an initial model with multiple training samples constituted by multiple groups of test data to obtain a pre-trained fault diagnosis model. According to a method for monitoring the health status of a water turbine guide vane provided by the present invention, the initial model is a support vector machine.

[0024] In a second aspect, the present invention provides a system for monitoring the health status of a water turbine guide vane, including: An acquisition unit for respectively acquiring an acoustic wave signal and a vibration signal generated during the operation of the guide vane through an acoustic wave sensor and a vibration sensor installed at the manhole of the spiral case of the water turbine; An extraction unit for extracting time-domain features, frequency-domain features, and time-frequency joint features of the acoustic wave signal and the vibration signal; A monitoring unit for inputting the time-domain features, frequency-domain features, and time-frequency joint features into a pre-trained fault diagnosis model, and monitoring the health status of the guide vane through the fault diagnosis model.

[0025] The technical solution of the present invention at least has the following technical effects: A method and a system for monitoring the health status of a water turbine guide vane provided by the present invention, the method includes: respectively acquiring an acoustic wave signal and a vibration signal generated during the operation of the guide vane through an acoustic wave sensor and a vibration sensor installed at the manhole of the spiral case of the water turbine; extracting time-domain features, frequency-domain features, and time-frequency joint features of the acoustic wave signal and the vibration signal; inputting the time-domain features, frequency-domain features, and time-frequency joint features into a pre-trained fault diagnosis model, and monitoring the health status of the guide vane through the fault diagnosis model. Through a non-invasive sensor layout design, collecting multi-modal signal features, it can monitor the operation state of the guide vane in real time, analyze the health condition of the guide vane, and timely detect abnormal situations such as water leakage, so as to provide maintenance suggestions. Description of the Drawings

[0026] To more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0027] In the drawings: Figure 1 It is a flowchart of the method for monitoring the health status of the turbine guide vane of the present invention. Specific embodiments

[0028] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will clearly and completely describe the technical solutions in the present invention in conjunction with the drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

[0029] The following will describe in detail some embodiments of the present invention in conjunction with the drawings. Without conflict, the following embodiments and the features in the embodiments can be combined with each other.

[0030] Please refer to Figure 1 , the embodiment of the present invention provides a method for monitoring the health status of a turbine guide vane, including: Step 1: Respectively collect the acoustic signal and vibration signal generated during the operation of the guide vane through an acoustic sensor and a vibration sensor installed at the manhole of the spiral case of the water turbine. It should be noted that during the power generation of the water turbine, a large amount of water flows into the runner chamber of the water turbine, with extremely large water volume and water pressure. Therefore, it is not possible to install sensors at the guide vane in the runner chamber, and monitoring devices such as sensors need to be installed in a place without moving water, such as outside the runner. The manhole of the spiral case is located outside the spiral case where there is no moving water. Therefore, when installing the acoustic sensor and the vibration sensor here in the present invention, they will not be affected by the water flow impact.

[0031] Specifically, the sampling rate of the acoustic sensor is 10 kHz, and the range is 30~130 dB. The vibration sensor can adopt a triaxial accelerometer, with a range of ±10g (g is the acceleration of gravity), and a sampling rate of 2 kHz. To avoid the direct impact of the water flow, the acoustic sensor and the vibration sensor can be installed outside the manhole of the spiral case.

[0032] In some embodiments, the collected acoustic signal and vibration signal can also be subjected to wavelet denoising and time-frequency alignment to eliminate environmental noise and the time deviation of the acoustic sensor and the vibration sensor.

[0033] Step 2: Extract the time-domain features, frequency-domain features, and time-frequency joint features of the acoustic wave signal and the vibration signal; Specifically, the time-domain features include kurtosis, waveform factor, and zero-crossing rate.

[0034] Among them, the calculation formula of kurtosis is:

[0035]

[0036] In the formula, is the kurtosis, represents the fourth-order central moment of the acoustic wave signal; represents the standard deviation of the acoustic wave signal; represents the mean value of the acoustic wave signal; N represents the number of sampling points within a single frame of the acoustic wave signal; represents the acoustic wave signal at the i-th sampling point.

[0037] Kurtosis is suitable for detecting transient shocks or abnormal pulses in the acoustic wave signal. Under normal operating conditions, the kurtosis of the guide vane acoustic wave signal is small. The leakage caused by guide vane failure will cause the kurtosis to gradually increase. In the case of a minor fault, a small number of transient shocks appear in the acoustic wave signal, and the kurtosis gradually increases; in the case of a serious fault, frequent transient shocks appear in the acoustic wave signal, with a sharp distribution, and the kurtosis rises to the highest. Therefore, the physical meaning of kurtosis is to detect the transient shock when the guide vane is damaged and leaking.

[0038] The calculation formula of the waveform factor is:

[0039] In the formula, C is the waveform factor, which is an important time-domain feature for measuring the relationship between the peak intensity and the overall energy of the vibration signal, and is defined as the ratio of the peak value to the effective value of the vibration signal. represents the i -th sampling point of the vibration signal; M represents the number of sampling points within a single frame of the vibration signal. Under normal operating conditions, the waveform of the guide vane vibration signal is relatively stable, the difference between the peak value and the effective value is small, and the waveform factor is small. When the guide vane fails and causes phenomena such as leakage, the mechanical structure will suddenly release energy, generating instantaneous high-amplitude impact signals (such as collision vibration, leakage vibration, etc.). At this time, the peak value increases significantly, while the effective value increases slowly due to the time-averaging effect, resulting in a sharp increase in the waveform factor. For example, the waveform factor of a flat square wave can be as low as 1.0 (unit waveform factor), but for other more peaked waveforms (such as triangular waves), the waveform factor can also be as high as 3 or 4. When the guide vane fails, the waveform factor will suddenly increase.

[0040] The calculation formula of the zero-crossing rate is:

[0041] In the formula, is the zero-crossing rate, which represents the number of times the vibration signal crosses the zero axis per unit time. represents the vibration signal at the i th sampling point; M represents the number of sampling points within a single frame of the vibration signal; represents 1 when the signs of the vibration signals at adjacent sampling points are opposite, and 0 otherwise. Under normal operating conditions, the water flow and mechanical vibration signals are relatively stable, and the zero-crossing rate is small and stable; when there is water leakage from the guide vane under fault conditions, the vibration signal oscillates rapidly, and the zero-crossing rate increases significantly.

[0042] Specifically, the frequency-domain features include the frequency-band energy ratio, such as the frequency-band energy ratio of the vibration signal in the 0 - 2000 Hz frequency band.

[0043] The calculation formula for the frequency-band energy ratio is:

[0044] In the formula, is the th frequency component after the acoustic signal undergoes fast Fourier transform, represents the lowest frequency, represents the highest frequency; N is the frame length of the acoustic signal, which represents the number of sampling points within a single frame of the acoustic signal. The numerator part is the sum of the energies of all frequency components within the target frequency band, and the denominator part is the total energy of the signal over the entire frequency band (0 - Nyquist frequency, i.e., half of the system sampling frequency).

[0045] is the frequency-band energy ratio of the acoustic signal, which calculates the proportion of the energy of a specific frequency interval (frequency band) of the acoustic signal in the total energy of the signal. This index can quantify the contribution of different frequency bands to the signal and is a key feature for distinguishing normal and fault states in the health monitoring of the guide vane. According to historical data, the main frequency of the leakage cavitation noise is usually at a relatively high frequency (500 Hz - 2 kHz), and a guide vane fault will excite the acoustic energy in a specific frequency band. Therefore, by detecting the proportion of high-frequency energy under the current operating conditions, normal and fault states can be distinguished. Moreover, this method has the characteristic of anti-noise interference because environmental noise (such as water flow turbulence) is usually distributed over the entire frequency band (0 - 2 kHz), while the fault energy is concentrated in a specific frequency band. By calculating the energy ratio, the fault characteristics can be amplified and the influence of noise can be suppressed. The frequency-band energy ratio of the guide vane is relatively low under normal conditions and increases significantly when the guide vane is faulty.

[0046] Specifically, the time-frequency joint features include the cross-correlation peak time delay between the acoustic signal and the vibration signal.

[0047] The cross-correlation function is used to measure the similarity between two signals at different time delays. The peak time delay is the time difference that makes the two signals most similar. Under normal circumstances, the vibration signal propagates through a fixed structural path; when there is a guide vane fault, the vibration wave changes to propagate through other paths, and at this time, the time delay of the vibration signal relative to the acoustic wave signal changes. By calculating the change in the peak time delay of the cross-correlation between the acoustic wave signal and the vibration signal, the mechanical state changes of the guide vane (such as loosening, wear, etc.) can be effectively captured, providing key time series features for fault diagnosis.

[0048] The mathematical expression of the cross-correlation function is:

[0049] where represents the acoustic wave signal (this signal is the reference signal), represents the vibration signal (the signal to be aligned), represents the time delay (an integer multiple of the sampling interval), represents the correlation at the time delay . When achieves the maximum value, it means that the two signals are most similar at this time delay, that is, this time delay is the peak time delay of the cross-correlation of the vibration signal relative to the acoustic wave signal. When the condition of the guide vane changes from healthy to unhealthy, the representing the peak time delay of the cross-correlation will mutate.

[0050] Therefore, the peak time delay of the cross-correlation between the acoustic wave signal and the vibration signal is:

[0051] where represents the meaning that when , achieves the maximum value, achieving the maximum value means that the two signals are most similar at this time delay, that is, this time delay is the propagation time delay of the vibration signal relative to the acoustic wave signal. When the condition of the guide vane changes from healthy to faulty, the representing the propagation time delay will mutate. For example, when the guide vane is damaged and leaking, shortens (for example, suddenly changes from the normal value of 5 ms to 2 ms).

[0052] Step 3: Input the time domain features, frequency domain features, and time-frequency joint features into a pre-trained fault diagnosis model, and monitor the health status of the guide vane through the fault diagnosis model.

[0053] Specifically, the training process of the pre-trained fault diagnosis model includes: Collect multiple groups of experimental data of the guide vane operating in a healthy state and a faulty state; among them, each group of experimental data includes an acoustic wave signal and a vibration signal; Extract historical time-domain features, historical frequency-domain features, and historical time-frequency joint features from the acoustic signals and vibration signals of each group of test data respectively; Take the historical time-domain features, historical frequency-domain features, and historical time-frequency joint features corresponding to each group of test data as a sample, and take the guide vane state corresponding to this group of test data as the label of this sample, thereby constructing a training sample; Train the initial model with multiple training samples composed of multiple groups of test data to obtain a pre-trained fault diagnosis model.

[0054] In some embodiments, the initial model can be a support vector machine.

[0055] It should be noted that the support vector machine (Support Vector Machine, SVM) is a commonly used machine learning algorithm, especially suitable for small-sample, non-linear binary classification problems. Its basic idea is to find an optimal classification hyperplane in the feature space so that samples of different classes can be separated with the maximum margin. For equipment state recognition, the healthy state and the fault state can be regarded as two classes, and the multi-dimensional feature quantities collected from the equipment are regarded as sample points, and an optimal state discrimination model is constructed by using SVM in the feature space.

[0056] Use the historical data in the healthy sample library to train the SVM model. The main steps are as follows: Select a certain number of samples from the healthy sample library as the training set. Each sample contains multiple feature variables and corresponding state labels. The healthy state label is -1, and the fault state label is +1; Normalize all feature variables so that their numerical ranges are scaled to [0, 1] or [-1, +1] to avoid the influence of features with different dimensions.

[0057] For the linearly separable case, the goal of SVM is to find an optimal hyperplane so that the two types of samples can be separated by the hyperplane with the maximum margin. For the linearly inseparable case, slack variables and penalty factors can be introduced to allow a small number of samples to be misclassified, and at the same time, the cost of misclassification is controlled by the penalty factor. The above problems can be solved by solving a convex quadratic programming problem to obtain the optimal solution, and the obtained optimal hyperplane parameters can be used to construct the decision function of SVM. For non-linear classification problems, the kernel trick can be used to map the samples from the original space to a high-dimensional feature space to make them linearly separable.

[0058] Commonly used kernel functions include polynomial kernels and Gaussian kernels (RBF kernels). Using the training set data, by solving the convex quadratic programming problem of the SVM, the optimal parameters of the model are obtained. For non-linear SVMs, the parameters of the kernel function need to be selected. The commonly used method is cross-validation combined with grid search to select the parameter combination with the optimal average performance. During the training process, it is also necessary to balance the fitting ability and generalization performance of the model, and control the complexity of the model by adjusting the penalty factor or kernel function parameters to avoid overfitting.

[0059] After the SVM model is trained, it can be used to perform real-time diagnosis on the online monitoring data of the guide vane health status. For the collected vibration signals and acoustic signals, multiple feature quantities are extracted in the same way as in the training stage. The feature quantities can be normalized to align with the numerical range of the training samples. The preprocessed feature vectors are substituted into the decision function of the SVM model to obtain the label of the current state of the guide vane (+1 or -1). Combining with the operating conditions of the guide vane, the state label is further interpreted: if is -1, the guide vane is operating normally; if is +1, there may be a fault or performance degradation in the guide vane. Of course, for the guide vane monitoring data identified as faulty or abnormal, the feature vectors can be compared with the support vectors of the SVM to find the most similar fault pattern and preliminarily infer the fault type and cause.

[0060] In practical applications, the fault diagnosis model trained by this SVM model can be embedded in the hydroturbine online monitoring system (such as an embedded industrial computer) to perform real-time diagnosis on the newly collected vibration signals and acoustic signals. When it is found that the state of the guide vane is predicted to be abnormal 5 times in a row, the system warning is triggered.

[0061] Based on the same inventive concept, another embodiment of the present invention provides a hydroturbine guide vane health status monitoring system, which corresponds to the method of the foregoing embodiment. The system includes: A collection unit for respectively collecting acoustic signals and vibration signals generated during the operation of the guide vane through an acoustic sensor and a vibration sensor installed at the manhole of the scroll case of the hydroturbine; An extraction unit for extracting the time-domain features, frequency-domain features, and time-frequency joint features of the acoustic signals and vibration signals; A monitoring unit for inputting the time-domain features, frequency-domain features, and time-frequency joint features into a pre-trained fault diagnosis model, and monitoring the health status of the guide vane through the fault diagnosis model.

[0062] The following is a specific embodiment of the present invention.

[0063] The experimental environment is the governor floor in a hydropower station powerhouse. The water turbine model is: Kaplan turbine, rated head 32.5m, flow rate 0 m³ / s (in normal shutdown state), the sampling rate of the acoustic wave sensor is 10 kHz, and the measuring range is 30 - 130 dB. The vibration sensor uses a triaxial accelerometer (X / Y / Z), the sampling rate is 2 kHz, and the measuring range is ±10g.

[0064] The guide vane operates in a healthy state (normal operating condition). At this moment, the guide vane is fully closed, and the acoustic wave sensor and vibration sensor operate continuously for 2 hours, collecting acoustic wave signals and vibration signals respectively, with the label .

[0065] The guide vane operates in a faulty state (faulty operating condition). Manually simulate the guide vane leakage fault, expand the guide vane opening to 2 mm (leakage, at this time the guide vane opening is about 1%), and the acoustic wave sensor and vibration sensor operate continuously for 1 hour, collecting acoustic wave signals and vibration signals respectively, with the label .

[0066] Multiple groups of dataset samples of the guide vane are obtained. Among them, the training set: 800 normal samples, 200 faulty samples (in a 4:1 ratio); the test set: 200 normal samples, 50 faulty samples.

[0067] Preprocess the acoustic wave signals and vibration signals, including: Denoising processing: Decompose the acoustic wave signal using the Sym5 wavelet basis for 5 layers, and perform SURE algorithm adaptive threshold denoising.

[0068] The vibration signal uses 4 - layer wavelet decomposition, soft threshold processing, and retains the frequency band of 0 - 2000 Hz.

[0069] Framing and normalization: Framing parameters: frame length 0.1 second (1000 points for acoustic wave signals, 200 points for vibration signals), 50% overlap. Z - score normalization: calculate the mean and standard deviation frame by frame.

[0070] Perform framing calculation every 0.1 second, and extract the following 5 features from each frame of the signal: 1. Time - domain features: Kurtosis : About 3.0 in normal condition, rising to 8.5 and above in faulty condition.

[0071] Form factor : About 3.1 in normal condition, rising to 5.2 and above in faulty condition.

[0072] Zero - crossing rate : 150 times per second in normal condition, rising to 300 and above in faulty condition.

[0073] 2. Frequency-domain features: Energy ratio of the vibration signal in the 0 - 2000 Hz frequency band : Among them, the energy ratio in the 500 - 2000 Hz frequency band is 5% under normal conditions and can reach 25% during a fault.

[0074] 3. Time-frequency combined features: Cross-correlation peak time delay between the acoustic signal and the vibration signal : It is 5 ms under normal conditions and can be shortened to 2 ms during a fault.

[0075] The prediction result of the present invention is:

[0076] In this embodiment, through the fault diagnosis model based on SVM combined with the above-mentioned multiple features, high-precision detection of the guide vane leakage fault is achieved (the recall rate is 90.5% and the F1-score is 92.3%), verifying the following advantages: Handling class imbalance: It can increase the recall rate of the fault class by 30%.

[0077] Real-time performance: The single prediction time < 1 ms, meeting the millisecond-level response requirements of the hydropower station.

[0078] Interpretability: Features such as cross-correlation time delay and kurtosis are key features, guiding the optimal deployment of sensors.

[0079] This solution can be directly integrated into the existing monitoring system of the hydropower station without modifying the guide vane structure, and uses non-intrusive sensors to provide reliable support for the health management of the guide vane.

[0080] Compared with the prior art, the beneficial effects of the technical solution of the present invention are as follows: 1. Improve monitoring accuracy: Through technical means such as vibration sensors, acoustic sensors, multiple features, and machine learning algorithms, the present invention can more accurately monitor the operating state of the guide vane and timely detect abnormal situations such as leakage.

[0081] 2. Reduce power generation losses: The present invention can detect the expansion of guide vane leakage at an early stage, promptly conduct hidden danger investigations, and reduce power generation losses caused by leakage.

[0082] 3. Improve the operating efficiency and safety of the unit: By promptly detecting abnormal situations of the guide vane and providing maintenance suggestions, the present invention helps to improve the operating efficiency and safety of the unit.

[0083] 4. Reduce maintenance costs and risks: The present invention can accurately analyze the health status of the guide vane and provide accurate maintenance suggestions, thereby reducing maintenance costs and risks.

[0084] 5. Implement preventive maintenance: Through quantitative data analysis, the present invention can predict potential failures and maintenance requirements, implement preventive maintenance, and reduce the impact of sudden failures on power generation.

[0085] Those skilled in the art will readily conceive of other embodiments of the present invention after considering the specification and practicing the embodiments disclosed herein. The present invention is intended to cover any variations, uses, or adaptations of the present invention, which follow the general principles of the present invention and include known common knowledge or conventional technical means in the technical field not disclosed by the present invention. It should be understood that the present invention is not limited to the exact structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present invention is only limited by the appended claims.

Claims

1. A method for monitoring the health status of a turbine guide vane, characterized in that, Including: Collecting acoustic signals and vibration signals generated during the operation of guide vanes respectively through an acoustic wave sensor and a vibration sensor installed at the manhole door of the spiral case of the water turbine; Extracting time-domain features, frequency-domain features and time-frequency joint features of the acoustic signals and vibration signals; Inputting the time-domain features, frequency-domain features and time-frequency joint features into a pre-trained fault diagnosis model, and monitoring the health status of the guide vanes through the fault diagnosis model.

2. The method for monitoring the health state of the water turbine guide vane according to claim 1, characterized in that The time-domain features include kurtosis, waveform factor and zero-crossing rate, the frequency-domain features include frequency band energy ratio, and the time-frequency joint features include the cross-correlation peak time delay between the acoustic signals and vibration signals.

3. The method for monitoring the health status of the water turbine guide vane according to claim 2, characterized in that The calculation formula of the kurtosis is: In the formula, is the kurtosis, represents the fourth-order central moment of the acoustic wave signal; represents the standard deviation of the acoustic wave signal; represents the mean value of the acoustic wave signal; N represents the number of sampling points within a single frame of the acoustic wave signal; represents the acoustic wave signal at the i-th sampling point.

4. The method for monitoring the health status of a water turbine guide vane according to claim 2, characterized in that, The calculation formula of the waveform factor is: In the formula, C is the waveform factor; represents the vibration signal at the i th sampling point; M represents the number of sampling points within a single frame of the vibration signal.

5. The method for monitoring the health status of the water turbine guide vane according to claim 2, characterized in that The calculation formula of the zero-crossing rate is: In the formula, is the zero crossing rate; represents the vibration signal at the i -th sampling point; represents the vibration signal at the i +1-th sampling point; M represents the number of sampling points within a single frame of the vibration signal; when the signs of the vibration signals at the i -th sampling point and the i +1-th sampling point are opposite, ; when the signs of the vibration signals at the i -th sampling point and the i +1-th sampling point are the same, .

6. The method for monitoring the health state of the water turbine guide vane according to claim 2, wherein The calculation formula of the frequency band energy ratio is: In the formula, is the frequency band energy ratio; is the -th frequency component after the acoustic wave signal undergoes fast Fourier transform, represents the lowest frequency, represents the highest frequency; N represents the number of sampling points within a single frame of the acoustic wave signal.

7. The method for monitoring the health state of the water turbine guide vane according to claim 2, wherein, The cross-correlation peak time delay between the acoustic signals and vibration signals is: In the formula, represents the peak time delay of the cross-correlation between the acoustic wave signal and the vibration signal; represents the acoustic wave signal, represents the vibration signal, represents the time delay, represents at the time delay the correlation between the acoustic wave signal and the vibration signal.

8. The method for monitoring the health status of the turbine guide vane according to claim 1, characterized in that, The training process of the pre-trained fault diagnosis model includes: Collecting multiple groups of test data when the guide vanes operate in a healthy state and a faulty state; wherein, each group of test data includes acoustic signals and vibration signals; Respectively extracting historical time-domain features, historical frequency-domain features and historical time-frequency joint features from the acoustic signals and vibration signals of each group of test data; Taking the historical time-domain features, historical frequency-domain features and historical time-frequency joint features corresponding to each group of test data as a sample, and taking the state of the guide vanes corresponding to this group of test data as the label of this sample, thereby constructing a training sample; Training an initial model with multiple training samples composed of multiple groups of test data to obtain a pre-trained fault diagnosis model.

9. The method for monitoring the health status of the water turbine guide vane according to claim 8, characterized in that, The initial model is a support vector machine.

10. A monitoring system for the health status of a turbine guide vane, characterized in that, Including: A collecting unit for collecting acoustic signals and vibration signals generated during the operation of guide vanes respectively through an acoustic wave sensor and a vibration sensor installed at the manhole door of the spiral case of the water turbine; An extracting unit for extracting time-domain features, frequency-domain features and time-frequency joint features of the acoustic signals and vibration signals; A monitoring unit for inputting the time-domain features, frequency-domain features and time-frequency joint features into a pre-trained fault diagnosis model, and monitoring the health status of the guide vanes through the fault diagnosis model.

Images