Fault diagnosis method based on double-layer fusion of multi-feature cloud model probability and boolean quantity

Abstract

The application discloses a fault diagnosis method of multi-feature cloud model probability and Boolean quantity double-level fusion, collects various vibration state signals of equipment which can represent the running state through vibration signals, obtains modeling group data, extracts several features for fault diagnosis from the modeling group data, calculates cloud model parameters of each feature under different equipment vibration states, collects signals to be diagnosed, obtains test group data, maps feature values of the test group data to each state cloud model, calculates probability values, generates a probability matrix, normalizes the probability matrix according to state characteristic weights, obtains a weighted probability vector representing state possibility through weighted summation, constructs a Boolean matrix based on the test group data, forms a Boolean indication vector, introduces a feature screening mechanism, and finally diagnoses the fault state through comprehensive weighted probability vector and Boolean indication vector. The application can jointly output the weighted probability vector and the Boolean indication vector to realize high-reliability fault state diagnosis.

Description

Technical Field

[0001] This invention relates to the field of equipment fault identification, and in particular to a fault diagnosis method that fuses multi-feature cloud model probabilities and Boolean values ​​at two levels. Background Technology

[0002] Efficient and reliable condition monitoring and fault diagnosis technologies have always been a research hotspot and challenge in the field of industrial equipment health management. Currently, mainstream fault diagnosis methods are mainly divided into three categories: fault diagnosis research based on mathematical models, fault diagnosis research based on knowledge, and fault diagnosis research based on data-driven approaches. Although these methods have achieved certain results in their respective application scenarios, a single model cannot simultaneously address the fuzziness and determinism of fault information. Therefore, it is necessary to develop a new fault diagnosis method that can overcome the limitations of existing single models or simple fusion, and achieve deep fusion of probabilistic uncertainty information and Boolean deterministic information from multi-source features. Summary of the Invention

[0003] Purpose of the invention: In order to overcome the shortcomings of the existing technology, the present invention provides a fault diagnosis method that fuses multi-feature cloud model probabilities and Boolean quantities at two levels. It uses an inverse cloud generator to construct cloud model parameters representing different states. During the testing phase, a probability matrix and a Boolean matrix are generated for the signal to be diagnosed. After fusion, a weighted probability vector and a Boolean indicator vector are output. The two vectors are combined to achieve high-reliability fault state diagnosis.

[0004] Technical Solution: To achieve the above objectives, the fault diagnosis method of the present invention, which integrates multi-feature cloud model probability and Boolean quantity at two levels, includes the following steps: S1: For equipment whose operating state can be characterized by vibration signals, various vibration state signals of the equipment are collected to obtain modeling group data, and the modeling group data is preprocessed; S2: Several features for fault diagnosis are extracted from the modeling group data; S3: Cloud model parameters of each feature under different equipment vibration states are calculated using a reverse cloud computing generator; S4: The signal to be diagnosed is collected to obtain test group data, and the test group data is preprocessed and features are extracted; S5: The feature values ​​of the test group data are mapped to each state cloud model, the probability values ​​are calculated, and a probability matrix is ​​generated; S6: The probability matrix is ​​normalized according to the state feature weights, and the weighted sum is obtained to obtain a weighted probability vector representing the possibility of the state; S7: A Boolean matrix is ​​constructed based on the test group data; S8: The number of features that uniquely support each state in the Boolean matrix is ​​counted to form a Boolean indicator vector; S9: A feature screening mechanism is introduced, and the fault state is finally determined by the combined weighted probability vector and the Boolean indicator vector.

[0005] Furthermore, in S2, the equipment vibration state signals in the modeling group data total [number missing]. Classification states, states Total Each sample signal is used for modeling, and each sample signal is calculated. Type of characteristics; for state Then the first The feature vectors obtained from the waveforms of each sample signal are as follows: ;in, Representing state Next, the The first sample signal was calculated to obtain the... One characteristic; for the state The following feature matrix can be obtained:

[0006] ;

[0007] Recombining the above feature array along the columns yields: ;in, , Representing state The first result calculated from all signals used for modeling A set of features.

[0008] Furthermore, in S3, for the first... Several states, using vectors As a data source, calculate the state. The next Cloud model parameters with certain characteristics ,in ; Indicates the first In the first state, the second The expected value of a feature set is given by the formula:

[0009] ;

[0010] Indicates the first In the first state, the second The entropy of a feature set is given by the formula:

[0011] ;

[0012] Indicates the first In the first state, the second The hyperentropy of a feature set is given by the following formula:

[0013] , ;

[0014] in, The second-order center distance;

[0015] So the state corresponding Cloud model parameters with certain characteristics .

[0016] Furthermore, in S4, the diagnostic signal is extracted from the test group data. From these features, we obtain the feature vector: .

[0017] Furthermore, in S5, the membership probability of each feature value in the test group to the probability of the fault state is calculated; for the i-th feature value in the test group... The feature values ​​to be verified for a certain type of feature Its application to the cloud model Characteristics under various fault conditions Membership probability The calculation formula is:

[0018] ;

[0019] Traverse all of the test group All features in the cloud model For each fault state, calculate the characteristic value. For each state Membership probability , generate 3D probability matrix .

[0020] Furthermore, in S6, the probability matrix is... Normalize along the column direction; for the first measurement group... Type of characteristics In its first Normalized weights under various fault conditions The calculation formula is:

[0021] ;

[0022] Then, probability aggregation is performed on each fault state to obtain the comprehensive weighted probability of the state, the th... Weighted probability values ​​of various fault states The calculation formula is: Through probability aggregation at the state level, we obtain The combined weighted probabilities of various fault states form a weighted probability vector. Then, the vectors are normalized to obtain the states to which the test group data belongs. probability vector Its formula is:

[0023] .

[0024] Furthermore, in S7, by defining the confidence interval of the eigenvalues, the confidence interval of the eigenvalues ​​in the test group is calculated. Eigenvalue pair The Boolean values ​​of various fault states are used to construct a Boolean matrix B; the confidence interval width coefficient is defined. For the first The first fault state This feature, whose confidence interval is determined by the expectation of the cloud model. Entropy Confirmed, indicated as: For the first data in the test group The feature values ​​to be detected for each type of feature Determine whether it falls into the first category. Within the confidence interval of various fault states, the Boolean value is determined. If and only if Falling into state When the confidence interval is reached, mark the Boolean value. True indicates that the feature supports this fault state; otherwise, it is marked False. Iterate through all test cases. All features in the cloud model Given a fault state, calculate the Boolean value of each characteristic value for each state. ,generate Boolean matrix of dimension .

[0025] Furthermore, in S8, for Boolean matrices... By performing a logical AND operation on Boolean values ​​with different characteristics under the same state, a one-dimensional Boolean indicator vector is obtained. In S9, the probability vector is based on the output of the probability layer. Boolean indicator vector output by the Boolean layer Determine the final fault state corresponding to the test group.

[0026] Furthermore, in S9, the Boolean indicator vector obtained from this test data is observed. From the elements in the array, determine the values ​​and indices of the elements that are True, and obtain their corresponding probability vectors. The probability value in; there are the following cases: if The number of True value elements is 1, and its corresponding index number is 1. If so, then the test data is considered to be the first. a state; if If the number of True value elements in Y is 0, then the test data is considered to be a fault state that has not occurred before; if the number of True value elements in Y is greater than 1, record the index sequence corresponding to the True value elements. If so, the test data is considered to be One or more of the following states.

[0027] Furthermore, in S9, if the number of True value elements in Y is greater than 1, the final probability vector corresponding to the index number is extracted. The probability values ​​in the vector form the candidate probability subvector. Normalize the candidate probability subvectors within the set to obtain the relative support probability of each candidate state: ;

[0028] Preset probability difference threshold With the threshold of main fault proportion Further classification and determination shall be made according to the following rules; if The highest relative probability exists in it. And satisfy If the test data corresponds to the single fault state with the highest probability, then the remaining candidate states are considered mismatches due to feature fluctuations and are discarded; if The relative probability difference between any two states does not exceed And the maximum relative probability Then it is determined that the test data exists simultaneously. The candidate states correspond to the composite concurrent faults. If the above two conditions are not met, the candidate states are sorted from high to low relative probability, and the top M states with a cumulative probability sum greater than 0.9 are taken as the primary and secondary fault sets for output. The first state is the primary fault, and the rest are secondary faults.

[0029] Beneficial effects: The fault diagnosis method of the present invention, which integrates multi-feature cloud model probability and Boolean quantity at two levels, extracts features in the modeling stage and uses an inverse cloud generator to construct cloud model parameters representing different states. In the testing stage, the signal to be diagnosed is mapped into the cloud model to generate a probability matrix and a Boolean matrix. Then, a decision is made through two-level fusion, and a feature screening mechanism is introduced. Finally, a highly reliable fault state diagnosis is achieved by combining the weighted probability vector and the Boolean indicator vector. This overcomes the problem in the prior art where a single model or simple fusion method cannot simultaneously take into account the ambiguity and certainty of fault information, thus leading to inaccurate fault diagnosis. Attached Figure Description

[0030] Appendix Figure 1 This is a flowchart of the method of the present invention;

[0031] Appendix Figure 2 This is a schematic diagram of the confidence interval. Detailed Implementation

[0032] The invention will now be further described with reference to the accompanying drawings.

[0033] As attached Figures 1 to 2 The fault diagnosis method based on the dual-level fusion of multi-feature cloud model probability and Boolean quantity includes the following steps.

[0034] S1: For equipment whose operating status can be characterized by vibration signals, collect various vibration status signals of the equipment to obtain modeling group data, and preprocess the modeling group data.

[0035] The industrial equipment for fault diagnosis described in this invention is industrial equipment whose operating state can be characterized by vibration signals, such as rolling bearings. Taking rolling bearings as an example, vibration signals of rolling bearings under various conditions are collected and preprocessed.

[0036] S2: Extract several features from the modeling group data for fault diagnosis.

[0037] In S2, the equipment vibration state signals in the modeling group data total [number missing]. Classification states, states Total Each sample signal is used for modeling, and each sample signal is calculated. Features. Taking the vibration signal of a rolling bearing as an example, time-domain, frequency-domain, and time-frequency-domain features are obtained for each sample signal. Then, for the state... Then the first The feature vectors obtained from the waveforms of each sample signal are as follows: .in, Representing state Next, the The first sample signal was calculated to obtain the... One feature; ; ; .

[0038] For state The following feature matrix can be obtained:

[0039] ;

[0040] Recombining the above feature array along the columns yields: .in, , , Representing state The first result calculated from all signals used for modeling A set of features.

[0041] For each state used for modeling, the following feature matrices can be obtained: Feature matrix corresponding to state 1 The feature matrix corresponding to state 2 ;……;state Corresponding feature matrix ;……;state Corresponding feature matrix .

[0042] S3: Utilize a reverse cloud computing generator to calculate cloud model parameters for each feature under different device vibration states, and construct Gaussian distribution cloud maps representing different states.

[0043] In S3, for the first data in the modeling group... Several states, using vectors As a data source, calculate the state. The next Cloud model parameters with certain characteristics ,in , The definitions and calculation formulas for each parameter are as follows.

[0044] Indicates the first In the first state, the second The expectation of a set of features reflects the state of that feature. The average level below is given by the following formula: .

[0045] Indicates the first In the first state, the second The entropy of a feature set reflects the degree of dispersion of that feature value, and its formula is: .

[0046] Indicates the first In the first state, the second The hyperentropy of a set of features reflects entropy. The uncertainty, i.e., the fluctuation in the degree of dispersion, is expressed by the following formula, where... The second-order center distance:

[0047] , .

[0048] By repeating the above steps for different states and features, the following cloud model set can be obtained for each state: State 1 corresponds to Cloud model parameters with certain characteristics State 2 corresponds to Cloud model parameters with certain characteristics ;……;state corresponding Cloud model parameters with certain characteristics ;……;state corresponding Cloud model parameters with certain characteristics .

[0049] S4: Collect the signal to be diagnosed, obtain the test group data, and perform preprocessing and feature extraction on the test group data.

[0050] In S4, the diagnostic signal is extracted from the test group data. From these features, we obtain the feature vector: .

[0051] S5: Map the feature values ​​of the test group data to each state cloud model, calculate the probability values, and generate a probability matrix.

[0052] In S5, based on the cloud model parameter set generated during the modeling phase. Calculate the membership probability of each feature value in the test group to the probability of the fault state. Specifically, for the i-th feature value in the test group... The feature values ​​to be verified for a certain type of feature Its application to the cloud model Characteristics under various fault conditions Membership probability Determined by the Gaussian membership function, its calculation formula is as follows:

[0053] .

[0054] Traverse all of the test group All features in the cloud model For each fault state, calculate the characteristic value. For each state Membership probability ,generate 3D probability matrix ,as follows:

[0055] .

[0056] probability matrix The generation process is as follows:

[0057] .

[0058] S6: Normalize the probability matrix according to the state-based feature weights, and sum the weights to obtain a weighted probability vector representing the likelihood of the state.

[0059] In S6, to eliminate the differences in dimensions or probability distributions of different features, the probability matrix needs to be modified. The probability values ​​of each feature are normalized to obtain the relative contribution weights of each feature to each state. Therefore, the probability matrix is... Normalize along the column direction. For the first measurement in the measurement group... Type of characteristics In its first Normalized weights under various fault conditions The calculation formula is: .

[0060] In the above formula, This is the fault status number. For feature serial number, This represents the total number of states. For characteristic number, ; . It is the first The first feature corresponds to the first Membership probability of each state. (Uppercase) It is a dummy variable; it only appears in the summation symbol "". The formula applies within the range of "" and is used to traverse all fault states. The above formula successfully converts the original probability into a relative contribution weight.

[0061] Based on the feature weight normalization, probability aggregation is performed on each fault state, that is, the normalized weights of all features in that state are summed to obtain the comprehensive weighted probability of the state. Weighted probability values ​​of various fault states The calculation formula is: ;in .

[0062] By probabilistic aggregation at the state level, we obtain The combined weighted probabilities of various fault states form a weighted probability vector. , ,in, It represents the weighted probability value of a single state. The first in One element, The larger the value, the better the test group feature set is for the th element. The higher the support for a particular fault state, the better. The vector is then normalized to obtain the states to which the test group data belongs. probability vector Its formula is: .

[0063] in, The first element of the weighted probability vector represents the... The element, i.e., the th element The combined weighted probability corresponding to each fault state. (Uppercase) It is a dummy variable for summation, also called a traversal dummy variable, which only appears in the summation symbol "". It operates within the scope of the operation and is a temporary traversal counting tool.

[0064] probability vector The generation process is as follows:

[0065] .

[0066] S7: Construct a Boolean matrix based on the test group data.

[0067] In S7, cloud model parameters generated during the modeling phase are used. By defining the confidence interval of the eigenvalues, the confidence interval of the eigenvalues ​​in the test group is calculated. Eigenvalue pair The Boolean values ​​of various fault states are used to construct a Boolean matrix B.

[0068] Specifically, based on the statistical characteristics of the cloud model's feature parameters, the confidence interval width coefficient is first defined. , This is used to quantify the "typical coverage" of characteristic values ​​for fault states. For the first... The first fault state This feature, whose confidence interval is determined by the expectation of the cloud model. Entropy Confirmed, indicated as: That is, when the eigenvalues ​​follow a Gaussian distribution, the region covers approximately... Typical sample, It is the standard normal cumulative distribution function.

[0069] For example, when At this time, the confidence interval is This covers approximately 68.27% of the samples. When When, the confidence interval is This covers approximately 95% of the samples, that is... Figure 2 As shown. When When, the confidence interval is It covers approximately 99.73% of the samples.

[0070] For the first in the test group data The feature values ​​to be detected for each type of feature Determine whether it falls into the first category. Within the confidence interval of various fault states, the Boolean value is determined. If and only if Falling into state When the confidence interval is reached, mark the Boolean value. True indicates that the feature supports this fault state; otherwise, it is marked False. Boundary intervals follow the closed interval principle, i.e. It is still considered a hit at this time. True.

[0071] Traverse all of the test group All features in the cloud model Given a fault state, calculate the Boolean value of each characteristic value for each state. ,generate Boolean matrix of dimension :

[0072] .

[0073] Boolean matrix The generation process is as follows:

[0074] .

[0075] S8: Count the number of unique features supporting each state in the Boolean matrix to form a Boolean indicator vector.

[0076] In S8, the Boolean layer is based on the distribution of the test feature set within the confidence interval of the cloud model, targeting the Boolean matrix. Performing a logical AND operation on Boolean values ​​with different characteristics under the same state yields a one-dimensional Boolean indicator vector. .

[0077] Boolean indicator vector The generation process is as follows:

[0078] .

[0079] S9: Introduce a feature filtering mechanism, and finally diagnose the fault state by combining the weighted probability vector and the Boolean indicator vector.

[0080] In S9, the probability vector is based on the output of the probability layer. Boolean indicator vector output by the Boolean layer The final fault state of the test group is determined through dual-threshold screening and optimal state determination rules. The specific decision-making process is as follows: Observe the Boolean indicator vector obtained from the test data. From the elements in the array, determine the values ​​and indices of the elements that are True, and obtain their corresponding probability vectors. The probability value is given. There are three possible scenarios.

[0081] In the first case, if The number of True value elements is 1, and its corresponding index number is 1. If so, then the test data is considered to be the first. A state.

[0082] In the second scenario, if If the number of True value elements is 0, then the test data is considered to be a fault state that has not occurred before.

[0083] In the third case, if the number of True value elements in Y is greater than 1, record the index sequence corresponding to the True value elements. , , If so, the test data is considered to be One or more of the following states.

[0084] In the third case, extract the final probability vector corresponding to the index number. The probability values ​​in the vector form the candidate probability subvector. Normalize the candidate probability subvectors within the set to obtain the relative support probability of each candidate state: .

[0085] Preset probability difference threshold With the threshold of main fault proportion Further classification and determination shall be made according to the following rules.

[0086] Scenario 1: Single-leader fault determination; if The highest relative probability exists in it. And satisfy If the test data corresponds to the single fault state with the highest probability, then the remaining candidate states are mismatches caused by feature fluctuations and are eliminated.

[0087] Scenario 2: Determining concurrent faults; if The relative probability difference between any two states does not exceed And the maximum relative probability Then it is determined that the test data exists simultaneously. Composite concurrent faults corresponding to various candidate states.

[0088] Scenario 3: Primary and secondary fault determination; if the above two conditions are not met, the candidate states are sorted from high to low relative probability, and the top M states with a cumulative probability sum greater than 0.9 are taken as the primary and secondary fault set for output. The first state is the primary fault, and the rest are secondary faults.

[0089] This invention proposes a device fault diagnosis framework based on two-level fusion decision-making, particularly suitable for rolling bearing fault diagnosis. In the modeling phase, multi-state signals are processed through overlapping framing and time-frequency domain features are extracted. A reverse cloud generator is used to construct cloud model parameters representing different states. In the testing phase, the signals undergo the same preprocessing to generate probability and Boolean matrices. Then, a two-level fusion decision-making process is implemented. The probabilistic fusion layer normalizes feature weights and performs a weighted summation to output a state probability vector; the Boolean fusion layer counts the number of features uniquely supporting each state, forming a Boolean indicator vector. Furthermore, a feature filtering mechanism is introduced, and finally, the weighted probability vector and the Boolean indicator vector are combined to achieve high-reliability fault state diagnosis.

[0090] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A fault diagnosis method that fuses multi-feature cloud model probabilities and Boolean values ​​at two levels, characterized in that: Includes the following steps: S1: For equipment whose operating status can be characterized by vibration signals, collect various vibration status signals of the equipment to obtain modeling group data, and preprocess the modeling group data; S2: Extract several features from the modeling group data for fault diagnosis; S3: Calculate the cloud model parameters of each feature under different device vibration states using a reverse cloud computing generator; S4: Collect the signal to be diagnosed, obtain the test group data, and preprocess and extract features from the test group data; S5: Map the feature values ​​of the test group data to each state cloud model, calculate the probability values, and generate a probability matrix; S6: Normalize the probability matrix according to the feature weights of the state, and sum them up to obtain a weighted probability vector representing the possibility of the state; S7: Construct a Boolean matrix based on the test group data; S8: Count the number of unique features supporting each state in the Boolean matrix to form a Boolean indicator vector; S9: Introduce a feature filtering mechanism, and finally diagnose the fault state by combining the weighted probability vector and the Boolean indicator vector.

2. The fault diagnosis method based on the dual-level fusion of multi-feature cloud model probability and Boolean quantity as described in claim 1, characterized in that: In S2, the equipment vibration state signals in the modeling group data total [number missing]. Classification states, states Total Each sample signal is used for modeling, and each sample signal is calculated. Type of characteristics; For state Then the first The feature vectors obtained from the waveforms of each sample signal are as follows: ； in, Representing state Next, the The first sample signal was calculated to obtain the... One feature; For state The following feature matrix can be obtained: ； Recombining the above feature array along the columns yields: ; in, , Representing state The first result of calculating all signals used for modeling A set of features.

3. The fault diagnosis method based on the dual-level fusion of multi-feature cloud model probability and Boolean quantity as described in claim 2, characterized in that: In S3, for the first Several states, using vectors As a data source, calculate the state. The next Cloud model parameters with certain characteristics ,in ; Indicates the first In the first state, the second The expected value of a feature set is given by the formula: ； Indicates the first In the first state, the second The entropy of a feature set is given by the formula: ； Indicates the first In the first state, the second The hyperentropy of a feature set is given by the following formula: ， ； in, The second-order center distance; So the state corresponding Cloud model parameters with certain characteristics .

4. The fault diagnosis method based on the dual-level fusion of multi-feature cloud model probability and Boolean quantity as described in claim 3, characterized in that: In S4, the diagnostic signal is extracted from the test group data. From these features, we obtain the feature vector: .

5. The fault diagnosis method based on the dual-level fusion of multi-feature cloud model probability and Boolean quantity as described in claim 4, characterized in that: In S5, the membership probability of each feature value in the test group to the probability of the fault state is calculated; for the i-th feature value in the test group... The feature values ​​to be verified for a certain type of feature Its application to the cloud model Characteristics under various fault conditions Membership probability The calculation formula is: ； Traverse all of the test group All features in the cloud model For each fault state, calculate the characteristic value. For each state Membership probability ,generate 3D probability matrix .

6. The fault diagnosis method based on the dual-level fusion of multi-feature cloud model probability and Boolean quantity as described in claim 5, characterized in that: In S6, the probability matrix Normalize along the column direction; for the first measurement group... Type of characteristics In its first Normalized weights under various fault conditions The calculation formula is: ； Then, probability aggregation is performed on each fault state to obtain the comprehensive weighted probability of the state, the th... Weighted probability values ​​of various fault states The calculation formula is: ； By probabilistic aggregation at the state level, we obtain The combined weighted probabilities of various fault states form a weighted probability vector. Then, the vectors are normalized to obtain the states to which the test group data belongs. probability vector Its formula is: 。 7. The fault diagnosis method based on the dual-level fusion of multi-feature cloud model probability and Boolean quantity as described in claim 6, characterized in that: In S7, the confidence intervals of eigenvalues ​​are defined to calculate the first eigenvalue in the test group. Eigenvalue pair The Boolean values ​​of various fault states are used to construct a Boolean matrix B; Define the confidence interval width coefficient For the first The first fault state This feature, whose confidence interval is determined by the expectation of the cloud model. Entropy Confirmed, expressed as: ; For the first in the test group data The feature values ​​to be detected for each type of feature Determine whether it falls into the first category. Within the confidence interval of various fault states, the Boolean value is determined. If and only if Falling into state When the confidence interval is defined, mark the Boolean value. True indicates that the feature supports this fault state; otherwise, it is marked False. Traverse all of the test group All features in the cloud model Given a fault state, calculate the Boolean value of each characteristic value for each state. ,generate Boolean matrix of dimension .

8. The fault diagnosis method based on the dual-level fusion of multi-feature cloud model probability and Boolean quantity as described in claim 7, characterized in that: In S8, for Boolean matrices By performing a logical AND operation on Boolean values ​​with different characteristics under the same state, a one-dimensional Boolean indicator vector is obtained. In S9, the probability vector is based on the output of the probability layer. Boolean indicator vector output by the Boolean layer Determine the final fault status corresponding to the test group.

9. The fault diagnosis method based on the dual-level fusion of multi-feature cloud model probability and Boolean quantity as described in claim 8, characterized in that: In S9, observe the Boolean indicator vector obtained from this test data. From the elements in the array, determine the values ​​and indices of the elements that are True, and obtain their corresponding probability vectors. The probability value in; The following situations exist: like The number of True value elements is 1, and its corresponding index number is 1. If so, then the test data is considered to be the first. a state; like If the number of True value elements is 0, then the test data is considered to be a fault state that has not appeared before. If the number of True value elements in Y is greater than 1, record the index sequence corresponding to the True value elements. If so, then the test data is considered to be One or more of the following states.

10. The fault diagnosis method based on the dual-level fusion of multi-feature cloud model probability and Boolean quantity according to claim 9, characterized in that: In S9, if the number of True values ​​in Y is greater than 1, extract the final probability vector corresponding to the index number. The probability values ​​in the vector form the candidate probability subvector. Normalize the candidate probability subvectors within the set to obtain the relative support probability of each candidate state: ; Preset probability difference threshold With the threshold of main fault proportion Further classification and determination shall be made according to the following rules; like The highest relative probability exists in it. And satisfy If the test data corresponds to the single fault state with the highest probability, the remaining candidate states are mismatches caused by feature fluctuations and are therefore eliminated. like The relative probability difference between any two states does not exceed And the maximum relative probability Then it is determined that the test data exists simultaneously. Composite concurrent faults corresponding to various candidate states; If the above two conditions are not met, the candidate states are sorted from high to low relative probability, and the top M states with a cumulative probability sum greater than 0.9 are taken as the primary and secondary fault sets for output. The first state is the primary fault, and the rest are secondary faults.

Images