CN103645052A - Wind turbine set gearbox remote online state monitoring and life assessment method - Google Patents

Abstract

The invention provides a wind turbine set gearbox remote online state monitoring and life assessment method. The method comprises the following steps: (1) a remote center computer collects a real-time operation parameter, temperature data and vibration data of each wind turbine through SCADA; (2) through analyzing the characteristic values of signal time domain and frequency domain, whether a vibration sensor has a fault or not is diagnosed; (3) the data is subjected to normalization process, a difference between data is reduced, and the diagnosis precision is raised; (4) when the fault of a wind generator set gearbox is diagnosed, different filtering modes are employed for different algorithms to extract the corresponding characteristic; (5) the collected data is subjected to characteristic extraction; (6) the fault diagnosis is carried out to provide the fact that whether the fault exists or not and the degree of the fault; (7) the residual life of the failed gearbox is estimated through a gray theory model, and an autoregression model is established for the residual error of the predication result to raise the predication precision. According to the method, the real-time estimation of the residual life is realized, and a basis is provided for the planned maintenance of the wind turbine set.

Description

A kind of gearbox of wind turbine remote online status monitoring and lifetime estimation method

Technical field

The present invention relates to a kind of gearbox of wind turbine remote online status monitoring and lifetime estimation method, belong to wind-driven generator group wheel box running status assessment technology field.Be specifically related to status monitoring, fault degree assessment and the residual life predictor method of wind-driven generator group wheel box.

Background technology

Along with the growth of the growth of the installed capacity of wind-driven power of China and wind power generating set working time, need to assess the running status of the gear case of wind power generating set, to better improve the transmission efficiency of wind-powered electricity generation unit, reduce corresponding maintenance cost.Because driving and the unstable error of traditional fault monitoring method diagnosis that causes of load of wind power generating set are large.If fault can not be found timely and process, can cause very large accident, lose huge; The running status of gear case can not correctly be assessed, and can cause too early maintenance or posterior maintenance, and the operation and maintenance of wind field is caused to tremendous influence.Therefore, the status monitoring of wind-driven generator group wheel box and life appraisal, to improving the reliability of gear case, reduce maintenance cost, have important practical significance and economic worth.

Wind-driven generator group wheel box is comprised of parts such as transmission shaft, bearing and the gear teeth.The fault of these parts is because inherent shortcoming, lubricated bad or overload cause, and can reflect by duty parameters such as the operational factor of wind power generating set and the vibration of corresponding component and temperature.The state estimation of wind power generating set mainly concentrates on the qualitative analysis of status monitoring and fault at present, can identify the fault of equipment and the type of fault and position, lacks residual life assessment under the quantitative evaluation of fault and operating condition.

Summary of the invention:

The present invention aims to provide a kind of gearbox of wind turbine remote online status monitoring and lifetime estimation method.By the state of gear case is carried out to remote real time monitoring and fault diagnosis, extract the fault factor of historical data and Real-time Monitoring Data, and then the residual life of gear case is carried out to real-time estimation and correction.It has overcome the model based on historical data in the past and has estimated the feature of corrected Calculation result in time, is a kind of lifetime estimation method more accurately.

Technical solution of the present invention is as follows:

Gearbox of wind turbine on-line condition monitoring disclosed by the invention and lifetime estimation method, is characterized in that: remote center's computing machine gathers the vibration data on real time execution parameter, temperature data and the gear case of wind power generating set in each wind energy turbine set by SCADA.By corresponding algorithm real-time analysis, give out of order analysis report and reliability thereof.The data handling procedure of remote center's computing machine is as follows:

(1) vibration data of collection is carried out to validity judgement, the data of sensor fault are rejected.Then it is stored in slip condition database together with real time execution parameter, temperature data etc., so that follow-up data processing.

(2) real time data of storage need to just can be carried out fault analysis after pre-service.Pre-service mainly adopts normalized method, and then it is different and cause the difference between different acquisition point or different blower fan data to reduce collecting device, data limit is arrived to the needed input data area of fault analysis simultaneously.

(3) data are carried out to analyzing and processing, judge whether gear case exists fault.Fault analysis comprises that filtering, the eigenwert of data are extracted and fault judges.Data filtering is mainly the interference component proposing in data.It is the effective constituent of extracting signal that eigenwert is extracted, and forms the proper vector of fault diagnosis.According to proper vector, carry out fault diagnosis, give the numerical value of out of order analysis report and the fault factor.Then by Trouble Report and the fault factor being carried out to the residual life of parts, analyze.

(4) residual life prediction is the trend analysis of carrying out according to real-time diagnosis report and the historical diagnosis report of equipment, and then judges the trend of this fault progression and the life-span of reliability operation, to arrange, reasonably safeguards.

In sum, see Fig. 1, the present invention's a kind of gearbox of wind turbine remote online status monitoring and lifetime estimation method, the method concrete steps are as follows:

Step 1: remote center's computing machine is by real time execution parameter, temperature data and the vibration data of each blower fan of SCADA Real-time Collection.Real time execution parameter comprises: the rotating speed of draught fan impeller, generator speed, active power and reactive power.Temperature data comprises the front and back bearings temperature of gear case oil temperature, heatsink temperature, cabin temperature, box bearing temperature and generator.Vibration data mainly comprises the vibration of gearbox input shaft, the vibration of planetary gear, the vibration of gear case slow-speed shaft, the vibration of high speed shaft of gearbox etc.After obtaining, data need the vibration data to gathering need to carry out validity judgement.

Step 2: it is the data that gather when rejecting vibration transducer fault that the vibration data gathering is carried out to data validity judgement.By the eigenwert of signal time domain and frequency domain is analyzed, whether diagnosis vibration transducer there is fault.When diagnosing out sensor to have fault, these data are rejected, reduce the diagnostic accuracy impact that misdata is carried out gearbox fault.

Step 3: data are normalized, reduce the difference between data, improve diagnostic accuracy.The method of the normalized adopting has amplitude normalization, statistics normalization, energy normalized and fundamental frequency normalization.Adopting different method for diagnosing faults, be to adopt different method for normalizing.

Step 4: when the diagnosing malfunction to wind-driven generator group wheel box, adopt different filtering modes for algorithms of different, then extract corresponding feature.The method of application wavelet filtering is carried out soft-threshold noise reduction to data, reduces the interference of white noise to data.Adopt comb filter to extract and obtain corresponding frequencies range content data vibration signal.

Step 5: the data that gather are carried out to feature extraction.The feature of vibration data comprises the proper vector that temporal signatures, frequency domain character and other algorithms extract.The feature of temperature data comprises rate of change, amplitude etc.Then according to the data characteristics of extracting, gear case is carried out to fault diagnosis.

Step 6: fault diagnosis has or not and the degree of fault giving to be out of order.The method that fault diagnosis adopts comprises: the mode based on model, by fan parameter and proper vector, calculate the theoretical numerical value of certain eigenwert under normal circumstances, and then by measured value and theoretical value, relatively carry out fault diagnosis.When judgement gear case exists fault, extract amplitude, effective value, kurtosis, peak-to-peak value, waveform index, pulse index, the fault factors such as frequency, 1/3 octave component, 1/2 frequency multiplication, 2 frequencys multiplication, 3 frequencys multiplication that turn.

Step 7: it is according to the diagnostic result of historical data and the fault factor that residual life is estimated, and by the residual life of Grey Model suspected fault gear case, the residual error that it is predicted the outcome is set up autoregressive model, improves precision of prediction.System default adopts effective value to carry out residual life prediction as state parameter, can change state parameter by relevant parameter is set.Residual life prediction steps block diagram as shown in Figure 2.

The step of grey modeling as shown in Figure 3.If the original series of prediction is x ⁽⁰⁾(k), k=1,2 ..., N.N is the number of original series.Grey generation is that this sequence is done to cumulative generation, cumulative formation sequence (1-AGO) x of forward ⁽¹⁾(k), k=1,2 ..., N.

$x^{\left(1\right)}\left(k\right)=\underset{m=1}{\overset{k}{\mathrm{\Σ}}}{x}^{\left(0\right)}\left(m\right)$

Set up GM(1,1) grey differential albefaction equation:

$\frac{{\mathrm{dx}}^{\left(1\right)}}{\mathrm{dt}}+{\mathrm{ax}}^{\left(1\right)}=b$

Argument sequence

$\hat{a}={[a,b]}^T$

Calculate

$\hat{a}={\left(B^{T}B\right)}^{-1}{B}^T{Y}_N$

In formula,

$B=\left[\begin{array}{cc}-\frac{1}{2}[x^{\left(1\right)}\left(1\right)+x^{\left(2\right)}\left(2\right)]& 1\\ -\frac{1}{2}[x^{\left(1\right)}\left(2\right)+x^{\left(1\right)}\left(3\right)]& 1\\ .& .\\ .& .\\ .& .\\ -\frac{1}{2}[x^{\left(1\right)}(k-1)+x^{\left(1\right)}\left(k\right)]& 1\end{array}\right]$

$Y_N={\left[\begin{array}{cccc}{x}^{\left(0\right)}\left(2\right)& x^{\left(0\right)}\left(3\right)& ...& x^{\left(0\right)}\left(k\right)\end{array}\right]}^T$

The predicted value of original data sequence is

${\hat{x}}^{\left(0\right)}\left(k\right)=a\{-\frac{1}{2}[{\hat{x}}^{\left(1\right)}(k-1)+{\hat{x}}^{\left(1\right)}\left(k\right)\left]\right\}+b$

By step above, obtained the predicted value of forecast model, then the residual error predicting the outcome has been set up to autoregressive model and obtain residual sequence

and then obtain the predicted value of residual life

$\hat{x}\left(k\right)={\hat{x}}^{\left(0\right)}\left(k\right)+\hat{e}\left(k\right)$

Advantage and effect: the present invention's a kind of gearbox of wind turbine remote online status monitoring and lifetime estimation method, its advantage is: on the basis of the long-range real-time state monitoring of gear case, realized the real-time estimation of residual life, foundation is provided to the schedule maintenance of wind-powered electricity generation unit.Concrete advantage is as follows:

1) data that gather are carried out to pre-service, the reliability of having rejected misdata and having strengthened data;

2) utilize the fault factor of gear case historical data and real time data and gray scale theory to estimate the life-span of gear case, having overcome the model based on historical data in the past and estimated the feature of corrected Calculation result in time, is a kind of lifetime estimation method more accurately.

Accompanying drawing explanation

Fig. 1 is workflow block diagram of the present invention

Fig. 2 life prediction FB(flow block)

Fig. 3 gray prediction block diagram

In figure, symbol description is as follows:

SCADA data acquisition and supervisor control

Embodiment

In conjunction with concrete example and accompanying drawing, the present invention is described in further details, but embodiments of the present invention are not limited to this.

As shown in Figure 1, a kind of method of wind-driven generator group wheel box remote online status monitoring and life appraisal, concrete steps are as follows:

Step 1: remote center's computing machine is by real time execution parameter, temperature data and the vibration data of each blower fan of SCADA Real-time Collection.Real time execution parameter comprises: the rotating speed of draught fan impeller, generator speed, active power and reactive power.Temperature data comprises the front and back bearings temperature of gear case oil temperature, heatsink temperature, cabin temperature, box bearing temperature and generator.Vibration data mainly comprises the vibration of gearbox input shaft, the vibration of planetary gear, the vibration of gear case slow-speed shaft, the vibration of high speed shaft of gearbox etc.After obtaining, data need the vibration data to gathering need to carry out validity judgement.

Step 2: it is the data that gather when rejecting vibration transducer fault that the vibration data gathering is carried out to data validity judgement.By the eigenwert of signal time domain and frequency domain is analyzed, whether diagnosis vibration transducer there is fault.When diagnosing out sensor to have fault, these data are rejected, reduce the diagnostic accuracy impact that misdata is carried out gearbox fault.

Step 3: data are normalized, reduce the difference between data, improve diagnostic accuracy.The method of the normalized adopting has amplitude normalization, statistics normalization, energy normalized and fundamental frequency normalization.Adopting different method for diagnosing faults, be to adopt different method for normalizing.

Step 4: when the diagnosing malfunction to wind-driven generator group wheel box, adopt different filtering modes for algorithms of different, then extract corresponding feature.The method of application wavelet filtering is carried out soft-threshold noise reduction to data, reduces the interference of white noise to data.Adopt comb filter to extract and obtain corresponding frequencies range content data vibration signal.

Step 5: the data that gather are carried out to feature extraction.The feature of vibration data comprises the proper vector that temporal signatures, frequency domain character and other algorithms extract.The feature of temperature data comprises rate of change, amplitude etc.Then according to the data characteristics of extracting, gear case is carried out to fault diagnosis.

Step 6: fault diagnosis has or not and the degree of fault giving to be out of order.The method that fault diagnosis adopts comprises: the mode based on model, by fan parameter and proper vector, calculate the theoretical numerical value of certain eigenwert under normal circumstances, and then by measured value and theoretical value, relatively carry out fault diagnosis.When judgement gear case exists fault, extract amplitude, effective value, kurtosis, peak-to-peak value, waveform index, pulse index, the fault factors such as frequency, 1/3 octave component, 1/2 frequency multiplication, 2 frequencys multiplication, 3 frequencys multiplication that turn.

Step 7: it is according to the diagnostic result of historical data and the fault factor that residual life is estimated, and by the residual life of Grey Model suspected fault gear case, the residual error that it is predicted the outcome is set up autoregressive model, improves precision of prediction.System default adopts effective value to carry out residual life prediction as state parameter, can change state parameter by relevant parameter is set.Residual life prediction steps block diagram as shown in Figure 2.

The step of grey modeling as shown in Figure 3.If the original series of prediction is x ⁽⁰⁾(k), k=1,2 ..., N.N is the number of original series.Grey generation is that this sequence is done to cumulative generation, cumulative formation sequence (1-AGO) x of forward ⁽¹⁾(k), k=1,2 ..., N.

$x^{\left(1\right)}\left(k\right)=\underset{m=1}{\overset{k}{\mathrm{\Σ}}}{x}^{\left(0\right)}\left(m\right)$

Set up GM(1,1) grey differential albefaction equation:

$\frac{{\mathrm{dx}}^{\left(1\right)}}{\mathrm{dt}}+{\mathrm{ax}}^{\left(1\right)}=b$

Argument sequence

$\hat{a}={[a,b]}^T$

Calculate

$\hat{a}={\left(B^{T}B\right)}^{-1}{B}^T{Y}_N$

In formula,

$B=\left[\begin{array}{cc}-\frac{1}{2}[x^{\left(1\right)}\left(1\right)+x^{\left(2\right)}\left(2\right)]& 1\\ -\frac{1}{2}[x^{\left(1\right)}\left(2\right)+x^{\left(1\right)}\left(3\right)]& 1\\ .& .\\ .& .\\ .& .\\ -\frac{1}{2}[x^{\left(1\right)}(k-1)+x^{\left(1\right)}\left(k\right)]& 1\end{array}\right]$

$Y_N={\left[\begin{array}{cccc}{x}^{\left(0\right)}\left(2\right)& x^{\left(0\right)}\left(3\right)& ...& x^{\left(0\right)}\left(k\right)\end{array}\right]}^T$

The predicted value of original data sequence is

${\hat{x}}^{\left(0\right)}\left(k\right)=a\{-\frac{1}{2}[{\hat{x}}^{\left(1\right)}(k-1)+{\hat{x}}^{\left(1\right)}\left(k\right)\left]\right\}+b$

By step above, obtained the predicted value of forecast model, then the residual error predicting the outcome has been set up to autoregressive model and obtain residual sequence

and then obtain the predicted value of residual life

$\hat{x}\left(k\right)={\hat{x}}^{\left(0\right)}\left(k\right)+\hat{e}\left(k\right)$ 。

Claims (1)

1. gearbox of wind turbine remote online status monitoring and a lifetime estimation method, is characterized in that: the method concrete steps are as follows:

Step 1: remote center's computing machine is by real time execution parameter, temperature data and the vibration data of each blower fan of SCADA Real-time Collection; Real time execution parameter comprises: the rotating speed of draught fan impeller, generator speed, active power and reactive power; Temperature data comprises the front and back bearings temperature of gear case oil temperature, heatsink temperature, cabin temperature, box bearing temperature and generator; Vibration data comprises the vibration of gearbox input shaft, the vibration of planetary gear, the vibration of gear case slow-speed shaft and the vibration of high speed shaft of gearbox; After obtaining, data need the vibration data to gathering need to carry out validity judgement;

Step 2: it is the data that gather when rejecting vibration transducer fault that the vibration data gathering is carried out to data validity judgement; By the eigenwert of signal time domain and frequency domain is analyzed, whether diagnosis vibration transducer there is fault; When diagnosing out sensor to have fault, these data are rejected, reduce the diagnostic accuracy impact that misdata is carried out gearbox fault;

Step 3: data are normalized, reduce the difference between data, improve diagnostic accuracy; The method of the normalized adopting has amplitude normalization, statistics normalization, energy normalized and fundamental frequency normalization; Adopting different method for diagnosing faults, be to adopt different method for normalizing;

Step 4: when the diagnosing malfunction to wind-driven generator group wheel box, adopt different filtering modes for algorithms of different, then extract corresponding feature; The method of application wavelet filtering is carried out soft-threshold noise reduction to data, reduces the interference of white noise to data, adopts comb filter to extract and obtain corresponding frequencies range content data vibration signal;

Step 5: the data that gather are carried out to feature extraction; The feature of vibration data comprises the proper vector that temporal signatures, frequency domain character and other algorithms extract; The feature of temperature data comprises rate of change, amplitude, then according to the data characteristics of extracting, gear case is carried out to fault diagnosis;

Step 6: fault diagnosis has or not and the degree of fault giving to be out of order; The method that fault diagnosis adopts comprises: the mode based on model, by fan parameter and proper vector, calculate the theoretical numerical value of certain eigenwert under normal circumstances, and then by measured value and theoretical value, relatively carry out fault diagnosis; When judgement gear case exists fault, extract amplitude, effective value, kurtosis, peak-to-peak value, waveform index, pulse index, turn frequency, 1/3 octave component, 1/2 frequency multiplication, 2 frequencys multiplication, the 3 frequency multiplication fault factors;

Step 7: it is according to the diagnostic result of historical data and the fault factor that residual life is estimated, and by the residual life of Grey Model suspected fault gear case, the residual error that it is predicted the outcome is set up autoregressive model, improves precision of prediction; System default adopts effective value to carry out residual life prediction as state parameter, by relevant parameter is set, changes state parameter;

If the original series of prediction is x ⁽⁰⁾(k), k=1,2 ..., N; N is the number of original series, and grey generation is that this sequence is done to cumulative generation, cumulative formation sequence (1-AGO) x of forward ⁽¹⁾(k), k=1,2 ..., N;

$x^{\left(1\right)}\left(k\right)=\underset{m=1}{\overset{k}{\mathrm{\Σ}}}{x}^{\left(0\right)}\left(m\right)$

Set up GM(1,1) grey differential albefaction equation:

$\frac{{\mathrm{dx}}^{\left(1\right)}}{\mathrm{dt}}+{\mathrm{ax}}^{\left(1\right)}=b$

Argument sequence

$\hat{a}={[a,b]}^T$

Calculate

$\hat{a}={\left(B^{T}B\right)}^{-1}{B}^T{Y}_N$

In formula,

$B=\left[\begin{array}{cc}-\frac{1}{2}[x^{\left(1\right)}\left(1\right)+x^{\left(2\right)}\left(2\right)]& 1\\ -\frac{1}{2}[x^{\left(1\right)}\left(2\right)+x^{\left(1\right)}\left(3\right)]& 1\\ .& .\\ .& .\\ .& .\\ -\frac{1}{2}[x^{\left(1\right)}(k-1)+x^{\left(1\right)}\left(k\right)]& 1\end{array}\right]$

$Y_N={\left[\begin{array}{cccc}{x}^{\left(0\right)}\left(2\right)& x^{\left(0\right)}\left(3\right)& ...& x^{\left(0\right)}\left(k\right)\end{array}\right]}^T$

The predicted value of original data sequence is

${\hat{x}}^{\left(0\right)}\left(k\right)=a\{-\frac{1}{2}[{\hat{x}}^{\left(1\right)}(k-1)+{\hat{x}}^{\left(1\right)}\left(k\right)\left]\right\}+b$

By step above, obtained the predicted value of forecast model, then the residual error predicting the outcome has been set up to autoregressive model and obtain residual sequence

and then obtain the predicted value of residual life

$\hat{x}\left(k\right)={\hat{x}}^{\left(0\right)}\left(k\right)+\hat{e}\left(k\right).$

Images