A fault diagnosis method and device based on knowledge extraction

Abstract

The present application discloses a fault diagnosis method and device based on knowledge extraction, which relates to the field of fault diagnosis technology. The method comprises: extracting knowledge from a fault sample set of a target device to obtain a plurality of compact confidence rules; each sample in the fault sample set includes a characteristic value of each fault feature; constructing a fault diagnosis model according to each of the compact confidence rules; optimizing the fault diagnosis model with the goal of maximizing the accuracy of fault diagnosis; and using the optimized fault diagnosis model to perform fault diagnosis on the target device. The present application can improve the readability of the fault diagnosis model.

Description

Technical Field

[0001] The present application relates to the technical field of fault diagnosis, and in particular to a fault diagnosis method and device based on knowledge extraction. Background Art

[0002] Fault diagnosis of equipment can timely discover existing faults, assist in making maintenance decisions, and ensure the reliable and safe operation of equipment. Among them, the most core fault diagnosis model is built based on fault knowledge such as fault data, historical information, expert experience and other information, which is an important basis for fault diagnosis. At present, there are many studies that include a large number of fault diagnosis models that extract fault knowledge and convert it into rule form. The expression form of rules conforms to the human thinking mode and has the characteristic of being explainable. Among the rule-based methods, the Belief Rule Base (BRB) stands out in the field of fault diagnosis as a modeling method that can integrate semi-quantitative information.

[0003] In the BRB modeling method, the framework of the fault diagnosis model, i.e., the rule antecedent, can be determined by the given fault feature reference value; then the rule weight, fault feature weight, and confidence in the rule consequent are determined by using the fault knowledge to complete the construction of the BRB fault diagnosis model. When the fault feature is input into the fault diagnosis model, the diagnosis result is generated by converting the fault feature, calculating the activation weight of the rule, and fusing the activation rule using the Evidential Reasoning (ER) algorithm. However, in the model framework determined by the BRB modeling method, the number of rules will show an exponential growth trend as the fault features increase. The large-scale BRB brought about by this, on the one hand, the large number of rules leads to poor readability of the model, and as the number of rules increases, the number of optimization parameters will surge, which will bring a burden to the optimization algorithm; on the other hand, there may be redundant rules in the added rules, which may cause over-optimization problems during the optimization process and weaken the interpretability of the model.

[0004] In existing research, the study on BRB scale mainly focuses on changing the connection method of rules, optimizing the structure of the model, and reducing the number of premise attributes, but the scale of BRB still needs to be optimized. Summary of the invention

[0005] The purpose of this application is to provide a fault diagnosis method and device based on knowledge extraction, which can improve the readability of the fault diagnosis model.

[0006] To achieve the above objectives, this application provides the following solutions:

[0007] In a first aspect, the present application provides a fault diagnosis method based on knowledge extraction, comprising:

[0008] Performing knowledge extraction on the fault sample set of the target device to obtain multiple compact confidence rules; each sample in the fault sample set includes the characteristic value of each fault feature; the compact confidence rules include boundary rules, center rules and isolation rules; the boundary rules are confidence rules generated by samples selected according to the maximum characteristic value of each fault feature and the minimum characteristic value of the fault feature in the fault sample set; the center rules are confidence rules generated by clustering the fault sample set and the samples corresponding to each cluster obtained by clustering; the boundary rules and the center rules constitute a rule base; the isolation rules are confidence rules generated by performing fault diagnosis on samples in the fault sample set through the rule base and the samples diagnosed incorrectly;

[0009] Constructing a fault diagnosis model according to each of the compact confidence rules;

[0010] Optimizing the fault diagnosis model with the goal of maximizing the accuracy of fault diagnosis;

[0011] The optimized fault diagnosis model is used to perform fault diagnosis on the target device.

[0012] In a second aspect, the present application provides a fault diagnosis device based on knowledge extraction, the fault diagnosis device based on knowledge extraction comprising:

[0013] A compact confidence rule extraction module is used to extract knowledge from a fault sample set of a target device to obtain a plurality of compact confidence rules; each sample in the fault sample set includes a characteristic value of each fault feature; the compact confidence rules include boundary rules, center rules and isolation rules; the boundary rules are confidence rules generated by samples selected according to the maximum characteristic value of each fault feature and the minimum characteristic value of each fault feature in the fault sample set; the center rules are confidence rules generated by clustering the fault sample set and corresponding to samples of each cluster obtained by clustering; the boundary rules and the center rules constitute a rule base; the isolation rules are confidence rules generated by performing fault diagnosis on samples in the fault sample set through the rule base and based on samples with incorrect diagnosis;

[0014] A fault diagnosis model building module, used to build a fault diagnosis model according to each of the compact confidence rules;

[0015] A fault diagnosis model optimization module, used to optimize the fault diagnosis model with the goal of maximizing the accuracy of fault diagnosis;

[0016] The optimized fault diagnosis model application module is used to perform fault diagnosis on the target device using the optimized fault diagnosis model.

[0017] According to the specific embodiments provided by the present application, the present application discloses the following technical effects: the present application provides a fault diagnosis method and device based on knowledge extraction, which extracts knowledge from a fault sample set to obtain a fault diagnosis model composed of boundary rules, central rules and isolation rules, thereby reducing the number of confidence rules for constructing the fault diagnosis model and realizing compact belief rule modeling (CBRM), thereby improving the readability of the fault diagnosis model. At the same time, knowledge extraction is used to solve the problem that the number of rules in the confidence rule base modeling method grows exponentially with the number of fault features, thereby reducing the burden of later model optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 This is an application environment diagram of a fault diagnosis method based on knowledge extraction in an embodiment of the present application;

[0020] Figure 2 A schematic diagram of a flow chart of a fault diagnosis method based on knowledge extraction provided in an embodiment of the present application;

[0021] Figure 3 A schematic diagram of a compact confidence rule modeling method provided in one embodiment of the present application;

[0022] Figure 4 A schematic diagram of rule antecedent distribution for compact confidence rule modeling provided in an embodiment of the present application;

[0023] Figure 5 A schematic diagram of relay fault diagnosis results provided in an embodiment of the present application;

[0024] Figure 6 A schematic diagram of functional modules of a fault diagnosis device based on knowledge extraction provided in one embodiment of the present application. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0026] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0027] The fault diagnosis method based on knowledge extraction provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be set up separately, integrated on the server 104, or placed on the cloud or other servers. The terminal 102 can send the sample to be diagnosed of the target device to the server 104. After the server 104 receives the sample to be diagnosed, for the sample to be diagnosed, the server 104 extracts knowledge from the fault sample set of the target device to obtain a plurality of compact confidence rules; each sample in the fault sample set includes the characteristic value of each fault feature; the compact confidence rules include boundary rules, center rules and isolation rules; with the goal of maximizing the accuracy of fault diagnosis, the fault diagnosis model is optimized; the target device is fault diagnosed using the optimized fault diagnosis model. The server 104 can feed back the obtained fault diagnosis results for the sample to be diagnosed to the terminal 102. In addition, in some embodiments, the knowledge extraction-based fault diagnosis method can also be implemented independently by the server 104 or the terminal 102. For example, the terminal 102 can directly perform knowledge extraction-based fault diagnosis on the sample to be diagnosed, or the server 104 can obtain the sample to be diagnosed from the data storage system and perform knowledge extraction-based fault diagnosis on the sample to be diagnosed.

[0028] The terminal 102 may be, but is not limited to, various desktop computers, laptop computers, smart phones, tablet computers, IoT devices, and portable wearable devices. The IoT devices may be smart speakers, smart TVs, smart air conditioners, smart vehicle-mounted devices, etc. The portable wearable devices may be smart watches, smart bracelets, head-mounted devices, etc. The server 104 may be implemented as an independent server or a server cluster consisting of multiple servers, or may be a cloud server.

[0029] In an exemplary embodiment, Figure 2 As shown, a fault diagnosis method based on knowledge extraction is provided. The method is executed by a computer device, and can be executed by a computer device such as a terminal or a server alone, or by a terminal and a server together. In the embodiment of the present application, the method is applied to Figure 1 The server 104 in the example is used for explanation, and the steps include the following steps 201 to 204.

[0030] Step 201: perform knowledge extraction on the fault sample set of the target device to obtain multiple compact confidence rules; each sample in the fault sample set includes the characteristic value of each fault feature; the compact confidence rules include boundary rules, center rules and isolation rules; the boundary rules are confidence rules generated by samples selected according to the maximum characteristic value of each fault feature and the minimum characteristic value of the fault feature in the fault sample set; the center rules are confidence rules generated by clustering the fault sample set and corresponding to the samples obtained by clustering; the boundary rules and the center rules constitute a rule base; the isolation rules are confidence rules generated by performing fault diagnosis on the samples in the fault sample set through the rule base and based on the samples with incorrect diagnosis.

[0031] Step 202: construct a fault diagnosis model according to each of the compact confidence rules.

[0032] Step 203: Optimizing the fault diagnosis model with the goal of maximizing the accuracy of fault diagnosis.

[0033] Step 204: Use the optimized fault diagnosis model to perform fault diagnosis on the target device.

[0034] This application improves the readability of the fault diagnosis model by extracting knowledge from the fault sample set and reducing the number of confidence rules. At the same time, knowledge extraction solves the problem that the number of rules in the confidence rule base modeling method grows exponentially with the number of fault features, thereby reducing the burden of subsequent model optimization.

[0035] In step 201 of the present application, the confidence rules extracted through knowledge extraction are compact confidence rules, wherein the isolation rules are confidence rules generated based on the samples in the fault sample set for fault diagnosis through the rule base and the samples with incorrect diagnosis in the fault diagnosis, such as Figure 3 As shown, the present application abandons the rule generation method based on Cartesian product, and uses key nodes in the system to generate compact confidence rules to complete the extraction of fault diagnosis knowledge.

[0036] This application mainly includes the conversion of fault features, the calculation of activation weights and the fusion of evidence-based reasoning algorithms; secondly, the existing fault knowledge is used to extract compact confidence rules, and the extracted compact confidence rules are divided into boundary rules, center rules and isolation rules; then, the projection covariance matrix adaptive evolutionary algorithm is used to optimize and update the parameters of the constructed fault diagnosis model.

[0037] In the fault diagnosis model of this application k Compact Confidence Rules It is expressed as: .

[0038] in, Indicates Fault characteristics, M is the number of fault characteristic parameters, Indicated in middle The reference value of Indicates Failure modes, Indicated in middle The confidence level, Indicates The weight of each fault feature.

[0039] In this application, each confidence rule is considered to be equally important, so the initial confidence rule weight is 1; similarly, the weight of the fault feature is also 1. Different from the BRB modeling method, the fault feature reference value of the CBRM modeling method is Fixed values ​​are no longer given initially, and there is no need to arrange the generation rules in the form of Cartesian products. Instead, they are generated by knowledge extraction.

[0040] The present application is directed to the form of confidence rules, and the activation weight of the confidence rules is calculated based on the hash algorithm. The distance between the input sample data to be diagnosed and the confidence rules is hashed to generate a binary hash value, and different hash values ​​represent different spaces. From different spaces, the nearest neighbor rule is activated to participate in the result fusion process. Among them, step 204 specifically includes steps 2041-2047.

[0041] Step 2041: For a confidence rule in the fault diagnosis model, calculate a first distance between each fault feature value in the to-be-diagnosed sample of the target device and a corresponding fault mode reference value in the confidence rule.

[0042] The first distance is expressed as: ,in, for and The distance between Samples to be diagnosed p Middle Fault characteristics, For the k The compact confidence rule Fault characteristic reference value, For the The compact confidence rule Fault characteristic reference value.

[0043] Step 2042: Hash each of the first distances to obtain a hash binary code of the confidence rule, specifically, Hash processing is performed to obtain the hash binary code of the space where the confidence rule is located.

[0044] The corresponding hash binary encoding is represented as: .

[0045] in, for The hash binary encoding of The hash binary encoding of It can be seen that each confidence rule generates a corresponding hash binary code.

[0046] Step 2043: Use the confidence rules of the same hash binary code as the confidence rules of the same space to obtain the confidence rules of multiple spaces.

[0047] Step 2044: assemble the confidence rules in each space that are closest to the sample to be diagnosed into an activation rule set.

[0048] Step 2045: Calculate the matching degree between the sample to be diagnosed and the fault feature reference value of the confidence rule in the activated rule set.

[0049] Step 2046: Calculate the activation weight of the corresponding confidence rule according to each matching degree to obtain the activated confidence rule.

[0050] Step 2047: Generate a diagnosis result by integrating the activated confidence rules using the evidence reasoning algorithm.

[0051] The calculation formula for calculating the matching degree between the sample to be diagnosed and the fault feature reference value of the confidence rule in the activated rule set is: , .

[0052] in, For the k Confidence rules and samples to be diagnosed p The matching degree, Samples to be diagnosed p and the fault feature reference vector k The distance between Samples to be diagnosed p and the fault feature reference vector n The distance between For the k Confidence rules, For the n Confidence rules, For the activation rule set, K is the number of confidence rules, M is the number of fault features, Samples to be diagnosed p, is the fault feature reference vector k , Samples to be diagnosed p Middle j Fault characteristics, for middle The reference value of for and The distance between For the j The fault feature weight of each fault feature.

[0053] The diagnostic results are generated by integrating the activated confidence rules using the evidence reasoning algorithm, including: p The generated diagnostic results are expressed as: .

[0054] in, For the t Failure modes, , T is the number of failure modes, Samples to be diagnosed p The diagnostic results generated Confidence of occurrence.

[0055] ;

[0056] ;

[0057] in, Samples to be diagnosed p The diagnostic results generated The confidence level of occurrence, for The rule activation weights are for The rule weight, for The rule weight, For the t Confidence rules, for middle The confidence level, for middle The confidence level, For the t Failure modes, For the i Failure modes, is the number of activated confidence rules.

[0058] This application is based on the fault sample set of the target device Perform knowledge extraction to obtain multiple compact confidence rules. ,The confidence rules obtained here are used as compact confidence rules.,Knowledge extraction is performed on the fault sample set of the target device, including: boundary rule extraction, center rule extraction and isolation rule extraction,,and three compact confidence rules are obtained.,The rules are no longer arranged in the form of Cartesian product,,which reduces the number of confidence rules. For sample p .

[0059] Among them, the rules generated by the sample points corresponding to the maximum eigenvalue and the minimum eigenvalue of each fault feature and the nearby sample points are called boundary rules. The boundary rule extraction specifically includes: 1) extracting the samples corresponding to the maximum eigenvalue of each fault feature in the fault sample set, the set number of samples closest to the sample corresponding to the maximum eigenvalue, the sample corresponding to the minimum eigenvalue and the set number of samples closest to the sample corresponding to the minimum eigenvalue to obtain boundary samples.

[0060] 2) Generate confidence rules according to each boundary sample, define the confidence rules generated by each boundary sample as boundary rules, and determine the fault feature reference value and fault mode confidence of each boundary rule according to the extracted boundary samples.

[0061] In an exemplary embodiment, the first attribute For example, the attribute is the fault feature, and its corresponding boundary rule Depends on data First, we need to determine the attributes in the sample The location of the maximum and minimum values ​​and their nearest Sample points.

[0062] ,in, and are all sample sets for generating boundary rules. express The sample point number where the maximum eigenvalue of express The number of the sample point where the minimum eigenvalue of is located. It is a Kd tree function, and its search strategy is similar to the divide-and-conquer method. It divides the sample space into multiple dimensions and can quickly index the sample points that are closest to the target point. , Respectively represent the distance samples searched in the sample space ,sample Recent A sample set, For the The feature value of the first attribute in the samples.

[0063] According to the sample set and Determine the antecedents of the boundary rule, i.e. the reference values ​​of the fault characteristics , the specific formula is: .

[0064] in, Representing a collection The cardinality, that is, the number of elements it contains, For the Samples The i The characteristic value of the fault feature. The rule consequence corresponding to the rule antecedent, that is, the confidence of the classification level , needs to be further calculated based on the sample set. In the process of determining the consequent of the boundary rule, it is necessary to comprehensively consider the sample set label and the number of times the label appears.

[0065] The calculation formula of the rule consequence, that is, the confidence, is: , .

[0066] in, Representation sample The sample labels of is the boundary rule antecedent vector and The normalized distance between For the sample set The label is A collection of samples.

[0067] The calculation formula of normalized distance is: .

[0068] in, is a constant, specifically a very small positive number, to prevent the meaningless situation of 0 appearing in the denominator. The difference between the maximum eigenvalue and the minimum eigenvalue of the fault feature in the sample set of the denominator is used to normalize the attribute sample value, uniformly scale the data to the standard range, and make the calculation process more reasonable.

[0069] The central rule extraction specifically includes: 1) dividing the fault sample set into multiple fault mode sample sets according to the fault labels of each sample in the fault sample set, and the fault labels of the samples in each fault mode sample set are the same; each sample in the fault sample set corresponds to a fault label.

[0070] 2) For each fault mode sample set, a density-based noise application spatial clustering method is used to cluster the fault mode sample set to obtain multiple clusters.

[0071] 3) Generate a confidence rule based on each cluster, and define the confidence rules generated by each cluster as central rules; the fault feature reference value and fault mode confidence of each central rule are generated based on the samples in the corresponding cluster.

[0072] In an exemplary embodiment, for a class Sample set , using the density-based spatial clustering of applications with noise (DBSCAN) method to perform clustering and obtain the clusters within the class Clusters and their border points (border), core points (core) and noise points (outlier).

[0073] First Taking the cluster generation center rule as an example, the center rule is generated using the core point information in the cluster. The specific process is as follows: The core point set of the cluster is , calculate the center point of the core point set to obtain the antecedent of the rule, that is, use The calculation formula solves the reference value of the fault characteristics of the rule, Indicates the core point, is the core sample. The confidence of the consequent of this rule is: .

[0074] Using this class Cluster data point generation The remaining classes are clustered in the same way to obtain the corresponding central rules. The remaining classes are clustered in turn to obtain the corresponding central rules. The calculated central rules are added to the rule base.

[0075] The isolation rule is a confidence rule that has an isolation rule scope to improve the model accuracy. In the fault feature definition domain, the sample range that the rule can participate in the fusion process is the rule scope.

[0076] For fault diagnosis, modeling using only boundary rules and center rules cannot achieve accurate classification requirements, especially for the classification of sample points near the decision boundary. Rules on the decision boundary, such as points near the boundary between two categories, are difficult to distinguish only by boundary rules and centers. Isolation rules can effectively supplement the initial fault diagnosis model and improve the accuracy of the fault diagnosis model.

[0077] The isolation rule extraction specifically includes: 1) dividing the fault sample set into a plurality of sample subsets according to the fault label, each of the sample subsets corresponding to a fault label.

[0078] 2) Performing fault diagnosis on samples in each of the sample subsets according to the rule base to obtain samples with fault diagnosis errors in each of the sample subsets.

[0079] 3) For the fault diagnosis error samples corresponding to each sample subset, classify them according to the fault mode of the error diagnosis to obtain at least one first error classification sample set.

[0080] 4) For each first misclassified sample set corresponding to a sample subset, a confidence rule of a fault mode corresponding to the first misclassified sample set with an importance level is selected from the rule base, and the selected confidence rules constitute a rule set of the fault mode; the importance level is the level of the maximum confidence distribution.

[0081] 5) constructing isolation rules according to each first error classification sample set and the corresponding rule set respectively, specifically comprising: for each confidence rule in the rule set, calculating the hash binary code value of the confidence rule and each sample in the first error classification sample set according to the second distance between each fault mode reference value in the confidence rule and each fault feature value in each sample in the first error classification sample set.

[0082] The samples corresponding to the same hash binary code in the hash binary codes generated by the second distance are taken as samples of the same space.

[0083] Traverse the samples in each space. If the number of samples in the currently traversed space is greater than the set threshold, generate a confidence rule based on the samples in the traversed space, and define the confidence rule generated by the samples in the currently traversed space as the isolation rule. Otherwise, traverse the next space until the traversal is completed.

[0084] In an exemplary embodiment, the specific generation process of the isolation rule includes: judging the sample point set that needs to be isolated based on the inference results of the boundary rule and the center rule. Here, the rule base composed of the boundary rule and the center rule is used to infer the samples to obtain the identification results of each category, which may include correctly classified samples and incorrectly classified samples. For example, the samples of this class can be divided into the correctly classified sample set and misclassified samples , the specific expression is: .

[0085] in, represents the recognition result obtained by inference of a compact rule model containing only boundary rules and center rules. The wrongly classified samples may include samples that are misclassified into other categories, so Further subdivision can get the wrong classification The first misclassified sample set , the specific form is:

[0086] .

[0087] Misclassified to Sample Set The generated isolation rule process is used as an example for analysis. First, we need to find out the important level corresponding to the existing rules. Ruleset , as shown below.

[0088] .

[0089] in Indicates the correspondence of importance levels in boundary rules and center rules The number of rules, Represents the rule base composed of boundary rules and central rules. Construct isolation rules for these rules and misclassified samples respectively.

[0090] In the confidence distribution, the importance level is the level of the maximum confidence distribution, which is the output category in the classification process, expressed as: , express The position of the important level, Indicated in middle The confidence level, For the failure mode.

[0091] To determine the Take the isolation rule as an example. To determine the antecedent of the isolation rule, we first need to determine Which samples have the fusion diagnosis results? information, that is On which samples did it work? According to the calculation method of the activation rule based on hash binary, the label can be calculated as by The second misclassified sample set affected As shown in the following formula.

[0092] .

[0093] in, Representation sample The activated rule set. Sample set generation In the sample set, for The isolation rules further require calculation The samples in Therefore, the space determination method in the rule-based activation method is used to calculate , and get samples of different binary spaces, expressed as: .

[0094] in, express Relative to The hash binary space of is numbered as Each space generates at most one isolation rule. space as an example, its corresponding binary code is , for The mth value in the space. Only when any space The number of samples in Greater than the specified threshold When the isolation rule is generated, it is necessary to judge the symbol of the binary code of the space. lie in The location of the sample. and Set isolation rules between them, and the antecedents of the isolation rules are set as follows: , , .

[0095] in, Represents the mth fault feature value in the sample.

[0096] Based on the existing rule antecedents, use the kd tree to search for samples corresponding to the rule antecedents Nearby point , using the confidence calculated above The formula is based on the labels of the adjacent points and the The distance generates the final consequence of the isolation rule.

[0097] Wherein, step 203 specifically includes: optimizing the fault diagnosis model based on a projection covariance matrix adaptive evolutionary algorithm with the goal of maximizing the accuracy of fault diagnosis.

[0098] The objective function of optimizing the fault diagnosis model is expressed as: .

[0099] in, It represents the accuracy of fault diagnosis. The rule weight and fault feature weight need to satisfy the constraint of not less than 0 and not greater than 1. The confidence needs to satisfy the constraint of not less than 0 and not greater than 1, and the sum of the confidence of each rule needs to satisfy the constraint of not greater than 1.

[0100] The projection covariance matrix adaptive evolutionary algorithm introduces projection operation on the basis of the covariance matrix evolutionary algorithm to meet the equality constraint. It is an intelligent evolutionary algorithm with good performance in solving high-dimensional nonlinear optimization problems. Therefore, this application adopts the projection covariance matrix adaptive evolutionary algorithm to optimize the parameters of the fault diagnosis model.

[0101] The target device is a relay, and the fault characteristics of the relay include pull-in voltage (PV), pull-in time (PT), resistance of no contact (RNC) and resistance of contact (RC). The fault modes of the relay for fault diagnosis include contact wear (CW), adhesion of normally open contact (ANOC), adhesion of normally closed contact (ANCC), loose spring (LS) and coil fault (CF). At the same time, the rule consequence includes the normal working (N) state. Therefore, the input and output of the compact confidence rule fault diagnosis model of the relay are clarified.

[0102] According to the above knowledge extraction algorithm, compact confidence rules that characterize the mapping relationship between fault features and fault modes are extracted. Based on the sample data of relay faults and normal conditions, 46 rules can be extracted, including 8 boundary rules, 6 center rules, and 32 isolation rules. The number of reference values ​​of fault features is shown in Table 1. The connection of fault feature reference values ​​in the rule antecedent is shown in Figure 4 According to the proposed hash activation rule calculation method and evidence reasoning algorithm, the fault feature input to the fault category is determined, and the reasoning result of the initial CBRM is given.

[0103] Table 1 Number of reference values ​​of fault characteristics

[0104]

[0105] The accuracy of the initial CBRM containing 46 rules is 91.78%. The relay fault diagnosis results based on the initial CBRM are as follows: Figure 5As shown in the figure, the missed alarm rate MAR is 1.07%, which is the ratio of the number of samples diagnosed as normal for faulty samples to the total number of faulty samples; the false alarm rate FAR is 14.33%, which is the ratio of the number of samples diagnosed as faulty for normal samples to the total number of normal samples; the detection rate FDR is 93%, which is the ratio of the number of samples diagnosed as faulty samples to the total number of faulty samples.

[0106] On the basis of the initial compact confidence rule model, the proposed objective function and constraints are used, and the projection covariance matrix adaptive evolutionary algorithm is used to adjust the model parameters to further improve the model accuracy. Due to the limited samples of relay failure or normal, in this application, a 5-fold cross-validation method will be used to train and optimize the selection model. First of all, it should be clear that there are 1,800 groups of samples in the relay failure or normal state, 6 categories, and each category contains 300 groups of samples. The samples of each category are randomly divided into 5 groups of samples, and the samples after grouping in other categories are randomly combined to generate 5 groups of samples, each containing 360 groups of samples for cross-validation. Use 4 groups of data in the divided data as training data, and the remaining group as test data to carry out parameter optimization.

[0107] After optimization, the accuracy of the relay fault diagnosis model can reach 98.83%. The diagnosis results of the optimized model are as follows: Figure 5 As shown in the figure, the accuracy of the initial CBRM model is improved by 7.68%, which realizes the effective diagnosis of relays and proves the good engineering application value of the fault diagnosis method based on compact confidence rules and knowledge extraction.

[0108] Based on the same inventive concept, the embodiment of the present application also provides a knowledge extraction-based fault diagnosis device for implementing the above-mentioned knowledge extraction-based fault diagnosis method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above-mentioned method, so the specific limitations in one or more knowledge extraction-based fault diagnosis device embodiments provided below can refer to the limitations of the knowledge extraction-based fault diagnosis method above, and will not be repeated here.

[0109] In an exemplary embodiment, Figure 6 As shown, a fault diagnosis device based on knowledge extraction is provided, comprising:

[0110] A compact confidence rule extraction module is used to extract knowledge from a fault sample set of a target device to obtain multiple compact confidence rules; each sample in the fault sample set includes a characteristic value of each fault feature; the compact confidence rules include boundary rules, center rules and isolation rules; the boundary rules are confidence rules generated by samples selected based on the maximum characteristic value of each fault feature in the fault sample set and the minimum characteristic value of the fault feature; the center rules are confidence rules generated by clustering the fault sample set and corresponding to samples of each cluster obtained by clustering; a rule base is composed of the boundary rules and the center rules; the isolation rules are confidence rules generated by performing fault diagnosis on samples in the fault sample set through the rule base and based on samples with incorrect diagnosis.

[0111] The fault diagnosis model building module is used to build a fault diagnosis model according to each of the compact confidence rules.

[0112] The fault diagnosis model optimization module is used to optimize the fault diagnosis model with the goal of maximizing the accuracy of fault diagnosis.

[0113] The optimized fault diagnosis model application module is used to perform fault diagnosis on the target device using the optimized fault diagnosis model.

[0114] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0115] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A fault diagnosis method based on knowledge extraction, characterized in that: The fault diagnosis method based on knowledge extraction includes: Performing knowledge extraction on the fault sample set of the target device to obtain multiple compact confidence rules; each sample in the fault sample set includes the characteristic value of each fault feature; the compact confidence rules include boundary rules, center rules and isolation rules; the boundary rules are confidence rules generated by samples selected according to the maximum characteristic value of each fault feature and the minimum characteristic value of the fault feature in the fault sample set; the center rules are confidence rules generated by clustering the fault sample set and the samples corresponding to each cluster obtained by clustering; the boundary rules and the center rules constitute a rule base; the isolation rules are confidence rules generated by performing fault diagnosis on samples in the fault sample set through the rule base and the samples diagnosed incorrectly; Constructing a fault diagnosis model according to each of the compact confidence rules; Optimizing the fault diagnosis model with the goal of maximizing the accuracy of fault diagnosis; Performing fault diagnosis on the target device using the optimized fault diagnosis model; The optimized fault diagnosis model is used to perform fault diagnosis on the target device, specifically including: For a confidence rule in the fault diagnosis model, calculating a first distance between each fault feature value in the to-be-diagnosed sample of the target device and a corresponding fault mode reference value in the confidence rule; Performing hash processing on each of the first distances to obtain a hash binary code of the confidence rule; Treat the confidence rules of the same hash binary code as the confidence rules of the same space; The confidence rules closest to the sample to be diagnosed in each space are combined into an activation rule set; Calculating the matching degree between the sample to be diagnosed and the fault feature reference value of the confidence rule in the activation rule set; Calculate the activation weight of the corresponding confidence rule according to each matching degree to obtain the activated confidence rule; The diagnostic results are generated by fusing the activated confidence rules using the evidence reasoning algorithm.

2. The fault diagnosis method based on knowledge extraction according to claim 1 is characterized in that: The calculation formula for calculating the matching degree between the sample to be diagnosed and the fault feature reference value of the confidence rule in the activation rule set is expressed as: ； ； in, For the k Confidence rules and samples to be diagnosed The matching degree, Samples to be diagnosed and the fault feature reference vector The distance between Samples to be diagnosed and the fault feature reference vector The distance between For the k Confidence rules, For the n Confidence rules, For the activation rule set, K is the number of confidence rules, M is the number of fault features, Samples to be diagnosed Middle j Fault characteristics, for middle The reference value of for and The distance between For the j The fault feature weight of each fault feature.

3. The fault diagnosis method based on knowledge extraction according to claim 2 is characterized in that: The diagnostic results are generated by integrating the activated confidence rules using the evidence reasoning algorithm, including: Samples to be diagnosed The generated diagnostic results are expressed as: ; in, For the t Failure modes, , T is the number of failure modes, Samples to be diagnosed The diagnostic results generated Confidence of occurrence; ； ； in, for The rule activation weights are for The rule weight, for The rule weight, For the t Confidence rules, for middle The confidence level, for middle The confidence level, For the t Failure modes, For the i Failure modes, is the number of activated confidence rules, For the k Confidence rules and samples to be diagnosed The matching degree, For the t Confidence rules and samples to be diagnosed degree of matching.

4. The fault diagnosis method based on knowledge extraction according to claim 1 is characterized in that: Knowledge is extracted from the fault sample set of the target device to obtain boundary rules, including: Extract the sample corresponding to the maximum eigenvalue of each fault feature in the fault sample set, the set number of samples closest to the sample corresponding to the maximum eigenvalue, the sample corresponding to the minimum eigenvalue, and the set number of samples closest to the sample corresponding to the minimum eigenvalue to obtain boundary samples; Confidence rules are generated according to each boundary sample, and the confidence rules generated by each boundary sample are defined as boundary rules. Fault feature reference values ​​and fault mode confidences of each boundary rule are determined according to the extracted boundary samples.

5. The fault diagnosis method based on knowledge extraction according to claim 4 is characterized in that: Knowledge is extracted from the fault sample set of the target device to obtain the central rules, including: According to the fault labels of each sample in the fault sample set, the fault sample set is divided into a plurality of fault mode sample sets, wherein the fault labels of the samples in each fault mode sample set are the same; each sample in the fault sample set corresponds to a fault label; For each fault mode sample set, a density-based noise application spatial clustering method is used to cluster the fault mode sample set to obtain multiple clusters; A confidence rule is generated according to each cluster, and the confidence rules generated by each cluster are defined as central rules; the fault feature reference value and fault mode confidence of each central rule are generated according to the samples in the corresponding cluster.

6. The fault diagnosis method based on knowledge extraction according to claim 5 is characterized in that: Knowledge is extracted from the fault sample set of the target device to obtain isolation rules, which include: Dividing the fault sample set into a plurality of sample subsets according to the fault label, each of the sample subsets corresponding to a fault label; Performing fault diagnosis on samples in each of the sample subsets according to the rule base to obtain samples with fault diagnosis errors in each of the sample subsets; For each sample subset, samples with fault diagnosis errors corresponding to the samples are classified according to the fault modes of the incorrect diagnosis to obtain at least one first incorrect classification sample set; For each first misclassified sample set corresponding to a sample subset, a confidence rule of a fault mode corresponding to the first misclassified sample set with an importance level is selected from the rule base, and the selected confidence rules constitute a rule set of the fault mode; the importance level is the level of the maximum confidence distribution; Constructing isolation rules according to each first error classification sample set and the corresponding rule set respectively, specifically including: for each confidence rule in the rule set, calculating the hash binary code value of the confidence rule and each sample in the first error classification sample set according to the second distance between each fault mode reference value in the confidence rule and each fault feature value in each sample in the first error classification sample set; taking the samples corresponding to the same hash binary code in each hash binary code generated by the second distance as samples of the same space; traversing the samples in each space, if the number of samples in the currently traversed space is greater than a set threshold, generating a confidence rule according to the samples in the traversed space, defining the confidence rule generated by the samples in the currently traversed space as the isolation rule, otherwise traversing the next space until the traversal is completed.

7. The fault diagnosis method based on knowledge extraction according to claim 1 is characterized in that: The fault diagnosis model is optimized with the goal of maximizing the accuracy of fault diagnosis, specifically including: Based on the projection covariance matrix adaptive evolutionary algorithm, the fault diagnosis model is optimized with the goal of maximizing the accuracy of fault diagnosis.

8. The fault diagnosis method based on knowledge extraction according to claim 1 is characterized in that: The target device is a relay, the fault characteristics of the relay include pull-in voltage, pull-in time, open point resistance and closed point resistance, and the fault modes of the relay for fault diagnosis include contact wear, normally open contact adhesion, normally closed contact adhesion, spring looseness and coil failure.

9. A fault diagnosis device based on knowledge extraction, characterized in that: The fault diagnosis device based on knowledge extraction comprises: A compact confidence rule extraction module is used to extract knowledge from a fault sample set of a target device to obtain a plurality of compact confidence rules; each sample in the fault sample set includes a characteristic value of each fault feature; the compact confidence rules include boundary rules, center rules and isolation rules; the boundary rules are confidence rules generated by samples selected according to the maximum characteristic value of each fault feature and the minimum characteristic value of each fault feature in the fault sample set; the center rules are confidence rules generated by clustering the fault sample set and corresponding to samples of each cluster obtained by clustering; the boundary rules and the center rules constitute a rule base; the isolation rules are confidence rules generated by performing fault diagnosis on samples in the fault sample set through the rule base and based on samples with incorrect diagnosis; A fault diagnosis model building module, used to build a fault diagnosis model according to each of the compact confidence rules; A fault diagnosis model optimization module, used to optimize the fault diagnosis model with the goal of maximizing the accuracy of fault diagnosis; An optimized fault diagnosis model application module is used to perform fault diagnosis on the target device using the optimized fault diagnosis model; The optimized fault diagnosis model is used to perform fault diagnosis on the target device, specifically including: For a confidence rule in the fault diagnosis model, calculating a first distance between each fault feature value in the to-be-diagnosed sample of the target device and a corresponding fault mode reference value in the confidence rule; Performing hash processing on each of the first distances to obtain a hash binary code of the confidence rule; Treat the confidence rules of the same hash binary code as the confidence rules of the same space; The confidence rules closest to the sample to be diagnosed in each space are combined into an activation rule set; Calculating the matching degree between the sample to be diagnosed and the fault feature reference value of the confidence rule in the activation rule set; Calculate the activation weight of the corresponding confidence rule according to each matching degree to obtain the activated confidence rule; The diagnostic results are generated by fusing the activated confidence rules using the evidence reasoning algorithm.

Images