Electric valve fault identification method and system based on feature dimension reduction

Abstract

The invention relates to the technical field of equipment health management and reliability engineering, in particular to an electrically operated valve fault identification method and system based on feature dimension reduction, and provides an electrically operated valve fault identification method under variable working conditions based on reliability growth tests, action life tests and the like of electrically operated valves. According to the method, a fusion index capable of representing the state of the electric valve is obtained from monitoring data through a data mining method and a dimensionality reduction and fusion technology, a state classification model of the electric valve is established based on the fusion index, and finally the state of the electric valve is recognized through the final model. And the classification model is adaptively updated according to the new action data. According to the method, the problem that fault features are difficult to identify under multiple working conditions is solved, and the difficulty that the action response data of the electric valve presents multiple modes due to different stresses is effectively overcome.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment health management and reliability engineering, and in particular to a method and system for identifying electric valve faults based on feature dimension reduction. Background Art

[0002] With the innovation and development of science and technology, electric valves are widely used in various systems in the nuclear power, petroleum, chemical and new energy industries. Whether the electric valve can complete the opening and closing action safely and reliably directly affects whether the system can operate normally. For example, in the nuclear power industry, electric valves are widely used in primary and secondary circuit systems. During the operation of the system, once the electric valve cannot complete the opening and closing normally, it will cause the leakage of radioactive materials and cause serious nuclear leakage incidents. In addition, electric valves have the characteristics of high reliability, long life, and small sample, and their working environment is harsh. It is difficult to check their faults in time under high temperature and high pressure conditions. There are the following difficulties in its health monitoring: 1) Health monitoring data for the entire life cycle is difficult to obtain; 2) There are very few negative samples for the entire life cycle, which is not enough to support fault classification and identification.

[0003] As the service life of electric valves increases, the frequency of their failures is also increasing. Therefore, how to establish a fault identification method for electric valves is an urgent problem to be solved. Summary of the invention

[0004] The purpose of the present invention is to provide a method and system for electric valve fault identification based on feature dimension reduction to solve the above problems in the prior art.

[0005] The embodiments of the present invention are implemented by the following technical solutions:

[0006] In a first aspect, the present invention provides an electric valve fault identification method based on feature dimension reduction, characterized in that it includes:

[0007] The original electric valve action monitoring data is obtained, the detection data is preprocessed, and multi-domain feature indicators are extracted to obtain the first feature indicator set of the time domain, frequency domain and time-frequency domain statistics of the preprocessed detection data:

[0008] Performing dimensionality reduction on the feature index set based on a local preservation projection algorithm to obtain a second feature index set;

[0009] The second feature index set after dimension reduction is fused to obtain a third feature index set;

[0010] A state adaptive classification model for electric valves is established through the third characteristic index set and Gaussian mixture model parameters, the original electric valve action monitoring data is divided into a training set and a test set, the state adaptive classification model is trained through the training set, and tested through the test set, and the state adaptive classification model with the best result is output;

[0011] The state of the electric valve is identified through a state adaptive classification model.

[0012] Preferably, the preprocessing of the detection data includes:

[0013] Standardize the test data to the interval [0, 1] or [-1, 1];

[0014]

[0015] Wherein, x is the detection data, min(X) is the minimum value in the detection data, max(X) is the maximum value in the detection data, and x′ is the standardized value.

[0016] Preferably, the dimensionality reduction of the first feature index set based on the local preservation projection algorithm includes:

[0017] Obtain the nearest neighbor of each data point in the first feature index set according to the determined distance metric, and construct an adjacency graph structure;

[0018] The loss function is used to measure the position change of the data points in the low-dimensional space after mapping;

[0019] Optimize the loss function and solve the dimension reduction projection matrix;

[0020] The obtained projection matrix is ​​used to map the original high-dimensional feature index set to a low-dimensional space.

[0021] Preferably, fusing the second feature index set after dimension reduction includes:

[0022] The second characteristic index set is clustered by using a Gaussian mixture model to obtain a probability matrix corresponding to the opening and closing actions of the electric valve;

[0023] The probability matrix is ​​fused based on Jensen-Rényi divergence to obtain the third characteristic index set that can characterize the state of the electric valve.

[0024] Preferably, clustering the second feature indicator set by using a Gaussian mixture model includes:

[0025] Determine the range of the number of clusters, input the data of the second feature index set, randomly select a data object from the second feature index set as the first cluster center, and perform clustering:

[0026]

[0027] In the formula, p(x) is the clustering result, is the Gaussian probability density function in N dimensions, μ k is the mean vector, ∑ k is the covariance matrix, π k is the mixed weight, and K is the number of the second feature index set.

[0028] Preferably, the establishment of a state adaptive classification model for the electric valve includes:

[0029]

[0030] In the formula, is the output result of the model, N is the number of components in the third characteristic index set, j is the jth input layer, k is the kth output layer, x j The weight from the jth input layer to the hidden layer, is the weight from the hidden layer to the k-th output layer, γ is the bias term, and T is the transpose.

[0031] Preferably, the training of the state adaptive classification model by means of a training set comprises:

[0032] The motion monitoring data in the training set is classified based on the state adaptive classification model, and the classification results are compared with the test set to obtain the accuracy used to characterize whether they are consistent;

[0033] Set the accuracy threshold. If the accuracy reaches the accuracy threshold, output the state adaptive classification model. If the accuracy does not reach the accuracy threshold, modify the parameters of the Gaussian mixture model and classify the motion monitoring data in the training set again until the accuracy reaches the accuracy threshold, and output the state adaptive classification model.

[0034] In a second aspect, the present invention further provides an electric valve fault identification system based on feature dimension reduction, comprising:

[0035] The data acquisition module is configured to acquire the original electric valve action monitoring data, preprocess the detection data, and extract multi-domain feature indicators to obtain the first feature indicator set of the time domain, frequency domain and time-frequency domain statistics of the preprocessed detection data:

[0036] The data processing module is configured to reduce the dimension of the feature index set based on the local preservation projection algorithm to obtain a second feature index set; and fuse the second feature index set after the dimension reduction to obtain a third feature index set;

[0037] The model building module is configured to establish a state adaptive classification model of the electric valve through the third feature indicator set and the Gaussian mixture model parameters, divide the original electric valve action monitoring data into a training set and a test set, train the state adaptive classification model through the training set, and test it through the test set, and output the state adaptive classification model with the best result; identify the state of the electric valve through the state adaptive classification model.

[0038] The technical solution of the embodiment of the present invention has at least the following advantages and beneficial effects:

[0039] The present invention mines the monitoring data of the opening and closing status of electric valves to obtain a parameter index set associated with the valve status, and obtains a fusion index by integrating the index set through existing methods, and establishes a mapping relationship between the valve status parameters and the fusion index, thereby realizing the status identification of the electric valve in service. Therefore, when performing status detection and health management on electric valves, the method of the present invention provides support for the subsequent prediction of the remaining life of electric valves, provides auxiliary decision support for the maintenance and guarantee of nuclear-grade electric valves and industrial electric valves, and realizes early warning of electric valve failures.

[0040] In addition, based on the reliability growth test and action life test of electric valves, a method for identifying electric valve faults under variable working conditions is proposed. This method uses data mining methods, dimension reduction and fusion technology to obtain fusion indicators that can characterize the state of electric valves from monitoring data, and establishes a state classification model of electric valves based on this. Finally, the final model is used to identify the state of the electric valve, and the classification model is adaptively updated according to the new action data. This method solves the problem of difficulty in identifying fault characteristics under multiple working conditions, and effectively overcomes the difficulty that the action response data of electric valves presents multi-modality due to different stresses. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0042] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0044] The division of modules in this application is a logical division. There may be other division methods when implemented in actual applications. For example, multiple modules can be combined or integrated into another system, or some features can be ignored or not executed.

[0045] The modules or submodules described independently may be physically separated or not: they may be implemented by software or hardware, and some modules or submodules may be implemented by software, and the processor may call the software to implement the functions of these modules or submodules, and other modules or submodules may be implemented by hardware, such as by hardware circuits. In addition, some or all of the modules may be selected according to actual needs to achieve the purpose of the present application.

[0046] Please refer to Figure 1 The present invention provides an electric valve fault identification method based on feature dimension reduction, which is characterized by comprising:

[0047] S101: Acquire original electric valve action monitoring data, preprocess the detection data, and extract multi-domain feature indicators to obtain a first feature indicator set of time domain, frequency domain, and time-frequency domain statistics of the preprocessed detection data:

[0048] Among them, the first characteristic index set can include 16 time domain characteristic indicators such as mean, peak, kurtosis, margin and root mean square; 13 characteristic indicators such as mean square frequency, frequency variance, mean frequency and first-order frequency centroid; time-frequency domain indicators such as empirical mode decomposition, wavelet transform and local mean decomposition. Since the correlation between the characteristic indicators and the valve status is unknown, it is necessary to extract as many redundant characteristic indicators as possible.

[0049] S102: reducing the dimension of the feature index set based on a local preservation projection algorithm to obtain a second feature index set;

[0050] S103: fusing the second feature index set after dimension reduction to obtain a third feature index set;

[0051] S104: establishing a state adaptive classification model for the electric valve by using the third characteristic index set and the Gaussian mixture model parameters, dividing the original electric valve action monitoring data into a training set and a test set, training the state adaptive classification model by using the training set, and testing it by using the test set, and outputting the state adaptive classification model with the best result;

[0052] The state of the electric valve is identified through a state adaptive classification model.

[0053] S105: Preferably, the preprocessing of the detection data includes:

[0054] Standardize the test data to the interval [0, 1] or [-1, 1];

[0055]

[0056] Wherein, x is the detection data, min(X) is the minimum value in the detection data, max(X) is the maximum value in the detection data, and x′ is the standardized value.

[0057] The monitoring data of the electric valve during operation is standardized, integrated and cleaned. The data is standardized according to the differentiated characteristics of each monitoring parameter under variable working conditions and the differences of different monitoring parameters under the same working conditions. For example, for parameters such as current, it is standardized to the interval [0, 1], and for vibration data, it is normalized to [-1, 1].

[0058] Of course, it also includes using a certain number of historical opening and closing action monitoring data of the electric valve as the training set of the state classification model, eliminating invalid and incomplete data in the training set, and assigning normal opening and closing and fault labels to the data valves in the training set; in order to eliminate the influence of different voltage conditions on the difference in current values, the action current of the electric valve is divided by the operating current of the action, that is, I(t) = I(t) / Irated, and the vibration data is standardized using the Z-score method.

[0059] In an exemplary embodiment of the present invention, the dimensionality reduction of the first feature index set based on the local preservation projection algorithm includes:

[0060] S201: Obtain the nearest neighbor of each data point in the first feature index set according to the determined distance metric, and construct an adjacency graph structure;

[0061] S202: measuring the position change of the data point in the low-dimensional space after mapping through the loss function;

[0062] The loss function is defined as minimizing the following objective:

[0063]

[0064] Where N is all the connected edges in the adjacency graph, Y i and Y j is the spatial representation of the i-th data point and the j-th data point after dimensionality reduction.

[0065] S203: Optimize the loss function and solve the dimension reduction projection matrix;

[0066] In order to minimize this loss, the following optimization problem can be obtained Among them, L is the graph Laplacian matrix, D is the degree matrix, Y is the dimension reduction projection matrix, and T is the transpose.

[0067] S204: Use the obtained projection matrix to map the original high-dimensional feature index set to a low-dimensional space.

[0068] Z=Y T X

[0069] Where Z is the data matrix represented in low dimension, and X is the data of the first feature index set.

[0070] Since it is not known in advance which characteristic indicators in the time domain, frequency domain and time-frequency domain can characterize the state of the electric valve, as many characteristic indicators as possible are obtained during feature extraction, resulting in a high dimension of the multi-domain characteristic indicator set, and there is information redundancy between some characteristic indicators, which has a certain impact on the accuracy of the state recognition model. Therefore, it is necessary to adopt a dimensionality reduction method to remove redundant information and features that are irrelevant to the state of the electric valve. The local preservation projection method is a classic dimensionality reduction method, which has the advantages of both the manifold learning method and the linear dimensionality reduction method. The present invention adopts this method to achieve dimensionality reduction of characteristic indicators.

[0071] In an exemplary embodiment of the present invention, fusing the second feature index set after dimension reduction includes:

[0072] The second characteristic index set is clustered by using a Gaussian mixture model to obtain a probability matrix corresponding to the opening and closing actions of the electric valve;

[0073] The probability matrix is ​​fused based on Jensen-Rényi divergence to obtain the third characteristic index set that can characterize the state of the electric valve.

[0074] The characteristic indicators obtained after the dimension reduction of the characteristic indicator set still have multiple indicators, which are difficult to establish a direct relationship with the state of the electric valve. Therefore, the method of the present invention uses the Gaussian mixture model method to cluster the characteristic indicators after the dimension reduction, so as to obtain the probability matrix corresponding to the opening and closing action of the electric valve, and finally fuses the probability matrix through the Jensen-Rényi divergence to obtain the fusion indicator that can characterize the state of the electric valve.

[0075] The Jensen-Rényi divergence is a measure for comparing probability distributions that can be extended to the fusion and combination of matrices and is defined as:

[0076]

[0077] Where P and Q are two probability distributions, α is a hyperparameter used to adjust the properties of the divergence, p(x) is the first probability distribution, q(x) is the second probability distribution, and x is the input of the probability matrix.

[0078] In an exemplary embodiment of the present invention, clustering the second feature indicator set by using a Gaussian mixture model includes:

[0079] Determine the range of the number of clusters, input the data of the second feature index set, randomly select a data object from the second feature index set as the first cluster center, and perform clustering:

[0080]

[0081] In the formula, p(x) is the clustering result, is the Gaussian probability density function in N dimensions, μ k is the mean vector, ∑ k is the covariance matrix, π k is the mixed weight, and K is the number of the second feature index set.

[0082] In an exemplary embodiment of the present invention, the establishment of a state adaptive classification model for an electric valve comprises:

[0083]

[0084] In the formula, is the output result of the model, N is the number of components in the third characteristic index set, j is the jth input layer, k is the kth output layer, x j The weight from the jth input layer to the hidden layer, is the weight from the hidden layer to the k-th output layer, γ is the bias term, and T is the transpose.

[0085] An electric valve state classification model is established based on the fusion index and the parameters of the Gaussian mixture model. As the opening and closing actions of the electric valve increase, the newly added opening and closing action data will also be used to update the parameters of the Gaussian mixture model, thereby realizing adaptive iteration of the model.

[0086] In an exemplary embodiment of the present invention, the training of the state adaptive classification model by using a training set includes:

[0087] The motion monitoring data in the training set is classified based on the state adaptive classification model, and the classification results are compared with the test set to obtain the accuracy used to characterize whether they are consistent;

[0088] Set the accuracy threshold. If the accuracy reaches the accuracy threshold, output the state adaptive classification model. If the accuracy does not reach the accuracy threshold, modify the parameters of the Gaussian mixture model and classify the motion monitoring data in the training set again until the accuracy reaches the accuracy threshold, and output the state adaptive classification model.

[0089] In this embodiment, the accuracy threshold is set to 100%. For the current state adaptive model, the electric valve action data in the training set is classified, and the classification results are compared with the real labels of the original data. If the real label of each action is consistent with the classification label, it indicates that the classification accuracy of the model is 100%, and the model can be used for subsequent electric valve state identification. Otherwise, the Gaussian mixture model parameters need to be updated.

[0090] In a second aspect, the present invention further provides an electric valve fault identification system based on feature dimension reduction, comprising:

[0091] The data acquisition module is configured to acquire the original electric valve action monitoring data, preprocess the detection data, and extract multi-domain feature indicators to obtain the first feature indicator set of the time domain, frequency domain and time-frequency domain statistics of the preprocessed detection data:

[0092] The data processing module is configured to reduce the dimension of the feature index set based on the local preservation projection algorithm to obtain a second feature index set; and fuse the second feature index set after the dimension reduction to obtain a third feature index set;

[0093] The model building module is configured to establish a state adaptive classification model of the electric valve through the third feature indicator set and the Gaussian mixture model parameters, divide the original electric valve action monitoring data into a training set and a test set, train the state adaptive classification model through the training set, and test it through the test set, and output the state adaptive classification model with the best result; identify the state of the electric valve through the state adaptive classification model.

[0094] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0095] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. The computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods of various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0096] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for identifying electric valve faults based on feature dimension reduction, characterized in that: include: The original electric valve action monitoring data is obtained, the detection data is preprocessed, and multi-domain feature indicators are extracted to obtain the first feature indicator set of the time domain, frequency domain and time-frequency domain statistics of the preprocessed detection data: Performing dimensionality reduction on the feature index set based on a local preservation projection algorithm to obtain a second feature index set; The second feature index set after dimension reduction is fused to obtain a third feature index set; A state adaptive classification model for electric valves is established through the third characteristic index set and Gaussian mixture model parameters, the original electric valve action monitoring data is divided into a training set and a test set, the state adaptive classification model is trained through the training set, and tested through the test set, and the state adaptive classification model with the best result is output; The state of the electric valve is identified through a state adaptive classification model.

2. The electric valve fault identification method based on feature dimension reduction according to claim 1 is characterized in that: The preprocessing of the detection data comprises: Standardize the test data to the interval [0, 1] or [-1, 1]; Wherein, x is the detection data, min(X) is the minimum value in the detection data, max(X) is the maximum value in the detection data, and x′ is the standardized value.

3. The electric valve fault identification method based on feature dimension reduction according to claim 1 is characterized in that: The dimensionality reduction of the first feature index set based on the local preservation projection algorithm includes: Obtain the nearest neighbor of each data point in the first feature index set according to the determined distance metric, and construct an adjacency graph structure; The loss function is used to measure the position change of the data points in the low-dimensional space after mapping; Optimize the loss function and solve the dimension reduction projection matrix; The obtained projection matrix is ​​used to map the original high-dimensional feature index set to a low-dimensional space.

4. The electric valve fault identification method based on feature dimension reduction according to claim 1 is characterized in that: The fusing of the second feature index set after dimension reduction includes: The second characteristic index set is clustered by using a Gaussian mixture model to obtain a probability matrix corresponding to the opening and closing actions of the electric valve; The probability matrix is ​​fused based on Jensen-Rényi divergence to obtain the third characteristic index set that can characterize the state of the electric valve.

5. The electric valve fault identification method based on feature dimension reduction according to claim 1 is characterized in that: The clustering of the second feature index set by using the Gaussian mixture model includes: Determine the range of the number of clusters, input the data of the second feature index set, randomly select a data object from the second feature index set as the first cluster center, and perform clustering: In the formula, p(x) is the clustering result, is the Gaussian probability density function in N dimensions, μ k is the mean vector, ∑ k is the covariance matrix, π k is the mixed weight, and K is the number of the second feature index set.

6. The electric valve fault identification method based on feature dimension reduction according to claim 1 is characterized in that: The state adaptive classification model of the electric valve is established including: In the formula, is the output result of the model, N is the number of components in the third characteristic index set, j is the jth input layer, k is the kth output layer, x j The weight from the jth input layer to the hidden layer, is the weight from the hidden layer to the k-th output layer, γ is the bias term, and T is the transpose.

7. The electric valve fault identification method based on feature dimension reduction according to claim 1 is characterized in that: The training of the state adaptive classification model by the training set includes: The motion monitoring data in the training set is classified based on the state adaptive classification model, and the classification results are compared with the test set to obtain the accuracy used to characterize whether they are consistent; Set the accuracy threshold. If the accuracy reaches the accuracy threshold, output the state adaptive classification model. If the accuracy does not reach the accuracy threshold, modify the parameters of the Gaussian mixture model and classify the motion monitoring data in the training set again until the accuracy reaches the accuracy threshold, and output the state adaptive classification model.

8. An electric valve fault identification system based on feature dimension reduction, Its characteristics include: The data acquisition module is configured to acquire the original electric valve action monitoring data, preprocess the detection data, and extract multi-domain feature indicators to obtain the first feature indicator set of the time domain, frequency domain and time-frequency domain statistics of the preprocessed detection data: The data processing module is configured to reduce the dimension of the feature index set based on the local preservation projection algorithm to obtain a second feature index set; and fuse the second feature index set after the dimension reduction to obtain a third feature index set; The model building module is configured to establish a state adaptive classification model of the electric valve through the third feature indicator set and the Gaussian mixture model parameters, divide the original electric valve action monitoring data into a training set and a test set, train the state adaptive classification model through the training set, and test it through the test set, and output the state adaptive classification model with the best result; identify the state of the electric valve through the state adaptive classification model.

Images