Large-scale circuit fault diagnosis method and system based on cloud theory and kernel density

Abstract

The present invention discloses a large-scale circuit fault diagnosis method and system based on a Bayesian model of Epanechnikov kernel density estimation based on cloud theory. This method belongs to the field of large-scale analog circuit fault diagnosis. The method collects fault information from modular integrated circuits after circuit tearing, uses a reverse generator to generate digital features of the cloud under each fault condition, and uses a forward cloud generator to obtain a number of cloud droplets under each fault mode. A standard cloud model is used to calculate the membership values ​​of each feature in the sample set and the corresponding fault state. The cloud droplets under each fault feature are determined based on the membership matrix to expand the fault sample set. The Bayesian model of Epanechnikov kernel density estimation is used to calculate the posterior probability density value for each fault category to obtain the fault diagnosis result. The present invention combines cloud theory and Bayesian kernel density methods to effectively address the problems of single state information and a small number of fault sample sets after network tearing. It integrates data information and prior knowledge, greatly improving fault accuracy.

Description

Technical Field

[0001] The present invention belongs to the technical field of large-scale analog circuit fault diagnosis, and more specifically, relates to a large-scale circuit fault diagnosis method and system based on a Bayesian model of Epanechnikov kernel density estimation based on cloud theory. Background Art

[0002] Since the development of integrated circuits, they have gone through small-scale, medium-scale, and large-scale development stages, and have now entered the ultra-large-scale integrated circuit stage. Therefore, industrial production has an urgent demand for large-scale circuit testing. Large-scale circuit fault diagnosis has not yet been fully mature in theory and methods.

[0003] Accurate module-level fault location and identification, with clear detection results, are pressing issues in electronic engineering and a key step in transitioning from theory to practical application. For large-scale analog circuits, the topology contains a huge number of devices, and the parameters of each device have continuity within their effective range. This leads to increasingly complex circuit operating conditions. Furthermore, after a large-scale circuit network is torn, the state information is single, making it difficult to obtain simulation samples. Furthermore, the fault sample set consisting of the torn node voltage under each fault mode is small, making it difficult to accurately diagnose all faults. Traditional fault analysis methods are no longer able to meet current needs. Summary of the Invention

[0004] In response to the above defects or improvement needs of the existing technology, the present invention proposes a large-scale circuit fault diagnosis method and system based on the Bayesian model of Epanechnikov kernel density estimation of cloud theory to improve the accuracy and efficiency of circuit diagnosis.

[0005] To achieve the above objectives, according to one aspect of the present invention, a large-scale circuit fault diagnosis method based on a Bayesian model of Epanechnikov kernel density estimation based on cloud theory is provided, comprising:

[0006] A large-scale circuit schematic was established, and the voltage of the modular tearing node was selected as the diagnostic signal. The fault categories were coded and classified according to the different faults of the components. The voltage of the tearing node under different faults was collected to form the initial sample set.

[0007] The voltage of the torn nodes under different faults constitutes the initial sample set, and the inverse generator of the cloud model is used to generate the digital feature expectation, entropy and super entropy of the cloud under each fault;

[0008] Based on the expected digital characteristics, entropy and super entropy of the cloud under each fault, a forward cloud generator is used to obtain several cloud droplets under each fault to restore the voltage values ​​of many measurable nodes.

[0009] The standard cloud model is used to calculate the membership values ​​of each voltage value and the corresponding fault state in the initial sample set to form a membership matrix. The cloud droplets under each fault voltage value are determined according to the fault membership matrix to expand the initial sample set.

[0010] The expanded initial sample set is divided into several folds for cross-validation, and the posterior probability density value of each fault category in the test set is calculated using the Bayesian model of Epanechnikov kernel density estimation;

[0011] The newly acquired test samples are classified into the category with the highest posterior probability density value to obtain the diagnosis result.

[0012] In some optional implementation schemes, selecting the voltage of the modular tearing node as the diagnostic signal includes:

[0013] Modular tearing is performed on large-scale circuits, and the tearing must be done on the premise of limited accessible nodes. The limited accessible nodes are torn into modules, and the voltage of the torn nodes under each fault mode is selected as the diagnostic signal to obtain the fault feature vector.

[0014] In some optional embodiments, the initial sample set consisting of the voltages of the torn nodes under different faults is generated using an inverse generator of the cloud model to generate the expected digital features, entropy, and hyperentropy of the cloud under each fault, including:

[0015] Let the initial sample set x of circuit faults be ij =[x i1 ,x i2 ,…,x in ](i=1,2,…n;j=1,2,…m), represents the jth type of fault of the i-th fault feature, n is the dimension of the feature vector under the fault mode, and m is the total number of fault modes, then Get the mean of the initial sample set of the jth type fault Depend on The first-order central moment K of the initial sample set of type j fault is obtained by Get the variance S of the initial sample set of the j-th type fault;

[0016] Since the degree of certainty is unknown, the reverse cloud generator CG without certainty is used. -1 Get the digital feature expectation E of the cloud model x , Entropy E n 、Super entropy H e for:

[0017] In some optional embodiments, the method of obtaining a number of cloud droplets under each fault condition by using a forward cloud generator includes:

[0018] In each type of failure cloud {E x ,E n ,H e}, the number of cloud droplets Num is set, and the forward cloud model CG(E x ,E n ,H e ,Num), first for the jth type of fault, generate E n For expectations, is a normal random number with variance E' ni ;

[0019] Using the generated random number, E x For expectations, Generate normal random numbers with variance x k , and calculate x k Membership Among them, u with certainty k x k Represents a cloud droplet in the universe of discourse;

[0020] Repeat the above steps until the set number of cloud droplets Num is generated, and obtain the cloud sample set drop(x k ,u k ).

[0021] In some optional implementation schemes, the method of using the standard cloud model to calculate the membership values ​​of each voltage value and the corresponding fault state in the initial sample set to form a membership matrix includes:

[0022] The cloud model is used to describe the membership of each fault feature vector corresponding to each fault type. The standard cloud model constructed under each fault is used to convert the initial fault sample x ij The standard cloud parameter corresponding to the j-th fault of the i-th eigenvector is substituted into Get x ij The corresponding membership degree, where u ij is the membership degree of the j-th type of fault of the i-th eigenvector of the initial fault sample, E nij ′ is the entropy E of the j-th fault of the i-th eigenvector nij and super entropy H eij 2 The generated normal random number, E xij is the numerical feature expectation of the j-th type of fault of the i-th feature vector;

[0023] After obtaining the membership of various faults of the initial fault sample under all eigenvectors, the membership moment is obtained: m represents the total number of failure modes;

[0024] Use fuzzy comprehensive evaluation uncertainty index to form a fuzzy vector, Obtain the degree of membership of uncertainty, n is the fuzzy comprehensive evaluation vector (u i1 ,u i2 ,…,u in ), the number of membership degrees of i=1,2,…n, β is (u i1 ,u i2 ,…,u in ), γ is the maximum membership in (u i1 ,u i2 ,…,u in ) in the second largest membership;

[0025] Calculate the uncertain membership degree under each fault under the membership degree and construct the uncertain membership fuzzy feature vector Represents the uncertain membership of the i-th eigenvector, and obtains the membership matrix of m rows and n+1 columns:

[0026] In some optional implementation schemes, the expanded initial sample set is divided into several folds for cross-validation, and the posterior probability density value of each fault category in the test set is calculated using the Bayesian model of Epanechnikov kernel density estimation, including:

[0027] The expanded initial sample set is divided into a training set and a test set by cross-validation;

[0028] In the Bayesian model, the likelihood function of the Epanechnikov kernel density estimation can be obtained by using several independent and identically distributed samples in the test set. According to the Bayesian theorem, the posterior distribution of the likelihood function of the Epanechnikov kernel density estimation is obtained, and then the posterior probability density value under each fault category is obtained.

[0029] In some optional embodiments, classifying the newly acquired test sample into the category with the highest posterior probability density value to obtain a diagnosis result includes:

[0030] The Bayesian model of Epanechnikov kernel density estimation is used to classify fault data. The voltage data points of the fault to be tested are regarded as a vector, and the probability density function of each fault category is estimated using the Epanechnikov kernel density estimation method.

[0031] The voltage data points of the fault to be tested are projected onto the probability density function of each fault category, the probability density value under each category is calculated, and the voltage data points of the fault to be tested are classified into the category with the highest probability density value.

[0032] In some optional implementation schemes, projecting the voltage data points of the fault to be tested onto the probability density function of each fault category, calculating the probability density value under each category, and classifying the voltage data points of the fault to be tested into the category with the highest probability density value includes:

[0033] The fault voltage data point to be tested is regarded as a vector containing several features. The features in the vector are subtracted from the posterior mean of the category respectively. Each element obtained is divided by the posterior variance of the category and substituted into the Epanechnikov kernel function as an independent variable. The average value of the new vector is taken to obtain the probability density value of the fault sample to be tested under this category.

[0034] Calculate the inverse of the square root of the determinant of the posterior variance of the category and multiply it by the obtained probability density value to normalize the probability value;

[0035] The posterior probability density value of the fault voltage data point to be tested under each fault category is calculated according to the Bayesian theorem, and the fault voltage data point to be tested is classified into the category with the highest probability density value.

[0036] According to another aspect of the present invention, a large-scale circuit fault diagnosis system based on a Bayesian model of Epanechnikov kernel density estimation based on cloud theory is provided, comprising:

[0037] The circuit acquisition module to be diagnosed is used to build a large-scale circuit schematic, select the voltage of the modular tear node as the diagnostic signal, encode and classify the fault category according to the different faults of the components, and collect the voltage of the tear node under different faults to form the initial sample set;

[0038] The cloud theory calculation module is used to generate the expected digital features, entropy, and hyperentropy of the cloud under each fault using the inverse cloud model generator from the initial sample set of voltages of the torn nodes under different faults. Based on the expected digital features, entropy, and hyperentropy of the cloud under each fault, the forward cloud generator is used to obtain a number of cloud droplets under each fault to restore the voltage values ​​of many measurable nodes.

[0039] The membership expansion sample module is used to use the standard cloud model to calculate the membership values ​​of each voltage value and the corresponding fault state under the initial sample set to form a membership matrix. The cloud droplet under each fault voltage value is determined according to the fault membership matrix to expand the initial sample set;

[0040] The fault diagnosis module is used to divide the expanded initial sample set into several folds for cross-validation, use the Bayesian model of Epanechnikov kernel density estimation to calculate the posterior probability density value of each fault category in the test set, classify the newly acquired test samples into the category with the highest posterior probability density value, and obtain the diagnosis result.

[0041] According to another aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of any one of the above methods are implemented.

[0042] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects compared with the prior art:

[0043] Large-scale circuit fault diagnosis based on the Bayesian model of Epanechnikov kernel density estimation in cloud theory uses cloud droplets generated by combining the forward cloud model and the inverse cloud model in cloud theory to expand the fault sample set. The generation of "cloud"-based samples integrates the randomness and fuzzy uncertainty theory of the cloud model, filters out characteristic data that deviates from the normal distribution, and has a certain noise resistance. The Bayesian model of the Epanechnikov kernel function does not require any assumptions about the distribution of the target variable. The circuit fault diagnosis model obtained on this basis can better adapt to different data sets and problems, and effectively improve the accuracy and efficiency of circuit fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 This is a flow chart of a circuit fault diagnosis method provided by an embodiment of the present invention;

[0045] Figure 2 This is a schematic diagram of a video amplifier circuit provided by an embodiment of the present invention;

[0046] Figure 3 This is a tear diagram of a video amplifier circuit provided by an embodiment of the present invention;

[0047] Figure 4 This is a fault cloud sample generation process provided by an embodiment of the present invention;

[0048] Figure 5 This is a comparison of cloud model graphs under normal and fault conditions provided by an embodiment of the present invention;

[0049] Figure 6 is a confusion matrix of a fault diagnosis result provided by an embodiment of the present invention;

[0050] Figure 7 It is a structural diagram of a circuit fault diagnosis device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0051] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely for the purpose of explaining the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.

[0052] like Figure 1 The figure is a flow chart of a method provided by an embodiment of the present invention, and the present invention is described in detail by taking a video amplifier circuit as an example. Figure 1 The method shown includes the following steps:

[0053] (1) Establish a video amplifier circuit simulation model, the simulation circuit diagram is as follows Figure 2 As shown in the figure, the video amplifier circuit is modularized under the condition of node tearing, as shown in the figure. Figure 3 As shown in Figure 1, the voltage state information of the tearing node is collected to obtain the fault feature vector, where the tolerance of the resistor in the circuit is 5%. Considering that semiconductor devices and passive devices are more prone to failure, and based on actual experience, a total of 9 fault modes including the normal state are set for the circuit, as shown in Table 1;

[0054] Table 1: Fault status and categories

[0055] Fault Label Fault status Fault code F0 Normal state 0 F1 <![CDATA[V1 open circuit]]> 1 F2 <![CDATA[Short circuit of R1]]> 2 F3 <![CDATA[Q4BE Extremely short circuit]]> 3 F4 <![CDATA[Q2CE extremely short circuit]]> 4 F5 <![CDATA[Short circuit of R9]]> 5 F6 <![CDATA[V2 open circuit]]> 6 F7 <![CDATA[V3 open circuit]]> 7 F8 <![CDATA[Q 10 CE is extremely short circuit]]> 8

[0056] In step (1), modular tearing is performed on large-scale circuits. A network should meet the following conditions when the nodes are torn: 1. The torn nodes must be accessible nodes, including public nodes; 2. There is no topological relationship and coupling between parameters between the sub-networks after tearing; 3. Each sub-network is the smallest sub-network relative to the torn node. Each sub-network is only used for accessible nodes and cannot be torn further. That is, the smallest sub-network size means fewer faults, fewer combinations of faults, and a higher diagnosis rate.

[0057] (2) The initial sample set consisting of the voltage of the torn node under each fault mode is sent to the reverse cloud generator to obtain the cloud {E x ,E n ,H e};

[0058] (3) Using the forward cloud generator algorithm to obtain several cloud droplets under each fault mode, we can restore the voltage values ​​of many measurable nodes;

[0059] (4) Using the standard cloud model to calculate the membership value of each voltage feature and the corresponding fault state under the initial sample set, the cloud droplet under each fault voltage feature is determined according to the membership matrix of the original fault to expand the initial sample set;

[0060] (5) Divide the expanded initial sample set into several folds for cross-validation, and use the Bayesian model of Epanechnikov kernel density estimation to calculate the posterior probability density value of each fault category in the test set;

[0061] (6) For the newly acquired test sample, classify it into the category with the highest posterior probability density value to obtain the diagnosis result.

[0062] In this embodiment, the above step (2) can be implemented in the following manner:

[0063] The generation of cloud samples requires both reverse cloud and forward cloud model algorithms. For each circuit fault, the cloud can be used to describe the uncertain transformation between the qualitative concept of the fault mode and the fixed value, so as to reflect the uncertainty of things in the natural world or human knowledge concepts: fuzziness and randomness. This is not only explained from the random theory and fuzzy set theory, but also reflects the correlation between fuzziness and randomness, forming a mapping between quantitative and qualitative, such as Figure 4 The figure shows the generation process of fault cloud samples. The elements in the sample set under the fault mode can be represented by cloud droplets. The sampled voltage is converted from a fixed value to a qualitative concept using the reverse cloud generator. The qualitative concept is represented by the digital feature {E x ,E n ,H e} to express, where the digital feature expectation E x , Entropy E n 、Super entropy H e .

[0064] The algorithm implementation of the one-dimensional inverse cloud generator is as follows:

[0065] Input: quantitative values ​​of the initial fault sample set;

[0066] Output: expected value E of the qualitative concept A represented by the cloud droplet x , entropy E n and super entropy H e ;

[0067] Let the initial sample set x of circuit faults be ij =[x i1 ,x i2 ,…,x in ](i=1,2,…,n;j=1,2,…,m), calculate the sample mean First-order sample absolute central moment Sample variance

[0068] In the universe space, the cloud droplet is the point that best represents the qualitative concept, and its expectation is the central value in the universe space.

[0069] Entropy is determined by the randomness and fuzziness of qualitative concepts, and represents the measurable granularity of a qualitative concept. n It is a measure of the randomness of qualitative concepts, reflecting the degree of discreteness of the cloud droplets, and also reflects the margin of qualitative concepts. It reflects the range of cloud droplets in the domain space that can be accepted by the qualitative concept. It is a measure of the fuzziness of qualitative concepts. Generally, the greater the entropy, the larger the range of cloud droplets acceptable to the qualitative concept, and the more vague the qualitative concept. This also reflects the correlation between randomness and fuzziness. Similarly, the entropy can be obtained from the sample mean.

[0070] Super entropy is a measure of the uncertainty of entropy, which reveals the cohesion of uncertainty of all points of language values ​​in the domain space and the relationship between fuzziness and randomness, and indirectly reflects the thickness of the cloud. From the sample variance and super entropy, we can get

[0071] In this embodiment, the above step (3) can be implemented in the following manner:

[0072] The forward cloud generator is a mapping from qualitative concepts to their quantitative representations. According to the digital features of the cloud obtained above {E x ,E n ,H e}Generate cloud droplets, each of which is a concrete implementation of this concept.

[0073] The algorithm implementation of the one-dimensional forward cloud generator is as follows:

[0074] Input: three numerical eigenvalues ​​E representing the qualitative concept A x ,E n ,H e And the number of cloud droplets Num;

[0075] Output: The quantitative values ​​of Num cloud droplets and the degree of certainty that each cloud droplet represents concept A;

[0076] Produced with E n For expectations, is a normal random number with variance E' ni ;

[0077] Using the generated random number, E x For expectations, Generate normal random numbers with variance x k , and calculate x k Membership Among them, u with certainty k x k Represents a cloud droplet in the universe of discourse;

[0078] Repeat the above steps until the set number of cloud droplets Num is generated, and obtain the cloud sample set drop(x k ,u k ).

[0079] In this example, the number of cloud droplets can be Num=10000.

[0080] like Figure 5 Shown are the cloud models and sample distributions for the fault labels F0, F3, F5, and F7, each with 10,000 cloud droplets.

[0081] By comparison, we can see that from the highest to lowest degree of cloud model analysis, the following progression is observed: expectation → certainty (membership) → randomness (discreteness). The range of cloud droplets accepted by each fault in the universe space and the degree of cloud droplet cohesion vary significantly, leading to significant differences between normal and faulty states.

[0082] In this embodiment, the above step (4) can be implemented in the following manner:

[0083] In order to avoid excessively lengthy feature vectors when expanding the fault sample set, the embodiment of the present invention selects the cloud membership matrix under the feature node to obtain cloud droplets with larger occurrence probability and certainty as feature vectors for the next step of fault mode identification.

[0084] In order to generate cloud sample sets without increasing the data dimension, the cloud model is used to describe the membership of each fault feature vector corresponding to each fault type. By using the standard cloud model constructed under each fault, the membership of the j-th type of fault of the i-th feature vector can be calculated. The initial fault sample x ij The standard cloud parameter corresponding to the j-th type of fault of the i-th feature is substituted into Get the corresponding membership, where u ij is the membership degree of the j-th type of fault of the i-th eigenvector of the initial fault sample, E nij ′ is the entropy E of the j-th fault of the i-th eigenvector nij and super entropy H eij 2 Generates normal random numbers.

[0085] At the same time, considering the uncertainty of data errors and external interference in the fault diagnosis process, a fuzzy vector is constructed using fuzzy comprehensive evaluation uncertainty indicators. The calculation formula for the uncertainty membership is as follows: n is the fuzzy comprehensive evaluation vector (u i1 ,u i2 ,…,u in ), the number of membership degrees of i=1,2,…n, β is (u i1 ,ui2 ,…,u in )’s maximum membership, γ is (u i1 ,u i2 ,…,u in ) in the second largest membership. Calculate the uncertain membership under each fault under the membership, and construct the uncertain membership fuzzy feature vector as shown in the formula In the formula Represents the uncertain membership of the i-th feature. Combining the above formula, we can regain the membership matrix of m rows and n+1 columns.

[0086] Taking into account the influence of external interference, 200 cloud droplets are collected at each membership degree u±0.01, and finally 11×9×200 fault samples with 11 fault characteristic attributes and 9 fault categories are obtained.

[0087] In this embodiment, the above step (5) can be implemented in the following manner:

[0088] The sample set is divided into training set and test set for cross validation.

[0089] Epanechnikov kernel density estimation is a non-parametric probability density estimation method that can be used to estimate any distribution, while the Bayesian model is a probability model based on Bayes' theorem that can be used to estimate the posterior distribution of unknown parameters. In the Bayesian model, the unknown parameters are regarded as random variables, and the posterior distribution of the parameters is updated by observing data. There are n independent and identically distributed samples X1, X2, ...X n , then the likelihood function expression of the Epanechnikov kernel density estimate can be obtained as follows:

[0090] Among them, h is the bandwidth parameter, which determines the smoothness of the kernel function. According to Bayes’ theorem, its posterior distribution is obtained: π(f|X1,…,X n )∞L(f|X1,…,X n )π(f);

[0091] In this Bayesian model, the variable y is assumed to be generated from an unknown probability distribution. The Epanechnikov kernel function is then used to construct a prior probability distribution, which represents the density distribution of y across the entire input space. Using Bayes' rule, the observed data x is combined with the prior probability distribution to produce a posterior probability distribution. Ultimately, this posterior probability distribution can be used to predict the target variable value for new data points. Unlike traditional parametric regression models, Bayesian models based on the Epanechnikov kernel function do not make any assumptions about the distribution of the target variable, making them more adaptable to diverse datasets and problems.

[0092] In this embodiment, step (6) can be implemented in the following manner:

[0093] The Bayesian model of Epanechnikov kernel density estimation is used to classify fault data. Each fault voltage data point is treated as a vector, and the Epanechnikov kernel density estimation method is used to estimate the probability density function of each fault category. Then, the data points to be classified are projected onto the probability density function of each fault category, and their probability density values ​​under each category are calculated. Finally, the data point is classified into the category with the highest probability density value. Specifically:

[0094] The fault data point to be tested is treated as a vector containing several features. Each feature in the vector is subtracted from the posterior mean of the category. Each resulting element is divided by the posterior variance of the category and substituted into the Epanechnikov kernel function as the independent variable. The average of the resulting new vector is the probability density value of the test sample under that category. The reciprocal square root of the determinant of the posterior variance of the category is calculated and multiplied by the probability density value obtained in the previous step to normalize the probability value. The posterior probability density value of the data point to be classified is calculated according to Bayes' theorem for each fault category, and the data point is classified into the category with the highest probability density value.

[0095] The confusion matrix of the fault classification results is as follows Figure 6 As shown, the diagnostic accuracy is 99.4%.

[0096] The present invention is based on a large-scale circuit fault diagnosis method of a Bayesian model of Epanechnikov kernel density estimation based on cloud theory. The method uses cloud droplets generated by combining the forward cloud model and the inverse cloud model in cloud theory to expand the fault sample set, effectively solving the problem of a small number of samples after network tearing or difficulty in obtaining simulation samples under specific circumstances. The Bayesian model of Epanechnikov kernel density estimation is used for pattern recognition, solving the overfitting problem caused by a small number of samples when estimating the probability density function of the class condition using Gaussian distribution, thereby improving the diagnostic accuracy of fault diagnosis.

[0097] like Figure 7 As shown, in another embodiment of the present invention, a large-scale circuit fault diagnosis system based on a Bayesian model of Epanechnikov kernel density estimation based on cloud theory is provided, comprising:

[0098] The circuit acquisition module to be diagnosed is used to build a large-scale circuit schematic, select the voltage of the modular tear node as the diagnostic signal, encode and classify the fault category according to the different faults of the components, and collect the voltage of the tear node under different faults to form the initial sample set;

[0099] The cloud theory calculation module is used to generate the expected digital features, entropy, and hyperentropy of the cloud under each fault using the inverse cloud model generator from the initial sample set of voltages of the torn nodes under different faults. Based on the expected digital features, entropy, and hyperentropy of the cloud under each fault, the forward cloud generator is used to obtain a number of cloud droplets under each fault to restore the voltage values ​​of many measurable nodes.

[0100] The membership expansion sample module is used to use the standard cloud model to calculate the membership values ​​of each voltage value and the corresponding fault state under the initial sample set to form a membership matrix. The cloud droplet under each fault voltage value is determined according to the fault membership matrix to expand the initial sample set;

[0101] The fault diagnosis module is used to divide the expanded initial sample set into several folds for cross-validation, use the Bayesian model of Epanechnikov kernel density estimation to calculate the posterior probability density value of each fault category in the test set, classify the newly acquired test samples into the category with the highest posterior probability density value, and obtain the diagnosis result.

[0102] The specific implementation of each module can refer to the description of the above method embodiment, and will not be repeated in this embodiment.

[0103] It should be pointed out that, according to the needs of implementation, the various steps / components described in this application can be split into more steps / components, or two or more steps / components or partial operations of steps / components can be combined into new steps / components to achieve the purpose of the present invention.

[0104] It will be easily understood by those skilled in the art that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A large-scale circuit fault diagnosis method based on a Bayesian model of Epanechnikov kernel density estimation based on cloud theory, characterized in that: include: A large-scale circuit schematic was established, and the voltage of the modular tearing node was selected as the diagnostic signal. The fault categories were coded and classified according to the different faults of the components. The voltage of the tearing node under different faults was collected to form the initial sample set. The voltage of the torn nodes under different faults constitutes the initial sample set, and the inverse generator of the cloud model is used to generate the digital feature expectation, entropy and super entropy of the cloud under each fault; Based on the expected digital features, entropy, and hyperentropy of the cloud under each fault, a forward cloud generator is used to obtain several cloud droplets under each fault to restore the voltage values ​​of many measurable nodes. For each circuit fault mode, a cloud is used to describe the randomness and fuzziness of the fault mode, as well as the correlation between the two. The elements in the fault sample set under the fault mode can be considered as the constituent cloud droplets. The standard cloud model is used to calculate the membership values ​​of each voltage value and the corresponding fault state in the initial sample set to form a membership matrix. The cloud droplets under each fault voltage value are determined according to the fault membership matrix to expand the initial sample set. The expanded initial sample set is divided into several folds for cross-validation, and the posterior probability density value of each fault category in the test set is calculated using the Bayesian model of Epanechnikov kernel density estimation; Classify the newly acquired test sample into the category with the highest posterior probability density value to obtain the diagnosis result; The expanded initial sample set is divided into several folds for cross-validation, and the posterior probability density value of each fault category in the test set is calculated using the Bayesian model of Epanechnikov kernel density estimation, including: The expanded initial sample set is divided into a training set and a test set by cross-validation; In the Bayesian model, the likelihood function of the Epanechnikov kernel density estimate can be obtained by using several independent and identically distributed samples in the test set. According to the Bayesian theorem, the posterior distribution of the likelihood function of the Epanechnikov kernel density estimate is obtained, and then the posterior probability density value under each fault category is obtained; The newly acquired test sample is classified into the category with the highest posterior probability density value to obtain a diagnosis result, including: The Bayesian model of Epanechnikov kernel density estimation is used to classify fault data. The voltage data points of the fault to be tested are regarded as a vector, and the probability density function of each fault category is estimated using the Epanechnikov kernel density estimation method. Project the voltage data points of the fault to be tested onto the probability density function of each fault category, calculate the probability density value under each category, and classify the voltage data points of the fault to be tested into the category with the highest probability density value; The method of projecting the voltage data points of the fault to be tested onto the probability density function of each fault category, calculating the probability density value under each category, and classifying the voltage data points of the fault to be tested into the category with the highest probability density value includes: The fault voltage data point to be tested is regarded as a vector containing several features. The features in the vector are subtracted from the posterior mean of the category respectively. Each element obtained is divided by the posterior variance of the category and substituted into the Epanechnikov kernel function as an independent variable. The new vector is averaged to obtain the probability density value of the fault sample to be tested under this category. Calculate the inverse of the square root of the determinant of the posterior variance of the category and multiply it by the obtained probability density value to normalize the probability value; The posterior probability density value of the fault voltage data point to be tested under each fault category is calculated according to the Bayesian theorem, and the fault voltage data point to be tested is classified into the category with the highest probability density value.

2. The fault diagnosis method according to claim 1, characterized in that: The selecting the voltage of the modular tearing node as the diagnostic signal includes: Modular tearing is performed on large-scale circuits, and the tearing must be done on the premise of limited accessible nodes. The limited accessible nodes are torn into modules, and the voltage of the torn nodes under each fault mode is selected as the diagnostic signal to obtain the fault feature vector.

3. The fault diagnosis method according to claim 1 or 2, characterized in that: The initial sample set consisting of the voltages of the torn nodes under different faults uses the inverse generator of the cloud model to generate the digital feature expectation, entropy and super entropy of the cloud under each fault, including: Let the initial circuit fault sample set x ij =[x i1 ,x i2 ,…,x in ](i=1,2,…n;j=1,2,…m) represents the jth type of fault of the i-th fault feature, n is the dimension of the feature vector under the fault mode, and m is the total number of fault modes, then Get the mean of the initial sample set of the jth type fault Depend on The first-order central moment K of the initial sample set of type j fault is obtained by Get the variance S of the initial sample set of the j-th type fault; Since the degree of certainty is unknown, the reverse cloud generator CG without certainty is used. -1 Get the digital feature expectation E of the cloud model x , Entropy E n 、Super entropy H e for:

4. The fault diagnosis method according to claim 3, characterized in that: The forward cloud generator is used to obtain a number of cloud droplets under each fault condition, including: In each type of failure cloud {E x ,E n ,H e }, the number of cloud droplets Num is set, and the forward cloud model CG(E x ,E n ,H e ,Num), first for the jth type of fault, generate E n For expectations, is a normal random number with variance E' ni ; Using the generated random number, generate x For expectations, Generate normal random numbers with variance x k , and calculate x k Membership Among them, u with certainty k x k Represents a cloud droplet in the universe of discourse; Repeat the above steps until the set number of cloud droplets Num is generated, and obtain the cloud sample set drop(x k ,u k ).

5. The fault diagnosis method according to claim 4, characterized in that: The standard cloud model is used to calculate the membership values ​​of each voltage value and the corresponding fault state under the initial sample set to form a membership matrix, including: The cloud model is used to describe the membership of each fault feature vector corresponding to each fault type. The standard cloud model constructed under each fault is used to convert the initial fault sample x ij The standard cloud parameter corresponding to the j-th fault of the i-th eigenvector is substituted into Get x ij The corresponding membership degree, where u ij is the membership degree of the j-th type of fault of the i-th eigenvector of the initial fault sample, E nij ′ is the entropy E of the j-th fault of the i-th eigenvector nij and super entropy H eij 2 The generated normal random number, E xij is the numerical feature expectation of the j-th type of fault of the i-th feature vector; After obtaining the membership of various faults of the initial fault sample under all eigenvectors, the membership moment is obtained: m represents the total number of failure modes; Use fuzzy comprehensive evaluation uncertainty index to form a fuzzy vector, Obtain the degree of membership of uncertainty, n is the fuzzy comprehensive evaluation vector (u i1 ,u i2 ,…,u in ), the number of membership degrees of i=1,2,…n, β is (u i1 ,u i2 ,…,u in ), γ is the maximum membership in (u i1 ,u i2 ,…,u in ) in the second largest membership; Calculate the uncertain membership degree under each fault under the membership degree and construct the uncertain membership fuzzy feature vector Represents the uncertain membership of the i-th eigenvector, and obtains the membership matrix of m rows and n+1 columns:

6. A large-scale circuit fault diagnosis system based on the Bayesian model of Epanechnikov kernel density estimation based on cloud theory, characterized in that: include: The circuit acquisition module to be diagnosed is used to build a large-scale circuit schematic, select the voltage of the modular tear node as the diagnostic signal, encode and classify the fault category according to the different faults of the components, and collect the voltage of the tear node under different faults to form the initial sample set; The cloud theory calculation module is used to generate the expected digital features, entropy, and hyperentropy of the cloud under each fault using the inverse cloud model generator from the initial sample set of voltages of the torn nodes under different faults. Based on the expected digital features, entropy, and hyperentropy of the cloud under each fault, the forward cloud generator is used to obtain a number of cloud droplets under each fault to restore the voltage values ​​of many measurable nodes. The membership expansion sample module is used to use the standard cloud model to calculate the membership values ​​of each voltage value and the corresponding fault state under the initial sample set to form a membership matrix. The cloud droplet under each fault voltage value is determined according to the fault membership matrix to expand the initial sample set; The fault diagnosis module is used to divide the expanded initial sample set into several folds for cross-validation, calculate the posterior probability density value for each fault category in the test set using the Bayesian model of Epanechnikov kernel density estimation, and classify the newly acquired test samples into the category with the highest posterior probability density value to obtain the diagnosis result; The expanded initial sample set is divided into several folds for cross-validation, and the posterior probability density value of each fault category in the test set is calculated using the Bayesian model of Epanechnikov kernel density estimation, including: The expanded initial sample set is divided into a training set and a test set by cross-validation; In the Bayesian model, the likelihood function of the Epanechnikov kernel density estimate can be obtained by using several independent and identically distributed samples in the test set. According to the Bayesian theorem, the posterior distribution of the likelihood function of the Epanechnikov kernel density estimate is obtained, and then the posterior probability density value under each fault category is obtained; The newly acquired test sample is classified into the category with the highest posterior probability density value to obtain a diagnosis result, including: The Bayesian model of Epanechnikov kernel density estimation is used to classify fault data. The voltage data points of the fault to be tested are regarded as a vector, and the probability density function of each fault category is estimated using the Epanechnikov kernel density estimation method. Project the voltage data points of the fault to be tested onto the probability density function of each fault category, calculate the probability density value under each category, and classify the voltage data points of the fault to be tested into the category with the highest probability density value; The method of projecting the voltage data points of the fault to be tested onto the probability density function of each fault category, calculating the probability density value under each category, and classifying the voltage data points of the fault to be tested into the category with the highest probability density value includes: The fault voltage data point to be tested is regarded as a vector containing several features. The features in the vector are subtracted from the posterior mean of the category respectively. Each element obtained is divided by the posterior variance of the category and substituted into the Epanechnikov kernel function as an independent variable. The new vector is averaged to obtain the probability density value of the fault sample to be tested under this category. Calculate the inverse of the square root of the determinant of the posterior variance of the category and multiply it by the obtained probability density value to normalize the probability value; The posterior probability density value of the fault voltage data point to be tested under each fault category is calculated according to the Bayesian theorem, and the fault voltage data point to be tested is classified into the category with the highest probability density value.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images