CN102072144A - Vibration and noise online monitoring and fault diagnosis system of scroll compressor - Google Patents

Abstract

The invention comprises a vibration and noise online monitoring device and a fault diagnosis system of a scroll compressor. The vibration and noise online monitoring device of the scroll compressor is composed of a scroll compressor, a piezoelectric-type accelerometer, a sound microphone, a preamplifier, an adapter, a signal conditioning module, an A/D (analog/digital) conversion module, a signal analysis module and a result displaying module, wherein the piezoelectric-type accelerometer is fixed on a shell of the scroll compressor; the sound microphone is arranged near the scroll compressor; the piezoelectric-type accelerometer and the sound microphone respectively collect the vibration amount and the noise amount of the scroll compressor; an analog signal is amplified and converted into a digital signal by the preamplifier, the adapter, the signal conditioning module and the A/D conversion module, and then the digital signal is sent to a computer and is analyzed by a computer software; and the analysis result is displayed. The fault diagnosis system of the scroll compressor is characterized in that the signal analysis module analyzes the vibration amount and the noise amount, and the operation stability and the faults of the scroll compressor device are judged through the result displaying module.

Description

Scroll compressor on-line vibration, noise monitoring and fault diagnosis system

Technical field

Scroll compressor on-line vibration of the present invention, noise monitoring and fault diagnosis system relate to a kind of detection diagnostic system, specifically scroll compressor on-line vibration, noise monitoring and fault diagnosis system.

Background technique

At present, the sealed refrigeration scroll compressor is used in the household electric appliance such as air conditioner, refrigerator, refrigerator-freezer more, and these electrical equipment play important effect in people's daily life, so the quality of its performance directly affects people's quality of life.Along with science and technology development and human to improving constantly that living environment requires, its noise more and more is subject to people's attention.So, scroll compressor must become the main noise source as the heart of such household electric appliance, also is the main object to such household electrical appliances vibration and noise reducing.Therefore, for scroll compressor Noise and Vibration Control and The Research of Relevant Technology thereof,,, extremely important meaning is arranged all still from refrigerator and air conditioner overall performance no matter from the performance of compressor own.

Because scroll compressor is than the compressor (especially reciprocal compressor) of other type, vibration and noise are less, so vibrate, Research on Noise never causes enough attention of people.Yet along with the develop rapidly of scroll compressor technology, vibration and noise as one of important indicator of scroll compressor performance more and more receive producer and client's concern.Present test and control technique research wretched insufficiency for scroll compressor vibration, noise, sampling observation before manufacturer also only limits to dispatch from the factory, and A-weighted sound level and linear sound level when test parameter generally is confined to the vibration amplitude of surface of shell and semianechoic room inside vortex compressor operating, and profound analysing and processing is not carried out in its vibration and noise.As analyzing its frequency characteristic, spatial distribution characteristic, vibration source and sound source position etc. by vibration and noise signal, this is unfavorable for the development in an all-round way of scroll compressor technology obviously.In addition, scroll compressor is difficult to the vibration and the noise of each main parts size are carried out independent measurement as totally enclosed type machinery, and this also is one of principal element of restriction test and control technique development.

Although the research for scroll compressor vibration and noise characteristic is less, and some scholars have done pilot study.Young Man Mr. Cho etc. utilizes state space theory that the compressor sound source has been carried out identification.Professor Liu Zhenquan has discussed by the spectrum analysis of oscillating signal and has caused scroll compressor vibration main cause.Wang doctor Zhen etc. by the vibration and the analytic demonstration of noise signal the coherence between them.People such as Zhang Tieshan and Yang Xiaofeng also are by related experiment and simple spectral analysis technology, and the scroll compressor noise source has been carried out identification.People such as Jiang Guoqing have carried out power spectrumanalysis and have proposed simple noise reduction measure with compressor noise refrigerator.Do not associate these documents as can be seen, only limit to surface signal, and not deep scroll compressor internal vibration and noise signal are detected for the test of scroll compressor vibration and noise.All adopt traditional spectral analysis technology but seldom to use the modern signal analytical method for the present major part of analytical method, as time frequency analysis, neuron network etc.At existing problem in the above-mentioned prior art, research and design a kind of novel scroll compressor on-line vibration, noise monitoring and fault diagnosis system, existing problem is very necessary in the prior art thereby overcome.

Summary of the invention

In view of existing problem in the above-mentioned prior art, the objective of the invention is research and design a kind of novel scroll compressor on-line vibration, noise monitoring and fault diagnosis system, thereby solve develop rapidly along with the scroll compressor technology, vibration and noise as one of important indicator of scroll compressor performance more and more receive producer and client's concern.Present test and control technique research wretched insufficiency for scroll compressor vibration, noise, sampling observation before manufacturer also only limits to dispatch from the factory, and A-weighted sound level and linear sound level when test parameter generally is confined to the vibration amplitude of surface of shell and semianechoic room inside vortex compressor operating, and profound analysing and processing is not carried out in its vibration and noise.All adopt traditional spectral analysis technology for the present major part of analytical method, but seldom use problems such as modern signal analytical method.

Scroll compressor on-line vibration of the present invention, noise monitoring and fault diagnosis system comprise scroll compressor on-line vibration, noise-monitoring equipment and scroll compressor fault diagnosis system.Described scroll compressor on-line vibration, noise-monitoring equipment show that by scroll compressor, piezoelectric accelerometer, sound Mike, preamplifier, ABAP Adapter signal condition module, A/D conversion, signal analysis, result institute forms.Described piezoelectric accelerometer is fixed on the shell of scroll compressor, sound Mike be placed on scroll compressor near, piezoelectric accelerometer harmony Mike tests the vibratory output and the noisiness of scroll compressor respectively and gathers, by preamplifier, ABAP Adapter, signal condition module and A/D conversion analogue signal is amplified and converted digital signal to then and give computer, carry out signal analysis by computer software, and result's demonstration is carried out in analysis.Described scroll compressor fault diagnosis system is to vibratory output and noisiness is analyzed and show by the result smoothness of operation and the fault of scroll compressor device are judged in signal analysis.

In showing, signal analysis of the present invention and result be provided with:

Be used to analyze and show the software module of scroll compressor surface vibration signal;

Be used to analyze and show the software module of scroll compressor noise signal;

Be used for analysis software module to the scroll compressor fault diagnosis;

(1), wherein the software module of scroll compressor surface vibration signal analysis has:

1) signal that surface vibration is sampled to scroll compressor carries out time-domain analysis and shows its oscillogram;

2) signal that obtains of sampling carries out amplitude spectrum, power spectrum, cepstrum, resonance and demodulation spectrum analysis and shows each analysis waveform figure, obtains the frequency information and the phase information of oscillating signal;

(2), the software module of scroll compressor noise signal analysis has:

1) the scroll compressor noise signal that sampling is obtained carries out time-domain analysis and carries out oscillogram showing.

2) noise signal is further analyzed, power spectrum, refinement spectrum, resonance and demodulation spectrum, third-octave analysis are arranged and show each analysis waveform figure; And demonstrate sound pressure level, sound power level or sound intensity level under the A weighted.

3) generation of form; The experiment Reporting Requirements of system is printed the compressor parameter information simultaneously and is tested the octave data information that obtains, and data form and figure show jointly.

(3), be used for its main contents of analysis software module to the scroll compressor fault diagnosis: the method for diagnosing faults of this module is the analysis of local ripple autoregressive spectrum, local ripple K-L quantity of information analysis and local ripple analysis of neural network method:

1) local ripple autoregressive spectrum analytical method:

Formula (1) be AR (n) model from spectral function:

In the formula

Be the residual error variance, Δ t is the sampling interval.

After correctly setting up the AR model, can utilize following method to find the solution the residual error variance: establishing sequential length is N, and model order is n, and model parameter is φ ₁, φ ₂..., φ _n, the residual error variance of model then

Can be expressed as formula (2):

${\mathrm{\σ}}_a^2=\frac{N-n}{N}{\left(\mathrm{X\Φ}\right)}^T\mathrm{X\Φ}---\left(2\right)$

In the formula

$X=\left[\begin{array}{c}{x}_{n+1},x_n,...,x_1\\ x_{n+2},x_{n+1},...,x_2\\ ......\\ x_N,x_{N-1},...,x_{N-n}\end{array}\right]$

T is the matrix transpose sign in the formula.

Can be expressed as a plurality of basic model component sums after wavelength-division is separated for measured data sequence X (t) local,

As the formula (3):

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

C in the formula _i(t) be I basic model component, each basic model component as a data sequence or only interested component is asked its AR spectrum by formula (1), and is analyzed it.

2) involve the method for diagnosing faults of K-L quantity of information based on local:

For identical two time serieses of data length: { x _t} _RBe reference sequential, { x _t} _TBe sequential to be checked, set up suitable AR (m respectively _R) model and AR (m _T) model, be respectively so can obtain the residual error of their correspondences:

With

Then sequential { x to be checked _t} _TAR (m by reference sequences _R) model testing, the residual sequence of output is designated as { a _t} _RTSo when when identical with time sequence status to be checked or close, existing with reference to time sequence status

Be white noise sequence, and have

When two sequence states do not belong to same state, then there is { a _t} _RTNot white noise, and

For reference sequential and the pairing residual error variance of sequential to be checked With

Can directly try to achieve by formula (2) according to model parameter and sequential separately, and for the residual error variance Be to try to achieve by formula (2) according to the model parameter and the sequential to be checked of reference sequential; K-L information distance function is:

$D_{\mathrm{KL}}^2(p_{\mathrm{RT}},p_R)=\∫p_{\mathrm{RT}}\left(a_t\right)\ln \frac{p_{\mathrm{RT}}\left(a_t\right)}{p_R\left(a_t\right)}{\mathrm{da}}_t---\left(4\right)$

P in the formula _RTAnd p _RBe respectively residual error { a _t} _RT{ a _t} _RProbability density function:

$p_{\mathrm{RT}}\left(a_t\right)=\frac{1}{\sqrt{2\mathrm{\π}}{\mathrm{\σ}}_T}\exp (-\frac{{\mathrm{\σ}}_{\mathrm{RT}}^2}{2{\mathrm{\σ}}_T^2})$

$p_R\left(a_t\right)=\frac{1}{\sqrt{2\mathrm{\π}}{\mathrm{\σ}}_R}\exp (-\frac{{(x_t-\mathrm{\μ})}^2}{2{\mathrm{\σ}}_R^2})$

Under one-dimensional case, consider { a _t} _RT{ a _t} _RAverage be zero, and wushu (4) integration changes discrete summation into and can get:

$D_{\mathrm{KL}}^2=\frac{1}{2}(\ln \frac{{\mathrm{\σ}}_R^2}{{\mathrm{\σ}}_T^2}+\frac{{\mathrm{\σ}}_{\mathrm{RT}}^2}{{\mathrm{\σ}}_R^2}-1)$

Consider that constant 1/2 does not influence the result by distance classification, so the K-L quantity of information can simply be represented an accepted way of doing sth (5):

$D_{\mathrm{KL}}^2=\ln \frac{{\mathrm{\σ}}_R^2}{{\mathrm{\σ}}_T^2}+\frac{{\mathrm{\σ}}_{\mathrm{RT}}^2}{{\mathrm{\σ}}_R^2}-1---\left(5\right)$

At first, reference sample and sample to be checked are carried out the local wavelength-division separate, and a plurality of basic model components after decomposing are set up different AR models as reference sequence and sequence to be checked, thereby obtain to accumulate in each the K-L quantity of information of mode component

Basis then

Value judge system mode, its value is more little, represents that then state to be tested and reference state are just approaching more, its value is big more, and to represent that then state to be checked departs from the reference state degree big more.So just can judge system failure type and fault degree according to the selection of reference state.Also can set in advance alarm threshold value δ in addition _i, when The time, implement to report to the police, thus the real time monitoring function of the system of realization.

3) local ripple neural network failure diagnostic method

The specific implementation step of this method is as follows:

The first step: set up network.Build neuron network according to actual needs.Determine input layer number of parameters, interlayer structure and output layer.

Second step: structure training sample.The vibration of the scroll compressor of typical fault and noise signal are carried out the local wavelength-division separate, thereby obtain a plurality of basic model components, extract each basic model component features relevant parameter, as amplitude, each temporal signatures parameter of energy, frequency.Output layer is for characterizing certain typical fault type space representation.

The 3rd step: training and corrective networks.According to actual conditions regulating networks structure, make it to reach the demand precision.

The 4th step: good neural network model carries out the identification of fault type to utilize foundation.

Scroll compressor on-line vibration of the present invention, noise monitoring and fault diagnosis system, on the basis of revolution type scroll compressor structure characteristic and working principle, utilizing hammering method that each of scroll compressor mainly formed component tests, obtained the frequency that has of each main parts size, for identification, the product improvement of scroll compressor vibration source and sound source are laid a good foundation, and put forward some scroll compressor vibration and noise reducing measure on this basis.

The present invention has obtained the space distribution of relation between scroll compressor surface vibration and the nearly acoustic noise characteristic and vibration thereof, noise by experiment, proved for fully-closed vortex compressor, its noise is mainly derived from the vibration of surface of shell, and according to scroll compressor vibration, noise behavior, propose to be applicable to several new method of scroll compressor fault diagnosis, effectively advanced the development of fault diagnosis technology.

The present invention is directed to the needs that compressor noise detects, online detection of four-way scroll compressor noise and analytical system have been developed, this system can show in real time to time domain, frequency domain, octave and the various weighted of the noise signal of four direction and the sound pressure level under the response, also have simultaneously identification of sound source function and testing journal sheet's output function, practical application shows, this system has satisfied the noise testing requirement, has certain practical value.

The remarkable result of scroll compressor on-line vibration of the present invention, noise monitoring and fault diagnosis system is: multichannel carries out on-line vibration, noise measurement to the scroll compressor housing, the signal relation that draws the noise of near sound field and compressor surface vibration acceleration is close, by to its The Characteristic Study, grasp causes the reason of scroll compressor vibration and noise, so that in improvement in performance, paid the utmost attention to, and make effective vibration and noise reducing measure, and can provide suitable mounting type for the client, to reduce the vibration and the noise of air-conditioning or refrigerator.

Description of drawings

The present invention has six width of cloth accompanying drawings, wherein:

Accompanying drawing 1 is scroll compressor on-line vibration, noise monitoring and fault diagnosis system schematic diagram;

Accompanying drawing 2 is scroll compressor on-line vibration, noise monitoring and fault diagnosis system path of propagation sketch;

Accompanying drawing 3 is scroll compressor on-line vibration, noise monitoring and fault diagnosis system structure stress schematic representation;

Accompanying drawing 4 is scroll compressor on-line vibration, noise monitoring and fault diagnosis system Fault Diagnosis Strategy structural drawing based on Local Wave Method and K-L quantity of information;

Accompanying drawing 5 is scroll compressor on-line vibration, noise monitoring and fault diagnosis system technology path sketch;

Accompanying drawing 6 is scroll compressor on-line vibration, noise monitoring and fault diagnosis system octave algorithm flow chart.

In the accompanying drawing: 1, scroll compressor, 2, piezoelectric accelerometer, 3, sound Mike, 4, preamplifier, 5, ABAP Adapter, 6 signal condition modules,, 7, the A/D conversion, 8, signal analysis, 9, the result shows, 10, vibratory output, 11, noisiness.

Embodiment

Specific embodiments of the invention as shown in drawings, accompanying drawing 1 is depicted as scroll compressor on-line vibration, noise monitoring and fault diagnosis system schematic diagram, and described scroll compressor on-line vibration, noise monitoring and fault diagnosis system comprise scroll compressor on-line vibration, noise-monitoring equipment and scroll compressor fault diagnosis system.Described scroll compressor on-line vibration, noise-monitoring equipment show that by scroll compressor 1, piezoelectric accelerometer 2, sound Mike 3, preamplifier 4, ABAP Adapter 5 signal condition modules 6, A/D conversion 7, signal analysis 8, result 9 are formed.Described piezoelectric accelerometer 2 is fixed on the shell of scroll compressor 1, sound Mike 3 be placed on scroll compressor 1 near, piezoelectric accelerometer 2 harmony Mikes 3 test the vibratory output 10 and the noisiness 11 of scroll compressor respectively and gather, by preamplifier 4, ABAP Adapter 5, signal condition module 6 and A/D conversion 7 analogue signal is amplified and converted digital signal to then and give computer, carry out signal analysis 8 by computer software, and the result is carried out in analysis show 9.Described scroll compressor fault diagnosis system is at 8 pairs of vibratory outputs 10 of signal analysis and noisiness 11 is analyzed and show that by the result smoothness of operation of 9 pairs of scroll compressor devices and fault judge.

Wherein signal analysis 8 and result show in 9 and are provided with:

Be used to analyze and show the software module of scroll compressor surface vibration signal;

Be used to analyze and show the software module of scroll compressor noise signal;

Be used for analysis software module to the scroll compressor fault diagnosis;

(1), the software module of scroll compressor surface vibration signal analysis:

1) signal that surface vibration is sampled to scroll compressor carries out time-domain analysis and shows its oscillogram;

2) signal that obtains of sampling carries out amplitude spectrum, power spectrum, cepstrum, resonance and demodulation spectrum analysis and shows each analysis waveform figure, obtains the frequency information and the phase information of oscillating signal.

The running of scroll compressor is by driven by motor, so power transmission and vibration transfer path relation that these mechanical vibration produce simply are expressed as shown in the accompanying drawing 2.From the scroll compressor structure, no matter how power transmit, also no matter the vibration propagation path how, the most final influence to housing is that we study the problem that vibration is concerned about most.In scroll compressor when work,, each motion component is operated in comparatively in the complex environment.Provided the stressed sketch of running part shown in the accompanying drawing 3, the active force that exists mainly contains the pulsation of air-flow body or high pressure to member force Fg, the support reaction FR of various connectors, major and minor bearing and moving vortex are to the support reaction of main shaft, the centrifugal force of major and minor equilibrium block and main shaft each several part also has each surface of contact to form the frictional force of friction pair in addition.Be in dynamic balance state during mechanical operation under the effect of above each power, the mutual alternation effect between each part forms vibration, and outwards transmits by certain path.

Because the purpose of vibration-testing is to judge whether the running of scroll compressor device is steady, analyze and solve the fault relevant with vibration etc. again, therefore, compressor dispatches from the factory and all will test its coupled vibration amount when preceding laboratory is inspected by random samples.The surveying that adopts, as shown in Figure 1, testable vibratory output has vibration acceleration speed and displacement.This module signal that surface vibration is sampled to scroll compressor carries out time-domain analysis and shows its oscillogram; And the signal that sampling obtains is carried out amplitude spectrum, power spectrum, cepstrum, resonance and demodulation spectrum analysis and shows each analysis waveform figure.

(2), the software module of scroll compressor noise signal analysis:

For the analysis of finishing the scroll compressor noise characteristic, evaluation, the identification of sound source and the putting on record of testing result of noise size,, specially formulate technology path shown in Figure 5 in conjunction with the existing measurement environment of company.

According to the demand analysis of system, the function that software systems need be finished is concluded, this compressor noise on-line monitoring system software section is divided into following module: sampling module; Digital signal processing module; The on-line analysis module; The off-line analysis module; Report display print module etc.Wherein sampling module is responsible for sending sampling or stopping sampling instruction, control sampling parameter, length etc. to the sampling hardware system; Digital signal processing module is responsible for the digital signal of sampling is carried out subsequent treatment, obtains genuine and believable signal; The on-line analysis module is used for the interfacial effect of control system; The off-line analysis module is responsible for the off-line analysis of sampled signal, comprising: time-domain analysis, frequecny domain analysis, resonance and demodulation, multiple scale analysis etc.The report display module is responsible for generted noise experiment form and prints filing.

The major function of sampling module: send sampling or stop the sampling instruction to the sampling hardware system; Regulate the control sampling parameter, and output to hardware; By serial ports and hardware device communication, gather signal data.

Digital signal processing module is an important module of this system, to the analysis of the characteristic of small-signal, takes digital signal corresponding treatment technology and digital signal filter technology that small-signal is handled, and directly has influence on the success and the failure of system.

The on-line analysis module has realized following function: 1, the signal time domain shows; 2, signal frequency-domain shows; 3, third-octave 1/1 octave of signal shows.According to the detailed algorithm of third-octave that the front is said in conjunction with the digital filter principle, being encapsulated in the dynamic link libraries dll file of the detailed algorithm of third-octave.The basic procedure of function as shown in Figure 6.

The off-line analysis module is used the function module in the original PDM2000 fault diagnosis software system, and it mainly can realize following function: 1, signal time-domain analysis; 2, signal frequency-domain analysis; 3, fault diagnosis.Selecting the analysis module submenu under the off-line analysis can enter the off-line analysis module under the main interface.This function module at first obtains and comes from the historical sample of preserving in the historical sample database; The time, can carry out various time and frequency domain analysis to this sample in the frequency analysis software object, comprise analytical methods such as integration, differential, self correlation, crosscorrelation, filtering, statistical analysis, axle center locus, auto-power spectrum, crosspower spectrum, resonance and demodulation, cepstral analysis, refinement spectrum analysis, coherence analysis, the analysis of biography letter, octave.In fault diagnosis software, utilize the analytical method in " time, frequency analysis " software, the state of compressor is assessed; Import diagnosis at last, form deagnostic report.

According to customer requirements, the report display print module should realize that several data shows in many ways, and native system uses crystal report Report Designer Component (RDC) and Visual Basic 6.0 to carry out the exploitation of Reports module jointly.Crystal report makes the very powerful and developing instrument efficiently of form exploitation, exactly because the mature technology of crystal report, Report Designer Component (RDC) embeds VisualBasic 6.0, make the programing work of form become simple, and multiple method of calling can be arranged between VB and the VC, make the form development time of system significantly reduce.

(3), be used for its main contents of analysis software module to the scroll compressor fault diagnosis: the method for diagnosing faults of this module is several different methods such as the analysis of local ripple autoregressive spectrum, local ripple K-L quantity of information analysis and local ripple analysis of neural network.

1) local ripple autoregressive spectrum analytical method

Formula (1) be AR (n) model from spectral function:

In the formula

Be the residual error variance, Δ t is the sampling interval.

After correctly setting up the AR model, can utilize following method to find the solution the residual error variance: establishing sequential length is N, and model order is n, and model parameter is φ ₁, φ ₂..., φ _n, the residual error variance of model then

Can be expressed as formula (2):

${\mathrm{\σ}}_a^2=\frac{N-n}{N}{\left(\mathrm{X\Φ}\right)}^T\mathrm{X\Φ}---\left(2\right)$

In the formula

$X=\left[\begin{array}{c}{x}_{n+1},x_n,...,x_1\\ x_{n+2},x_{n+1},...,x_2\\ ......\\ x_N,x_{N-1},...,x_{N-n}\end{array}\right]$

T is the matrix transpose sign in the formula.

Can be expressed as a plurality of basic model component sums after wavelength-division is separated for measured data sequence X (t) local, as the formula (8):

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

C in the formula _i(t) be I basic model component, each basic model component as a data sequence or only interested component is asked its AR spectrum by formula (1), and is analyzed it.

2) involve the method for diagnosing faults of K-L quantity of information based on local

For identical two time serieses of data length: { x _t} _RBe reference sequential, { x _t} _TBe sequential to be checked, set up suitable AR (m respectively _R) model and AR (m _T) model, be respectively so can obtain the residual error of their correspondences: With

Then sequential { x to be checked _t} _TAR (m by reference sequences _R) model testing, the residual sequence of output is designated as { a _t} _RTSo when when identical with time sequence status to be checked or close, existing with reference to time sequence status

Be white noise sequence, and have When two sequence states do not belong to same state, then there is { a _t} _RTNot white noise, and

For reference sequential and the pairing residual error variance of sequential to be checked

With Can directly try to achieve by formula (2) according to model parameter and sequential separately, and for the residual error variance Be to try to achieve by formula (2) according to the model parameter and the sequential to be checked of reference sequential.

K-L information distance function is:

$D_{\mathrm{KL}}^2(p_{\mathrm{RT}},p_R)=\∫p_{\mathrm{RT}}\left(a_t\right)\ln \frac{p_{\mathrm{RT}}\left(a_t\right)}{p_R\left(a_t\right)}{\mathrm{da}}_t---\left(4\right)$

P in the formula _RTAnd p _RBe respectively residual error { a _t} _RT{ a _t} _RProbability density function:

$p_{\mathrm{RT}}\left(a_t\right)=\frac{1}{\sqrt{2\mathrm{\π}}{\mathrm{\σ}}_T}\exp (-\frac{{\mathrm{\σ}}_{\mathrm{RT}}^2}{2{\mathrm{\σ}}_T^2})$

$p_R\left(a_t\right)=\frac{1}{\sqrt{2\mathrm{\π}}{\mathrm{\σ}}_R}\exp (-\frac{{(x_t-\mathrm{\μ})}^2}{2{\mathrm{\σ}}_R^2})$

Under one-dimensional case, consider { a _t} _RT{ a _t} _RAverage be zero, and wushu (4) integration changes discrete summation into and can get:

$D_{\mathrm{KL}}^2=\frac{1}{2}(\ln \frac{{\mathrm{\σ}}_R^2}{{\mathrm{\σ}}_T^2}+\frac{{\mathrm{\σ}}_{\mathrm{RT}}^2}{{\mathrm{\σ}}_R^2}-1)$

Consider that constant 1/2 does not influence the result by distance classification, so the K-L quantity of information can simply be represented an accepted way of doing sth (5):

$D_{\mathrm{KL}}^2=\ln \frac{{\mathrm{\σ}}_R^2}{{\mathrm{\σ}}_T^2}+\frac{{\mathrm{\σ}}_{\mathrm{RT}}^2}{{\mathrm{\σ}}_R^2}-1---\left(5\right)$

At first, reference sample and sample to be checked are carried out the local wavelength-division separate, and a plurality of basic model components after decomposing are set up different AR models as reference sequence and sequence to be checked, thereby obtain to accumulate in each the K-L quantity of information of mode component

Basis then

Value judge system mode, its value is more little, represents that then state to be tested and reference state are just approaching more, its value is big more, and to represent that then state to be checked departs from the reference state degree big more.So just can judge system failure type and fault degree according to the selection of reference state.Also can set in advance alarm threshold value δ in addition _i, when

The time, implement to report to the police, thus the real time monitoring function of the system of realization.The Fault Diagnosis Strategy that adopts as shown in Figure 4.

3) local ripple neural network failure diagnostic method:

The specific implementation step of this method is as follows:

The first step: set up network.Build neuron network according to actual needs.Determine input layer number of parameters, interlayer structure and output layer (representative fault type).

Second step: structure training sample.The vibration of the scroll compressor of typical fault and noise signal are carried out the local wavelength-division separate, thereby obtain a plurality of basic model components, extract each basic model component features relevant parameter, as the amplitude of energy, a certain frequency, each temporal signatures parameter etc.Output layer is for characterizing certain typical fault type space representation.

The 3rd step: training and corrective networks.According to actual conditions regulating networks structure, make it to reach the demand precision.

The 4th step: good neural network model carries out the identification of fault type to utilize foundation.

Claims (2)

1. a scroll compressor on-line vibration, noise monitoring and fault diagnosis system is characterized in that comprising scroll compressor on-line vibration, noise-monitoring equipment and scroll compressor fault diagnosis system; Described scroll compressor on-line vibration, noise-monitoring equipment show that by scroll compressor (1), piezoelectric accelerometer (2), sound Mike (3), preamplifier (4), ABAP Adapter (5) signal condition module (6), A/D conversion (7), signal analysis (8), result (9) are formed; Described piezoelectric accelerometer (2) is fixed on the shell of scroll compressor (1), sound Mike (3) be placed on scroll compressor (1) near, piezoelectric accelerometer (2) harmony Mike (3) tests the vibratory output (10) and the noisiness (11) of scroll compressor respectively and gathers, by preamplifier (4), ABAP Adapter (5), signal condition module (6) and A/D conversion (7) analogue signal is amplified and converted digital signal to then and give computer, carry out signal analysis (8) by computer software, and result's demonstration (9) is carried out in analysis; Described scroll compressor fault diagnosis system is to vibratory output (10) and noisiness (11) is analyzed and show that by the result (9) judge the smoothness of operation and the fault of scroll compressor device in signal analysis (8).

2. scroll compressor on-line vibration according to claim 1, noise monitoring and fault diagnosis system is characterized in that being provided with in described signal analysis (8) and the result demonstration (9):

Be used to analyze and show the software module of scroll compressor surface vibration signal;

Be used to analyze and show the software module of scroll compressor noise signal;

Be used for analysis software module to the scroll compressor fault diagnosis;

(1), wherein the software module of scroll compressor surface vibration signal analysis has:

1) signal that surface vibration is sampled to scroll compressor carries out time-domain analysis and shows its oscillogram;

2) signal that obtains of sampling carries out amplitude spectrum, power spectrum, cepstrum, resonance and demodulation spectrum analysis and shows each analysis waveform figure, obtains the frequency information and the phase information of oscillating signal;

(2), the software module of scroll compressor noise signal analysis has:

1) the scroll compressor noise signal that sampling is obtained carries out time-domain analysis and carries out oscillogram showing;

2) noise signal is further analyzed, power spectrum, refinement spectrum, resonance and demodulation spectrum, third-octave analysis are arranged and show each analysis waveform figure; And demonstrate sound pressure level, sound power level or sound intensity level under the A weighted;

3) generation of form; The experiment Reporting Requirements of system is printed the compressor parameter information simultaneously and is tested the octave data information that obtains, and data form and figure show jointly;

(3), be used for its main contents of analysis software module to the scroll compressor fault diagnosis: the method for diagnosing faults of this module is the analysis of local ripple autoregressive spectrum, local ripple K-L quantity of information analysis and local ripple analysis of neural network method:

1) local ripple autoregressive spectrum analytical method

Formula (1) be AR (n) model from spectral function:

In the formula

Be the residual error variance, Δ t is the sampling interval;

After correctly setting up the AR model, can utilize following method to find the solution the residual error variance: establishing sequential length is N, and model order is n, and model parameter is φ ₁, φ ₂..., φ _n, the residual error variance of model then Can be expressed as formula (2):

${\mathrm{\σ}}_a^2=\frac{N-n}{N}{\left(\mathrm{X\Φ}\right)}^T\mathrm{X\Φ}---\left(2\right)$

In the formula

$X=\left[\begin{array}{c}{x}_{n+1},x_n,...,x_1\\ x_{n+2},x_{n+1},...,x_2\\ ......\\ x_N,x_{N-1},...,x_{N-n}\end{array}\right]$

T is the matrix transpose sign in the formula;

Can be expressed as a plurality of basic model component sums after wavelength-division is separated for measured data sequence X (t) local,

As the formula (3):

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

C in the formula _i(t) be I basic model component, each basic model component as a data sequence or only interested component is asked its AR spectrum by formula (1), and is analyzed it;

2) involve the method for diagnosing faults of K-L quantity of information based on local:

For identical two time serieses of data length: { x _t} _RBe reference sequential, { x _t} _TBe sequential to be checked, set up suitable AR (m respectively _R) model and AR (m _T) model, be respectively so can obtain the residual error of their correspondences:

With

Then sequential { x to be checked _t} _TAR (m by reference sequences _R) model testing, the residual sequence of output is designated as { a _t} _RTSo when when identical with time sequence status to be checked or close, existing with reference to time sequence status

Be white noise sequence, and have

When two sequence states do not belong to same state, then there is { a _t} _TRNot white noise, and For reference sequential and the pairing residual error variance of sequential to be checked

With

Can directly try to achieve by formula (2) according to model parameter and sequential separately, and for the residual error variance

Be to try to achieve by formula (2) according to the model parameter and the sequential to be checked of reference sequential;

K-L information distance function is:

$D_{\mathrm{KL}}^2(p_{\mathrm{RT}},p_R)=\∫p_{\mathrm{RT}}\left(a_t\right)\ln \frac{p_{\mathrm{RT}}\left(a_t\right)}{p_R\left(a_t\right)}{\mathrm{da}}_t---\left(4\right)$

P in the formula _RTAnd p _RBe respectively residual error { a _t} _RT{ a _t} _RProbability density function:

$p_{\mathrm{RT}}\left(a_t\right)=\frac{1}{\sqrt{2\mathrm{\π}}{\mathrm{\σ}}_T}\exp (-\frac{{\mathrm{\σ}}_{\mathrm{RT}}^2}{2{\mathrm{\σ}}_T^2})$

$p_R\left(a_t\right)=\frac{1}{\sqrt{2\mathrm{\π}}{\mathrm{\σ}}_R}\exp (-\frac{{(x_t-\mathrm{\μ})}^2}{2{\mathrm{\σ}}_R^2})$

Under one-dimensional case, consider { a _t} _RT{ a _t} _RAverage be zero, and wushu (4) integration changes discrete summation into and can get:

$D_{\mathrm{KL}}^2=\frac{1}{2}(\ln \frac{{\mathrm{\σ}}_R^2}{{\mathrm{\σ}}_T^2}+\frac{{\mathrm{\σ}}_{\mathrm{RT}}^2}{{\mathrm{\σ}}_R^2}-1)$

Consider that constant 1/2 does not influence the result by distance classification, so the K-L quantity of information can simply be represented an accepted way of doing sth (5):

$D_{\mathrm{KL}}^2=\ln \frac{{\mathrm{\σ}}_R^2}{{\mathrm{\σ}}_T^2}+\frac{{\mathrm{\σ}}_{\mathrm{RT}}^2}{{\mathrm{\σ}}_R^2}-1---\left(5\right)$

At first, reference sample and sample to be checked are carried out the local wavelength-division separate, and a plurality of basic model components after decomposing are set up different AR models as reference sequence and sequence to be checked, thereby obtain to accumulate in each the K-L quantity of information of mode component

Basis then

Value judge system mode, its value is more little, represents that then state to be tested and reference state are just approaching more, its value is big more, and to represent that then state to be checked departs from the reference state degree big more; So just can judge system failure type and fault degree according to the selection of reference state; Also can set in advance alarm threshold value δ in addition _i, when

The time, implement to report to the police, thus the real time monitoring function of the system of realization;

3) local ripple neural network failure diagnostic method

The specific implementation step of this method is as follows:

The first step: set up network.Build neuron network according to actual needs.Determine input layer number of parameters, interlayer structure and output layer;

Second step: structure training sample.The vibration of the scroll compressor of typical fault and noise signal are carried out the local wavelength-division separate, thereby obtain a plurality of basic model components, extract each basic model component features relevant parameter, as amplitude, each temporal signatures parameter of energy, frequency.Output layer is for characterizing certain typical fault type space representation;

The 3rd step: training and corrective networks.According to actual conditions regulating networks structure, make it to reach the demand precision;

The 4th step: good neural network model carries out the identification of fault type to utilize foundation.

Images