Self-data-driven wind turbine generator online monitoring and multistage early warning method

The self-data-driven online monitoring method for wind turbines addresses the limitations of fixed thresholds by adaptively adjusting alarm levels based on real-time data, enhancing fault detection accuracy and responsiveness.

CN119062528APending Publication Date: 2024-12-03XI AN JIAOTONG UNIV
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411377143.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-30
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

Existing wind turbine fault monitoring methods lack flexibility and adaptability, often resulting in false alarms or missed detections due to fixed thresholds and static strategies, failing to accurately distinguish fault severity and respond to changing operating conditions.

Method used

A self-data-driven online monitoring method that includes data preprocessing, feature extraction, operating condition conversion, and multi-level fault warning logic to adaptively adjust alarm thresholds based on real-time data and environmental changes.

Benefits of technology

Enhances the accuracy and adaptability of fault monitoring, reducing false alarms and ensuring timely responses to equipment conditions, thereby improving operational safety and efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119062528A_ABST
    Figure CN119062528A_ABST
Patent Text Reader

Abstract

A self-data-driven wind turbine generator online monitoring and multistage early warning method comprises the following steps: aiming at a plurality of wind turbine generators in a wind field, indexing and acquiring a vibration data set of a specific measuring point of a single generator generator in real time on the basis of a generator generator number and a measuring point position; inputting the vibration data set into a data preprocessing module in sequence, and cleaning the data; extracting two characteristic indexes, namely spectral kurtosis and an effective value, from the preprocessed vibration data; the effective value is subjected to working condition conversion, so that the fan is kept consistent under different rotating speed working conditions; and finally, fault early warning and positioning are carried out, early warning logic is based on step-by-step judgment, early warning of different fault levels can be realized, and comprehensiveness and accuracy of early warning information are ensured. According to the method, the efficiency and accuracy of online monitoring and early warning of the wind turbine generator are improved, and feature changes under different working conditions are effectively identified.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of wind turbine fault monitoring and health management, and in particular relates to a self-data driven wind turbine online monitoring and multi-level early warning method. Background Art

[0002] Reducing operation and maintenance costs is one of the effective ways to improve the profitability of wind farms. In particular, condition monitoring of important components such as gearboxes and generators can detect early faults in a timely manner, which can effectively prevent further deterioration of faults and ultimately reduce maintenance costs, thereby reducing equipment downtime, reducing maintenance costs, and ensuring the continued stability of power generation efficiency.

[0003] In recent years, with the development of big data and artificial intelligence technology, data-driven fault diagnosis methods have gradually received widespread attention. The monitoring of wind power equipment operation data has achieved relatively comprehensive coverage, providing a rich foundation for fault diagnosis. However, most of the existing technical solutions use simple vibration signal preprocessing and fixed threshold alarm methods. These methods usually rely on basic filtering and noise reduction and simple feature parameter extraction to monitor the equipment status and judge whether the equipment is abnormal based on the set fixed threshold.

[0004] However, this method lacks flexibility in actual application and is difficult to effectively distinguish the severity of faults and dynamic responses. When an early fault occurs in the equipment, the system either cannot capture it in time or frequently reports false alarms, which increases the maintenance burden and reduces the reliability of the system. For serious faults, existing methods also find it difficult to automatically upgrade the alarm level and cannot effectively distinguish between minor and serious faults, limiting the safety assurance capabilities of equipment and production. In addition, existing hierarchical warning methods lack adaptability under complex working conditions and changing environments. They usually rely on static thresholds or single models, and it is difficult to adjust the alarm strategy according to equipment status changes and environmental conditions. In the case of frequent changes in working conditions, this static strategy is prone to false alarms or response delays, and cannot reflect the true status of the equipment in a timely manner. Therefore, it is very necessary to propose a self-data-driven online monitoring and multi-level warning method for wind turbines. Summary of the invention

[0005] In order to overcome the shortcomings of the above-mentioned prior art, the purpose of the present invention is to provide a self-data-driven online monitoring and multi-level early warning method for wind turbines, improve the efficiency and accuracy of online monitoring and early warning of wind turbines, and effectively identify characteristic changes under different working conditions.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is:

[0007] A self-data-driven online monitoring and multi-level early warning method for wind turbines first uses a data preprocessing module to clean the raw vibration data collected in real time; then uses the cleaned data to extract features, and performs working condition conversion on the extracted features to eliminate the influence of time-varying working conditions; finally, the features are input into a fault warning module, and different fault warning levels are given through continuous jump warnings.

[0008] A self-data driven wind turbine online monitoring and multi-level early warning method comprises the following steps:

[0009] 1) Obtain vibration signals: For multiple wind turbines in a wind farm, based on the unit number and measurement point location, the vibration data set of a specific measurement point of a single unit is indexed and obtained in real time;

[0010] 2) Data preprocessing: The vibration data sets obtained in step 1) are sequentially input into the data preprocessing module to clean the data;

[0011] 3) Feature extraction and index analysis: Extract two characteristic indexes, spectral kurtosis and effective value, from the vibration data preprocessed in step 2);

[0012] 4) Operating condition conversion: The effective value extracted in step 3) is converted to an operating condition so that the fan remains consistent under different speed conditions;

[0013] 5) Fault warning and location: The benchmark effective value sequence after the working condition conversion is input into the component warning module, and threshold calculation and warning judgment are performed on the vibration data of multiple components, multiple measuring points and multiple directions. The warning logic is based on step-by-step judgment, which can realize warnings of three different fault levels to ensure the comprehensiveness and accuracy of the warning information; the final generated warning description covers all components, measuring points and directions that trigger the warning, and records the over-limit situation of each abnormal point in detail.

[0014] The step 2) is specifically as follows: first, data is screened based on the fan speed to remove data whose speed is not in working state; second, the vibration data set is de-meaned to ensure that the signal fluctuates around zero; next, the vibration data set is filtered to remove noise and signal components of non-target frequency bands, retain valid information of specific frequency bands, and output filtered signal data; finally, the signal data after filtering is used as the preprocessing result.

[0015] In the step 3), the effective value is used as a time domain indicator to reflect the amplitude and energy level of the vibration data. The vibration intensity of the equipment and possible abnormal conditions are determined by calculating the root mean square value of the vibration data. In the feature extraction of spectral kurtosis, the vibration data is decomposed into multiple levels by designing a filter group, the features of different frequency bands are extracted layer by layer, and the kurtosis value of each frequency band is calculated, so as to quantify the non-Gaussian characteristics of the vibration data in each frequency band, select the frequency band with the largest kurtosis value in each frequency band, locate the frequency band that may contain pulse or fault information, and output its center frequency and bandwidth information.

[0016] The step 4) is specifically as follows: first, determine different reference speed operating ranges according to actual conditions and calculate the reference speed band; second, construct an interpolation function, smoothly connect each characteristic point in turn, optimize the fitting parameters by the least squares method to minimize the fitting error, and establish a reference operating condition conversion model; finally, convert the characteristic values ​​under the working speed range into reference characteristic values ​​through the reference operating condition conversion model to achieve operating condition conversion.

[0017] The specific steps of step 5) are as follows:

[0018] a) Online data reading: traverse in the order of wind farm, wind turbine, component, measuring point and direction, and realize online monitoring according to the processing date, filtering out valid vibration data in units of days;

[0019] b) Comprehensive early warning logic: First, determine the fluctuation range of the healthy operation characteristic value; second, set the continuous trigger acc num The historical maximum trigger max num Record the current number of times exceeding the fluctuation range and the maximum number of times exceeding the fluctuation range before; then traverse the observed values ​​of the benchmark effective value and the benchmark spectral kurtosis respectively, and increase the continuous trigger acc if it exceeds the healthy fluctuation range num , if the new observation is in the healthy range again, reset acc num And update max num ; Finally, the "OR" logic warning is used. When any of the continuous trigger counts of the benchmark effective value or the benchmark spectrum kurtosis exceeds num Compared with the historical maximum trigger max num When there is a significant jump, acc num -max num >kk∈{1, 2, 3, ...}, the higher level of the two is used for early warning;

[0020] c) Warning level discrimination: Three levels of warning thresholds are set to achieve hierarchical monitoring from early degradation to serious failures. The first threshold is calculated by the data mean and standard deviation, and the healthy signal range boundary is adaptively adjusted. The thresholds corresponding to the second and third level warnings are obtained according to industry standards. According to the continuous warning criteria of the comprehensive warning logic, the sequences of the benchmark spectrum kurtosis and the benchmark effective value are compared with the three thresholds respectively, and the warning level is continuously improved in sequence. The basic data of "warning time", "warning component" and "warning measurement point" are saved in the warning data table;

[0021] d) Warning information description: First, all monitoring points and their directions are traversed to extract warning-related channels, characteristic values ​​and threshold information; second, according to the comprehensive warning level, combined with the specific measuring points and directions of the warning, a detailed warning description is generated, including the triggering cause, over-limit characteristic values ​​and threshold conditions, and three different warning descriptions of "severe alarm", "moderate alarm" and "minor alarm" are generated; finally, all warning information is integrated into a multi-dimensional warning result table, and complete warning data containing detailed information of wind farms, units, components, measuring points, directions and warning levels is output.

[0022] Compared with the prior art, the present invention has the following beneficial effects:

[0023] The present invention adopts an intelligent strategy of multi-level early warning and online monitoring, and refines and clarifies the entire monitoring process, so it has higher sensitivity and adaptability in equipment status monitoring and fault early warning. The present invention sets multiple alarm levels such as mild, moderate, and severe, and adaptively adjusts the alarm threshold, effectively solving the problems of false alarms and missed alarms caused by fixed thresholds in the prior art, ensuring early warning of faults and timely response to serious faults. At the same time, the present invention adopts a refined signal processing method and a dynamic adjustment mechanism of the operating condition conversion model, so that the system can optimize the alarm strategy in real time according to actual operating condition changes, overcoming the limitation of the prior art that it cannot cope with the changing operating conditions, and significantly improving the safety and efficiency of equipment operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a flow chart of the present invention.

[0025] Figure 2 : is a characteristic trend diagram of the horizontal direction of the free end of the generator in the embodiment, wherein the mutation alarm point is marked by a black box, and the solid line is the reference effective value after the operating condition is converted.

[0026] Figure 3 It is the horizontal vibration signal, spectrum and envelope spectrum of the free end of the generator of the faulty unit in the embodiment.

[0027] Figure 4It is the horizontal vibration signal, spectrum and envelope spectrum of the free end of the normal generator of the embodiment at the same time and similar speed. DETAILED DESCRIPTION

[0028] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0029] Reference Figure 1 , a self-data driven wind turbine online monitoring and multi-level early warning method, comprising the following steps:

[0030] 1) Obtain vibration signals: Vibration signals and historical characteristics are the basis for fault diagnosis analysis and processing. For multiple wind turbines in a wind farm, the vibration data set of a specific measuring point of a single turbine is indexed and obtained in real time based on the turbine number and measuring point location;

[0031] 2) Data preprocessing: The vibration data sets obtained in step 1) are sequentially input into the data preprocessing module to perform operations such as data cleaning to improve data quality and ensure the accuracy of subsequent diagnostic steps;

[0032] The specific steps are as follows: first, data is screened based on the fan speed to remove data whose speed is not in working state, so as to avoid analyzing such invalid data; second, the vibration data set is de-meaned to ensure that the signal fluctuates around zero value to eliminate DC offset; next, the vibration data set is filtered to remove noise and signal components of non-target frequency bands, retain valid information of specific frequency bands, and output filtered signal data; finally, the filtered signal data is returned as the preprocessing result to provide high-quality data for subsequent analysis;

[0033] 3) Feature extraction and index analysis: Extract two characteristic indexes, spectral kurtosis and effective value, from the vibration data preprocessed in step 2);

[0034] The effective value is a time domain indicator used to reflect the amplitude and energy level of the vibration data. By calculating the root mean square value of the vibration data, the vibration intensity of the equipment and possible abnormal conditions can be determined.

[0035] In the feature extraction of spectral kurtosis, the vibration data is decomposed at multiple levels by designing a filter bank, the features of different frequency bands are extracted layer by layer, and the kurtosis value of each frequency band is calculated, so as to quantify the non-Gaussian characteristics of the vibration data in each frequency band, select the frequency band with the largest kurtosis value in each frequency band, locate the frequency band that may contain pulse or fault information, and output its center frequency and bandwidth information, which provides an important basis for subsequent fault analysis;

[0036] 4) Operating condition conversion: The effective value extracted in step 3) is converted to an operating condition so that the fan remains consistent under different speed conditions;

[0037] Specifically, firstly, different reference speed operating ranges are determined according to actual conditions, and the reference speed band is calculated; secondly, an interpolation function is constructed to smoothly connect each characteristic point in sequence to provide a basis for subsequent fitting; on this basis, the fitting parameters are optimized by the least squares method to minimize the fitting error and establish a reference operating condition conversion model; finally, the characteristic values ​​under the working speed range are converted into reference characteristic values ​​through the reference operating condition conversion model to realize the operating condition conversion;

[0038] 5) Fault warning and location: The reference effective value sequence after the working condition conversion is input into the component warning module, and the threshold value calculation and warning judgment are performed for the vibration data of multiple components, multiple measuring points and multiple directions. The warning logic is based on step-by-step judgment, which can realize warnings of three different fault levels to ensure the comprehensiveness and accuracy of the warning information; the warning description finally generated covers all components, measuring points and directions that trigger the warning, and records the over-limit situation of each abnormal point in detail;

[0039] The specific steps are as follows:

[0040] a) Online data reading: After completing the data processing of steps 1) to 4) above, traverse in the order of wind farm, wind turbine, component, measuring point, and direction, and implement online monitoring based on the processing date, filtering out valid vibration data in units of days to avoid introducing interference from historical data;

[0041] b) Comprehensive early warning logic: Here, the main logic of the subsequent early warning behavior is clarified. First, the fluctuation range of the healthy operation characteristic value is determined; secondly, continuous triggers acc are set respectively. num The historical maximum trigger max num Record the current number of times exceeding the fluctuation range and the maximum number of times exceeding the fluctuation range before; then traverse the observed values ​​of the benchmark effective value and the benchmark spectral kurtosis respectively, and increase the continuous trigger acc if it exceeds the healthy fluctuation range num , if the new observation is in the healthy range again, reset acc num And update max num ; Finally, the "OR" logic warning is used. When any of the continuous trigger counts of the benchmark effective value or the benchmark spectrum kurtosis exceeds num Compared with the historical maximum trigger max num When there is a significant jump, acc num -max num >kk∈{1, 2, 3, ...}, the higher level of the two is used for early warning;

[0042] c) Warning level discrimination: In order to more accurately reflect the equipment status and improve the flexibility of equipment monitoring, three levels of warning thresholds are set to achieve hierarchical monitoring from early degradation to serious failures. The first threshold is calculated by the data mean and standard deviation, and the healthy signal range boundary is adaptively adjusted. The thresholds corresponding to the second and third level warnings are obtained according to industry standards. On this basis, according to the continuous warning criterion of the comprehensive warning logic, the sequence of the benchmark spectrum kurtosis and the benchmark effective value is compared with the three thresholds respectively, and the warning level is continuously improved in sequence, and the basic data such as "warning time", "warning component", and "warning measurement point" are saved in the warning data table;

[0043] d) Warning information description: First, all monitoring points and their directions are traversed to extract warning-related channels, characteristic values ​​and threshold information; secondly, according to the comprehensive warning level, combined with the specific measuring points and directions of the warning, a detailed warning description is generated, including the triggering cause, over-limit characteristic values ​​and threshold conditions, and three different warning descriptions of "serious alarm", "moderate alarm" and "minor alarm" are generated; finally, all warning information is integrated into a multi-dimensional warning result table, and complete warning data containing detailed information such as wind farms, units, components, measuring points, directions and warning levels is output to provide support for further fault analysis and troubleshooting decisions.

[0044] In this embodiment, according to the above process steps, taking the doubly-fed wind turbine as an example, a total of 39 sampling channels are set in the wind turbine as shown in Table 1. The actual data collected by the wind turbine gearbox_first-level free_horizontal direction sensor will be verified below.

[0045] Table 1 Sampling channel settings for double-fed wind turbine

[0046]

[0047]

[0048] After obtaining targeted data through sensors, the vibration data collected by the wind turbines are continuously processed according to the steps of preprocessing, feature extraction and index analysis, working condition conversion, fault warning and location. According to the relevant industry standards of gearbox effective value, threshold 2 alarm2 = 7.5 and threshold 3 alarm3 = 12 are selected. Finally, the early warning system issues a warning level of "moderate alarm" based on the benchmark effective value after working condition conversion. By observing Figure 2 It can be seen from the characteristic trend diagram shown (the warning point is marked with a black box) that the characteristic value after the working condition conversion meets the continuous jump warning logic in step 5), and the warning is correct.

[0049] Reference Figure 3 and Figure 4By comparing the vibration data of normal units and faulty units at the warning moment, it can be seen that the time domain amplitude of the warning unit is significantly larger than that of the normal unit, and has obvious periodic impact. The frequency components in the frequency domain are complex, and unknown frequency components appear, and there is a possibility of abnormality.

Claims

1. A self-data driven wind turbine online monitoring and multi-level early warning method, characterized in that: First, the data preprocessing module is used to clean the raw vibration data collected in real time; then the cleaned data is used to extract features, and the extracted features are converted into working conditions to eliminate the influence of time-varying working conditions; finally, the features are input into the fault warning module, and different fault warning levels are given through continuous jump warnings.

2. A self-data driven wind turbine online monitoring and multi-level early warning method, characterized in that: The following steps are involved: 1) Obtain vibration signals: For multiple wind turbines in a wind farm, based on the unit number and measurement point location, the vibration data set of a specific measurement point of a single unit is indexed and obtained in real time; 2) Data preprocessing: The vibration data sets obtained in step 1) are sequentially input into the data preprocessing module to clean the data; 3) Feature extraction and index analysis: Extract two characteristic indexes, spectral kurtosis and effective value, from the vibration data preprocessed in step 2); 4) Operating condition conversion: The effective value extracted in step 3) is converted to an operating condition so that the fan remains consistent under different speed conditions; 5) Fault warning and location: The benchmark effective value sequence after the working condition conversion is input into the component warning module, and threshold calculation and warning judgment are performed on the vibration data of multiple components, multiple measuring points and multiple directions. The warning logic is based on step-by-step judgment, which can realize warnings of three different fault levels to ensure the comprehensiveness and accuracy of the warning information; the final generated warning description covers all components, measuring points and directions that trigger the warning, and records the over-limit situation of each abnormal point in detail.

3. The method according to claim 2, characterized in that The step 2) is specifically as follows: first, data is screened based on the fan speed to remove data whose speed is not in working state; second, the vibration data set is de-meaned to ensure that the signal fluctuates around zero; next, the vibration data set is filtered to remove noise and signal components of non-target frequency bands, retain valid information of specific frequency bands, and output filtered signal data; finally, the signal data after filtering is used as the preprocessing result.

4. The method according to claim 2, characterized in that: In the step 3), the effective value is used as a time domain indicator to reflect the amplitude and energy level of the vibration data. The vibration intensity of the equipment and possible abnormal conditions are determined by calculating the root mean square value of the vibration data. In the feature extraction of spectral kurtosis, the vibration data is decomposed into multiple levels by designing a filter group, the features of different frequency bands are extracted layer by layer, and the kurtosis value of each frequency band is calculated, so as to quantify the non-Gaussian characteristics of the vibration data in each frequency band, select the frequency band with the largest kurtosis value in each frequency band, locate the frequency band that may contain pulse or fault information, and output its center frequency and bandwidth information.

5. The method according to claim 2, characterized in that: The step 4) is specifically as follows: first, determine different reference speed operating ranges according to actual conditions and calculate the reference speed band; second, construct an interpolation function, smoothly connect each characteristic point in turn, optimize the fitting parameters by the least squares method to minimize the fitting error, and establish a reference operating condition conversion model; finally, convert the characteristic values ​​under the working speed range into reference characteristic values ​​through the reference operating condition conversion model to achieve operating condition conversion.

6. The method according to claim 2, characterized in that The specific steps of step 5) are as follows: a) Online data reading: traverse in the order of wind farm, wind turbine, component, measuring point and direction, and realize online monitoring according to the processing date, filtering out valid vibration data in units of days; b) Comprehensive early warning logic: First, determine the fluctuation range of the healthy operation characteristic value; second, set the continuous trigger accnum and the historical maximum trigger maxnum to record the current number of times exceeding the fluctuation range and the maximum number of times exceeding the fluctuation range before; Next, the observed values ​​of the benchmark effective value and the benchmark spectral kurtosis are traversed respectively. If they exceed the healthy fluctuation range, the continuous trigger accnum is increased. If the new observed value is in the healthy interval again, accnum is reset and maxnum is updated. Finally, the "or" logic warning is adopted. When any continuous trigger count accnum of the benchmark effective value or the benchmark spectral kurtosis shows a significant jump compared with the historical maximum trigger maxnum, that is, accnum-maxnum>kk∈{1,2,3,…}, the higher level of the two is used for warning. c) Warning level discrimination: Three levels of warning thresholds are set to achieve hierarchical monitoring from early degradation to serious failures. The first threshold is calculated by the data mean and standard deviation, and the healthy signal range boundary is adaptively adjusted. The thresholds corresponding to the second and third level warnings are obtained according to industry standards. According to the continuous warning criteria of the comprehensive warning logic, the sequences of the benchmark spectrum kurtosis and the benchmark effective value are compared with the three thresholds respectively, and the warning level is continuously improved in sequence. The basic data of "warning time", "warning component" and "warning measurement point" are saved in the warning data table; d) Warning information description: First, traverse all monitoring points and their directions to extract warning-related channels, characteristic values ​​and threshold information; second, according to the comprehensive warning level, combined with the specific measurement points and directions of the warning, generate a detailed warning description, including the triggering cause, over-limit characteristic values ​​and threshold conditions, and generate three different warning descriptions of "severe alarm", "moderate alarm" and "minor alarm"; finally, integrate all warning information into a multi-dimensional warning result table, and output complete warning data containing detailed information on wind farms, units, components, measurement points, directions and warning levels.

Citation Information

Cited By

  • Fan tower health management method based on modal model and storage medium

    CN121139286A