CN104635146A - Analog circuit fault diagnosis method based on random sinusoidal signal test and HMM (Hidden Markov Model) - Google Patents

Abstract

The invention discloses an analog circuit fault diagnosis method based on a random sinusoidal signal test and an HMM (Hidden Markov Model). The analog circuit fault diagnosis method comprises the following steps: A. stimulating a to-be-measured analog circuit by adopting a random sinusoidal signal, wherein the amplitude, phase and frequency of the random sinusoidal signal are random variables which satisfy Gaussian distribution; B. collecting output data samples of the to-be-measured analog circuit, and extracting time domain features and spectrum features of the data samples, so as to form feature components; C. respectively inputting an HMM diagnosis system with a random time sequence to each type feature component, and fusing a plurality of diagnosis results by adopting an ECOC (Error Correcting Output Codes) method, so as to realize fault diagnosis. According to the method, the random sinusoidal signal is used as test stimulation of the analog circuit, so that the frequency components of the output samples can be increased, and the superposition of a fuzzy fault group is reduced; a time sequence analysis method of the HMM is combined, so that the diagnosis precision of analog circuit fuzzy faults can be increased.

Description

Based on the analog-circuit fault diagnosis method of the test of random sinusoidal signal and HMM

Technical field

The present invention relates to a kind of analog-circuit fault diagnosis method, especially a kind of analog-circuit fault diagnosis method based on the test of random sinusoidal signal and the Hidden Markov Model (HMM) based on time series analysis.

Background technology

Along with electronic equipment develops towards intelligent and flexibility, its complicacy and from strength to strength functional.In complex electronic equipment design and test, reliability consideration accounts for extremely important status.Electronic equipment has penetrated into the every field of society, and wherein mimic channel is pith indispensable in electronic equipment.But due to analog circuit fault state complex, failure symptom is fuzzy, test node is limited, fault element value has tolerance, circuit is the problems such as nonlinear effect, analog circuit fault diagnosing job very difficult.

Can be divided into according to the order of emulation in test process: emulation and survey post-simulation before surveying.Can be divided into according to methodology: fault dictionary method, failure proof method, Parameter Identification, artificial intelligence diagnosis's method, information fusion method and expert system approach etc.Wherein, fault dictionary method and artificial intelligence method belong to surveys front simulation method, and information fusion method and expert system approach belong to mixed method, and other method belongs to surveys post-simulation method.In recent years, the achievement in research that failure proof method, Parameter Identification are relevant with expert system approach was less, and information fusion method forms diagnostic system in conjunction with artificial intelligence method usually, and fault dictionary method, information fusion method and artificial intelligence method are current study hotspots.

Existing analog-circuit fault diagnosis method is all based on deterministic signal test, and namely the test signal of circuit under test can represent by clear and definite funtcional relationship.For this kind of method of testing, there is the deficiency of following two aspects.The first, deterministic signal comprises one or more parameter.In order to obtain concrete test signal, need the parameter determining test signal, the final test signal adopted has uncertainty, and test signal can affect the output response of circuit, directly affects diagnostic result.The second, the frequency component that deterministic signal contains is limited.In analog circuit fault test and diagnosis, the frequency component that test signal comprises is more, and the quantity of information that circuit output respond packet contains is abundanter, and this will improve the examining property of fault signature greatly.

Summary of the invention

The object of the invention is to overcome traditional unknown parameter based on existing in deterministic signal test and the limited shortcoming of frequency component, reduce the plyability of fuzzy fault group, improve the examining property of fault sample, propose a kind of analog-circuit fault diagnosis method based on random sinusoidal signal test and HMM (Hidden Markov Model (HMM)).

Compared to existing technologies, following technical scheme is adopted:

A. random sinusoidal signal X (t)={ x is adopted ₁(t), x ₂(t) ..., x _n(t) } encourage mimic channel to be measured, x _nt the amplitude of (), phase place and frequency meet Gaussian distribution;

B. output data sample Y (t)={ y of mimic channel to be measured is gathered ₁(t), y ₂(t) ..., y _n(t) }, extract the temporal signatures and spectrum signature constitutive characteristic component that export data sample, every category feature component is all time series; Wherein temporal signatures component is mathematical expectation m _y(t), variance coefficient R _y(τ), spectrum signature component is power spectrum S _y(ω);

C1. four category feature components are inputted four HMM as four class time serieses, construct four Hidden Markov diagnostic models, four Hidden Markov diagnostic models are all diagnosed test data, and each test data obtains four diagnostic results:

C101. the mathematical expectation m of output sample will be obtained in step B _yt (), as sequence input time, training HMM diagnostic model one, obtains diagnostic result one;

C102. the variance of output sample will be obtained in step B as sequence input time, training HMM diagnostic model two, obtains diagnostic result two;

C103. the coefficient R of output sample will be obtained in step B _y(τ) as sequence input time, training HMM diagnostic model three, obtains diagnostic result three;

C104. the power spectrum S of output sample will be obtained in step B _y(ω) as sequence input time, training HMM diagnostic model four, obtains diagnostic result four;

C2. adopt ECOC error correcting output codes method to merge four diagnostic results, realize the final judgement of fault: construct a sparse matrix by four diagnostic results obtained in step C1, and encode according to ECOC theory, obtain codeword vector; According to D-S evidence theory, codeword vector is decoded, obtain the diagnostic result merged by four Hidden Markov diagnostic models.

In described steps A, random sinusoidal signal X (t) obtains by the following method:

A1. random sinusoidal signal X (t) meets X (t)=A (t) cos [Ω (t) t+ Φ (t)], wherein, A (t), Ω (t) and Φ (t) are respectively the amplitude stochastic variable, the phase place random sum frequency accidental variable that meet Gaussian distribution;

A2. produce the sample of the random sinusoidal signal of n group, be designated as X (t)={ x ₁(t), x ₂(t) ..., x _n(t) }, respectively by each sample signal x _nt () encourages mimic channel to be measured.

Mathematical expectation m described in described step B _y(t), variance coefficient R _y(τ), power spectrum S _y(ω) be theoretical based on random signal analysis, adopt mathematical statistics method to obtain:

$m_Y\left(t\right)=E\left[Y\left(t\right)\right]={\∫}_{-\∞}^{\∞}{\mathrm{yf}}_Y(y,t)\mathrm{dy}$

${\mathrm{\σ}}_Y^2\left(t\right)=D\left[Y\left(t\right)\right]={\∫}_{-\∞}^{\∞}{(y-m_Y\left(t\right))}^2{f}_Y(y,t)\mathrm{dy}$

$R_Y\left(\mathrm{\τ}\right)=E\left[Y\left(t_1\right)Y\left(t_2\right)\right]={\∫}_{-\∞}^{\∞}{\∫}_{-\∞}^{\∞}{y}_1{y}_2{f}_Y(y_1,y_2,t_1,t_2){\mathrm{dy}}_1{\mathrm{dy}}_2$

$S_Y\left(\mathrm{\ω}\right)={\∫}_{-\∞}^{\∞}{R}_Y\left(\mathrm{\τ}\right)e^{-\mathrm{j\ω\τ}}\mathrm{d\τ}$

Wherein Y (t) is stochastic process, f _ythe One-dimensional probability function that (y, t) is Y (t), f _y(y ₁, y ₂, t ₁, t ₂) be two probability density functions of Y (t), Y (t ₁) and Y (t ₂) be at t respectively ₁and t ₂the stochastic variable that moment observation Y (t) obtains, τ is t ₁and t ₂interval between moment.

During structure Hidden Markov diagnostic model, Hidden Markov diagnostic model can describe (Ω by 5 tuples _n, Ω ₀, A, π, O), wherein Ω _nthe finite aggregate of state, Ω _nvalue is M, M is the number of faults that mimic channel to be measured needs diagnosis; Ω ₀represent the finite aggregate of observed reading, Ω in 4 HMM models ₀value be mathematical expectation m respectively _y(t), variance coefficient R _y(τ), power spectrum S _y(ω) time series; A represents the transition probability matrix that state one step shifts, and transition probability matrix A produces at random in simulated program; π is initial state distribution probability, and initial state distribution probability π is set to [1,0,0 ..0], namely thinks when original state, and the probability of circuit normal condition is 1, and the probability broken down is 0; O represents the probability of the observation of sequence, adopts forward-backward algorithm to calculate.

Beneficial effect of the present invention:

(1) adopt random sinusoidal signal as test signal, add the frequency component of fault sample, the examining property of fault sample can be improved.

(2) because the signal of test is random sinusoidal signal, the output response of circuit is a random function of time, extraction temporal signatures and spectrum signature are as characteristic component, and fault signature comprises temporal signatures and spectrum signature simultaneously, adds the Information Meter of fault sample feature.

(3) because test signal is Random time sequence, construct the Hidden Markov Model (HMM) based on time series analysis, and adopt the ECOC coding method of sparse random array to merge diagnostic result, the precision of fault diagnosis can be improved.

Accompanying drawing explanation

Fig. 1 is FB(flow block) of the present invention.

Fig. 2 is the diagnostic system figure that in the inventive method, many HMM merge.

Embodiment

Below in conjunction with embodiment, the invention will be further described, but protection scope of the present invention is not limited thereto:

Embodiment 1, composition graphs 1, based on the analog-circuit fault diagnosis method of the test of random sinusoidal signal and HMM, comprises following steps:

A. random sinusoidal signal X (t)={ x is adopted ₁(t), x ₂(t) ..., x _n(t) } encourage mimic channel to be measured, x _nt the amplitude of (), phase place and frequency meet Gaussian distribution;

B. output data sample Y (t)={ y of mimic channel to be measured is gathered ₁(t), y ₂(t) ..., y _n(t) }, extract the temporal signatures and spectrum signature constitutive characteristic component that export data sample, every category feature component is all time series; Wherein temporal signatures component is mathematical expectation m _y(t), variance coefficient R _y(τ), spectrum signature component is power spectrum S _y(ω);

Four category feature components are inputted four HMM as four class time serieses by C1. composition graphs 2, and construct four Hidden Markov diagnostic models, four Hidden Markov diagnostic models are all diagnosed test data, and each test data obtains four diagnostic results:

C101. the mathematical expectation m of output sample will be obtained in step B _yt (), as sequence input time, training HMM diagnostic model one, obtains diagnostic result one;

C102. the variance of output sample will be obtained in step B as sequence input time, training HMM diagnostic model two, obtains diagnostic result two;

C103. the coefficient R of output sample will be obtained in step B _y(τ) as sequence input time, training HMM diagnostic model three, obtains diagnostic result three;

C104. the power spectrum S of output sample will be obtained in step B _y(ω) as sequence input time, training HMM diagnostic model four, obtains diagnostic result four;

C2. adopt ECOC error correcting output codes method to merge four diagnostic results, realize the final judgement of fault: construct a sparse matrix by four diagnostic results obtained in step C1, and encode according to ECOC theory, obtain codeword vector; According to D-S evidence theory, codeword vector is decoded, obtain the diagnostic result merged by four Hidden Markov diagnostic models.

Embodiment 2, as the analog-circuit fault diagnosis method based on the test of random sinusoidal signal and HMM of embodiment 1, in steps A, random sinusoidal signal X (t) obtains by the following method:

A1. random sinusoidal signal X (t) meets X (t)=A (t) cos [Ω (t) t+ Φ (t)], wherein, A (t), Ω (t) and Φ (t) are respectively the amplitude stochastic variable, the phase place random sum frequency accidental variable that meet Gaussian distribution;

A2. produce the sample of the random sinusoidal signal of n group with simulation software, be designated as X (t)={ x ₁(t), x ₂(t) ..., x _n(t) }, respectively by each sample signal x _nt () encourages mimic channel to be measured.

Embodiment 3, as the analog-circuit fault diagnosis method based on the test of random sinusoidal signal and HMM of embodiment 1 or 2, mathematical expectation m described in step B _y(t), variance coefficient R _y(τ), power spectrum S _y(ω) be theoretical based on random signal analysis, adopt mathematical statistics method to obtain:

$m_Y\left(t\right)=E\left[Y\left(t\right)\right]={\∫}_{-\∞}^{\∞}{\mathrm{yf}}_Y(y,t)\mathrm{dy}$

${\mathrm{\σ}}_Y^2\left(t\right)=D\left[Y\left(t\right)\right]={\∫}_{-\∞}^{\∞}{(y-m_Y\left(t\right))}^2{f}_Y(y,t)\mathrm{dy}$

$R_Y\left(\mathrm{\τ}\right)=E\left[Y\left(t_1\right)Y\left(t_2\right)\right]={\∫}_{-\∞}^{\∞}{\∫}_{-\∞}^{\∞}{y}_1{y}_2{f}_Y(y_1,y_2,t_1,t_2){\mathrm{dy}}_1{\mathrm{dy}}_2$

$S_Y\left(\mathrm{\ω}\right)={\∫}_{-\∞}^{\∞}{R}_Y\left(\mathrm{\τ}\right)e^{-\mathrm{j\ω\τ}}\mathrm{d\τ}$

Wherein Y (t) is stochastic process, f _ythe One-dimensional probability function that (y, t) is Y (t), f _y(y ₁, y ₂, t ₁, t ₂) be two probability density functions of Y (t), Y (t ₁) and Y (t ₂) be at t respectively ₁and t ₂the stochastic variable that moment observation Y (t) obtains, τ is t ₁and t ₂interval between moment.

In embodiment 1 ~ 3, the hidden Markov model based on Random time sequence performs by the following method:

First, 4 HMM diagnostic models are trained with 4 kinds of characteristic sequences.HMM can describe (Ω by 5 tuples _n, Ω ₀, A, π, O), wherein Ω _nthe finite aggregate of state, Ω ₀represent the finite aggregate of observed reading, A represents the transition probability matrix that state one step shifts, and π is initial state distribution probability, and O represents the probability of the observation of sequence.The present invention constructs 4 HMM diagnostic models, wherein Ω _nvalue is M, M is the number of faults that mimic channel to be measured needs diagnosis.Ω in 4 HMM models ₀value be mathematical expectation m respectively _y(t), variance coefficient R _y(τ), power spectrum S _y(ω) time series.Transition probability matrix A produces at random in simulated program.It has been generally acknowledged that circuit to be diagnosed is normal condition, therefore, initial state distribution probability π is set to [1,0,0 ..0], namely thinks when original state, and the probability of circuit normal condition is 1, and the probability broken down is 0.Observation probability O adopts forward-backward algorithm to calculate.

Then, diagnostic test data, merge diagnostic result.Gather the test data of faulty circuit, calculate the mathematical expectation m of test data _t(t), variances sigma _t(t), coefficient R _t(t), power spectrum S _tt (), input 4 HMM models respectively, the input value that the observation probability value [O1, O2, O3, O4] obtaining 4 characteristic sequences is encoded as ECOC, carries out Error Correction of Coding, and adopts D-S evidence theory to decode, obtain final diagnostic result.

The HMM training algorithm used in embodiment, ECOC error correcting output codes method, D-S evidence theory are prior art, particular content can list of references (< random signal analysis (the 2nd edition) >, Zhao Shuqing, Electronic Industry Press; " a kind of error correcting code SVM diagnostic method of mimic channel ", Hu Yingcen, computer-aided design (CAD) and graphics journal, 2011; " analog circuit fault diagnosing based on SVDD and D-S theory ", Tang Jingyuan, observation and control technology, 2008).

Specific embodiment described herein is only illustrate spirit of the present invention.Those skilled in the art can make various amendment or supplement or adopt similar mode to substitute to described specific embodiment, but can't depart from spirit of the present invention or surmount the scope that appended claims defines.

Claims (4)

1., based on the analog-circuit fault diagnosis method of the test of random sinusoidal signal and HMM, it is characterized in that comprising following steps:

A. random sinusoidal signal X (t)={ x is adopted ₁(t), x ₂(t) ..., x _n(t) } encourage mimic channel to be measured, x _nt the amplitude of (), phase place and frequency meet Gaussian distribution;

B. output data sample Y (t)={ y of mimic channel to be measured is gathered ₁(t), y ₂(t) ..., y _n(t) }, extract the temporal signatures and spectrum signature constitutive characteristic component that export data sample, every category feature component is all time series; Wherein temporal signatures component is mathematical expectation m _y(t), variance coefficient R _y(τ), spectrum signature component is power spectrum S _y(ω);

C1. four category feature components are inputted four HMM as four class time serieses, construct four Hidden Markov diagnostic models, four Hidden Markov diagnostic models are all diagnosed test data, and each test data obtains four diagnostic results:

C101. the mathematical expectation m of output sample will be obtained in step B _yt (), as sequence input time, training HMM diagnostic model one, obtains diagnostic result one;

C102. the variance of output sample will be obtained in step B as sequence input time, training HMM diagnostic model two, obtains diagnostic result two;

C103. the coefficient R of output sample will be obtained in step B _y(τ) as sequence input time, training HMM diagnostic model three, obtains diagnostic result three;

C104. the power spectrum S of output sample will be obtained in step B _y(ω) as sequence input time, training HMM diagnostic model four, obtains diagnostic result four;

C2. adopt ECOC error correcting output codes method to merge four diagnostic results, realize the final judgement of fault: construct a sparse matrix by four diagnostic results obtained in step C1, and encode according to ECOC theory, obtain codeword vector; According to D-S evidence theory, codeword vector is decoded, obtain the diagnostic result merged by four Hidden Markov diagnostic models.

2. a kind of analog-circuit fault diagnosis method based on the test of random sinusoidal signal and HMM according to claim 1, is characterized in that in steps A, random sinusoidal signal X (t) obtains by the following method:

A1. random sinusoidal signal X (t) meets X (t)=A (t) cos [Ω (t) t+ Φ (t)], wherein, A (t), Ω (t) and Φ (t) are respectively the amplitude stochastic variable, the phase place random sum frequency accidental variable that meet Gaussian distribution;

A2. produce the sample of the random sinusoidal signal of n group with simulation software, be designated as X (t)={ x ₁(t), x ₂(t) ..., x _n(t) }, respectively by each sample signal x _nt () encourages mimic channel to be measured.

3. a kind of analog-circuit fault diagnosis method based on the test of random sinusoidal signal and HMM according to claim 1, is characterized in that mathematical expectation m described in step B _y(t), variance coefficient R _y(τ), power spectrum S _y(ω) be theoretical based on random signal analysis, adopt mathematical statistics method to obtain:

$m_Y\left(t\right)=E\left[Y\left(t\right)\right]={\&Integral;}_{-\&infin;}^{\&infin;}y{f}_Y(y,t)\mathrm{dy}$

${\mathrm{\&sigma;}}_Y^2\left(t\right)=D\left[Y\left(t\right)\right]={\&Integral;}_{-\&infin;}^{\&infin;}{(y-m_Y\left(t\right))}^2{f}_Y(y,t)\mathrm{dy}$

$R_Y\left(\mathrm{\&tau;}\right)=E\left[Y\left(t_1\right)Y\left(t_2\right)\right]={\&Integral;}_{-\&infin;}^{\&infin;}{\&Integral;}_{-\&infin;}^{\&infin;}{y}_1{y}_2{f}_Y(y_1,y_2,t_1,t_2){\mathrm{dy}}_1{\mathrm{dy}}_2$

$S_Y\left(\mathrm{\&omega;}\right)={\&Integral;}_{-\&infin;}^{\&infin;}{R}_Y\left(\mathrm{\&tau;}\right)e^{-\mathrm{j\&omega;\&tau;}}\mathrm{d\&tau;}$

Wherein Y (t) is stochastic process, f _ythe One-dimensional probability function that (y, t) is Y (t), f _y(y ₁, y ₂, t ₁, t ₂) be two probability density functions of Y (t), Y (t ₁) and Y (t ₂) be at t respectively ₁and t ₂the stochastic variable that moment observation Y (t) obtains, τ is t ₁and t ₂interval between moment.

4. according to claims 1 to 3 any one is based on the analog-circuit fault diagnosis method of the test of random sinusoidal signal and HMM, and when it is characterized in that constructing Hidden Markov diagnostic model, Hidden Markov diagnostic model can with 5 tuple Ω _n, Ω ₀, A, π, O describe, wherein Ω _nthe finite aggregate of state, Ω _nvalue is M, M is the number of faults that mimic channel to be measured needs diagnosis; Ω ₀represent the finite aggregate of observed reading, Ω in 4 HMM models ₀value be mathematical expectation m respectively _y(t), variances sigma _y ²(t), coefficient R _y(τ), power spectrum S _y(ω) time series; A represents the transition probability matrix that state one step shifts, and transition probability matrix A produces at random in simulated program; π is initial state distribution probability, and initial state distribution probability π is set to [1,0,0 ..0], namely thinks when original state, and the probability of circuit normal condition is 1, and the probability broken down is 0; O represents the probability of the observation of sequence, adopts forward-backward algorithm to calculate.