Online Condition Monitoring and Warning System and Method for Pitch Reducer of Wind Turbine

Abstract

The present invention provides an on-line condition monitoring and early warning system and method for a pitch reducer of a wind turbine, including: collecting real-time vibration signals of three pitch reducers by using vibration sensors, and transmitting the vibration signals to a micro industrial control computer in the nacelle through a wireless transmission acquisition card; the micro industrial control computer is used for receiving the real-time vibration signals and identifying the effective vibration signals caused by pitch actions, so as to realize on-line condition monitoring of the pitch reducer; the early warning system first receives the effective vibration signals sent by the nacelle radio station, and realizes fault early warning of the pitch reducer by constructing a comprehensive pitch vibration index and a dynamic threshold.

Description

Technical Field

[0001] The present invention relates to the technical field of wind turbine fault warning, and particularly to an on-line condition monitoring and warning system and method for a pitch reducer of a wind turbine. Background Art

[0002] As a huge whole, a wind turbine is composed of multiple systems. As an actuator for pitch control, the reducer is a subsystem with a relatively high failure rate of the generating set. If a fault occurs and is not discovered and repaired in time, it will endanger the safe operation of the unit, form an irreparable fault, and even lead to the out-of-control, collapse of the unit and other serious consequences. Recently, pitch reducers of multiple units on site have failed, which has had a serious impact on the safety and economic operation and maintenance of the wind farm.

[0003] The fault of a wind turbine is a process of gradually increasing deterioration. At first, the degree of deterioration is not large. As the wind turbine operates, the deterioration becomes more and more serious, and serious faults such as broken teeth will occur at a certain moment, resulting in the shutdown of the wind turbine. During the process from the appearance of deterioration to the occurrence of a fault, it is possible to determine whether a wind turbine is about to have a fault through index analysis.

[0004] At present, the technology of vibration monitoring and fault diagnosis is already very mature, but its application on pitch reducers is rarely seen. Therefore, the fault warning of pitch reducers based on vibration monitoring is feasible and has high on-site application value. Summary of the Invention

[0005] The purpose of the present invention is to solve at least one of the above technical defects.

[0006] To this end, the purpose of the present invention is to provide an on-line condition monitoring and warning system and method for a pitch reducer of a wind turbine to solve the problems mentioned in the background art and overcome the deficiencies in the prior art.

[0007] To achieve the above purpose, an embodiment of the present invention provides an on-line condition monitoring and warning system for a pitch reducer of a wind turbine, including:

[0008] A vibration sensor is respectively installed on each pitch reducer of the wind turbine. The vibration sensor is installed on the surface of the housing of the secondary gear ring of the corresponding pitch reducer. The output end of each vibration sensor is connected to a wireless transmission acquisition card, and the wireless transmission acquisition card is further connected to a wireless receiving micro industrial computer inside the nacelle. The wireless receiving micro industrial computer is bidirectionally connected to the nacelle radio station, the nacelle radio station is wirelessly connected to the centralized control room radio station, and the centralized control room radio station is bidirectionally connected to the status warning system, wherein,

[0009] Each of the vibration sensors is used to collect the real-time vibration signals on the corresponding pitch reducer, and send the real-time vibration signals to the wireless receiving micro industrial control computer in the nacelle through a wireless transmission acquisition card;

[0010] The wireless receiving micro industrial control computer is used to receive the collected real-time vibration signals and process them by the following steps:

[0011] 1) Decompose the vibration signal by variational mode decomposition (VMD) to obtain multiple components;

[0012] 2) Eliminate the blade rotation frequency component. For the IMF containing the blade rotation frequency trend, take the frequency value within the range of 0.15 - 0.3 Hz, and this frequency dominates in the spectrum of its blade rotation frequency component; search for the frequency with the largest amplitude within the frequency range [0, 20] Hz of each component. If it falls within the range of 0.15 - 0.3 Hz, then determine that this component is the blade rotation frequency component and eliminate it; at the same time, the non-linear trend component is approximately a DC component, and all components are eliminated after eliminating the blade rotation frequency component, and the remaining components are summed and reconstructed into the remaining signal;

[0013] 3) Use the Hilbert transform method to obtain the envelope signal of the remaining signal. The envelope signal contains randomly occurring high-amplitude pulses;

[0014] 4) Sliding median filtering for noise reduction: Smooth this envelope signal through sliding median filtering, suppress the possible high-amplitude pulse interference, and retain the effective vibration components;

[0015] 5) Identify the pitch action: Take the real-time data received by the micro industrial control computer as the input, execute steps 1) to 4), and calculate and determine the threshold for pitch using the 3σ criterion; compare the smoothed envelope signal with this threshold, lock the effective pitch action time period, and save the data of this time period;

[0016] 6) On the basis of pitch action recognition, use support vector data description (SVDD) to perform hypersphere modeling on the time-domain vibration characteristics of the reducer, so as to realize the fault warning of the pitch reducer;

[0017] The status warning system synchronously monitors the running status and communication status of each pitch reducer according to the warning signal. If the reducer status is abnormal, analyze whether the reducer has impact-type or wear-type faults through comprehensive indicators.

[0018] Preferably, according to any of the above solutions, on the basis of pitch action recognition, the warning system uses support vector data description (SVDD) to perform hypersphere modeling on the time-domain and frequency-domain vibration characteristics of the reducer under normal conditions, and uses the distance between the characteristic data and the center of the hypersphere as the basis for fault warning.

[0019] Preferably, according to any of the above solutions, if the wireless receiving micro industrial control computer determines that the current running state of the speed reducer is normal, updates the data set, and recalculates the threshold, thereby realizing a dynamic threshold.

[0020] Preferably, according to any of the above solutions, the wireless receiving micro industrial control computer uses SVDD to perform a hypersphere modeling on the comprehensive characteristics of the speed reducer to realize the fault warning of the pitch speed reducer, including:

[0021] 1) Offline feature extraction: Perform VMD decomposition on the pitch vibration signal, remove the periodic fluctuation component of the vibration signal caused by the blade rotation, and extract the comprehensive characteristic indicators composed of traditional time domain indicators and time-frequency domain indicators;

[0022] 2) Offline comprehensive index dimensionality reduction: Dimensionality reduction is performed on the comprehensive characteristic indicators composed of traditional time domain indicators and time-frequency domain indicators to obtain two-dimensional indicators representing the main change trends of each indicator;

[0023] 3) Offline model training: Use the two-dimensional indicators in the normal state as the training samples of the SVDD model to obtain the optimal hypersphere containing normal samples, and the radius of the hypersphere is R. This process is the training process;

[0024] 4) Establish an SVDD warning model: Extract features and perform comprehensive index dimensionality reduction on the online pitch action signal, input the online two-dimensional indicators into the established SVDD model, and obtain the distance D from it to the center of the hypersphere;

[0025] 5) Establish an adaptive alarm threshold: When setting the threshold, regard the distance index obtained based on SVDD as a separate parameter for observation, and consider that the data change form of the index conforms to the normal distribution; Since the 3σ criterion in probability statistics is defined as, for a variable that conforms to a normal distribution with a mean of μ and a variance of σ 2 when taking values, the probability of taking values in the interval (μ - σ 2 , μ + σ 2 ) is 99.73%; when a certain distance index value does not belong to this range, it is considered an abnormal point, that is, it is in an abnormal state. At the same time, to prevent the influence of external interference factors, if a continuous number of distance index values exceed the 3σ value range defined by the previous distance index value, it is considered that an abnormal value has occurred and an alarm is issued.

[0026] Preferably, according to any of the above solutions, the time domain index parameter set includes: maximum value, minimum value, peak value, peak-to-peak value, average value, root mean square, kurtosis, impulse factor, peak factor, and crest factor; the frequency domain index parameter set includes: average frequency, root mean square of frequency, and standard deviation of frequency.

[0027] The present invention also provides an online state monitoring and warning method for a pitch reducer of a wind turbine, including the following steps:

[0028] Step S1: Use a vibration sensor to collect the real-time vibration signal on the corresponding pitch reducer, and send the real-time vibration signal to the wireless receiving micro industrial computer in the nacelle through a wireless transmission acquisition card.

[0029] Step S2: The wireless receiving micro industrial computer is used to receive the collected real-time vibration signal and process it using the following steps:

[0030] 1) Decompose the real-time vibration signal using variational mode decomposition (VMD) to obtain multiple components.

[0031] 2) Blade rotation frequency component rejection: Select the IMF components in the range of 0.15 - 0.3 Hz caused by blade rotation, and confirm that the blade rotation frequency dominates in the spectrum of this component. Search for the frequency with the maximum amplitude within the frequency range [0, 20] Hz of each component. If it falls within the range of 0.15 - 0.3 Hz, then determine that this component is the blade rotation frequency component and reject it. The low-frequency trend component is approximately a DC component, and this component is rejected together with the blade rotation frequency component. The remaining components are summed and reconstructed into the remaining signal.

[0032] 3) Use the Hilbert transform method to obtain the envelope signal of the remaining signal. The envelope signal contains randomly occurring high-amplitude pulses.

[0033] 4) Sliding median filtering for noise reduction: Smooth this envelope signal through sliding median filtering, suppress the possible high-amplitude pulse interference, and retain the effective vibration components.

[0034] 5) Identify the pitch action: Use the real-time data received by the micro industrial computer as the input, execute steps 1) to 4), and calculate and determine the threshold for pitch using the 3σ criterion. Compare the smoothed envelope signal with this threshold, lock the effective pitch action time period, and save the data for this time period.

[0035] 6) On the basis of pitch action recognition, use support vector data description (SVDD) to perform hypersphere modeling on the comprehensive characteristics of the reducer to achieve fault warning for the pitch reducer.

[0036] Step S3: The status warning system synchronously monitors the operating status and communication status of each pitch reducer according to the warning signal. If the reducer status is abnormal, analyze whether the reducer has impact-type or wear-type faults through indicators.

[0037] Preferably, based on the pitch action recognition, the warning system uses support vector data description (SVDD) to perform hypersphere modeling on the time-domain and frequency-domain vibration characteristics of the reducer under normal conditions, and uses the distance between the characteristic data and the center of the hypersphere as the warning index for fault warning.

[0038] Preferably, according to any of the above solutions, if the wireless receiving micro industrial control computer determines that the current operating state of the speed reducer is normal, updates the data set, and recalculates the threshold, thereby implementing a dynamic threshold.

[0039] Preferably, according to any of the above solutions, in step S2, the wireless receiving micro industrial control computer uses SVDD to perform a hypersphere modeling on the comprehensive characteristics of the speed reducer to achieve a pitch speed reducer fault warning, including:

[0040] 1) Offline feature extraction: Perform VMD decomposition on the pitch vibration signal, remove the periodic fluctuation component of the vibration signal caused by the blade rotation, and extract the comprehensive characteristic indicators composed of traditional time domain indicators and time-frequency domain indicators;

[0041] 2) Offline comprehensive index dimensionality reduction: Dimensionality reduction is performed on the comprehensive characteristic indicators composed of traditional time domain indicators and time-frequency domain indicators to obtain two-dimensional indicators representing the main change trends of each indicator;

[0042] 3) Offline model training: Use the two-dimensional indicators in the normal state as the training samples of the SVDD model to obtain the optimal hypersphere containing normal samples, and the radius of the hypersphere is R. This process is the model training process;

[0043] 4) Establish an SVDD warning model: Extract features and perform dimensionality reduction on the comprehensive index of the online pitch action signal, input the online two-dimensional index into the established SVDD model, and obtain the distance D between it and the center of the hypersphere;

[0044] 5) Establish an adaptive alarm threshold: When setting the threshold, regard the distance index obtained based on SVDD as a single parameter for observation, and consider that the data change form of the index conforms to a normal distribution; Since the 3σ criterion in probability statistics is defined as, for a variable that conforms to a normal distribution with a mean of μ and a variance of σ 2 take values for the variable, and the probability of taking values in the interval (μ - σ 2 , μ + σ 2 ) is 99.73%; when a certain distance index value does not belong to this range, it is considered an abnormal point, that is, it is in an abnormal state. At the same time, to prevent the influence of external interference factors, if a continuous number of distance index values exceed the 3σ value range defined by the previous distance index value, it is considered that an abnormal value has occurred and an alarm is issued.

[0045] Preferably, according to any of the above solutions, the time domain index parameter set includes: maximum value, minimum value, peak value, peak-to-peak value, average value, root mean square, kurtosis, impulse factor, peak factor, and crest factor, peak value, average value, root mean square, kurtosis, impulse factor, peak factor, and crest factor; the frequency domain index parameter set includes: average frequency, root mean square of frequency, and standard deviation of frequency.

[0046] Compared with the prior art, the present invention has the following beneficial effects over the prior art:

[0047] Based on the key indicators of the speed reducer, the present invention realizes early warning and can detect the status and potential hazards of the pitch speed reducer in advance. To avoid serious mechanical failures that may cause the pitch system to malfunction, the project adopts vibration monitoring and wireless transmission technologies to achieve online status monitoring and fault warning of the pitch reducer without occupying the optical fibers for the operation / standby of the unit.

[0048] Through outdoor wireless testing, laboratory full-machine testing, and on-site full-machine testing, the present invention proves that the vibration monitoring of the pitch speed reducer based on wireless transmission is feasible.

[0049] The faults of the pitch speed reducer increase the equipment replacement costs in the wind farm and the power generation losses during that period. Calculated at 50,000 yuan per speed reducer, the economic losses brought by the unplanned replacement of three speed reducers are far higher than 150,000 yuan. In addition, the pitch speed reducer of the wind turbine is an important actuator for the normal start / stop, constant power operation, and emergency shutdown of the unit. After a serious fault (tooth breakage) of the speed reducer, the safety of the whole machine will be endangered. Conducting research on its status monitoring and fault warning has important theoretical and engineering value for the safe operation of the whole machine and pitch optimization control.

[0050] After the system is put into operation, the equipment maintenance method will change from fault maintenance and regular maintenance to predictive maintenance. Jilin has a national wind power base of tens of millions of kilowatts. The project results can significantly reduce the operation and maintenance costs of new energy, promote the construction of grid-friendly wind farms, and improve the ability of the power grid to absorb wind power, thus generating good economic and social benefits.

[0051] The additional aspects and advantages of the present invention will be partly given in the following description, partly will become obvious from the following description, or will be understood through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The above and / or additional aspects and advantages of the present invention will become obvious and easy to understand from the description of the embodiments in conjunction with the following drawings, wherein:

[0053] Figure 1 FIG. is a structural diagram of an online status monitoring and warning system for a pitch speed reducer of a wind turbine generator according to an embodiment of the present invention;

[0054] Figure 2 FIG. is a flowchart of pitch action recognition based on VMD-MMF according to an embodiment of the present invention;

[0055] Figure 3 FIG. is a flowchart of pitch speed reducer fault warning based on SVDD according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0056] Embodiments of the present invention will be described in detail below. Examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary and are intended to explain the present invention and should not be construed as limiting the present invention.

[0057] As Figure 1 shown, the online condition monitoring and early warning system for the pitch reducer of a wind turbine in an embodiment of the present invention includes: a vibration sensor is respectively installed on each pitch reducer of the wind turbine, and the vibration sensor is installed on the surface of the secondary gear ring housing of the corresponding pitch reducer. The output end of each vibration sensor is connected to a wireless transmission acquisition card, and the wireless transmission acquisition card is further connected to a wireless receiving micro industrial computer inside the nacelle. The wireless receiving micro industrial computer is bidirectionally connected to the nacelle radio station, the nacelle radio station is wirelessly connected to the centralized control room radio station, and the centralized control room radio station is bidirectionally connected to the condition early warning system.

[0058] In an embodiment of the present invention, three pitch reducers are installed on the wind turbine, and a vibration sensor is installed on each of the three pitch reducers. The vibration sensor is installed on the surface of the secondary gear ring housing of the corresponding pitch reducer. The wireless transmission acquisition card is responsible for transmitting the data collected by the vibration sensor in the hub to the industrial computer in the nacelle, the wireless receiving micro industrial computer.

[0059] After the wireless receiving micro industrial computer receives the data sent by the acquisition card, the industrial computer will preprocess the data. The remote transmission unit includes serial communication and a radio station. The serial communication and the radio station are responsible for data interaction between the wind turbine and the centralized control room, and adopt a wireless station transmission method with a master-slave communication mode. The master station sends access requests to the slave stations in sequence, and each slave station responds in turn and sends data to the master station in sequence.

[0060] Specifically, each vibration sensor is used to collect the real-time vibration signal on the corresponding pitch reducer, and send the real-time vibration signal to the wireless receiving micro industrial computer in the nacelle through the wireless transmission acquisition card.

[0061] As Figure 2 shown, the wireless receiving micro industrial computer is used to receive the collected real-time vibration signal and processes it by the following steps:

[0062] 1) Use variational mode decomposition (VMD) to decompose the vibration signal to obtain multiple components;

[0063] 2) Remove the blade passing frequency component. The frequency of the IMF containing the blade passing frequency trend is generally in the range of 0.15 - 0.3 Hz, and this frequency dominates in the spectrum of its blade passing frequency component. Therefore, search for the frequency with the largest amplitude within the frequency range [0, 20] Hz of each component. If it falls within the range of 0.15 - 0.3 Hz, then determine that this component is the blade passing frequency component and remove it. At the same time, the non-linear trend component is approximately a DC component, which can be removed by all components after removing the blade passing frequency component, and the remaining components are summed and reconstructed into the remaining signal.

[0064] 3) Use the Hilbert transform method to obtain the envelope signal of the remaining signal. At this time, the envelope signal contains randomly occurring high-amplitude pulses.

[0065] 4) Sliding median filtering for noise reduction: Smooth this envelope signal through sliding median filtering, suppress possible large high-amplitude pulse interferences, and retain the effective vibration components;

[0066] 5) Identify the pitch action: Use the effective pitch data as the input, execute step1 - step3 to calculate 3σ and use it as the threshold for judging the pitch. Compare the smoothed envelope signal with this threshold, accurately lock the time period of the effective pitch action, and save the data of this time period.

[0067] 6) Based on the identification of the pitch action, use SVDD (Support Vector Data Description) to perform hypersphere modeling on the time domain vibration characteristics of the reducer, so as to realize the fault warning of the pitch reducer. The specific implementation process is as Figure 3 shown.

[0068] 6.1) Offline feature extraction. Perform VMD decomposition on the original vibration signal, remove the periodic fluctuation component of the vibration signal caused by the blade rotation, and extract the traditional time domain indicators and time-frequency domain indicators to form comprehensive feature indicators.

[0069] 6.2) Offline comprehensive index dimension reduction. Reduce the dimension of the comprehensive feature indicators composed of the traditional time domain indicators and time-frequency domain indicators to obtain two-dimensional indicators representing the main change trends of each indicator.

[0070] 6.3) Offline model training. Use the two-dimensional indicators in the normal state as the training samples of the SVDD model to obtain the optimal hypersphere containing the normal samples. The radius of the hypersphere is R. This process is the model training process.

[0071] 6.4) Establish an SVDD warning model. Perform feature extraction and comprehensive index dimension reduction on the online pitch action signal, input the online two-dimensional indicators into the established SVDD model, and obtain the distance D from it to the center of the hypersphere.

[0072] 6.5) Establish an adaptive alarm threshold. When setting the threshold, the distance metric obtained based on SVDD needs to be regarded as a single parameter for observation, and it can be considered that the data change form of the metric conforms to a normal distribution. Since the 3σ criterion in probability statistics is defined as follows: for a variable that conforms to a normal distribution with a mean of μ and a variance of σ 2 and taking values for the variable, the probability of obtaining values within the interval (μ - σ 2 , μ + σ 2 ) is 99.73%. Therefore, once a distance metric value does not belong to this range, it is considered an outlier, that is, it is in an abnormal state. At the same time, to prevent the influence of external interference factors, if a continuous number of distance metric values exceed the 3σ value range defined by the previous distance metric value, it is considered that an abnormal value has occurred and an alarm is issued.

[0073] Based on the recognition of the pitch control action, the early warning system uses support vector data description (SVDD) to perform hypersphere modeling on the time-domain and frequency-domain vibration characteristics of the reducer under normal conditions, and uses the distance between the characteristic data and the center of the hypersphere as the early warning index for fault early warning.

[0074] If the wireless receiving micro industrial control computer determines that the current operating state of the reducer is normal, it updates the data set and recalculates the threshold, thereby realizing a dynamic threshold.

[0075] In the embodiment of the present invention, the time-domain index parameter set includes: maximum value, minimum value, peak value, peak-to-peak value, average value, root mean square, kurtosis, impulse factor, peak factor, and crest factor. Among them, the impulse factor, peak factor, and crest factor are all dimensionless characteristics obtained by the mutual ratio of other indexes. The vibration signal is subjected to a fast Fourier transform to transform the time-domain signal into a frequency-domain signal.

[0076] In the embodiment of the present invention, the frequency-domain index parameter set includes: average frequency, root mean square of frequency, and standard deviation of frequency. Among all the indexes used in the present invention, the indexes of kurtosis, peak factor, and impulse factor are highly sensitive to impact-type gear faults, such as broken teeth, missing teeth, and severe pitting, etc., while the margin and effective value indexes are often used to detect the wear condition of mechanical equipment.

[0077] The status early warning system is used to receive early warning signals through the centralized control room radio station, synchronously monitor the operating status and communication status of each pitch reducer according to the early warning signals. If the reducer status is abnormal, it analyzes whether the reducer has an impact-type or wear-type fault through indexes. If the wireless receiving micro industrial control computer determines that the current operating state of the reducer is normal, it updates the data set and recalculates the threshold, thereby realizing a dynamic threshold.

[0078] A set of servers is installed in the centralized control room to receive and process the data transmitted back by the industrial control computers in each wind turbine. The operating status and communication status of each wind power pitch reducer are synchronously monitored through DataSocket technology.

[0079] The present invention uses vibration sensors to collect real-time vibration signals of three pitch reducers, and sends the vibration signals to a micro industrial control computer in the nacelle through a wireless transmission acquisition card; the micro industrial control computer is used to receive the real-time vibration signals and identify the effective vibration signals caused by pitch actions, so as to realize the online status monitoring of the pitch reducer; the early warning system first receives the effective vibration signals sent by the nacelle radio station, and realizes the fault early warning of the pitch reducer by constructing a comprehensive pitch vibration index and a dynamic threshold.

[0080] The reliability of fault early warning is the primary concern of operation and maintenance personnel. False alarms will instead reduce the utilization rate of the unit and increase the labor and time costs of operation and maintenance. Common time-domain threshold early warning methods are often affected by the operating conditions of the unit. The threshold needs to change continuously with the conditions, and the early warning line depends on operating experience and is not easy to promote and use; similarly, the early warning methods using frequency-domain indicators also have high requirements for condition identification, while the on-site early warning methods based on vibration trend analysis only require manual selection of data under the same conditions, or only collect data at the rated speed. Therefore, considering the influence of operating conditions is an effective means to improve the reliability of drive train early warning. This project will carry out research on pitch reducer early warning based on vibration monitoring, based on vibration signals and combined with fault repair records.

[0081] The existing technologies have both indicators sensitive to faults and indicators insensitive to faults, and different indicators may have certain similarities in time-domain or frequency-domain characteristic parameters, and their characteristics are redundant, which is not conducive to subsequent fault early warning. Therefore, it is necessary to reduce the dimension of the extracted multiple index parameters to obtain indicators sensitive to faults. The present invention uses 10 time-domain and 3 frequency-domain indicators.

[0082] The embodiment of the present invention also proposes an online status monitoring and early warning method for a wind turbine pitch reducer, including the following steps:

[0083] Step S1, using vibration sensors to collect real-time vibration signals on the corresponding pitch reducer, and sending the real-time vibration signals to a wireless receiving micro industrial control computer in the nacelle through a wireless transmission acquisition card.

[0084] Step S2, the wireless receiving micro industrial control computer is used to receive the collected real-time vibration signals and process them using the following steps:

[0085] 1) Use variational mode decomposition (VMD) to decompose the vibration signal to obtain multiple components;

[0086] 2) Remove the blade passing frequency component. The frequency of the IMF containing the blade passing frequency trend is generally within the range of 0.15 - 0.3 Hz, and this frequency dominates in the spectrum of its blade passing frequency component. Therefore, search for the frequency with the maximum amplitude within the frequency range [0, 20] Hz of each component. If it falls within the range of 0.15 - 0.3 Hz, then determine that this component is the blade passing frequency component and remove it. At the same time, the non - linear trend component is approximately a DC component, which can be removed by all components after removing the blade passing frequency component, and the remaining components are summed and reconstructed into the remaining signal.

[0087] 3) Use the Hilbert transform method to obtain the envelope signal of the remaining signal. At this time, the envelope signal contains randomly occurring high - amplitude pulses.

[0088] 4) Sliding median filtering for noise reduction: Smooth this envelope signal through sliding median filtering, suppress the possible high - amplitude pulse interference, and retain the effective vibration components;

[0089] 5) Identify the pitch action: Take the pitch action data as the input, execute step1 - step3 to calculate 3σ and use it as the threshold for judging the pitch. Compare the smoothed envelope signal with this threshold, accurately lock the effective pitch action time period, and save the data of this time period.

[0090] 6) Based on the identification of the pitch action, use SVDD to perform a hypersphere modeling on the comprehensive characteristics of the reducer, so as to realize the fault warning of the pitch reducer. The specific implementation process is as Figure 3 shown.

[0091] 6.1) Off - line feature extraction. Perform VMD decomposition on the original vibration signal, remove the periodic fluctuation component of the vibration signal caused by the blade rotation, and extract the traditional time - domain indicators and time - frequency domain indicators to form comprehensive characteristic indicators.

[0092] 6.2) Off - line comprehensive index dimensionality reduction. Dimensionality reduction is performed on the comprehensive characteristic indicators composed of traditional time - domain indicators and time - frequency domain indicators to obtain two - dimensional indicators representing the main change trends of each indicator.

[0093] 6.3) Off - line model training. Use the two - dimensional indicators in the normal state as the training samples of the SVDD model to obtain the optimal hypersphere containing normal samples, and the radius of the hypersphere is R. This process is the model training process.

[0094] 6.4) Establish an SVDD warning model. Perform feature extraction and comprehensive index dimensionality reduction on the on - line pitch action signal, input the on - line two - dimensional indicators into the established SVDD model, and obtain the distance D from it to the center of the hypersphere.

[0095] 6.5) Establish an adaptive alarm threshold. When setting the threshold, the distance index obtained based on SVDD needs to be regarded as a single parameter for observation, and it can be considered that the data change form of the index conforms to a normal distribution. Since the 3σ criterion in probability statistics is defined as follows: for a variable that conforms to a normal distribution with a mean of μ and a variance of σ 2 and takes values within the interval (μ - σ 2 , μ + σ 2 ), the probability is 99.73%. Therefore, once a distance index value does not belong to this range, it is considered an outlier, that is, it is in an abnormal state. At the same time, to prevent the influence of external interference factors, if a continuous number of distance index values exceed the 3σ value range defined by the previous distance index value, it is considered that an abnormal value has occurred and an alarm is issued.

[0096] Based on the pitch action recognition, the early warning system uses the support vector data description (SVDD) to perform a hypersphere modeling on the time-domain and frequency-domain vibration characteristics of the reducer under normal conditions, and uses the distance between the characteristic data and the center of the hypersphere as the basis for fault early warning.

[0097] In the embodiment of the present invention, the time-domain index parameter set includes: maximum value, minimum value, peak value, peak-to-peak value, average value, root mean square, kurtosis, impulse factor, peak factor, and crest factor. Among them, the impulse factor, peak factor, and crest factor are all dimensionless characteristics obtained by the mutual ratio of other indexes. The vibration signal is subjected to a fast Fourier transform to transform the time-domain signal into a frequency-domain signal.

[0098] In the embodiment of the present invention, the frequency-domain index parameter set includes: average frequency, root mean square of frequency, and standard deviation of frequency. Among all the indexes used in the present invention, the indexes of kurtosis, peak factor, and impulse factor are highly sensitive to impact-type gear faults, such as broken teeth, missing teeth, and severe pitting, etc., while the margin and effective value indexes are often used to detect the wear condition of mechanical equipment.

[0099] Step S3: The state early warning system synchronously monitors the running state and communication state of each pitch reducer according to the early warning signal. If the reducer state is abnormal, it analyzes whether the reducer has an impact-type or wear-type fault through indexes. If the wireless receiving micro industrial computer determines that the current reducer running state is normal, it updates the data set and recalculates the threshold, thereby realizing a dynamic threshold, including the following steps:

[0100] S1) Use the time scale to limit the sliding window size, divide the historical data, and perform data standardization processing;

[0101] S2) Obtain the healthy state threshold;

[0102] S3) Use the next data set to re-obtain the threshold.

[0103] The present invention utilizes vibration sensors to collect real-time vibration signals of three pitch-changing speed reducers, and transmits the vibration signals to a micro industrial control computer in the nacelle through a wireless transmission acquisition card; the micro industrial control computer is used to receive the real-time vibration signals and identify the effective vibration signals caused by pitch-changing actions, thereby realizing on-line condition monitoring of the pitch-changing speed reducer; the early warning system first receives the effective vibration signals sent by the nacelle radio station, and realizes fault early warning of the pitch-changing speed reducer by constructing a comprehensive pitch-changing vibration index and a dynamic threshold.

[0104] Compared with the prior art, the present invention has the following beneficial effects over the prior art:

[0105] The present invention realizes early warning based on the key indicators of the speed reducer, and can discover the state and hidden dangers of the pitch-changing speed reducer in advance. To avoid serious mechanical failures that cause the pitch-changing system to malfunction, the project adopts vibration monitoring and wireless transmission technologies to realize on-line condition monitoring and fault early warning of the pitch-changing speed reducer without occupying the fiber optic cables for the operation / standby of the unit.

[0106] The present invention proves the feasibility of vibration monitoring of the pitch-changing speed reducer based on wireless transmission through outdoor wireless tests, laboratory full-machine tests, and on-site full-machine tests.

[0107] Faults of the pitch-changing speed reducer increase the equipment replacement cost of the wind farm and the power generation loss during the period. Calculated at 50,000 yuan per speed reducer, the economic loss caused by the unplanned replacement of three speed reducers is far higher than 150,000 yuan. In addition, the pitch-changing speed reducer of the wind turbine is an important actuator for the normal start-stop, constant power operation, and emergency shutdown of the unit. After a serious fault of the speed reducer (tooth breakage), the safety of the whole machine will be endangered. Conducting research on its condition monitoring and fault early warning has important theoretical and engineering value for the safe operation of the whole machine and the optimized control of pitch-changing.

[0108] After the system is put into operation, the equipment maintenance method will change from fault maintenance and regular maintenance to predictive maintenance. Jilin has a national wind power base of tens of millions of kilowatts. The project results can significantly reduce the operation and maintenance costs of new energy, promote the construction of power grid-friendly wind farms, and improve the ability of the power grid to absorb wind power, thus generating good economic and social benefits.

[0109] In the description of this specification, the descriptions with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples", etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples.

[0110] It is not difficult for those skilled in the art to understand that the present invention includes any combination of the invention content and the specific implementation part of the above specification and each part shown in the drawings. Due to space limitations and to make the specification concise, the various solutions formed by these combinations are not described one by one. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

[0111] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, replacements, and variations to the above embodiments within the scope of the present invention without departing from the principle and purpose of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An on-line condition monitoring and early warning system for the pitch reducer of a wind turbine, characterized in that, Including: A vibration sensor is respectively installed on each pitch reducer of the wind turbine. The vibration sensor is installed on the surface of the secondary gear ring housing of the corresponding pitch reducer. The output end of each vibration sensor is connected to a wireless transmitting and collecting card, and the wireless transmitting and collecting card is further connected to a wireless receiving micro industrial computer inside the nacelle. The wireless receiving micro industrial computer is bidirectionally connected to the nacelle radio station, the nacelle radio station is wirelessly connected to the centralized control room radio station, and the centralized control room radio station is bidirectionally connected to the status warning system. Among them, Each vibration sensor is used to collect the real-time vibration signal on the corresponding pitch reducer and send the real-time vibration signal to the wireless receiving micro industrial computer in the nacelle through the wireless transmitting and collecting card; The wireless receiving micro industrial computer is used to receive the collected real-time vibration signal and perform pitch action recognition and processing by the following steps: 1) Use variational mode decomposition (VMD) to decompose the real-time vibration signal to obtain multiple components; 2) Blade rotation frequency component removal: Select the 0.15 - 0.3 Hz IMF component caused by blade rotation and confirm that the blade rotation frequency dominates in the spectrum of this component. Search for the frequency with the largest amplitude within the frequency range [0, 20] Hz of each component. If it falls within the 0.15 - 0.3 Hz range, then determine that this component is the blade rotation frequency component and remove it; The low-frequency trend component is approximately a DC component, and this component is removed together with the blade rotation frequency component. The remaining components are summed and reconstructed into a remaining signal; 3) Use the Hilbert transform method to obtain the envelope signal of the remaining signal. The envelope signal contains randomly occurring high-amplitude pulses; 4) Sliding median filtering for noise reduction: Smooth this envelope signal through sliding median filtering, suppress possible high-amplitude pulse interference, and retain the effective vibration components; 5) Identify the pitch action: Take the real-time data received by the micro industrial computer as input, execute steps 1) to 4), and use the criterion to calculate and determine the threshold for pitch; Compare the smoothed envelope signal with the threshold, lock the effective pitch action time period, and save the data of this time period; The warning system synchronously monitors the operating status and communication status of each pitch reducer according to the warning signal. If the reducer status is abnormal, analyze whether the reducer has an impact-type or wear-type fault through indicators; Based on the pitch action recognition, the warning system uses support vector data description (SVDD) to perform hypersphere modeling on the time-domain and frequency-domain vibration characteristics under the normal state of the reducer, and uses the distance between the feature data and the hypersphere center as the basis for fault warning; If the wireless receiving micro industrial computer determines that the current reducer is operating normally, update the data set and recalculate the threshold to achieve a dynamic threshold.

2. The on-line condition monitoring and early warning system for the pitch reducer of a wind turbine according to claim 1, characterized in that, The wireless receiving micro industrial computer uses SVDD to perform hypersphere modeling on the time-domain vibration characteristics of the reducer to realize pitch reducer fault warning, including: 1) Offline feature extraction: Perform VMD decomposition on the pitch action vibration signal, remove the periodic fluctuation component of the vibration signal caused by blade rotation, and construct a comprehensive feature index set through time-domain and frequency-domain feature extraction; 2) Offline comprehensive index dimensionality reduction: Combine traditional time-domain indexes and time-frequency domain indexes to form comprehensive feature indexes for dimensionality reduction, and obtain two-dimensional indexes representing the main change trends of each index; 3) Offline model training: Using the two-dimensional indicators in the normal state as the training samples of the SVDD model to obtain the optimal hypersphere containing normal samples, and the radius of the hypersphere is , and this process is the model training process; 4) Establish an SVDD early warning model: Extract features from the online pitch action signals and reduce the dimensionality of the comprehensive indicators. Input the online two-dimensional indicators into the established SVDD model to obtain the distance from them to the center of the hypersphere. ; 5) Establish an adaptive alarm threshold: When setting the threshold, regard the distance index obtained based on SVDD as a single parameter for observation, and the change form of the index data conforms to the normal distribution; Since in probability statistics The criterion is defined as, for a variable that conforms to a normal distribution with a mean of , and a variance of , when taking values, the probability of taking values within the interval is 99.73%; When a certain distance index value does not belong to this range, it is considered an outlier, that is, it is in an abnormal state. At the same time, to prevent the influence of external interference factors, it is set that if multiple consecutive distance index values exceed the value range defined by the previous distance index value, it is considered that an abnormal value has occurred and an alarm is issued.

3. The on-line condition monitoring and early warning system for the pitch reducer of a wind turbine according to claim 2, characterized in that, The time-domain index parameter set includes: maximum value, minimum value, peak value, peak-to-peak value, average value, root mean square, kurtosis, impulse factor, peak factor, and crest factor. The frequency-domain index parameter set includes: average frequency, root mean square of frequency, and standard deviation of frequency.

4. An on-line condition monitoring and early warning method for the pitch reducer of a wind turbine, characterized in that, It includes the following steps: Step S1: Use a vibration sensor to collect the real-time vibration signal on the corresponding pitch reducer, and send the real-time vibration signal to the wireless receiving micro industrial control computer in the nacelle through a wireless transmission acquisition card. Step S2: The wireless receiving micro industrial control computer is used to receive the collected real-time vibration signal and process it using the following steps: 1) Use variational mode decomposition (VMD) to decompose the vibration signal to obtain multiple components. 2) Perform blade rotation frequency component rejection. For the IMF containing the blade rotation frequency trend, take the frequency value within the range of 0.15 - 0.3 Hz, and this frequency dominates in the spectrum of its blade rotation frequency component. Search for the frequency with the largest amplitude within the frequency range [0, 20] Hz of each component. If it falls within the range of 0.15 - 0.3 Hz, then determine that this component is the blade rotation frequency component and reject it. At the same time, the non-linear trend component is approximately a DC component. After rejecting the blade rotation frequency component, all other components are rejected, and the remaining components are summed and reconstructed into the remaining signal. 3) Use the Hilbert transform method to obtain the envelope signal of the remaining signal. The envelope signal contains randomly occurring high-amplitude pulses. 4) Sliding median filtering for noise reduction: Smooth this envelope signal through sliding median filtering to suppress possible high-amplitude pulse interference and retain the effective vibration components. 5) Identify the pitch action: Take the real-time data received by the micro industrial computer as the input, execute steps 1) to 4), and use the criterion to calculate and judge the threshold of pitch; Compare the smoothed envelope signal with the threshold, lock the effective pitch action time period, and save the data of this time period; 6) Based on the pitch action recognition, use support vector data description (SVDD) to perform hypersphere modeling on the time-domain vibration characteristics of the reducer, so as to realize the fault warning of the pitch reducer. Step S3: The warning system synchronously monitors the operating state and communication state of each pitch reducer according to the warning signal. If the reducer state is abnormal, analyze whether the reducer has impact-type or wear-type faults through comprehensive indicators. Based on the pitch action recognition, the warning system uses support vector data description (SVDD) to perform hypersphere modeling on the time-domain and frequency-domain vibration characteristics of the reducer under normal conditions, and uses the distance between the characteristic data and the hypersphere center as the warning index for fault warning. If the wireless receiving micro industrial control computer determines that the current reducer operating state is normal, update the data set and recalculate the threshold, so as to realize the dynamic threshold.

5. The on-line condition monitoring and early warning method for the pitch reducer of a wind turbine according to claim 4, characterized in that, In the step S2, the wireless receiving micro industrial control computer uses SVDD to perform hypersphere modeling on the time-domain vibration characteristics of the reducer to realize the fault warning of the pitch reducer, including: 1) Offline feature extraction: Perform VMD decomposition on the pitch vibration signal, remove the periodic fluctuation component of the vibration signal caused by blade rotation, and extract traditional time-domain indicators and time-frequency domain indicators to form comprehensive characteristic indicators. 2) Offline comprehensive index dimensionality reduction: Perform LLE dimensionality reduction on the comprehensive characteristic indicators composed of traditional time-domain indicators and time-frequency domain indicators to obtain two-dimensional indicators representing the main change trends of each indicator. 3) Offline model training: Using the two-dimensional indicators in the normal state as the training samples for the SVDD model to obtain the optimal hypersphere containing normal samples, and the radius of the hypersphere is , and this process is the model training process; 4) Establish an SVDD early warning model: extract features from the online pitch action signals and reduce the dimensionality of the comprehensive indicators, and input the online two-dimensional indicators into the established SVDD model to obtain the distance from it to the center of the hypersphere ; 5) Establish an adaptive alarm threshold: When setting the threshold, regard the distance index obtained based on SVDD as a single parameter for observation, and assume that the data change form of the index conforms to a normal distribution; Since in probability statistics The criterion is defined as, for a variable that conforms to a normal distribution with a mean of , and a variance of , when taking values, the probability of taking values within the interval is 99.73%; When a certain distance index value does not belong to this range, it is considered an outlier, that is, it is in an abnormal state. At the same time, to prevent the influence of external interference factors, it is set that if multiple consecutive distance index values exceed the value range defined by the previous distance index value, it is considered that an abnormal value has occurred and an alarm is issued.

6. The online condition monitoring and early warning method for the pitch reducer of a wind turbine as claimed in claim 5, wherein The time-domain index parameter set includes: maximum value, minimum value, peak value, peak-to-peak value, average value, root mean square, kurtosis, impulse factor, peak factor, and crest factor. The frequency-domain index parameter set includes: average frequency, root mean square of frequency, and standard deviation of frequency.

Images