Analog circuit fault diagnosis method using single testable node

Abstract

An analog circuit fault diagnosis method using a single testable node comprises the following steps: (1) obtaining prior sample data vectors under each fault mode; (2) computing a statistical average of the prior sample data vectors under each of the fault modes; (3) decomposing a signal by an orthogonal Haar wavelet filter set; (4) extracting the feature factor of the prior sample fault modes; (5) extracting a fault-mode-to-be-tested feature factor; (6) computing a correlation coefficient matrix and correlation metric parameters between the feature factor of the prior sample fault modes and the feature factor of the fault-mode-to-be-tested; and (7) determining a fault mode according to a maximal correlation principle by comparing the correlation metric parameters. The method can convert a single signal into a plurality of signals without losing original measurement information, and extract an independent fault mode feature factor reflecting variations of a circuit structure in different fault modes, can be used to study an associated mode determination rule and successfully complete classification of circuit fault modes.

Description

CROSS-REFERENCE TO RELATED APPLICATION

This application is a 371 application of the International PCT application serial no. PCT/CN2015/095475, filed on Nov. 25, 2015, which claims the priority benefit of China application no. 201510482994.X, filed on Aug. 7, 2015. The entirety of each of the abovementioned patent applications is hereby incorporated by reference herein and made a part of this specification.

TECHNICAL FIELD

The present invention relates to an analog circuit fault diagnosis method using a single testable node.

BACKGROUND

Analog circuit fault diagnosis is equivalent to mode recognition in essence, and the key is to seek a relationship between feature extraction and mode criterion function. Therefore, how to seek potential fault feature factors from seemingly complicated test data and conduct correct judgment and recognition on the fault mode on this basis is deemed as a significant research subject in the field of analog circuit testing.

Through development of the analog circuit fault diagnosis over the past decade, the achievement which has been achieved is varied and shows diversity, and the new research results continually come out. To summarize the technologies adopted in fault diagnosis, a fault feature extracting method based on statistical theory and wavelet analysis, and a fault mode recognition method based on neural network and support vector machine are widely applied at present. These aspects have great impact on promoting the development of the analog circuit fault diagnosis technologies. However, the analog circuit fault diagnosis technologies are still in development so far due to the large knowledge scope of analog circuit fault diagnosis design, deficiency of fault models and the binding character of the methods thereof.

Because the measurable nodes of the analog circuit are limited, and only one node at an output terminal serves as the measurable node under most circumstances. The test data collected under this case are usually mixed data of various independent sources, and the feature factors thereof are implied deeply. When diagnosing these circuits, a very expensive amount of computations for classification is caused usually if the original data collected is directly fed to a classifier for classification, which is difficult to implement, and also has poor classification effect, and high erroneous judgment ratio. Independent factors from system bottom are not single; moreover, these are invisible independent sources (i.e., blind sources) for a ten final user. A technology regarding blind source processing at current is mainly applied to the field of voice recognition, and its application prerequisite is multi-channel measured data source. Therefore, extracting the feature factors of the circuit measured with using a single testable node cannot be directly implemented by a blind source separation technology.

SUMMARY

A technical problem to be solved by the present invention is to provide an analog circuit fault diagnosis method using a single testable node with low misjudgment ratio so as to overcome the above defects of the prior art.

The essence of the analog circuit fault diagnosis method using a single testable node according to the present invention is to convert a single testable node signal into a plurality of signals by using an orthogonal wavelet analysis filter set, so as to extract fault features via using a blind source separation technology, and estimate a fault-mode-to-be-tested of a specific circuit node based on correlation metric parameters by computing a correlation coefficient matrix between the specific node independent fault-mode-to-be-tested feature of the circuit and a known independent sample fault feature. Known from statistics knowledge, the larger the correlation coefficient of two random vectors is, then the larger the correlation of the two random vectors is, i.e., the higher the similarity of the two random vectors is. In an allowable error scope, we take one approximation to approximately deem the fault mode of the known sample as the fault-mode-to-be-tested.

A technical solution to be adopted by the present invention to solve the technical problems thereof is as follows:

An analog circuit fault diagnosis method using a single testable node comprises the following steps:

(1) obtaining prior sample data vectors under each of fault modes: obtaining M groups of voltage sample vectors V_ij of an analog circuit under test under each of fault modes F_i by using computer simulation software, wherein i=1, 2, . . . , N, j=1, 2, 3, . . . , M, N is a total number of the fault modes of the circuit, i represents that the circuit works under a i^th fault mode, j is a j^th collected samples, and V_ij represents a j^th voltage sample vector collected when the circuit works under the i^th fault mode;

(2) computing a statistical average

$V_i=\sum _{j=1}^M{V}_{\mathrm{ij}}\text{/}M$ of the prior sample data vectors under each of the fault modes, wherein i=1, 2, . . . , N, and V_i is voltage sample statistical average vectors when the circuit works under the fault modes F_i;

(3) decomposing a signal by an orthogonal Haar wavelet filter set: decomposing the voltage sample statistical average vectors V_i (i=1, 2, . . . , N) under each of the fault modes into (K+1) filter output signals by a K-layer orthogonal Haar wavelet filter set;

(4) extracting feature factors of prior sample fault modes: extracting (K+1) feature factors s_i,d of the prior sample fault modes through processing the (K+1) filter output signals under the fault modes F_i by using a blind source processing technology, wherein d represents a serial number of fault feature factors, and d=1, 2, . . . , K+1, and s_i,d represents a d^th feature factor of the prior sample fault modes of a voltage sample signal under the fault mode F_i;

(5) extracting feature factors of a fault-mode-to-be-tested: collecting M groups of voltage testable vectors under the fault-mode-to-be-tested, computing a statistical average of the voltage testable vectors, decomposing by the orthogonal Haar wavelet filter set in the step (3), and obtaining (K+1) feature factors s_T,h of the voltage testable vectors under the fault-mode-to-be-tested through the blind source processing technology in step (4), wherein T represents to-be-tested, which is the first letter of Test, and is intended to distinguish the fault-mode-to-be-tested and the prior fault modes; h represents a serial number of the feature factors, h=1, 2, . . . , K+1, and s_T,h represents a h^th feature factor of a voltage testable signal under the fault-mode-to-be-tested;

(6) computing a correlation coefficient matrix R_i and correlation metric parameters δ_i between the feature factors of the fault-mode-to-be-tested and the feature factors of the prior sample fault modes of all the fault modes F_i (i=1, 2, . . . , N);

$R_i=\left[\begin{array}{cccc}{\rho }_{11}& {\rho }_{12}& \dots & {\rho }_{1\left(K+1\right)}\\ {\rho }_{21}& {\rho }_{22}& \dots & {\rho }_{2\left(K+1\right)}\\ ⋮& ⋮& ⋮& ⋮\\ {\rho }_{\left(K+1\right)1}& {\rho }_{\left(K+1\right)2}& \dots & {\rho }_{\left(K+1\right)\left(K+1\right)}\end{array}\right],{\delta }_i=\sum _{h=1}^{K+1}\left(\underset{d}{\mathrm{Max}}\left({\rho }_{\mathrm{hd}}\right)\right),$ wherein ρ_hd=E((s_T,h−E(s_T,h))·(s_i,d))), i=1, 2, . . . , E(⋅) represents to determine an expected value, s_T,h (h=1, 2, . . . , K+1) represents the h^th feature factor of the voltage testable signal under the fault-mode-to-be-tested, s_i,d (d=1, 2, . . . , K+1) represents the d^th feature factor of the prior sample fault modes of the voltage sample signal under the fault mode F_i, the physical meaning of ρ_hd is a correlation coefficient between the h^th feature factor of the voltage testable signal under the fault-mode-to-be-tested and the d^th feature factor of the prior sample fault modes of the voltage sample signal under the fault mode F_i;

$\underset{d}{\mathrm{Max}}\left({\rho }_{\mathrm{hd}}\right)$ represents the maximum ρ_hd when h is constant and d=1, 2, . . . , (K+1); and

(7) comparing all the δ_i, and determining a k^th fault mode if

$k=\mathrm{Index}\left(\underset{i}{\mathrm{Max}}\left({\delta }_i\right)\right),$ wherein i=1, 2, . . . , N, and Index(⋅) represents to calculate an index.

Further, a feature of the K-layer orthogonal Haar wavelet analysis filter set in the step (3) is that each of the layers of the filter set consists of a low-pass filter g(n) and a high-pass filter h(n), output portion of the high-pass filter h(n) is subjected to double downsampling to enter next layer of the wavelet filter set, and output of the low-pass filter g(n) at each layer is subjected to double downsampling and then outputted directly, the low-pass filter g(n)={1/√{square root over (2)}, 1/√{square root over (2)}}, and the high-pass filter h(n)={1/√{square root over (2)}, −1/√{square root over (2)}}.

Further, a method of determining a layer number K of the K-layer orthogonal Haar wavelet analysis filter set in the step (3) is: setting an input signal of the filter set as x, and the outputs of the high-pass filter h(n) and the low-pass filter g(n) at the layer K as y_K,H and y_K,L respectively, then steps of determining the K value are as follows:

(3.1) initializing: K=1, β₀=Th, wherein K is the layer number of filters, β₀ is an energy ratio threshold, and Th is a preset original value of the energy ratio threshold which can be any real number greater than 0 but less than 1;

(3.2) computing the energy ratio

$\beta =\frac{\langle y_{K,H},y_{K,H}\rangle }{\langle x,x\rangle },$ wherein <,> represents to compute an inner product; and

(3.3) if β>β₀, then K=K+1, returning to execute the step (3.2); otherwise, outputting the layer number of filters K.

Further, a method of extracting the feature factor by using a blind source processing technology in the step (4) is: setting a signal matrix needing to be processed by the blind source technology as Y_i, wherein an extracted feature factor matrix is S_i=[s_i,1 s_i,2 . . . s_i,d . . . s_i,(K+1)], s_i,d represents the d^th feature factor of the prior sample fault of the voltage sample signal under the fault mode F_i; and the dimensionalities of Y_i and S_i are equal;

(4.1) initializing a feature extracting matrix W₀ and an update step-length μ, and letting W₁←W₀, wherein “←” represents to assign the value of W₀ to W₁, W₀ is any unit matrix, and μ is a real number between (0, 0.3);

(4.2) computing: S_i=W₀Y_i;

(4.3) updating W₁: W_i←W₀+μ[I−f[S_i]]g^T[S_i], wherein forms of functions f(⋅) and g(⋅) herein are respectively f[S_i]=S_i and g[S_i]=S_i³, g^T(⋅) represents matrix transposition; and I represents a standard unit matrix;

(4.4) standardizing W₁:

$W_1\leftarrow \frac{W_1}{W_1},$ wherein ∥⋅∥ represents a matrix norm,

$\frac{W_1}{W_1}$ represents to standardize W₁, and “←” represents to standardize W₁ and then assign a value to W₁; and

(4.5) determining convergence: determining whether W₁W₁^T→I, i.e., determining whether a product of W₁W₁^T is infinitely approaching to I, “→” represents “infinitely approaching”; if yes, then outputting S_i; otherwise, W₀←W₁, wherein “←” represents assigning, and returning to the step (4.2); and I represents a standard unit matrix.

The present invention can convert a single signal into a plurality of signals without losing original measurement information, and extract an independent fault mode feature factor reflecting variations of a circuit structure in different fault modes, can be used to study an associated mode determination rule and successfully complete classification of circuit fault modes.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1A and FIG. 1B are block diagrams of an analog circuit fault diagnosis method using a single testable node.

FIG. 2 is a structural diagram of a wavelet filter set.

FIG. 3 is a flow of determining a layer number of the filter set.

FIG. 4 is a flow of extracting a feature factor using a single testable node.

DETAILED DESCRIPTION

The invention is explained in details hereinafter with reference to the drawings.

Referring to FIG. 1A and FIG. 1B, an analog circuit fault diagnosis method using a single testable node comprises the following steps:

(1) obtaining prior sample data vectors under each fault mode: obtaining M groups of voltage sample vectors V_ij of an analog circuit under test under each of fault modes F_i by using computer simulation software, wherein i=1, 2, . . . , N, j=1, 2, 3, . . . , M, N is a total number of the fault modes of the circuit, i represents that the circuit works under a i^th fault mode, j is a j^th collected samples, and V_ij represents a j^th voltage sample vector collected when the circuit works under the i^th fault mode; in FIG. 1A and FIG. 1B, it is expressed as: collecting M groups of the voltage sample vectors under the first type fault mode, collecting M groups of the voltage sample vectors under the second type fault mode, . . . , and collecting M groups of the voltage sample vectors under the N^th type fault mode;

(2) computing a statistical average

$V_i=\sum _{j=1}^M{V}_{\mathrm{ij}}/M$ of the prior sample data vectors under each of the fault modes, wherein i=1, 2, . . . , N, and V_i is voltage sample statistical average vectors of the voltage sample when the circuit works under the fault modes F_i;

(3) decomposing a signal by an orthogonal Haar wavelet analysis filter set: decomposing the voltage sample statistical average vector V_i (i=1, 2, . . . , N) under each of the fault modes into (K+1) filter output signals by a K-layer orthogonal Haar wavelet filter set;

(4) extracting feature factors of prior sample fault modes: extracting (K+1) feature factors s_i,d of the prior sample fault modes through processing the (K+1) filter output signals under the fault modes Fi by using the blind source processing technology under the fault mode F_i, wherein d represents a serial number of fault feature factors, and d=1, 2, . . . , K+1, and s_i,d represents a d^th feature factor of the prior sample fault modes of a voltage sample signal under the fault mode F_i;

(5) extracting feature factors of a fault-mode-to-be-tested: collecting M groups of voltage testable vectors under the fault-mode-to-be-tested, computing a statistical average of the voltage testable vectors, decomposing by the orthogonal Haar wavelet filter set in the step (3), and obtaining (K+1) feature factors s_T,h of the voltage testable vectors under the fault-mode-to-be-tested through the blind source processing technology in step (4), wherein T represents to-be-tested, which is the first letter of Test, and is intended to distinguish the fault-mode-to-be-tested and the prior fault modes; h represents a serial number of the feature factors, h=1, 2, . . . , K+1, and s_T,h represents a h^th feature factor of a voltage testable signal under the fault-mode-to-be-tested;

(6) computing a correlation coefficient matrix R_i and correlation metric parameters δ_i between the feature factor of the fault-mode-to-be-tested and the feature factor of the prior sample fault modes of all the fault modes F_i (i=1, 2, . . . , N);

$R_i=\left[\begin{array}{cccc}{\rho }_{11}& {\rho }_{12}& \cdots & {\rho }_{1\left(K+1\right)}\\ {\rho }_{21}& {\rho }_{22}& \cdots & {\rho }_{2\left(K+1\right)}\\ ⋮& ⋮& ⋮& ⋮\\ {\rho }_{\left(K+1\right)1}& {\rho }_{\left(K+1\right)2}& \cdots & {\rho }_{\left(K+1\right)\left(K+1\right)}\end{array}\right],{\delta }_i=\sum _{h=1}^{K+1}\left(\underset{d}{\mathrm{Max}}\left({\rho }_{\mathrm{hd}}\right)\right)$ wherein ρ_hd=E((s_T,h−E(s_T,h))·(s_i,d−E(s_i,d))), i=1, 2, . . . , N, E(⋅) represents to determine an expected value, s_T,h (h=1, 2, . . . , K+1) represents the h^th feature factor of the voltage testable signal under the fault-mode-to-be-tested, s_i,d (d=1, 2, . . . , K+1) represents the d^th feature factor of the prior sample fault modes of the voltage sample signal under the fault mode F_i, the physical meaning of ρ_hd is a correlation coefficient between the h^th feature factor of the voltage testable signal under the fault-mode-to-be-tested and the d^th feature factor of the prior sample fault modes of the voltage sample signal under the prior fault mode F_i,

$\underset{d}{\mathrm{Max}}\left({\rho }_{\mathrm{hd}}\right)$ represents the maximum ρ_hd when h is constant and d=1, 2, . . . , (K+1); and

(7) comparing all the δ_i, and determining a k^th fault mode if

$k=\mathrm{Index}\left(\underset{i}{\mathrm{Max}}\left({\delta }_i\right)\right),$ wherein i=1, 2, . . . , N, and Index(⋅) represents to calculate an index.

Referring to FIG. 2, a feature of the K-layer orthogonal Haar wavelet analysis filter set in the step (3) is that each of the layers of the filter set consists of a low-pass filter g(n) and a high-pass filter h(n), output portion of the high-pass filter h(n) is subjected to double downsampling to enter next layer of wavelet filter set, and the output of the low-pass filter g(n) at each layer is subjected to double downsampling and then outputted directly, the low-pass filter g(n)={1/√{square root over (2)}, 1/√{square root over (2)}}, and the high-pass filter h(n)={1/√{square root over (2)}, −1/√{square root over (2)}}. In FIG. 2, y_cL (c=1, 2, . . . , K) represents the output of a low pass filter at a c^th layer of the filter set, subscript c represents the serial number of the filter layer of the filter, L represents low-pass; y_KH represents output of a high-pass filter at a k^th layer of the filter set, subscript K represents the serial number of the filter layer of the filter, and H represents high-pass.

Referring to FIG. 3, a method of determining a layer number K of the K-layer orthogonal Haar wavelet analysis filter set in the step (3) is: setting an input signal of the filter set as x, and the outputs of the high-pass filter h(n) and the low-pass filter g(n) at the layer K as y_K,H and y_K,L respectively, then steps of determining the K value are as follows:

(3.1) initializing: K=1, β₀=Th, wherein K is the layer number of filters, β₀ is an energy ratio threshold, and Th is a preset original value of the energy ratio threshold which can be any real number greater than 0 but less than 1;

(3.2) computing the energy ratio

$\beta =\frac{\langle y_{K,H},y_{K,H}\rangle }{\langle x,x\rangle },$ wherein <,> represents to compute an inner product; and

(3.3) if β>β₀, then K=K+1, returning to execute the step (3.2); otherwise, outputting the layer number of filters K.

Referring to FIG. 4, a method of extracting the feature factor using a blind source processing technology in the step (4) is: setting a signal matrix needing to be processed by the blind source technology as Y_i, wherein an extracted feature factor matrix is S_i=[s_i,1 s_i,2 . . . s_i,d . . . s_i,(K+1)], s_i,d represents the d^th feature factor of the prior sample fault of the voltage sample signal under the fault mode F_i; and the dimensionalities of Y_i and S_i are equal;

(4.1) initializing a feature extracting matrix W₀ and an update step-length μ, and letting W₁←W₀, wherein “←” represents to assign the value of W₀ to W₁, W₀ is any unit matrix, and μ is a real number between (0, 0.3);

(4.2) computing: S_i=W₀Y_i;

(4.3) updating W₁: W₁←W₀+μ[I−f[S_i]]g^T[S_i], wherein forms of functions f(⋅) and g(⋅) herein are respectively f[S_i]=S_i and g[S_i]=S_i³, g^T(⋅) represents matrix transposition; and I represents a standard unit matrix;

(4.4) standardizing W₁:

$W_1\leftarrow \frac{W_1}{W_1},$ wherein ∥⋅∥ represents a matrix norm

$\frac{W_1}{W_1}$ represents to standardize W₁, and “←” represents to standardize W₁ and then assign a value to W₁; and

(4.5) determining convergence: determining whether W₁W₁^T→I, i.e., determining whether a product of W₁W₁^T is infinitely approaching to I, “→” represents “infinitely approaching”; if yes, then outputting S_i; otherwise, W₀←W₁, wherein “←” represents assigning, and returning to the step (4.2); and I represents a standard unit matrix.

Claims

1. An analog circuit fault diagnosis method using a single testable node, comprising: (1) obtaining prior sample data vectors under each of fault modes: obtaining M groups of voltage sample vectors Vij of an analog circuit under test under each of the fault modes Fi by using computer simulation software, wherein i=1, 2, . . . , N, j=1, 2, 3, . . . , M, N is a total number of the fault modes of the circuit, i represents that the circuit works under a ith fault mode, j is a jth collected sample, and Vij represents a jth voltage sample vector collected when the circuit works under a ith fault mode;(2) computing a statistical average

2. The analog circuit fault diagnosis method using a single testable node according to claim 1, which characterized in, a feature of the K-layer orthogonal Haar wavelet analysis filter set in the step (3) is: each of layers of the filter set consists of a low-pass filter g(n) and a high-pass filter h(n), output portion of the high-pass filter h(n) is subjected to double downsampling to enter next layer of the wavelet filter set, and output of the low-pass filter g(n) at each layer is subjected to double downsampling and then outputted directly, the low-pass filter g(n)={1/√{square root over (2)}, 1/√{square root over (2)}}, and the high-pass filter h(n)={1/√{square root over (2)}, 1/√{square root over (2)}}.

3. The analog circuit fault diagnosis method using a single testable node according to claim 2, wherein a method of determining a layer number K of the K-layer orthogonal Haar wavelet analysis filter set in the step (3) comprises: setting an input signal of the filter set as x=Vi, wherein Vi is the statistical average

4. The analog circuit fault diagnosis method using a single testable node according to claim 1, wherein a method of extracting the feature factor by using the blind source processing technology in the step (4) is: setting a signal matrix needing to be processed by the blind source technology as Yi, wherein an extracted feature factor matrix is Si=[si,1 si,2 . . . si,d . . . si,(K+1)], represents the dth feature factor of the prior sample fault mode under the fault mode Fi, and the dimensionalities of Yi and Si are equal; (4.1) initializing a feature extracting matrix W0 and an update step-length μ, and letting W1←W0, wherein “←” represents to assign the value of W0 to W1, W0 is any unit matrix, and μ is a real number between (0, 0.3);(4.2) computing: Si=W0Yi;(4.3) updating W1: W1←W0+μ[I−f[Si]]gT[Si], wherein forms of functions f(⋅) and g(⋅) herein are respectively f[Si]=Si and g[Si]=Si3, gT(⋅) represents matrix transposition; and I represents a standard unit matrix;(4.4) standardizing W1:

5. The analog circuit fault diagnosis method using a single testable node according to claim 1, wherein a method of determining a layer number K of the K-layer orthogonal Haar wavelet analysis filter set in the step (3) comprises: setting an input signal of the filter set as x=Vi, wherein Vi is the statistical average