CN104406792A - Solid rocket engine fault diagnosis method - Google Patents

Abstract

The invention relates to a solid rocket engine fault diagnosis method, and particularly relates to application of vibration signal analysis and a constant fault alarm detection algorithm in solid rocket engine fault diagnosis. The fault diagnosis method is based on a vibration signal analysis technology, and the constant fault alarm detection algorithm is applied to the feature extraction process to perform state identification on specific frequency points on the basis of a frequency fault diagnosis method so that sudden frequency signals submerged in clutter energy spectrum can be effectively detected and accurate frequency information can be obtained. The solid rocket engine fault diagnosis method has important meaning on engine fault decision and provides reliable reference and support for engine design personnel and test personnel.

Description

A kind of Solid Rocket Motor Fault Diagnosis method

Technical field

The present invention relates to a kind of Solid Rocket Motor Fault Diagnosis method, particularly analysis of vibration signal and the CFAR detection algorithm application in Solid Rocket Motor Fault Diagnosis.

Background technology

Solid propellant rocket ground run is a complicated systems engineering, has excessive risk, high cost, test period short and fault and the features such as rapid occur.Vibration signal is one of important parameter of engine run measurement, can reflect the performance change of engine than Steady-state Parameters quickly; Fault diagnosis mainly carries out failure mode analysis (FMA) to solid propellant rocket ground run data, and the selected algorithm adapted with it, to engine, makes Performance Evaluation accurately; CFAR detection is a kind of effective signal analysis recognizer, has have very large advantage to the dash forward detection of frequency of frequency-region signal.At present, the system being carried out Diagnosis on Engine fault by vibration parameters measurement is also in theoretical research and technology exploration stage, and diagnostic method is single, and frequency fault diagnosis, as a kind of important diagnostic method, cannot differentiate signal frequency of dashing forward effectively.

Summary of the invention

The technical matters that the present invention solves is a kind of Solid Rocket Motor Fault Diagnosis method of invention, this method for diagnosing faults is based on analysis of vibration signal technology, the basis of original frequency fault diagnostic method adds CFAR detection algorithm, compensate for deficiency of the prior art.

First this method for diagnosing faults carries out the pre-service of signal time domain to the engine luggine signal collected, and then carries out frequency domain FFT conversion, then carries out feature extraction, carries out state recognition, finally carry out fault decision-making to characteristic frequency point.In characteristic extraction procedure, use CFAR detection algorithm to carry out detection to prominent frequency signal to the signal spectrum after frequency domain process and identify, be specially: suppose detected unit x _mbe positioned at the m after Fast Fourier Transform (FFT) capable; the protected location that removing spectrum energy is revealed; extract L point at the row at detected target place and enter detection reference unit; detect reference unit to L to sort from big to small; remove r maximal value afterwards, mean value computation is carried out to the amplitude of residue unit, then is multiplied by scale factor k; its result as detection threshold, by tested measuring point x _mamplitude compared with detection threshold. when the amplitude of tested measuring point is greater than detection threshold, output coordinate information, i.e. frequency information and amplitude information.

Signal Pretreatment process comprises data cutout, and zero line adjusts, trend abstraction with get rid of and abnormally to get rid of.

In described characteristic extraction procedure, after CFAR detection identification is carried out to prominent frequency signal, carry out fundamental frequency identification and state recognition process.

Being described as of this fault decision-making technique: amorphous solid rocket engine, case material and Grain structure are fixing, and during engine operation, cardinal process frequency vibration position is also fixing; The engine of stable performance, prominent frequency unique point in particular job section does not change, when breaking down in engine working process, also will there is significantly change in vibration amplitude and rumble spectrum, and main fault signature is that the signal in spectrogram dashes forward frequency component can showed increased.

Beneficial effect of the present invention:

Effectively can detect the prominent frequency signal be submerged in clutter energy spectrum, obtain accurate frequency information, for follow-up Analysis on Fault Diagnosis, for engine designer, tester provide reliable support.

Accompanying drawing explanation

Below in conjunction with accompanying drawing and instantiation, the present invention is described in detail.

Fig. 1 is the fault diagnosis flow scheme schematic diagram of example of the present invention, comprises Signal Pretreatment, frequency domain process, feature extraction, fault decision-making;

Fig. 2 is the CFAR detection algorithm schematic diagram of example of the present invention, comprises extracting detecting reference unit, and sequence is averaging, and determines detection threshold, relatively also output characteristic information four steps.

Embodiment

Example of the present invention: a kind of Solid Rocket Motor Fault Diagnosis method.

As shown in Figure 1, Solid Rocket Motor Fault Diagnosis method mainly comprises Signal Pretreatment, frequency domain process, feature extraction, fault decision-making four part.

Signal Pretreatment mainly carries out time domain data process to the original vibration data that solid propellant rocket ground run collects, and mainly comprises data cutout, and zero line adjusts, trend abstraction and eliminating, and abnormal point is got rid of.Below Signal Pretreatment process is described:

Step one: data cutout

According to signal effective length, intercept from engine luggine data valid data.The data length intercepted is N, and starting point is k1 input parameter is time range (t1, t2);

Step 2: zero line adjusts

Signal after intercepting is averaging, then from original signal, deducts average.Computing formula:

x(k)＝x(k)-x _mean

K=k1, k1+N, N are the number of data points removing average,

$x_{\mathrm{mean}}=\underset{k=n1}{\overset{n1+L}{\mathrm{\Σ}}}x\left(k\right)$

L is the data length of computation of mean values, and n1 is the starting point of computation of mean values;

Step 3: trend abstraction and eliminating

Use running mean method to carry out trend abstraction and eliminating to data, repeatedly can carry out trend abstraction.Computing formula:

$x\left(k\right)=\frac{1}{L}\underset{m=n1}{\overset{n1+L}{\mathrm{\Σ}}}x(k+m)$

Trend extraction, L is that calculating mean value is counted, and get rid of trend term, x (n)=x (n)-x (k), n=n1, n1+N, in formula, n1 is removal trend starting point, and N is that removal trend is always counted;

Step 4: abnormal point is got rid of

To the data in appointed area, data point absolute value being greater than threshold value is set to the average of its left and right data or puts constant.

Frequency domain process mainly carries out Fast Fourier Transform (FFT) (FFT) to pretreated time-domain signal, in the present invention, this algorithm is specifically expressed as: to the sinusoidal signal result of spectrum analysis blocked complete cycle, and the amplitude of its frequency spectrum equals sinusoidal signal amplitude.Computing formula:

$F\left(f\right)=\left\{\begin{array}{c}2X\left(f\right)/\mathrm{\&Delta;f},f>0\\ X\left(f\right)/\mathrm{\&Delta;f},f=0\\ 0,f<0\end{array}\right.$

In formula, $X\left(f\right)=\frac{1}{N}\underset{n=1}{\overset{N}{\mathrm{\&Sigma;}}}x\left(n\right)e^{-j2\mathrm{\&pi;nf\&Delta;t}}$ For Fourier transform.The present invention can calculate average frequency spectrum, when average time is n _dtime, whole data are divided into n _dsection, every segment data length is N, N is that FFT counts, and in the present invention, N gets 1024, then carries out population mean to the frequency spectrum of every segment data:

$\stackrel{\&OverBar;}{X}\left(f\right)=\frac{1}{n_d}\underset{n=1}{\overset{n_d}{\mathrm{\&Sigma;}}}{X}_n\left(f\right)$

Signal is after frequency domain conversion, and need to carry out feature extraction, its process mainly comprises CFAR detection, fundamental frequency identification, state recognition.Below characteristic extraction procedure is described:

Step one: CFAR detection

CFAR detection CMLD-CFAR is carried out for the signal after frequency domain FFT converts.This algorithm realization is mainly described as: suppose detected unit x _mbe positioned at the m after Fast Fourier Transform (FFT) capable; the protected location that removing spectrum energy is revealed; extract L point at the row at detected target place and enter detection reference unit; detect reference unit to L to sort from big to small; remove r maximal value afterwards, mean value computation is carried out to the amplitude of residue unit, then is multiplied by scale factor k; its result as detection threshold, by tested measuring point x _mamplitude compared with detection threshold. when the amplitude of tested measuring point is greater than detection threshold, output coordinate information, i.e. frequency information and amplitude information.；

Step 2: fundamental frequency identification

The engine of often kind of model has fixing vibration frequently prominent, these prominent frequently with the housing of engine and Grain structure closely related, when engine breaks down, the position of these cardinal process frequencies and amplitude will change, by the position generalized case finding relevant main fundamental frequency also just to have found fault in spectrogram, often increase the distinguishing rule as malfunction using fundamental frequency signal amplitude;

Step 3: state recognition

Vibration performance data refer to by coherent signal process means such as FFT, the related data extracted from the fundamental frequency signal of vibration, comprising the value of always shaking of vibration signal, and the amplitude etc. of engine dominant frequency and each frequency multiplication thereof.The amplitude on fundamental frequency and relevant characteristic frequency point is extracted in the frequency spectrum of vibration signal.

Fault decision-making relates generally to frequency fault decision-making technique, the present invention's being described as this decision-making technique: amorphous solid rocket engine, and case material and Grain structure are fixing, and during engine operation, cardinal process frequency vibration position is also fixing; The engine of stable performance, prominent frequency unique point in particular job section does not change substantially, when breaking down in engine working process, also will there is significantly change in vibration amplitude and rumble spectrum, main fault signature is that the signal in spectrogram dashes forward frequency component can showed increased.

Under normal circumstances, the prominent frequency value on characteristic frequency point can change within the specific limits, and its variation range bound is added up by all previous firing test data and obtained:

x _tmin(k)≤x(k)≤x _tmax

If continuous 3 times or 4 times exceed this scope, then show that engine there occurs fault.

As shown in Figure 2, CFAR detection CMLD-CFAR is carried out for the signal after frequency domain FFT conversion.This algorithm realization step is as follows:

Step one: extract and detect reference unit

Suppose detected unit x _mbe positioned at the m after Fast Fourier Transform (FFT) capable, the protected location that removing spectrum energy is revealed, extract L point enter detection reference unit at the row at detected target place, relating to unit is from x _m-L/2-1to x _m-1, and x _m+1to x _m+L/2+1;

Step 2: sequence is averaging

Detect reference unit to L to sort from big to small, remove r maximal value afterwards, mean value computation is carried out to the amplitude of residue unit;

Step 3: determine detection threshold

The average of step 2 being tried to achieve is multiplied by scale factor k, and its result is as detection threshold;

Step 4: relatively also output characteristic information

By tested measuring point x _mamplitude compared with detection threshold. when the amplitude of tested measuring point is greater than detection threshold, output coordinate information, i.e. frequency information and amplitude information.

In sum, present invention achieves the efficient diagnosis of solid propellant rocket fault.

Claims (4)

1. a Solid Rocket Motor Fault Diagnosis method, comprise Signal Pretreatment, frequency domain process, feature extraction, fault decision-making four part, wherein said Signal Pretreatment process carries out time domain data process to the original vibration data that solid propellant rocket ground run collects; Described frequency domain process carries out Fast Fourier Transform (FFT) to pretreated time-domain signal; Described fault decision process relates to frequency fault decision-making technique, it is characterized in that: in described characteristic extraction procedure, uses CFAR detection algorithm to carry out detection to prominent frequency signal identify, be specially for the signal spectrum after frequency domain Fourier transform:

Suppose detected unit x _mbe positioned at the m after Fast Fourier Transform (FFT) capable; the protected location that removing spectrum energy is revealed; extract L point at the row at detected target place and enter detection reference unit; detect reference unit to L to sort from big to small; remove r maximal value afterwards, mean value computation is carried out to the amplitude of residue unit, then is multiplied by scale factor k; its result as detection threshold, by tested measuring point x _mamplitude compared with detection threshold, when the amplitude of tested measuring point is greater than detection threshold, output coordinate information, i.e. frequency information and amplitude information.

2. Solid Rocket Motor Fault Diagnosis method according to claim 1, is characterized in that: described Signal Pretreatment process comprises data cutout, and zero line adjusts, trend abstraction with get rid of and abnormally to get rid of.

3. Solid Rocket Motor Fault Diagnosis method according to claim 1, is characterized in that: in described characteristic extraction procedure, after carrying out CFAR detection identification to prominent frequency signal, carries out fundamental frequency identification and state recognition process.

4. Solid Rocket Motor Fault Diagnosis method according to claim 1, it is characterized in that: being described as of described fault decision-making technique: amorphous solid rocket engine, case material and Grain structure are fixing, and during engine operation, cardinal process frequency vibration position is also fixing; The engine of stable performance, prominent frequency unique point in particular job section does not change, when breaking down in engine working process, also will there is significantly change in vibration amplitude and rumble spectrum, and main fault signature is that the signal in spectrogram dashes forward frequency component can showed increased.