A wind turbine tower bolt loosening identification method and system

Abstract

The application discloses a wind turbine tower bolt loosening identification method and system, comprising the following steps: S1, obtaining unit operation data; S2, calculating the actual tower frequency, and then generating the corresponding time sequence; S3, performing stability analysis on the generated time sequence, and marking the suspected bolt loosening abnormal points of the tower; and S4, determining the loosening position of the tower bolt. Without using special monitoring hardware equipment, the application is based on the existing general operation data fusion driving of the unit, constructs the corresponding algorithm, realizes the reliable monitoring and early warning of the tower bolt loosening, and has the advantages of high accuracy, high reliability, low cost, and the like.

Description

Technical Field

[0001] This invention relates to the technical field of wind power generation, and in particular to a method and system for identifying loose bolts on wind turbine towers. Background Technology

[0002] The tower is the key structure supporting the entire wind turbine, typically made of steel. Currently, the most common type is the conical steel pipe tower, although some use truss towers, concrete towers, or hybrid tower designs. Its height is adjusted according to wind resources and site conditions, generally ranging from 80 to 170 meters, or even higher. The foundation secures the tower to the ground, ensuring the entire wind turbine stands stably under various operating conditions and transferring the enormous load generated by the rotor rotation to the soil or rock of the foundation. Tower bolts are crucial connecting components between the tower and the foundation, between different sections of the tower, and between the tower and the nacelle and other superstructures. Pre-tightening provides sufficient clamping force to ensure no relative displacement between the flanges of the connecting components, reliably transferring the mechanical loads and wind loads generated by wind energy conversion to the foundation, maintaining the geometric stability of the tower under different wind speed conditions, and preventing structural instability or fatigue failure. The tightness of the bolts directly affects the safety of the tower and even the entire wind turbine. Loose or broken bolts will lead to structural connection failure, potentially causing serious accidents such as tower overturning and turbine damage. Wind turbine tower collapse not only causes huge direct economic losses to wind farms, but also results in significant power loss due to prolonged shutdowns. It can also incur high indirect costs such as the construction of access roads for large turbine components and land acquisition. Therefore, it is crucial to detect potential loosening of tower bolts in advance and promptly notify the project site to conduct tower bolt torque checks.

[0003] Currently, tower bolt condition monitoring mainly relies on periodic on-site inspections, a traditional timed maintenance model. Loosening is assessed by checking anti-loosening marks, manual tightening tests, or random checks with torque wrenches, but these torque checks are conducted over a long period. This method cannot reflect bolt condition in real time, and under the complex and variable operating conditions of wind turbines, it is difficult to detect potential bolt loosening in a timely manner, posing a significant safety risk.

[0004] Existing technologies also include solutions using dedicated monitoring equipment. These solutions utilize sensors such as acoustic sensors, image sensors, displacement sensors, strain sensors, ultrasonic sensors, and pressure sensors, or devices such as visual inspection and satellite positioning, to collect data and enable bolt loosening monitoring and early warning. While these methods allow for real-time monitoring, they suffer from complex wiring, difficult maintenance, and high equipment costs. Furthermore, the relevant technologies are still under development and cannot fully meet the wind power industry's actual needs for high reliability and low-cost management.

[0005] In conclusion, without considering the use of proprietary monitoring hardware, effectively monitoring and early warning of tower bolt loosening using existing general operating data of wind turbines has become a technical problem that the industry urgently needs to solve. Summary of the Invention

[0006] The primary objective of this invention is to overcome the shortcomings of existing technologies and provide a method for identifying loose tower bolts in wind turbine units. Without employing dedicated monitoring hardware, this method utilizes existing general operating data of the turbine unit to drive the fusion of relevant algorithms and achieve reliable monitoring and early warning of loose tower bolts.

[0007] The second objective of this invention is to provide a system for identifying loose bolts on wind turbine towers.

[0008] A third objective of this invention is to provide a storage medium.

[0009] A fourth objective of this invention is to provide a computing device.

[0010] The first objective of this invention is achieved through the following technical solution:

[0011] A method for identifying loose bolts on wind turbine towers, comprising the following steps:

[0012] S1. Acquire unit operating data; the unit operating data includes main control timing data and SCADA data;

[0013] S2. Based on the unit operation data, filter the data segments that meet the preset conditions for tower frequency analysis, construct the operating condition classification feature dataset and determine the theoretical tower frequency; remove interference data according to the operating conditions, calculate the impeller frequency and the characteristic frequency of three times the impeller frequency based on the average impeller speed of the removed data segments; when the impeller frequency and the characteristic frequency of three times the impeller frequency exceed the preset range of the theoretical tower frequency, use the NKDE nonparametric kernel density estimation tower frequency adaptive estimation algorithm to obtain the actual tower frequency, and then generate the corresponding time series.

[0014] S3. The stability analysis of the generated time series is performed using the single-value control chart test method. If the test determines that the time series is unstable, the corresponding data point is marked as a suspected loosening abnormal point of the tower bolts.

[0015] S4. Combining the time series and the unit operating condition classification feature dataset, an outlier detection algorithm is used to identify suspected loosening anomalies of the marked tower bolts. Based on the concentrated distribution characteristics of the anomalies in the unit operating condition classification feature dataset, the location of the loosening of the tower bolts is determined.

[0016] Furthermore, in step S1, based on the Nyquist sampling theorem, in the unit's SCADA system, the SCADA data sampling interval is set to 1 second, and the SCADA data recording interval is consistent with the sampling interval; in the unit's main control system, the main control timing data sampling interval is set to 10ms, and at least one main control timing data file is recorded daily by timed or fault alarm triggers.

[0017] Acquire a list of all IO points for SCADA data and main control timing data. Through spectrum analysis of the data, determine the operating parameters required to identify tower bolt loosening, including but not limited to: time, operating mode, wind speed, impeller speed, generator speed, blade angle, active power, ambient temperature, wind direction, nacelle northward angle, nacelle lateral vibration, nacelle longitudinal vibration, gearbox oil level, and yaw system pressure telemetry signals.

[0018] Perform time synchronization processing on the master control timing data and SCADA data.

[0019] Furthermore, step S2 includes the following steps:

[0020] S2.1 Calculate the number of data points required for spectrum analysis based on the tower's frequency resolution requirements. As shown in the following formula:

[0021] ;

[0022] in, For data sampling frequency, Frequency resolution;

[0023] According to the number of data points The SCADA data is cut and truncated using the equal-length rule, and the data integrity ratio is calculated. The data integrity ratio = actual data points / total number of data points. ;

[0024] Determine whether the data integrity ratio meets the preset conditions for tower frequency analysis, and then filter out the data segments that meet the preset conditions for tower frequency analysis;

[0025] S2.2 Construct a dataset of unit operating condition classification features based on the selected data segments;

[0026] S2.3 To improve the signal-to-noise ratio of the frequency band in which the tower frequency is located, a bandpass filter or HP filter is used to perform detrending processing on the time series signal of the operating condition classification feature dataset.

[0027] S2.4 Perform Fourier transform on the detrended time-series signal and plot the spectrum. Obtain the theoretical tower estimated frequency based on the preset selected data segments. By comparing and analyzing the theoretical tower estimated frequencies obtained from the data segments, the final theoretical tower frequency of the unit is determined.

[0028] S2.5. Based on the SCADA data under operating conditions, perform spectrum analysis on the unit operating parameters such as nacelle left-right vibration, nacelle forward-backward vibration, and impeller speed, and automatically remove interference data according to the unit operating conditions; calculate the impeller frequency and the characteristic frequency of three times the impeller frequency based on the average impeller speed in the data segment after removing interference, and determine whether the impeller frequency and the characteristic frequency of three times the impeller frequency exceed the preset allowable range of the theoretical tower frequency;

[0029] S2.6 If the frequency exceeds the limit, the tower frequency adaptive estimation algorithm based on NKDE nonparametric kernel density estimation is used to obtain the actual tower frequency, and then the corresponding time series is generated; if the frequency does not exceed the limit, the corresponding data segment is discarded.

[0030] Furthermore, the selected data segments are time-series index data of unit operating parameters, including unit operating mode, rotor speed, nacelle northward angle, active power, and blade angle.

[0031] Furthermore, the HP filtering time-series signal processing method is used to detrend the operating condition classification feature dataset, as detailed below:

[0032] The time series of the operational condition classification feature dataset is decomposed into trend and periodic components, which is achieved by minimizing the following objective function, as shown in the following equation:

[0033] ;

[0034] in, The time series to be decomposed , , Trend components are respectively , , Data at any given time The second difference of the trend component is used, and its sum of squares is used to measure the smoothness of the trend component. As a periodic component, For smoothing parameters.

[0035] Furthermore, in step S2.4, the theoretical tower estimated frequency obtained based on SCADA data and the theoretical tower estimated frequency obtained based on main control timing data are compared and verified. If the frequency consistency under different unit speed conditions is met, the theoretical tower estimated frequency is further compared and verified with the multi-mode frequency of the tower output by the unit simulation. If the multi-mode frequency of the tower output by the simulation is within the preset range of the theoretical tower estimated frequency, the first-order mode frequency of the tower output by the simulation is determined as the final theoretical tower frequency of the unit.

[0036] Furthermore, in step S2.6, the mathematical expression for the nonparametric kernel density estimation of NKDE using the Gaussian kernel function is shown below:

[0037] ;

[0038] in, This represents the number of spectral lines in the spectrum. Let be the amplitude of the i-th spectral line; In order to be in The estimate of the point based on nonparametric kernel density; The optimal window width is determined based on the asymptotic integral mean square error criterion, and its value is determined by the corresponding mathematical expression, as shown in the following formula:

[0039] ;

[0040] in, The standard deviation of the total amplitude of the spectral lines;

[0041] Next, within the preset frequency band of the tower frequency, the tower spectrum data is fitted and the peak value is located to obtain the actual tower frequency of the unit.

[0042] Finally, based on the number of data points The SCADA data is segmented and cut using the equal-length rule. The actual tower frequencies are obtained by analyzing each segment, and then a time series of the actual tower frequencies is generated.

[0043] The second objective of this invention is achieved through the following technical solution:

[0044] A wind turbine tower bolt loosening identification system, used to implement the aforementioned wind turbine tower bolt loosening identification method, includes:

[0045] The data acquisition module is used to acquire unit operating data; the unit operating data includes main control timing data and SCADA data;

[0046] The theoretical tower frequency determination module is used to filter data segments that meet the preset conditions for tower frequency analysis based on unit operation data, construct an operating condition classification feature dataset and determine the theoretical tower frequency.

[0047] The module for generating the actual tower frequency and its time series is used to remove interference data according to the operating conditions, and calculate the impeller frequency and the characteristic frequency of three times the impeller frequency based on the average impeller speed of the data segment after removal; when the impeller frequency and the characteristic frequency of three times the impeller frequency exceed the preset range of the theoretical tower frequency, the tower frequency adaptive estimation algorithm based on NKDE nonparametric kernel density estimation is used to obtain the actual tower frequency, and then generate the corresponding time series.

[0048] The bolt suspected loosening anomaly marking module is used to perform stability analysis on the generated time series using the single-value control chart test method. If the test determines that the time series is unstable, the corresponding data point is marked as a tower bolt suspected loosening anomaly.

[0049] The bolt loosening location determination module combines time series data with unit operating condition classification feature datasets and uses an outlier detection algorithm to identify suspected loosening anomalies of marked tower bolts. Based on the concentrated distribution characteristics of the anomalies in the unit operating condition classification feature dataset, the loosening location of the tower bolts is determined.

[0050] The third objective of this invention is achieved through the following technical solution:

[0051] A storage medium storing a program, which, when executed by a processor, implements the aforementioned method for identifying loose tower bolts in wind turbine units.

[0052] The fourth objective of this invention is achieved through the following technical solution:

[0053] A computing device includes a processor and a memory for storing processor-executable programs, wherein when the processor executes the program stored in the memory, it implements the above-described method for identifying loose tower bolts of wind turbine units.

[0054] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0055] 1. Cost Savings: By integrating existing common operating data from generating units, data-driven modeling is achieved to identify abnormal tower bolt loosening. This method eliminates the need for dedicated monitoring hardware, avoiding problems such as complex hardware installation and wiring, difficult subsequent maintenance, and high investment costs.

[0056] 2. High precision: Select operating parameter data that directly or indirectly contains the inherent frequency information of the unit tower. Through data interception, operating condition identification, detrending processing, anti-interference processing, spectrum analysis and adaptive estimation of tower frequency, the actual tower frequency can be obtained with high precision. The obtained tower frequency resolution can reach the level of one ten-thousandth of a hertz.

[0057] 3. High stability: Compared with dedicated monitoring hardware equipment, the unit's general operation data monitoring dimensions are more comprehensive, covering parameters such as angle, pressure, temperature, speed, vibration, and electrical parameters; it can integrate information such as wind speed, wind direction, nacelle northward angle, ambient temperature, and impeller speed from SCADA and main control time series data to construct a classification feature dataset of unit operating conditions, and use big data algorithms to identify and verify the tower frequency stability, so as to achieve reliable identification and early warning of tower bolt loosening.

[0058] 4. High ease of use: This method uses a data-driven algorithm to identify loose tower bolts in wind turbines. The packaged algorithm can run on current mainstream standard computers or laptops. It only needs to connect to existing common operating data of the units to carry out analysis. The process is simple, efficient and easy to use.

[0059] 5. Good compatibility: The first-order natural frequency of mainstream grid-connected wind turbine towers is 0.1–0.5 Hz, while the sampling interval for the second-level operation data of the turbines using general SCADA systems is 1 second. Based on the Nyquist sampling theorem, this method can perform high-precision monitoring of the first-order natural frequency of various mainstream tower types, such as steel towers, flexible towers, and hybrid towers, and can also identify bolt loosening, demonstrating good compatibility. Attached Figure Description

[0060] Figure 1 This is a flowchart of the method of the present invention.

[0061] Figure 2 This is a flowchart for obtaining the high-precision actual tower frequency according to the present invention.

[0062] Figure 3 This is a time-series data diagram of the tower frequency in this invention.

[0063] Figure 4 This is a graph showing the frequency data of the generator tower under different north-facing angles of the engine room in this invention. Detailed Implementation

[0064] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0065] Example 1:

[0066] like Figure 1 As shown, this embodiment provides a method for identifying loose bolts on wind turbine towers, including the following steps:

[0067] S1. Obtain unit operation data; the unit operation data includes main control timing data Tracelog and SCADA data; specifically...

[0068] The master control timing data Tracelog is a full-variable timing data continuously recorded by the unit's master control system during operation, triggered by timed events, faults / alarms, or manual actions, and sliced ​​at fixed time lengths. It can completely reproduce the unit's real operating parameters and status before, during, and after a fault, and pinpoint the root cause of the fault.

[0069] Its common sampling periods are 10ms and 20ms, which can cover almost all key input / output IO, intermediate variables, status words, etc., such as wind speed, power, blade angle, speed, and nacelle vibration. Because it has a timestamp accurate to the millisecond level, it can be correlated with SCADA data after synchronization.

[0070] Because the master control timing data Tracelog has a high sampling frequency and is directly recorded by the master control system, the data has high reliability. Based on the nacelle vibration and other data recorded under specific operating conditions, complete timing waveforms can be seen, and the accuracy of its spectrum analysis results is high, which is beneficial for determining the theoretical tower frequency of the unit. However, the data recording is a slice-style, non-full-condition continuous data recording, and the data recording length is limited, which cannot meet the required frequency resolution. Therefore, it is necessary to fuse SCADA data to obtain high-precision tower frequency data.

[0071] SCADA data refers to the operational data of wind turbines collected and recorded by the SCADA system. The SCADA system is a key technology platform for wind farm operation and management, ensuring the safe, efficient, and economical operation of the wind farm through real-time monitoring, data acquisition, analysis, and control of wind turbines and related equipment. SCADA systems typically employ a layered distributed architecture. The SCADA system installed in the central control room of the substation communicates with the main controllers (main control systems) of each wind turbine in the wind farm, forming the field device layer, enabling turbine status awareness and equipment operation. The data transmission layer achieves data communication and ensures network security through communication media (such as fiber optics), standardized communication protocols (such as Modbus, Profibus, etc.), and communication equipment such as switches and longitudinal isolation devices. The data processing and control layer runs SCADA client software on the application server, allowing maintenance personnel to view turbine status and operation control commands in real time, while simultaneously storing the collected turbine operation data in a historical data server. The main operating parameters of wind turbines monitored by SCADA include:

[0072] Angular data: wind direction, nacelle northward angle, blade angle, etc.;

[0073] Pressure data: yaw brake pressure, hydraulic system pressure, coolant pressure, etc.

[0074] Temperature data: gearbox bearing temperature, generator winding temperature, converter power module temperature, ambient temperature, etc.

[0075] Speed ​​data: wind speed, impeller speed, generator speed, yaw speed, etc.;

[0076] Vibration data: lateral vibration of the nacelle, longitudinal vibration of the nacelle, etc.;

[0077] Electrical parameter data: grid voltage, grid frequency, active power, reactive power, etc.

[0078] SCADA data uses a fixed sampling time interval and is continuously recorded 24 hours a day, meeting the requirements for high-resolution tower frequency analysis. Simultaneously, it monitors comprehensive unit operating parameters, including angle, pressure, temperature, velocity, vibration, and electrical parameters. The acquired tower frequency data can be combined with wind speed, wind direction, nacelle northward angle, ambient temperature, and impeller speed to create a multidimensional dataset. Big data algorithms are used to identify and verify the stability of the unit's tower frequency and to reliably identify and warn of loose tower bolts.

[0079] SCADA data is massive, with each MW-level unit generating nearly or more than 1000 operating parameter data points. Assuming a sampling interval of 1 second, one unit would generate 86,400 records per day. Considering the large data volume, SCADA data files in specific formats (such as CSV, ZIP archives, or proprietary serialization formats) can be exported daily from the SCADA system. Tracelog data is recorded in text file format in the main controller and can be directly accessed via FTP.

[0080] Based on the existing operating data of the wind turbine, in order to achieve high-precision monitoring of the tower's natural frequency, considering that the first-order natural frequency of the tower of mainstream grid-connected wind turbines is 0.1~0.5 Hz, according to the Nyquist sampling theorem, in the unit's SCADA system, the SCADA data sampling interval is set to 1 second, and the SCADA data recording interval is consistent with the sampling interval; in the unit's main control system, the main control timing data sampling interval is set to 10ms, and at least one main control timing data file is recorded daily by timed or fault alarm triggers.

[0081] Acquire a list of all IO points for SCADA data and main control timing data. Through spectrum analysis of the data, determine the operating parameters required to identify tower bolt loosening, including but not limited to: time, operating mode, wind speed, impeller speed, generator speed, blade angle, active power, ambient temperature, wind direction, nacelle northward angle, nacelle lateral vibration, nacelle longitudinal vibration, gearbox oil level, and yaw system pressure telemetry signals.

[0082] Perform time synchronization processing on the master control timing data and SCADA data.

[0083] S2. Based on the unit operation data, filter the data segments that meet the preset conditions for tower frequency analysis, construct the operating condition classification feature dataset and determine the theoretical tower frequency; remove interference data according to the operating conditions, calculate the impeller frequency and the characteristic frequency of three times the impeller frequency based on the average impeller speed of the removed data segments; when the impeller frequency and the characteristic frequency of three times the impeller frequency exceed the preset range of the theoretical tower frequency, use the NKDE nonparametric kernel density estimation tower frequency adaptive estimation algorithm to obtain the actual tower frequency, and then generate the corresponding time series.

[0084] like Figure 2 As shown, the specific steps include the following:

[0085] S2.1, Integrity assessment of data quality:

[0086] Considering that the first-order natural frequency of the tower of mainstream generating units is below 0.5 Hz, the frequency resolution of the tower to be monitored needs to reach the level of one ten-thousandth of a Hz in order to monitor the slight changes in the tower frequency.

[0087] To ensure the required frequency resolution, the number of data points needed for spectrum analysis is calculated based on the tower's frequency resolution requirements. As shown in the following formula:

[0088] ;

[0089] in, The data sampling frequency is set to 1 in this embodiment; Frequency resolution;

[0090] According to the number of data points The SCADA data is cut and truncated using the equal-length rule, and the data integrity ratio is calculated. The data integrity ratio = actual data points / total number of data points. ;

[0091] The integrity ratio of the data quality is judged to meet the preset conditions for tower frequency analysis. In this embodiment, it is set to 90%, and then the data segments that meet the preset conditions for tower frequency analysis are selected.

[0092] S2.2 Constructing a dataset of unit operating condition classification features:

[0093] A dataset of unit operating condition classification features was constructed based on the selected data segments. The selected data segments consist of time-series index data of unit operating parameters, including unit operating mode, rotor speed, nacelle northward angle, active power, and blade angle. The time-series index data includes: statistical indicators such as maximum, minimum, mean, standard deviation, median, quartiles, range, absolute sum, skewness, and kurtosis; trend indicator indicators such as linear trend slope, linear trend correlation, first-order difference statistics, and trend change variables; time-series indicators such as autocorrelation characteristics, number of repetitions of values, number of times the mean is crossed, and fluctuation persistence; frequency domain indicators such as Fourier coefficients, maximum power spectral density, spectral centroid, and spectral entropy; nonlinear indicator indicators such as sample entropy, fractal dimension, and Lyapunov exponent; and other indicators of the time-series data.

[0094] S2.3 Detrending improves signal-to-noise ratio:

[0095] Considering the significant DC components in parameters such as impeller speed, generator speed, gearbox oil level, and yaw system pressure, detrending processing (i.e., removing low-frequency DC components) is necessary to improve the signal-to-noise ratio in the tower frequency band (0.1~0.5 Hz). Bandpass filters or HP filters can be used for time-series signal processing to detrend the time-series signals of the operating condition classification feature dataset.

[0096] Hodrick-Prescott (HP) filtering is a data smoothing technique for time series analysis. Its core objective is to decompose a time series data into a trend component (reflecting long-term trends) and a periodic component (reflecting short-term fluctuations), achieved by minimizing the following objective function. This embodiment uses HP filtering for time series signal processing to detrend a dataset of operational condition classification features, as detailed below:

[0097] The time series of the operational condition classification feature dataset is decomposed into trend and periodic components, which is achieved by minimizing the following objective function, as shown in the following equation:

[0098] ;

[0099] in, The time series to be decomposed; , , Trend components are respectively , , Data at any given moment is used to reflect long-term trends in data changes; It is a periodic component (reflecting short-term fluctuations in the data); This is a smoothing parameter used to balance goodness of fit and smoothness. A larger value... Values ​​that make the trend component smoother, while smaller values ​​will make the trend component smoother. The value allows the trend component to more closely approximate the fluctuations of the original data; the first item This refers to the degree of deviation between the original data and the trend component, ensuring that the trend component fits the original data well. (Second item) It is the sum of squares of the second differences of the trend components, used to measure the smoothness of the trend components. The larger the second difference, the more drastic the changes in the trend components. Minimizing this term can yield a smoother trend component.

[0100] S2.4 Determination of theoretical tower frequency:

[0101] Fourier transform is performed on the detrended time-series signal, and a spectrum diagram is plotted. The theoretical tower estimated frequency is obtained based on multiple pre-selected data segments. By comparing and analyzing the theoretical tower estimated frequencies obtained from multiple data segments, the final theoretical tower frequency of the unit is determined. Specifically, since the main control time-series data Tracelog has a high sampling frequency and good data reliability, and the recorded nacelle vibration and other data show complete time-series waveforms, its spectrum analysis results are highly accurate. The theoretical tower estimated frequency obtained based on SCADA data and the theoretical tower estimated frequency obtained based on main control time-series data are compared and verified. If the frequency consistency is met under different unit speed conditions, the above theoretical tower estimated frequency is further compared and verified with the multi-mode frequency of the tower output by the unit simulation. If the multi-mode frequency of the tower output by the simulation is within ±10% of the preset theoretical tower estimated frequency, the first-order mode frequency of the tower output by the simulation is determined as the final theoretical tower frequency of the unit.

[0102] S2.5 Interference Data Processing:

[0103] In SCADA data, under operating conditions, the spectrum of unit operating parameter data (such as nacelle lateral vibration, nacelle longitudinal vibration, impeller speed, etc.) will contain characteristic frequencies such as 1P (impeller frequency) and 3P (third harmonic of impeller frequency, unique to mainstream three-bladed units). It is necessary to automatically remove the interfering data according to the unit operating conditions, obtain the data segment after removal, and calculate the impeller frequency and the characteristic frequency of the third harmonic of impeller frequency based on the average impeller speed in the data segment after removing interference. Then, it is necessary to determine whether the impeller frequency and the characteristic frequency of the third harmonic of impeller frequency exceed the preset allowable range (i.e., ±10%) of the theoretical tower frequency.

[0104] S2.6, Tower frequency adaptive estimation:

[0105] If the data does not exceed the limit, the corresponding data segment is removed. If it exceeds the limit, the tower frequency adaptive estimation algorithm based on NKDE nonparametric kernel density estimation is used to obtain the actual tower frequency, and then the corresponding time series is generated. In the spectrum diagram, within the frequency band of the actual tower frequency, the frequency corresponding to the maximum amplitude is generally taken as the actual tower frequency point.

[0106] NKDE is a nonparametric probability density estimation method. It considers that spectrum data is essentially a type of data similar to probability density function. The core is to use a kernel function to generate smooth "hills" around each sample point, and after superposition, a continuous and smooth curve is obtained. This can avoid the offset and jump of the actual tower frequency point under different operating conditions.

[0107] Since the spectral data within a specific frequency band can be fitted using two core parameters, namely the kernel function and the window width, the fitting effect is quite detailed. Furthermore, the optimal window width value can be optimized based on the Asymptotic Mean Integral Square Error (AMISE) criterion, thereby achieving adaptive estimation of the tower frequency. Therefore, it has considerable versatility.

[0108] The mathematical expression for the NKDE estimate is shown below:

[0109] ;

[0110] in, This represents the number of spectral lines in the spectrum. Let be the amplitude of the i-th spectral line; This is the window width value; For kernel functions; In order to be in The estimate of the point based on nonparametric kernel density.

[0111] Commonly used kernel functions include: Gaussian kernel, Epanechnikov kernel, uniform kernel, trigonometric kernel, and cosine kernel. This embodiment uses the widely applicable Gaussian kernel for estimation, and its expression is:

[0112]

[0113] In summary, the mathematical expression for nonparametric kernel density estimation of NKDE using the Gaussian kernel function is as follows:

[0114] ;

[0115] Among them, the window width value The optimal window width is determined based on the asymptotic integral mean square error criterion, and its value is determined by the corresponding mathematical expression, as shown in the following formula:

[0116] ;

[0117] in, The standard deviation of the total amplitude of the spectral lines;

[0118] Based on the above method, the tower spectrum data can be fitted within the actual tower frequency band. Then, the peak value can be automatically detected and located using the existing Python library package scipy.signal.findpeaks to obtain a high-precision actual tower frequency.

[0119] According to the number of data points The SCADA data is segmented and truncated using equal-length rules. By analyzing each segment individually, high-precision actual tower frequencies are obtained, and a time series sequence of the actual tower frequencies is generated.

[0120] S3. The stability analysis of the generated time series is performed using the single-value control chart test method. If the test determines that the time series is unstable, it indicates that the tower frequency has undergone a slight non-random change. The unit can be marked as having a suspected loose tower bolt abnormality, which requires close attention and further analysis and confirmation. Therefore, the corresponding data point is marked as the suspected loose tower bolt abnormality point.

[0121] The rules and specific K values ​​for the single-value control chart test are shown in Table 1 below:

[0122] Rule 1 A point is more than K standard deviations away from the center line 3 Rule 2 K consecutive points on the same side of the center line 9 Rule 3 K consecutive points are either increasing or decreasing 6 Rule 4 K consecutive points, alternating up and down 14 Rule 5 Of the K+1 points, K points are more than 2 standard deviations away from the center line (on the same side). 2 Rule 6 K consecutive points, within 1 standard deviation of the center line (on either side). 15 Rule 7 K consecutive points are located at a distance greater than 1 standard deviation from the center line (on either side). 8

[0123] Table 1. Rules and specific K-values ​​for the single-value control chart test.

[0124] Figure 3 The actual tower frequency time series data of a certain unit is given. For example... Figure 3 As shown, the tower frequency remained stable at 0.258 Hz from November to December 2024, fluctuating by 0.002. On December 24, 2024, the tower frequency showed a slight change, dropping to 0.240 Hz, and fluctuated between 0.260 and 0.238 Hz for three consecutive months. The stability test of this data sequence failed the value control chart verification method. The decrease in the unit's tower frequency indicates a decrease in tower stiffness and suspected loosening of tower bolts.

[0125] S4. Combining the time series and the unit operating condition classification feature dataset, outlier detection algorithms (such as Local Outlier Factor) are used to identify suspected loose tower bolt anomalies. This identifies whether there are concentrated anomalies in the unit operating condition classification feature dataset. If the actual tower frequency is lower than the theoretical tower frequency, and the abnormal tower frequencies are concentrated within a certain nacelle angle range, such as... Figure 4As shown, the outlier detection algorithm shows that the data points with low actual tower frequency are concentrated between 120° and 200° north of the nacelle. This indicates that the tower bolts of the unit are loose in this direction, thus determining the location of the loose tower bolts.

[0126] Finally, information about abnormal units with loose tower bolts was fed back to the project site, and on-site personnel were arranged to promptly check the tower bolt torque. The findings were then fed back to facilitate model review and optimization.

[0127] This invention is based on the principle that loose tower bolts will cause a slight change in the tower frequency. By integrating unit operating parameter data, it sequentially performs data interception, operating condition identification, detrending processing, anti-interference processing, spectrum analysis, and adaptive estimation of tower frequency to achieve high-precision acquisition of the actual tower frequency. Subsequently, the stability of the unit tower frequency is verified, and the loosening of tower bolts is analyzed and confirmed, thereby achieving reliable identification and early warning of loose tower bolts.

[0128] In summary, this invention does not require dedicated monitoring hardware equipment. It can obtain high-precision actual tower frequency in real time by integrating existing general operating data of the unit. When the algorithm detects signs of loosening of tower bolts, it issues an early warning in a timely manner, and then arranges on-site personnel to check and re-inspect the tower bolt torque to ensure the safe and stable operation of the unit.

[0129] Example 2:

[0130] This embodiment provides a wind turbine tower bolt loosening identification system to implement the wind turbine tower bolt loosening identification method described in Embodiment 1, including:

[0131] The data acquisition module is used to acquire unit operating data; the unit operating data includes main control timing data and SCADA data;

[0132] The theoretical tower frequency determination module is used to filter data segments that meet the preset conditions for tower frequency analysis based on unit operation data, construct an operating condition classification feature dataset and determine the theoretical tower frequency.

[0133] The module for generating the actual tower frequency and its time series is used to remove interference data according to the operating conditions, and calculate the impeller frequency and the characteristic frequency of three times the impeller frequency based on the average impeller speed of the data segment after removal; when the impeller frequency and the characteristic frequency of three times the impeller frequency exceed the preset range of the theoretical tower frequency, the tower frequency adaptive estimation algorithm based on NKDE nonparametric kernel density estimation is used to obtain the actual tower frequency, and then generate the corresponding time series.

[0134] The bolt suspected loosening anomaly marking module is used to perform stability analysis on the generated time series using the single-value control chart test method. If the test determines that the time series is unstable, the corresponding data point is marked as a tower bolt suspected loosening anomaly.

[0135] The bolt loosening location determination module combines time series data with unit operating condition classification feature datasets and uses an outlier detection algorithm to identify suspected loosening anomalies of marked tower bolts. Based on the concentrated distribution characteristics of the anomalies in the unit operating condition classification feature dataset, the loosening location of the tower bolts is determined.

[0136] Example 3:

[0137] This embodiment provides a storage medium, characterized in that, when the program is executed by a processor, it implements the wind turbine tower bolt loosening identification method described in Embodiment 1.

[0138] Example 4:

[0139] This embodiment provides a computing device, including a processor and a memory for storing processor-executable programs. The processor executes the program stored in the memory to implement the wind turbine tower bolt loosening identification method described in Embodiment 1.

[0140] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope disclosed in the present invention, based on the technical solution and inventive concept of the present invention, shall fall within the scope of protection of the present invention.

Claims

1. A method for identifying loose bolts on wind turbine towers, characterized in that, Including steps, S1. Acquire unit operating data; the unit operating data includes main control timing data and SCADA data; S2. Based on the unit operation data, filter the data segments that meet the preset conditions for tower frequency analysis, construct the operating condition classification feature dataset and determine the theoretical tower frequency; remove the interference data according to the operating conditions, and calculate the impeller frequency and the characteristic frequency of three times the impeller frequency based on the average impeller speed of the removed data segments. When the impeller frequency and the characteristic frequency of three times the impeller frequency exceed the preset range of the theoretical tower frequency, the tower frequency adaptive estimation algorithm based on NKDE nonparametric kernel density estimation is used to obtain the actual tower frequency, and then the corresponding time series is generated. S3. The stability analysis of the generated time series is performed using the single-value control chart test method. If the test determines that the time series is unstable, the corresponding data point is marked as a suspected loosening abnormal point of the tower bolts. S4. Combining the time series and the unit operating condition classification feature dataset, an outlier detection algorithm is used to identify suspected loosening anomalies of the marked tower bolts. Based on the concentrated distribution characteristics of the anomalies in the unit operating condition classification feature dataset, the location of the loosening of the tower bolts is determined.

2. The method for identifying loose bolts on a wind turbine tower according to claim 1, characterized in that, In step S1, based on the Nyquist sampling theorem, the SCADA data sampling interval is set to 1 second in the unit's SCADA system, and the SCADA data recording interval is consistent with the sampling interval; in the unit's main control system, the main control timing data sampling interval is set to 10ms, and at least one main control timing data file is recorded daily by timed or fault alarm triggers. Acquire a list of all IO points for SCADA data and main control timing data. Through spectrum analysis of the data, determine the operating parameters required to identify tower bolt loosening, including but not limited to: time, operating mode, wind speed, impeller speed, generator speed, blade angle, active power, ambient temperature, wind direction, nacelle northward angle, nacelle lateral vibration, nacelle longitudinal vibration, gearbox oil level, and yaw system pressure telemetry signals. Perform time synchronization processing on the master control timing data and SCADA data.

3. The method for identifying loose bolts on wind turbine towers according to claim 1, characterized in that, Step S2 includes the following steps: S2.1 Calculate the number of data points required for spectrum analysis based on the tower's frequency resolution requirements. As shown in the following formula: ； in, For data sampling frequency, Frequency resolution; According to the number of data points The SCADA data is cut and truncated using the equal-length rule, and the data integrity ratio is calculated. The data integrity ratio = actual data points / total number of data points. ; Determine whether the data integrity ratio meets the preset conditions for tower frequency analysis, and then filter out the data segments that meet the preset conditions for tower frequency analysis; S2.2 Construct a dataset of unit operating condition classification features based on the selected data segments; S2.3 To improve the signal-to-noise ratio of the frequency band in which the tower frequency is located, a bandpass filter or HP filter is used to perform detrending processing on the time series signal of the operating condition classification feature dataset. S2.4 Perform Fourier transform on the detrended time-series signal and plot the spectrum. Obtain the theoretical tower estimated frequency based on the preset selected data segments. By comparing and analyzing the theoretical tower estimated frequencies obtained from the data segments, the final theoretical tower frequency of the unit is determined. S2.

5. Based on the SCADA data under operating conditions, perform spectrum analysis on the unit operating parameters such as nacelle left-right vibration, nacelle forward-backward vibration, and impeller speed, and automatically remove interference data according to the unit operating conditions; calculate the impeller frequency and the characteristic frequency of three times the impeller frequency based on the average impeller speed in the data segment after removing interference, and determine whether the impeller frequency and the characteristic frequency of three times the impeller frequency exceed the preset allowable range of the theoretical tower frequency; S2.6 If the frequency exceeds the limit, the tower frequency adaptive estimation algorithm based on NKDE nonparametric kernel density estimation is used to obtain the actual tower frequency, and then the corresponding time series is generated; if the frequency does not exceed the limit, the corresponding data segment is discarded.

4. The method for identifying loose bolts on a wind turbine tower according to claim 3, characterized in that, The selected data segments are time-series index data of unit operating parameters, including unit operating mode, rotor speed, nacelle northward angle, active power, and blade angle.

5. The method for identifying loose bolts on a wind turbine tower according to claim 3, characterized in that, The HP filtering time-series signal processing method is used to detrend the operating condition classification feature dataset, as detailed below: The time series of the operational condition classification feature dataset is decomposed into trend and periodic components, which is achieved by minimizing the following objective function, as shown in the following equation: ； in, The time series to be decomposed , , Trend components are respectively , , Data at any given time The second difference of the trend component is used, and its sum of squares is used to measure the smoothness of the trend component. It is a periodic component. For smoothing parameters.

6. The method for identifying loose bolts on a wind turbine tower according to claim 3, characterized in that, In step S2.4, the theoretical tower estimated frequency obtained based on SCADA data and the theoretical tower estimated frequency obtained based on main control timing data are compared and verified. If the frequency consistency under different unit speed conditions is met, the theoretical tower estimated frequency is further compared and verified with the multi-mode frequency of the tower output by the unit simulation. If the multi-mode frequency of the tower output by the simulation is within the preset range of the theoretical tower estimated frequency, the first-order mode frequency of the tower output by the simulation is determined as the final theoretical tower frequency of the unit.

7. The method for identifying loose bolts on a wind turbine tower according to claim 3, characterized in that, In step S2.6, the mathematical expression for the nonparametric kernel density estimation of NKDE using the Gaussian kernel function is shown below: ； in, This represents the number of spectral lines in the spectrum. Let be the amplitude of the i-th spectral line; In order to be in The estimate of the point based on nonparametric kernel density; The optimal window width is determined based on the asymptotic integral mean square error criterion, and its value is determined by the corresponding mathematical expression, as shown in the following formula: ； in, The standard deviation of the total amplitude of the spectral lines; Next, within the preset frequency band of the tower frequency, the tower spectrum data is fitted and the peak value is located to obtain the actual tower frequency of the unit. Finally, based on the number of data points The SCADA data is segmented and cut using the equal-length rule. The actual tower frequencies are obtained by analyzing each segment, and then a time series of the actual tower frequencies is generated.

8. A system for identifying loose bolts on wind turbine towers, characterized in that, A method for identifying loose tower bolts of a wind turbine as described in any one of claims 1 to 7 includes: The data acquisition module is used to acquire unit operating data; the unit operating data includes main control timing data and SCADA data; The theoretical tower frequency determination module is used to filter data segments that meet the preset conditions for tower frequency analysis based on unit operation data, construct an operating condition classification feature dataset and determine the theoretical tower frequency. The module for generating the actual tower frequency and its time series is used to remove interference data according to the operating conditions, and calculate the impeller frequency and the characteristic frequency of three times the impeller frequency based on the average impeller speed of the data segment after removal; when the impeller frequency and the characteristic frequency of three times the impeller frequency exceed the preset range of the theoretical tower frequency, the tower frequency adaptive estimation algorithm based on NKDE nonparametric kernel density estimation is used to obtain the actual tower frequency, and then generate the corresponding time series. The bolt suspected loosening anomaly marking module is used to perform stability analysis on the generated time series using the single-value control chart test method. If the test determines that the time series is unstable, the corresponding data point is marked as a tower bolt suspected loosening anomaly. The bolt loosening location determination module combines time series data with unit operating condition classification feature datasets and uses an outlier detection algorithm to identify suspected loosening anomalies of marked tower bolts. Based on the concentrated distribution characteristics of the anomalies in the unit operating condition classification feature dataset, the loosening location of the tower bolts is determined.

9. A storage medium storing a program, characterized in that, When the program is executed by the processor, it implements a method for identifying loose tower bolts of a wind turbine as described in any one of claims 1 to 7.

10. A computing device, comprising a processor and a memory for storing a processor-executable program, characterized in that, When the processor executes the program stored in the memory, it implements the method for identifying loose tower bolts of a wind turbine as described in any one of claims 1 to 7.

Images