CN104392082A - Diagnosis method for initial failure of gearbox of wind generating set based on vibration monitoring - Google Patents

Abstract

The invention provides a diagnosis method for an initial failure of a gearbox of a wind generating set based on vibration monitoring. The method comprises the following steps: firstly, collecting an original vibrating acceleration signal of each monitoring point of the gearbox of the wind generating set; adopting an LMD (Local mean decomposition) method for decomposing the original vibrating acceleration signal into a plurality of PF (Product function) components with certain physical significances; selecting the PF components containing failure information, such as amplitude modulation and frequency modulation, for reconstructing a signal; performing Hilbert conversion on the reconstructed signal and further extracting fault characteristic quantity; lastly, judging the fault type and the fault degree by calculating a KL-divergence value between a to-be-tested sample and a standard sample.

Description

A kind of wind-driven generator group wheel box Incipient Fault Diagnosis method based on vibration monitoring

Technical field

The present invention relates to wind power generation field, be specifically related to a kind of rotating machinery Incipient Fault Diagnosis method based on vibration monitoring, particularly relate to and be a kind ofly applied to the diagnostic method of wind-driven generator group wheel box initial failure based on vibration monitoring.

Background technology

Renewable as one, the free of contamination new green power of wind energy, is extensively thought a kind of novel energy Land use systems with huge development and utilization prospect both at home and abroad.When energy shortage is day by day serious in the world at present, greatly developing wind power technology is one of the effective way and inexorable trend that solve energy shortage and problem of environmental pollution.2013, China's (not comprising Taiwan), adding new capacity 16088.7MW, increases by 24.1% on a year-on-year basis; Accumulative installed capacity 91412.89MW, increases by 21.4% on a year-on-year basis.Newly-increased installation and accumulative installation two item number are according to all ranking first in the world.

In recent years, while wind-power electricity generation fast development, wind power generating set safety, stable operation cause great attention both domestic and external gradually.At present, wind power generating set mainly comprises three types: double feed wind power generator group, direct-driving type wind power generation unit and half direct-driving type wind power generation machine.Double feed wind power generator group relies on gear case to realize speedup, and its production technology is more ripe, and this kind of aerogenerator is current mainstream model.Directly drive type and there is no gear case, it avoid the decline of high failure rate and the maintenance cost caused due to gear case, but directly drive type and also have its inevitable shortcoming: the motor weight that is few, that directly drive unit of the rare earth resources needed for permanent magnet generator type causes greatly transport assembling very difficult, ENERCON company, at a 6WM type in hamburger, lifts and consuming timely reaches 3 months.Semi-direct driving wind machine combines double-fed type and directly drives the advantage of type, and while meeting transmission and load design, structure is more compact, lightweight.The up-to-date G10X-4.5MW wind turbine of Gamesa Wind release in 2011, adopts two grades of wheel boxes to add magneto alternator; The gear case of the up-to-date V164-7.0MW wind turbine that Vestas is released also adopts 3 grades of speedups to change 2 grades of speedups into, and motor technology also uses permanent magnet technology; At home, the wind turbine of golden wind science and technology 3MW also uses and partly directly drives technology, and consider the trend that Wind turbines maximizes, the epoch of partly directly driving may arrive.Therefore, carry out studying important theoretical direction for double-fed type and the gearbox fault that partly directly drives type to be worth and engineer applied meaning.

Because wind turbine multidigit is in field; stand the impact of the factors such as various inclement weather, changeable the produced shock load of wind speed and direction and operating mode constantly change; according to the data of 1997-2005 period four wind fields (Sweden's two wind fields, Finland's wind field and Germany wind field); gear case becomes one of higher parts of wind power generating set failure rate; proportion stop time caused by gearbox fault is maximum, and accounting for total down-time is 32%.Wind-driven generator wheel-box is once break down, and its dismounting, transport and maintenance cost are up to nearly 1,000,000 yuan, and the dismounting of marine windmill also will employ large ship and tank crane, even helicopter, its maintenance and maintenance cost higher.Therefore, enough attention must be caused to the on-line monitoring and fault diagonosing of wind-driven generator group wheel box, the maintenance cost of wind energy turbine set could be reduced further, increase the efficiency of wind energy turbine set.Visible, the stop time having gearbox fault to cause in wind power generating set is the longest, so it is significant to carry out research to the gearbox fault in wind power generating set.

In recent years, comparatively extensive and fruitful research has been carried out in the fault diagnosis of bearing, but be still in the Primary Study stage for the Incipient Fault Diagnosis problem of bearing, and the achievement in research that Bearing Initial Fault Diagnosis method is effectively applied in wind power generating set is still comparatively lacked.This is mainly due to when the damage class fault that bearing occurs is in commitment, the shock characteristic characterizing failure message is usually fainter, there is normal vibration and undesired signal noise in the mechanical system due to wind turbine, makes the fault signature of signal be difficult to extract and highlight; Simultaneously due to wind power generating set work under bad environment, wind speed variation range are large and blower fan start and stop and bearing fault time the special circumstances such as shock load, be all that early detection and the diagnosis of bearing fault brings great challenge.

In CN, disclosed on 07 11st, 2012, application number is 102564568A, name is called a kind of initial failure searching method of the large rotating machinery equipment for industries such as the energy, iron and steel, coal, transports of the Invention Announce of " the initial failure searching method under large rotating machinery complex working condition ", and this system is by carrying out pre-service, extracting the search that the link such as fault factor, failure symptom coupling realizes initial failure dangerous point to time of vibration sequence.Because wind power generating set multidigit is in field, stand the impact of the factors such as various inclement weather, changeable the produced shock load of wind speed and direction and operating mode constantly change, the vibration signal gathered contains a lot of ground unrests, and therefore the effect of conventional rotating machinery fault monitoring method in wind power generating set is undesirable.

Summary of the invention

The present invention is directed to current wind-driven generator group wheel box Incipient Fault Diagnosis method shortcoming accurately not prompt enough, vibration monitoring is carried out to wind-driven generator group wheel box, propose a kind of wind-driven generator group wheel box Incipient Fault Diagnosis method, the method comprises the following steps:

Step 1: adopt vibration acceleration sensor to gather original vibration signal, measuring point comprises: the horizontal and vertical position of gearbox input shaft bearing, gearbox planetary gear side, gearbox intermediate shaft side, gearbox high-speed axle bearing totally four measuring points;

Step 2: to the original vibration signal collected the temporally N number of sample point of tag extraction, N >=4096, carry out the decomposition of LMD method to the original vibration signal x (t) extracted, and can obtain the PF component that several instantaneous frequencys have physical significance, namely $x\left(t\right)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{PF}}_i+r---\left(4\right)$

In formula, x (t)-original vibration signal Time Domain Amplitude ( m/s ² );

PF _i-the i-th component;

R-survival function.

The essence of LMD method is the modulation signal of several simple components by multicomponent signal decomposition, and wind-driven generator group wheel box fault vibration signal has the feature of the many modulation of multicarrier, therefore LMD method can improve signal to noise ratio (S/N ratio), and effectively extract fault signature;

Step 3: the PF component comprising the failure messages such as amplitude modulationfrequency modulation is used for signal reconstruction, i.e. x ' (t)=∑ PF _i;

Step 4: Hilbert conversion is carried out to reconstruction signal, and extracts fault characteristic value.Modulation intelligence in signal can extract by Hilbert conversion, and intensity and the frequency namely by analyzing modulation intelligence can judge that gear case produces position and the degree of injury of fault.The fault signature that the Hilbert of signal converts extracting cycle impacts composition and severe degree thereof, then need a suitable index to carry out the working condition of accurate description gear case.When gear case breaks down, in Hilbert spectrogram, corresponding fault characteristic frequency place there will be spectrum peak, fault characteristic frequency mainly comprises: the meshing frequency of the rotational frequency of each axle of gear and harmonic wave thereof, each axle and harmonic wave, rolling bearing inner ring characteristic frequency and harmonic wave thereof, outer shroud characteristic frequency and harmonic wave, rolling body characteristic frequency and harmonic wave thereof, so the amplitude choosing above-mentioned each fault characteristic frequency place that can characterize fault type is characteristic quantity.

Step 5: gather the sample of wind-driven generator group wheel box under different conditions as master sample, calculate the fault characteristic value of master sample and sample to be tested successively, and calculate the KL-divergence between sample to be tested and master sample, namely

Wherein, D _klthe KL-divergence of (P Q)-between sample to be checked and master sample;

P (i)-the i-th specimen reconstruct signal fault to be checked characteristic quantity;

Q (i)-the i-th master sample reconstruction signal fault characteristic value.

It is become symmetric form, obtains D _kl(P Q)=[D _kl(P Q)+D _kl(Q P)]/2 (6)

Finally provide trouble location and fault degree according to the KL-divergence value between sample to be tested and master sample.

Beneficial effect of the present invention: 1, the present invention proposes a kind of denoising method for wind-driven generator group wheel box initial failure vibration signal, effectively can improve signal to noise ratio (S/N ratio); 2, the present invention not only can position residing for failure judgement point, also can make relative judgement to fault degree, provide foundation to the on-condition maintenance of wind power generating set; 3, effectively can be monitored the state of wind-driven generator group wheel box by the method, thus find fault early, reduce time and the expense of shutdown and the replacing parts caused by catastrophic failure; 4, the present invention is also applicable to the Incipient Fault Diagnosis of other rotating machinery.

Accompanying drawing explanation

Fig. 1 is wind-driven generator group wheel box Incipient Fault Diagnosis method flow diagram provided by the invention;

Fig. 2 is from the vibration acceleration signal time domain waveform illustrated example that experiment porch gathers in specific implementation process of the present invention;

Fig. 3 is for application LMD method is to this decomposition result of wave pattern shown in Fig. 2 A;

The Hilbert that Fig. 4 is specimen reconstruct signal shown in Fig. 2 A converts spectrogram;

Fig. 5 is from the vibration signal time domain beamformer that wind energy turbine set experimental prototype gathers in specific implementation process of the present invention.

Embodiment

First, specific embodiment of the invention step is described in detail from the vibration data of testing table collection by analyzing, then analyze the vibration data gathered from actual wind energy turbine set experimental prototype, thus verify validity and the using value of method for diagnosing faults proposed by the invention further.

This patent test figure used is all from the rolling bearing fault simulated experiment platform in U.S.'s Case Western Reserve University electrical engineering laboratory, and this experiment table comprises the motor of 2 horsepowers, a torque sensor and a power test meter.Bearing to be measured is positioned at the two ends of motor, and drive end bearing model is SKF6205, and fan end bearing designation is SKF6203, and bearing fault point electric spark processes, and the diameter of impaired loci is respectively 0.1778mm, 0.3556mm, 0.5334mm.Wherein, the impaired loci of bearing outer ring is respectively at clock: 3 o'clock, 6 o'clock, 12 o'clock three directions, vibration data is collected by the vibration acceleration sensor be arranged on motor casing, and sample frequency is 12kHz, and power and rotating speed are recorded by torque sensor/code translator.

Specific embodiment of the invention step is as described below:

Step 1: above-mentioned testing table data acquisition vibration acceleration sensor gathers original vibration signal, the data gathered all are stored as * .mat form (Matlab file), be sample frequency at 12kHz shown in Fig. 2 A, the vibration signal fault sample that trouble spot is positioned at drive end inner ring, trouble spot diameter is 0.1778mm, motor speed collects when being 1730rpm; Fig. 2 B is take 12kHz as sample frequency, the vibration signal fault sample that trouble spot is positioned at drive end inner ring, trouble spot diameter is 0.3556mm, motor speed collects when being 1797rpm; Fig. 2 C is taking 12kHz as sample frequency, and the vibration signal fault sample that trouble spot is positioned at drive end inner ring, trouble spot diameter is 0.5334mm, motor speed collects when being 1797rpm, all the other samples used in the present invention are not listed one by one.Step 2: the decomposition of LMD method is carried out to the original vibration signal collected in step 1,

In formula, x (t)-original vibration signal Time Domain Amplitude ( m/s ² );

PF _i-the i-th component;

R-survival function.

For sample shown in Fig. 2 A, obtain 2 PF components and survival function r, result as shown in Figure 3.

Step 3: the component containing failure messages such as amplitude modulationfrequency modulations is used for signal reconstruction.As shown in Figure 3, the waveform of PF1 has the feature of amplitude modulationfrequency modulation, therefore obtains reconstruction signal

Step 4: carry out Hilbert conversion to reconstruction signal x ' (t), result as shown in Figure 4, extracts the fault characteristic value in reconstruction signal Hilbert transformation results, and namely the fault characteristic frequency of rolling bearing and two frequencys multiplication thereof (are expressed as: f successively _i, 2f _i, 3f _i, f _o, 2f _o, 3f _o, f _r, 2f _r, 3f _r) amplitude at place, be expressed as C=[A successively _fi, A _2fi, A _3fi, A _fo, A _2fo, A _3fo, A _fr, A _2fr, A _3fr], as data point in Fig. 4 (161.9,2587), the horizontal ordinate of (323.7,1310) and (485.6,1019) is respectively inner ring fault characteristic frequency f _iand two frequency multiplication 2f _iwith frequency tripling 3f _i, so, the fault characteristic value C=[25.87,1310,1019,20.4,5.5,15.7,24.9,64.6,55.2] of sample shown in Fig. 4.

Step 5: the sample of bearing under normal condition and various malfunction first collecting this model, for Criterion sample: n-097, i-105, o-130, r-3006; Then calculate the fault characteristic value of master sample and sample to be tested successively by above-mentioned steps 1-4, result is as shown in table 1, as space is limited reason, and this patent only lists the characteristic quantity of part sample, wherein, is normal sample with the sample of n beginning; The sample of i beginning is the sample that inner ring has pitting fault; With the sample of o beginning for outer shroud has the sample of pitting fault; With the sample of r beginning for rolling body has the sample of pitting fault; Finally calculate the KL-divergence between sample to be checked and master sample, result is as shown in table 2, characteristic quantity under different faults pattern (as inner ring fault) and the KL-divergence value between the characteristic quantity of corresponding master sample will be significantly less than the divergence value between the characteristic quantity of other standards sample, diagnostic result conforms to completely with actual fault point position, all obtain correct result, illustrate that the method effectively can identify the trouble location of rolling bearing.Sample i-106, i-107, i-108, | i-109 is the fault sample that inner ring has spot corrosion, and trouble spot diameter is 0.1778mm; Sample i-209, i-210, i-211, | the trouble spot diameter of i-212 is 0.5334mm, from table 2, from sample i-106 to i-212, the value of KL-divergence increases tens times, can push away thus, KL-divergence is more responsive to fault degree of injury, relatively can characterize fault degree.Other fault types are added up, same conclusion can be drawn, do not list one by one in table.

Table 1. different faults type characteristic quality of sample

KL-divergence between table 2 sample to be tested and master sample

Below by the vibration data that same step analysis gathers from actual wind energy turbine set experimental prototype, thus verify validity and the using value of method for diagnosing faults proposed by the invention further.

On-line monitoring is carried out to certain wind energy turbine set domestic 1.5MWvestas V66 wind power generating set, in July, 2010 there is vibration, the defective mode that noise is large in gear case, require to arrange according to measuring point and install acceleration transducer, the sample frequency of sensor is set to 12KHz.See shown in accompanying drawing 5, A figure is the sample time-domain oscillogram gathered in July, 2010; Figure B is the sample time-domain oscillogram gathered in May, 2011.Specific embodiment of the invention step is roughly the same with above-mentioned test platform vibration data treatment step, therefore no longer enumerates waveform and the data of producing property of computation process at this:

Step 1: adopt vibration acceleration sensor to gather original vibration signal, measuring point comprises: the horizontal and vertical position of gearbox input shaft bearing, gearbox planetary gear side, gearbox intermediate shaft side, gearbox high-speed axle bearing totally four measuring points.Vibration acceleration sensor is arranged near gearbox shaft bearing, so that the situation of change of sensitive detection gear case internal vibration, and reduces the loss in signaling pathways.Fig. 2 A and 2B is the wind-driven generator group wheel box high speed shaft side that the collects time domain beamformer at the vibration acceleration of different time points.

Step 2: to the original vibration signal collected temporally label carry 4096 sample points, the decomposition of LMD method is carried out to the original vibration signal x (t) extracted, $x\left(t\right)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{PF}}_i+r---\left(8\right)$

In formula, x (t)-original vibration signal Time Domain Amplitude ( m/s ² );

PF _i-the i-th component;

R-survival function.

Step 3: the PF component comprising the failure messages such as amplitude modulationfrequency modulation is used for signal reconstruction.

Step 4: carry out Hilbert conversion to reconstruction signal, extracts the modulation intelligence in signal, to extract fault signature.Herein means out, this gearbox high-speed end bearing model is NJ2326E, its inner ring fault characteristic frequency f _i=8.3794f _n; Outer shroud fault characteristic frequency f _o=5.6522f _n; Rolling body fault characteristic frequency f _r=4.9802f _n; f _nfor the gyro frequency of high speed shaft of gearbox.

KL divergence value between table 3 sample to be checked and master sample

Step 5: collection bearing normal condition sample 2010.7 and inner ring have the sample 2010.7 under spot corrosion malfunction, for Criterion sample, then the characteristic quantity of master sample and sample to be tested (sample 2010.11 and sample 2011.5) is calculated successively by above-mentioned steps 1-4, and the KL-divergence between sample to be tested and master sample, the results are shown in accompanying drawing table 3.By the known sample of data 2010.11, sample 2011.5 exists and sample 2010.7 is same fault type in table, namely inner ring is peeled off, and degree of peeling off becomes large successively, so attendant has changed bearing in May, 2011, find that bearing inner ring seriously peels off, peel off the about 50mm × 5mm of area, and lubricating oil is seriously polluted by metallic particles, the gearbox high-speed end vibration values after replacing recovers normal value.

Claims (1)

1., based on a wind-driven generator group wheel box Incipient Fault Diagnosis method for vibration monitoring, it is characterized in that it comprises the following steps:

Step 1: gather original vibration signal by acceleration transducer, gear case Sensor is: the horizontal and vertical position of gearbox input shaft bearing, gearbox planetary gear side, gearbox intermediate shaft side, gearbox high-speed axle bearing totally four measuring points.

Step 2: to the original vibration signal collected the temporally N number of sample point of tag extraction, N >=4096, carry out the decomposition of LMD method to the original vibration signal x (t) extracted, and can obtain the PF component that several instantaneous frequencys have physical significance, namely

$x\left(t\right)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{PF}}_i+r---\left(1\right)$

In formula, x (t)-original vibration signal Time Domain Amplitude;

PF _i-the i-th component;

R-survival function.

Step 3: choose comprise the failure messages such as amplitude modulationfrequency modulation PF component for signal reconstruction, i.e. x ' (t)=∑ PF _i.

Step 4: Hilbert conversion is carried out to reconstruction signal, and extracts fault characteristic value.

Step 5: gather the sample of wind-driven generator group wheel box under different conditions as master sample, calculate the fault characteristic value of master sample and sample to be tested successively, and calculate the KL-divergence between sample to be tested and master sample, namely

Wherein, D _klthe KL-divergence of (P Q)-between sample to be checked and master sample;

P (i)-the i-th specimen reconstruct signal fault to be checked characteristic quantity;

Q (i)-the i-th master sample reconstruction signal fault characteristic value.

It is become symmetric form,

D _kl(P□Q)＝[D _kl(P□Q)+D _kl(Q□P)]/2 (3)

Finally provide trouble location and fault degree.