CN104165925B - The centrifugal compressor half-opened impeller crack fault detection method of accidental resonance - Google Patents

Abstract

The invention belongs to field of diagnosis about equipment fault, the centrifugal compressor half-opened impeller crack fault detection method of a kind of accidental resonance.The present invention uses data collecting system, sound pressure sensor, strain data wireless acquisition system and foil gauge to be acquired compressor pressure pulsation and blade strain.And pressure fluctuation signal is carried out accidental resonance and empirical mode decomposition, counter stress signal has carried out spectrum analysis.The present invention studies and defines a kind of novel centrifugal compressor half-opened impeller crack fault recognition methods, it is possible to achieve double opentype compressor blade cracks detects, and can avoid the accident caused due to the crack fault of blade.

Description

The centrifugal compressor half-opened impeller crack fault detection method of accidental resonance

Technical field

The invention belongs to field of diagnosis about equipment fault, the centrifugal compressor semi-open type of a kind of accidental resonance Impeller crack fault detection method.

Background technology

Centrifugal compressor can carry out lifting fluid pressure by the rotation of impeller, is the visual plant of petrochemical factory.Existing , people increasingly pay close attention to high efficiency and the high reliability of centrifugal compressor.Blade be one of centrifugal compressor non- The most important but the weakest parts, its working environment very severe is subjected to the impact of fluid, noise and high temperature.Due to The working environment of centrifugal compressor is complicated and changeable, and therefore compressor often runs under off design point, and blade cracks is frequent Occur, leaf destruction may finally be caused, cause unit to damage.But, the detection of blade cracks is complex, always puzzlement Equipment fault diagnosis engineering circles and a difficult problem for academia both at home and abroad.

Now the method that crackle Non-Destructive Testing is conventional mainly there are ultrasonic inspection, radiographic inspection, eddy current inspection, magnetic powder inspection With gold penetrant inspection etc., although blade cracks can be detected by these methods, but these methods mostly can not realize Real-time monitoring to blade cracks, re-test after can only shutting down, lack certain ageing.

When characteristic information is the faintest, it is often flooded by noise, and traditional method is to remove noise, and special Reference breath extracts, and accidental resonance is to utilize the noise in signal.Signal through bistable system, is utilized noise energy by it Amount make signal with the frequency transition of characteristic information thus by feature information extraction out.Bistable system expression formula

$\stackrel{\·}{x}=\mathrm{ax}-{\mathrm{bx}}^3+s\left(t\right)+n\left(y\right)---\left(1\right)$

Wherein s (t) is characterized signal n (t) is white noise, and s (t)+n (t) is actual signal, and bistable system has two gesture Trap, when signal is fainter, signal can only shake in a potential well back and forth.When there being noise, and its energy reaches certain Value, signal can transition back and forth in two potential wells.Its escape rate is

$r_k=\frac{a}{\sqrt{2}\mathrm{\π}}\exp \left(\frac{-\mathrm{\ΔV}}{D}\right)---\left(2\right)$

Wherein D is noise intensity,For barrier height.Just can occur during the half allowing signal frequency be escape rate Resonance.Due toAnd r_kReduce with the increase of a.Therefore signal can resonate at low frequency, it is therefore desirable to right Signal carries out time domain stretching, makes signal can resonate at high frequency treatment.The necessary bar can restrained after additionally signal entered bistable system Part is ah ＜ 1,Just the value of a can be obtained with this, it addition, when signal frequency f=0.5r_kIt is to can reach Resonance, then can try to achieve b according to formula 2, so that system resonance, can solve out by characteristic frequency.

The basic thought of EMD method is the combination that primary signal resolves into a series of intrinsic mode function IMF,

$X\left(t\right)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{c}_i\left(t\right)+r_n\left(t\right)---\left(3\right)$

The most according to actual needs, each IMF utilization is carried out follow-up Treatment Analysis and feature extraction, such as warp The features such as the instantaneous frequency of mode component, instantaneous amplitude or time-frequency spectrum are asked in Hilbert conversion.

IMF has procedure below to obtain:

First find all maximum points and the minimum point of signal, respectively maximum and minimum are carried out cubic spline Function Fitting obtains maximum and minimizing envelope, two envelope summations is taken average and just can get first eigen mode State function C₁T (), then deducts C by original signal₁(t), then carry out first step operation, just can get C₂(t).The most reciprocal. Until n-th order IMF component C_n(t) or its surplus r_nT () is less than preset value；Or as residual components r_nT () is monotonic function or normal During amount, EMD catabolic process stops.

X(t)-C₁(t)=r₁(t)

r₁(t)-C₂(t)=r₂(t) (4)

……

r_n-1(t)-C_n(t)=r_n(t)

Summary of the invention

For above-mentioned the problems of the prior art and accidental resonance and the feature of empirical mode decomposition, the purpose of the present invention According to centrifugal compressor and the structure of half-opened impeller thereof and the feature of working condition, study and define a kind of novel based on The centrifugal compressor half-opened impeller crack fault frequency extraction method of the accidental resonance of accidental resonance and empirical mode decomposition, Thus solve the dynamic test problems of centrifugal compressor half-opened impeller crackle, it is to avoid cause due to the crack fault of blade Accident occur.

The centrifugal compressor of the technical scheme is that a kind of accidental resonance of the present invention is half-open Formula impeller crack fault detection method, comprises the following steps:

Sound pressure sensor is installed respectively at centrifugal compressor impeller entrance, diffuser inlet, diffuser exit, passes through The pressure fluctuation signal of sound pressure sensor described in data acquisition system；

Described pressure fluctuation signal is carried out empirical mode decomposition, obtains described centrifugal compressor feature at low frequency Frequency f_c；Meanwhile, described pressure fluctuation signal is carried out accidental resonance process, make bistable system resonate at low frequency, the most right Data after accidental resonance processes carry out spectrum analysis, obtain the frequency spectrum after accidental resonance；

Multiple scale analysis: search characteristic frequency f in the frequency spectrum after accidental resonance_cAnd frequency multiplication；

Judging characteristic frequency f_cAnd whether frequency multiplication exists, if characteristic frequency f_cDo not exist with its frequency multiplication, then judge blade Flawless；

If it is present judging characteristic frequency f_cWhether it is axle frequency, if it is, judge blade flawless, otherwise judges There is crackle in blade.

Described described pressure fluctuation signal is carried out empirical mode decomposition, obtain described centrifugal compressor at low frequency Characteristic frequency f_c, comprise the following steps:

At blade passing frequency, described pressure fluctuation signal is carried out bandpass filtering；

Signal after bandpass filtering is carried out envelope processing；

Empirical mode decomposition；

The IMF signal obtained through empirical mode decomposition is carried out Fourier transformation, it is thus achieved that centrifugal compressor is at low frequency Characteristic frequency f_c。

Described described pressure fluctuation signal is carried out accidental resonance process so that it is resonate at low frequency, comprise the following steps:

Described pressure fluctuation signal is carried out cubic spline interpolation and carries out resampling；

Signal after above-mentioned process carries out time domain stretching, and the time interval i.e. expanded between data makes characteristic frequency fall low Frequently district；

Integral pressure microseismic data is divided by pressure fluctuation data valid；

According to condition of convergence ah ＜ 1,Determine a, according to resonance condition F=0.5r_kDetermine that b, resonant frequency f are escape rate r_k1/2 can reach resonance, wherein, a, b are the structure ginseng of bistable system Number, x₀For first value of data, h is iteration step length,For barrier height, D is for estimating noise intensity；

Parameter a, b are brought into afterwards and are finely adjusted and make bistable system reach resonance.

The present invention has the following advantages and beneficial effect:

1. the present invention passes through the research to centrifugal compressor half-opened impeller crack fault recognition methods of the method for test, To centrifugal compressor impeller entrance, diffuser inlet, diffuser exit pressure fluctuation data analyzed, and pass through Contrasting with blade strain signal, blade can be split by the pressure fluctuation signal at final certification diffuser inlet and diffuser exit Stricture of vagina is monitored.

2. the present invention can pass through the simple and quick centrifugal compressor vibration signal real-time data acquisition platform built, right The pressure fluctuation signal data of centrifugal compressor diffuser entrance and outlet are acquired, and then carry out data at data Reason, by the presence or absence of failure judgement frequency and then centrifugal compressor half-opened impeller crack fault can be identified, and then The accident caused because of impeller crack fault can be prevented effectively from occur.

Accompanying drawing explanation

Fig. 1 is the inventive method flow chart；

Fig. 2 is accidental resonance process chart；

Fig. 3 is empirical mode decomposition flow chart；

Fig. 4 is crackle and the foil gauge location drawing；

Fig. 5 is 4500rpm strain data spectrogram；

Fig. 6 is pressure fluctuation signal time domain and frequency domain figure；

Fig. 7 is the EMD exploded view to envelope signal；

Fig. 8 is the first rank IMF and spectrogram thereof；

Fig. 9 is rear time-domain diagram after accidental resonance；

Figure 10 is frequency domain figure after accidental resonance；

Figure 11 is frequency domain enlarged drawing after accidental resonance；

In figure, mark point is characterized frequency and frequency multiplication thereof.

Detailed description of the invention

Below in conjunction with the accompanying drawings and embodiment the present invention is described in further detail.

The present invention is embodied as route such as shown in Fig. 1, and it specifically comprises the following steps that

(1) building centrifugal compressor pressure fluctuation signal data acquisition platform, this platform features is:

(1) there is the data collection and analysis instrument of multiple input path and data processing function；

(2) by data collection and analysis instrument real-time acquisition and display centrifugal compressor pressure fluctuation signal.

What the present invention was studied is industrial centrifugal compressor, to centrifugal compressor diffuser entrance and outlet Pressure fluctuation signal data are acquired, and then data are carried out data process, it was demonstrated that diffuser inlet and the pressure of outlet Blade cracks can effectively be reflected by microseismic data.

(2) building blade stress acquisition platform, this platform features is:

(1) there is the blade stress wireless test system of wireless transmissions function；

(2) to the normal of compressor with have crackle blade to carry out stress test, to obtain the feature frequency of reflection blade fault Rate.

(3) data process

Based on accidental resonance and the ultimate principle of empirical mode decomposition, and according to centrifugal compressor and semi-open type thereof The structure of impeller and the feature of running parameter, be analyzed empirical mode decomposition and be total at random compressor pressure microseismic data Shake process.Its detailed process is as follows:

First, pressure fluctuation signal carrying out empirical mode decomposition, concrete handling process is shown in Fig. 4.Pass through empirical mode Decompose the characteristic frequency f obtained at low frequency_c, then pressure fluctuation signal is carried out accidental resonance process and makes it be total at low frequency Shaking, method and the flow process of accidental resonance are shown in Fig. 3.Wherein, bandpass filtering concrete signal needs concrete analysis, and different signals has not Same requirement.Then the data after processing accidental resonance carry out spectrum analysis, obtain the frequency spectrum after accidental resonance.The most again with In machine resonance spectrum, find f_cAnd frequency multiplication, if there is then thinking f_cIt is characteristic frequency to be looked for, otherwise it is assumed that without this Characteristic frequency.Last counter stress signal carries out spectrum analysis and obtains failure-frequency the characteristic frequency with pressure fluctuation signal gained Carry out contrasting to verify the accuracy of the method.

In bistable system, can only resonate at low frequency, and actual signal is often positioned in high frequency treatment, it is therefore desirable to will Characteristic frequency when signal carries out time domain stretching namely frequency domain compression enters low frequency and reaches resonance, this process need dt namely Sampling interval expansion obtains h, and this will cause used bistable system iteration step length h to increase, and this system can be caused to dissipate, for Making h less, carrying out the resampling of higher frequency, dt namely sampling interval will diminish herein, and the h obtained after expansion is the most corresponding Diminish.The resonance of bistable system the most also with two other parameter a of system, b is correlated with, can determine a according to the condition of convergence, according to Resonance condition determines b, so that system reaches resonance.

The method carried has been carried out experimental verification, first operating simulation crack length 70mm on blade, and to blade Stress is measured, and foil gauge pastes position and crack position is shown in Fig. 4, wherein foil gauge a, and b, c paste crackle leaf On sheet, foil gauge d pastes on flawless blade, and the frequency spectrum that the stress data recorded is analyzed obtaining stress data is shown in figure 5, it is apparent that have the b of crackle blade, c to be in about 53HZ have an obvious characteristic frequency, and nothing at the d of normal blade This frequency, it is believed that 53HZ is the failure-frequency of blade cracks.Checking to this extracting method is as follows, first to pressure fluctuation Carry out Fourier transformation to obtain its spectrogram and see Fig. 6, wherein without the information the most relevant with 53HZ, then to pressure fluctuation number See Fig. 3 according to carrying out EMD decomposition idiographic flow, obtained 53HZ left characteristic fault frequency and seen Fig. 7, Fig. 8, but do not carried out stress survey The accuracy of the method cannot be verified during examination, then utilize accidental resonance that pressure fluctuation signal is resonated specifically at low frequency and flow Journey is shown in Fig. 2, and the result obtained is shown in Fig. 9-11, search characteristics frequency and frequency multiplication thereof in the pressure fluctuation data after accidental resonance, Due in the drawings it appeared that significantly characteristic frequency and frequency multiplication thereof (Figure 10, Figure 11 mark shown in), it is believed that 53HZ is Failure-frequency is consistent with stress test.Spline interpolation in Fig. 2 is the premise of resampling, and resampling is to make sampling interval dt Reduce, so that iteration step length reduces, to ensure system convergence.

The formulation process of crackle criterion is:

1) respectively in centrifugal compressor impeller entrance, diffuser inlet, sound pressure sensor is installed at diffuser exit, Respectively near blade cracks and without there being installation foil gauge on the blade of crackle；

2) by sound pressure sensor and the compressor pressure fluctuating signal of data acquisition system, and pressure fluctuation is believed Number carry out accidental resonance and empirical mode decomposition, it is thus achieved that pass judgment on characteristic frequency with presence or absence of blade cracks.And utilize strain wireless Blade stress is acquired by acquisition system.

3) with axle frequency, the pressure fluctuation signal characteristic frequency that contrast obtains can obtain whether blade cracks exists.Counter stress Signal carries out spectrum analysis and obtains Blade Crack Fault frequency.Contrast 2) in the characteristic frequency that obtains and the fault frequency in stress Rate, can be to 2) proposed in method verify.

Claims (2)

1. the centrifugal compressor half-opened impeller crack fault detection method of an accidental resonance, it is characterised in that include with Lower step:

Sound pressure sensor is installed respectively at centrifugal compressor impeller entrance, diffuser inlet, diffuser exit, passes through data Acquisition system gathers the pressure fluctuation signal of described sound pressure sensor；

Described pressure fluctuation signal is carried out empirical mode decomposition, obtains described centrifugal compressor feature frequency at low frequency Rate；Meanwhile, described pressure fluctuation signal is carried out accidental resonance process, make bistable system resonate at low frequency, then at random Data after resonance processes carry out spectrum analysis, obtain the frequency spectrum after accidental resonance；

Multiple scale analysis: search characteristic frequency f in the frequency spectrum after accidental resonance_cAnd frequency multiplication；

Judging characteristic frequency f_cAnd whether frequency multiplication exists, if characteristic frequency f_cDo not exist with its frequency multiplication, then judge that blade is without splitting Stricture of vagina；

If it is present judging characteristic frequency f_cWhether it is axle frequency, if it is, judge blade flawless, otherwise judges that blade is deposited At crackle；

Described described pressure fluctuation signal is carried out empirical mode decomposition, obtain described centrifugal compressor feature at low frequency Frequency f_c, comprise the following steps:

At blade passing frequency, described pressure fluctuation signal is carried out bandpass filtering；

Signal after bandpass filtering is carried out envelope processing；

Empirical mode decomposition；

The IMF signal obtained through empirical mode decomposition is carried out Fourier transformation, it is thus achieved that centrifugal compressor spy at low frequency Levy frequency f_c。

The centrifugal compressor half-opened impeller crack fault detection method of accidental resonance the most according to claim 1, its It is characterised by, described described pressure fluctuation signal is carried out accidental resonance process so that it is resonate at low frequency, including following step Rapid:

Described pressure fluctuation signal is carried out cubic spline interpolation and carries out resampling；

Signal after above-mentioned process carries out time domain stretching, and the time interval i.e. expanded between data makes characteristic frequency fall at low frequency District；

Integral pressure microseismic data is divided by pressure fluctuation data valid；

According to condition of convergence ah ＜ 1,Determine a, according to resonance conditionF= 0.5r_kDetermine that b, resonant frequency f are escape rate r_k1/2 can reach resonance, wherein, a, b are the structural parameters of bistable system, x₀For first value of data, h is iteration step length,For barrier height, D is for estimating noise intensity；

Bistable system is made to reach resonance after parameter a, b are brought into and be finely adjusted.