Multi-frequency fan blade health state online monitoring method based on vibration acceleration

Abstract

A multi-frequency fan blade health state on-line monitoring method based on vibration acceleration comprises the steps that vibration acceleration information of a blade in the fan operation process is extracted through a sensor arranged on the fan blade, the frequency band characteristic of the vibration acceleration is obtained through Fourier decomposition, the characteristic value is calculated, the fan blade operation state is diagnosed through a CURE method, and the fan blade health state is monitored. And judging whether operation fails or not. According to the method, on the premise that normal operation of the fan blade is guaranteed, the operation state of the fan blade can be recognized through the representative point-based clustering method (CURE) according to the multi-band characteristic of the vibration acceleration of the fan blade, and the method has good reliability and rapidity.

Description

Technical Field

[0001] The present invention relates to a technology in the field of fan fault diagnosis, in particular to an online monitoring method for the health status of fan blades based on multi-frequency vibration acceleration. Background Art

[0002] Wind power generation is a form of renewable energy that accounts for a relatively high proportion in my country. However, during the operation of wind turbines, the blades may rotate abnormally due to factors such as aging, impact, rain and snow, which will affect the stability of power generation and easily cause grid fluctuations. In this context, it is very important to monitor whether the blades are operating normally. Summary of the invention

[0003] In view of the shortcomings of the prior art that the specific fault location of the fan blade cannot be determined and the defect that misjudgment is prone to occur, the present invention proposes an online monitoring method for the health status of the fan blade based on the multi-frequency property of vibration acceleration. Under the premise of ensuring the normal operation of the fan blade, the operating status of the fan blade can be identified by a clustering method based on representative points (CURE) according to the multi-band characteristics of its vibration acceleration, and has good reliability and rapidity.

[0004] The present invention is achieved through the following technical solutions:

[0005] The present invention relates to an online monitoring method for the health status of fan blades based on vibration acceleration multi-frequency. The method comprises the following steps: a sensor arranged on the fan blades is used to extract the vibration acceleration information of the blades during the operation of the fan; a frequency band characteristic of the vibration acceleration is obtained through Fourier decomposition and a characteristic value is calculated; and a CURE method is used to diagnose the operation status of the fan blades to determine whether a fault occurs during the operation.

[0006] The sensors are specifically four sensors deployed on each blade of each wind turbine from the tail to the head.

[0007] The operating states of the fan blades include: normal, i.e. the fan runs smoothly and without faults; abnormal, i.e. the fault state when the fan blades vibrate abnormally due to aging, ice and snow, impact, etc.

[0008] The present invention relates to a system for implementing the above method, comprising: an information acquisition unit, an information storage unit, a signal processing unit and a result diagnosis unit, wherein: the information acquisition unit uploads the instantaneous value of vibration acceleration to the information storage unit through a sensor via a network cable, the information storage unit can store historical data and real-time data as well as diagnostic results, the signal processing unit performs Fourier decomposition based on the real-time data in the information storage unit and calculates based on the frequency band characteristics to obtain characteristic values, and the result diagnosis unit obtains the operating status of the fan blades through the CURE algorithm based on the characteristic values. Technical Effects

[0009] The present invention places sensors at different positions of the fan blades, combines the historical operation data of the fan and the new operation data, and realizes online threshold-free adaptive fault monitoring of the fan blade fault through cure clustering. Compared with the prior art, the present invention can realize adaptive fan blade fault monitoring by combining historical data and new operation data. This process does not require setting thresholds, and effectively prevents the problem of poor accuracy caused by threshold setting due to vibration differences at different positions of the fan blades. At the same time, accurate positioning of fan blade faults is achieved by arranging sensors at different positions. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 It is a flow chart of the present invention;

[0011] Figure 2 This is a schematic diagram of the normal operating spectrum of the fan blades;

[0012] Figure 3 This is a schematic diagram of the abnormal operation spectrum of the fan blades;

[0013] Figure 4 Schematic diagram of the clustering results of 1CURE implementation for the case;

[0014] Figure 5 Schematic diagram of the 2CURE clustering results for the case. DETAILED DESCRIPTION

[0015] like Figure 1 As shown, this embodiment relates to an online monitoring method for the health status of a fan blade based on multi-frequency vibration acceleration, comprising:

[0016] Step 1: Arrange sensors on the three blades of each wind turbine in the wind farm, install four sensors on each blade, and number each wind turbine blade as wt_id. At the same time, the ID of each sensor is set to SN. The sensor is connected to the wind turbine power supply through the network cable, powered by the network cable and uploads the data to the database in real time.

[0017] The sensor is used to collect vibration acceleration, sound and other data during the operation of the fan blade. This method uses the vibration acceleration during the operation of the fan blade for judgment, where the vibration acceleration data is uploaded every 5 seconds, and the data sampling frequency is set to f s In this embodiment, the frequency is set to 205 Hz, that is, 1024 pieces of data are uploaded in 5 seconds.

[0018] The database is used to store the operating data of all sensor IDs corresponding to the fan blades during all operating hours and the results of identification by the health status online monitoring technology. There are many data types in the sensor, among which the data name of vibration acceleration is recorded as ff00 for the fault identification method to call.

[0019] Step 2: Get a list of all sensors in the database. According to the sensor number, it can correspond to the corresponding wind turbine blade in the wind farm. Then select a blade, such as wt_id01, whose corresponding sensor ID is SN00007, SN00005, SN00002, SN00029. Get all the data of all sensors on the wind turbine blade within a certain time interval t1. The method of obtaining data is to first obtain the current time t2, and then subtract the time interval t1 to call the required time interval data in the database.

[0020] In this embodiment, it is set to 1 minute, and can be set arbitrarily according to actual needs.

[0021] Step 3, the data obtained is the vibration acceleration data on the x, y, and z axes uploaded by all sensors in the t1 time interval. Since the data is a large array, it cannot be input into the algorithm for processing. Therefore, a loop is set: the multiple data in the array are split, and one data is the vibration acceleration data on the x, y, and z axes uploaded by a sensor. Each time a data is input, the algorithm processes the fault result and then inputs the next data for identification.

[0022] Step 4: After extracting the three-axis data of vibration acceleration x, y, and z, perform Fourier decomposition on each set of data. The Fourier decomposition sampling frequency f z With the data sampling frequency f s The same, specifically: F(ω) = ∫f(t)·e -iωt dt, where: f(t) is a non-periodic function, F(ω) is the representation of the function in the frequency domain, e -iωt is a complex exponential function, and ω is the angular frequency.

[0023] Step 5. Perform Fourier decomposition on the data of the three axes x, y, and z, and extract the frequency domain results into 6 different frequency bands. The frequency band selection principle is to select 6 peaks and m points near them according to the frequency band during normal operation. Set the corresponding weights according to the different characteristics of each frequency band. Calculate the corresponding calculation results based on the weights and the energy of the m points in the frequency band, and add the multi-band calculation results of the vibration acceleration of the three axes x, y, and z to get the final eigenvalue. Where: α i Indicates the weight of each frequency band, i=1,2…6; pows=∑ m |F(n j ) 2 , j=1,2…m.

[0024] Step 6: Use the random sampling algorithm to sample the calculated data to obtain the sample set S. Random sampling ensures that representative samples are obtained in different areas of the data space and isolated points are eliminated.

[0025] Step 7: Perform hierarchical clustering on the sampled points, using either single-link or complete-link hierarchical clustering methods. If a cluster grows slowly, it is removed, and initial compact clusters are created based on the proximity between points in this smaller data set.

[0026] Step 8: For each group clustered in the previous step, select k (for example, 4) representative points in each group. The selection criteria for these four points are to be as dispersed as possible in the group and to cover the entire group as much as possible. The method is to randomly select a point in the group that is farthest from the center point of the group, the second point is the point farthest from the first point among all the points in the group, the third point is the point farthest from the first and second points among all the points in the group, and so on. The purpose of this step is to try to find the shape of each group, define the appearance of each group, and form each cluster. The distance is calculated using the Euclidean distance.

[0027] Step 9: Move the representative points in each cluster to the center of the group by a fixed percentage α (set to 10% in this example). The farther the point is, the greater the movement, that is, the greater the shrinkage. This step can shrink each cluster by the same proportion, so the CURE algorithm is less sensitive to outliers and can reduce the impact of outliers.

[0028] Step 10: Rescan all data points. For each data point p, use the following formula to find the cluster closest to p. closest =min(dist(p,C i )), dist(p,C i )=min||p-p'||,C i For clusters.

[0029] Step 11: Place p in the nearest cluster according to the calculation result of the above formula.

[0030] Step 12: Identify the operating status of the fan blades through the clusters obtained by CURE clustering and output the color label corresponding to each cluster. Adaptive monitoring of the operating status of the fan blades can be achieved based on the color label.

[0031] After specific experiments, 80 normal data and 20 abnormal data were set in scene 1, totaling 100 data. The abnormal data was obtained by adding a 44g load to the wind turbine blades. Figure 2 This is the frequency spectrum diagram when the fan blades are operating normally. Figure 3 This is the frequency spectrum diagram when the fan blades are operating abnormally. Figure 4 The recognition of 100 data points after CURE clustering. The color labels can clearly distinguish normal data from abnormal data, where normal data points are marked in orange and abnormal data points are marked in green. By comparing with the initial data, only one abnormal data point was identified incorrectly, with an accuracy rate of 99%. Without setting any threshold, high-precision monitoring of the wind turbine operating status can be achieved.

[0032] Further, 80 pieces of normal data and 20 pieces of abnormal data are set in scenario 2. The abnormal data are obtained by setting a 5-cm-long crack on the wind turbine blade. Figure 5 This is the identification of 100 data points after CURE clustering. The color labels clearly distinguish normal data from abnormal data, where normal data points are marked in orange and abnormal data points are marked in green. By comparing with the initial data, only two abnormal data points were identified incorrectly, with an accuracy rate of 98%.

[0033] The algorithm analysis results are matched with the sensor ID, and the analysis results are saved and written into the database. The background can view the sensor fault identification results on the corresponding fan blade at the corresponding time at any time. Since multiple sensors are installed on each fan blade, by comparing the data of different sensors on each blade, the specific fault location of the fan blade can be determined, and the fault can be accurately located.

[0034] In summary, the present invention obtains the operating data of the fan blades in real time, constructs a monitoring model, performs Fourier decomposition on the vibration acceleration, splits the result into multiple frequency bands, and draws conclusions by weight calculation based on the characteristics of different frequency bands. Through the cure clustering method, it can effectively distinguish whether the fan blades are operating normally or not, and the reliability is greatly improved, avoiding misjudgment caused by tiny vibrations. At the same time, it has very good speed, can timely judge the operating status of the fan blades and accurately locate the fault position.

[0035] The above-mentioned specific implementation can be partially adjusted in different ways by those skilled in the art without departing from the principle and purpose of the present invention. The protection scope of the present invention shall be based on the claims and shall not be limited by the above-mentioned specific implementation. Each implementation scheme within its scope shall be subject to the constraints of the present invention.

Claims

1. A method for identifying fan blade faults based on multi-band characteristics of vibration acceleration, characterized in that: The vibration acceleration information of the fan blades during the operation of the fan is extracted by sensors installed on the fan blades. The frequency band characteristics of the vibration acceleration are obtained through Fourier decomposition and the characteristic values ​​are calculated. The operating status of the fan blades is diagnosed by the CURE method to determine whether there is a fault in the operation.

2. The method for identifying fan blade faults based on multi-band characteristics of vibration acceleration according to claim 1 is characterized in that: The sensors are specifically four sensors deployed on each blade of each wind turbine from the tail to the head.

3. The method for identifying fan blade faults based on multi-band characteristics of vibration acceleration according to claim 1 is characterized in that: The operating states of the fan blades include: normal, i.e. the fan runs smoothly and without faults; abnormal, i.e. the fault state when the fan blades vibrate abnormally due to aging, ice and snow, or impact.

4. The method for identifying fan blade faults based on multi-band characteristics of vibration acceleration according to claim 1 is characterized in that: The extraction of the vibration acceleration information of the blades during the operation of the fan specifically includes: Step 1: Arrange sensors on three blades of each wind turbine in the wind farm, install four sensors on each blade, number each wind turbine blade as wt_id, and set the ID of each sensor as SN. Connect the sensor to the wind turbine power supply through the network cable, power it through the network cable and upload the data to the database in real time; Step 2, obtain a list of all sensors in the database, and correspond to the corresponding wind turbine blades in the wind farm according to the sensor number, then select a certain blade, and obtain all the data of all sensors on the wind turbine blade within a certain time interval t1, that is, first obtain the current time t2, and then subtract the time interval t1, and then call the required time interval data in the database; Step 3, the data obtained are the vibration acceleration data on the x, y, and z axes uploaded by all sensors multiple times within the t1 time interval. Since the data is a large array, it cannot be input into the algorithm for processing, so a loop is set up: the multiple data in the array are split, and one piece of data is the vibration acceleration data on the x, y, and z axes uploaded by a sensor once. Each time a piece of data is input, the algorithm is processed to obtain the fault result, and then the next piece of data is input for identification.

5. The method for identifying fan blade faults based on multi-band characteristics of vibration acceleration according to claim 1 is characterized in that: The characteristic value is obtained by: Step a: After extracting the three-axis data of vibration acceleration x, y, and z, each set of data is subjected to Fourier decomposition. The Fourier decomposition sampling frequency f z With the data sampling frequency f s The same, specifically: F(ω) = ∫f(t)·e -iωt dt, where: f(t) is a non-periodic function, F(ω) is the representation of the function in the frequency domain, e -iωt is a complex exponential function, ω is the angular frequency; Step b, the frequency domain results of the data of the three axes of x, y, and z are respectively Fourier decomposed and cut into 6 different frequency bands. The principle of frequency band selection is to select 6 peaks and a total of m points near them according to the frequency band during normal operation, and set corresponding weights according to the different characteristics of each frequency band. The corresponding calculation results are calculated based on the weights and the energy of the m points in the frequency band, and the multi-band calculation results of the vibration acceleration of the three axes of x, y, and z are added to obtain the final eigenvalue. Where: α i Indicates the weight of each frequency band, i=1,2…6; pows=∑ m |F(n j )| 2 , j=1,2…m.

6. The method for identifying fan blade faults based on multi-band characteristics of vibration acceleration according to claim 1 is characterized in that: The CURE method for diagnosing the operating status of the fan blades specifically includes: Step i: Sample the calculated data using a random sampling algorithm to obtain a sample set S. Random sampling ensures that representative samples are obtained in different areas of the data space and isolated points are removed. Step ii: Perform hierarchical clustering on the sampled points, using single-link or complete-link hierarchical clustering methods. If a class grows slowly, it is removed, and initial compact clusters are created based on the proximity between points in this smaller data set. Step iii: For each group clustered in the previous step, select k representative points in each group to form a cluster; Step iv, move the representative points in each cluster to the center point of the group by a fixed percentage α to reduce the influence of outliers; Step v: Rescan all data points. For each data point p, find the cluster closest to p and place p in the nearest cluster. Specifically: p closest =min(dist(p,C i )), dist(p,C i )=min||p-p'||,C i for clusters; Step vi: Identify the operating status of the fan blades through the clusters obtained by CURE clustering and output the color label corresponding to each cluster. Adaptive monitoring of the operating status of the fan blades can be achieved based on the color label.

7. The method for identifying fan blade faults based on multi-band characteristics of vibration acceleration according to claim 6 is characterized in that: In step iii, four representative points are selected to form a cluster, specifically: a point farthest from the center point of the group is randomly selected in the group, the second point is the point farthest from the first point among all the points in the group, the third point is the point farthest from the first and second points among all the points in the group, and so on; The distance is the Euclidean distance 8. A multi-band characteristic wind turbine blade fault identification system implementing the method described in any one of claims 1 to 7, characterized in that: include: An information collection unit, an information storage unit, a signal processing unit and a result diagnosis unit, wherein: the information collection unit uploads the instantaneous value of vibration acceleration to the information storage unit via a sensor through a network cable; the information storage unit can store historical data, real-time data and diagnosis results; the signal processing unit performs Fourier decomposition based on the real-time data in the information storage unit and calculates based on the frequency band characteristics to obtain characteristic values; the result diagnosis unit obtains the operating status of the fan blades through the CURE algorithm based on the characteristic values.

Images