A distributed fault detection method and device for rail transit vehicle pantograph

Abstract

The present invention discloses a distributed fault detection method and device for a pantograph of a rail transit vehicle. The method comprises: collecting original data on each pantograph device; performing dimensionality reduction processing on the original data of a local node and an adjacent node to obtain the reduced-dimensional data; constructing a system matrix using the reduced-dimensional data and estimating a covariance matrix using the system matrix, performing singular value decomposition on the covariance matrix to obtain a singular matrix; designing a residual generator using the singular matrix and constructing a test statistic T 2 Perform online monitoring; solve the threshold J in real time th , if T 2 ≤J th , then the pantograph of the rail transit vehicle has no fault. If T 2 >J th , then the pantograph of the rail transit vehicle fails; the advantage of the present invention is that the accuracy of fault detection is high.

Description

Technical Field

[0001] The present invention relates to the field of vehicle fault detection, and more specifically to a distributed fault detection method and device for a pantograph of a rail transit vehicle. Background Art

[0002] Rail transit is a type of transportation system that uses electricity as energy and operates on the ground or underground using wheel-rail operation. The advantages of this system include high speed, large passenger volume, safety, convenience, cleanliness and environmental protection, etc. Therefore, rail transit occupies a very important position in my country's transportation system.

[0003] With the rapid development of my country's rail transit and the continuous improvement of rail transit vehicle technology, the composition of rail transit vehicles is becoming more and more complex. Rail transit is developing in the direction of scale, intelligence, and automation. These factors have also led to frequent failures of rail transit equipment, which has put forward higher requirements for the safety of my country's rail transit operating equipment.

[0004] Rail transit vehicles are prone to various faults during long-term operation, which will affect the normal operation of the vehicle, causing delays, etc. In serious cases, it may even threaten the driving safety of the vehicle and cause great harm to the safety of passengers' lives and property. Rail transit is a complex system with many key equipment and subsystems, such as electric traction power supply system, door control system, air conditioning and ventilation system, on-board signal system, steering control system, etc. According to statistics of accidents at home and abroad, the traction power supply system of rail transit is a key equipment that affects the safe operation of rail transit. The rail transit traction power supply system mainly includes pantographs, high-speed circuit breakers, converters and other equipment. The coordination and unity of these equipment in work is the key to ensuring that rail transit vehicles have reliable and sufficient power.

[0005] The test results of many rail vehicles show that the dynamic process of electric locomotive power supply is obtained by the vibration between the high-speed pantograph and the contact wire. The sliding friction between the pantograph and the contact wire will increase as the train travels at high speed, and it will become more difficult to maintain good contact performance between the pantograph and the catenary. Long-term sliding friction will cause the vehicle pantograph sliding strip to wear, and in severe cases, the pantograph will detach from the contact wire. At the moment when the pantograph detaches from the contact wire, an arc will occur, which will cause serious traffic accidents and cause serious harm to the property and life safety of passengers. Therefore, it is very necessary to conduct real-time monitoring and fault diagnosis of the traction power supply system of rail transit, and on this basis, accurate analysis and rapid decision-making to maintain the safety of rail transit operation is one of the major issues that must be solved at present.

[0006] Chinese patent publication number CN108680104A discloses a pantograph fault monitoring device and method for urban rail trains. The device includes a field control unit, a field data acquisition unit, a remote transmission unit, a data analysis unit, a data storage and publishing unit, and a client access unit; wherein the field data acquisition unit includes a camera, a flash, and a photoelectric sensor; the field control unit includes a wheel axle position sensor, a PLC, and a vehicle number recognition device AEI; the remote transmission unit includes a switch and an optical fiber; the data analysis unit includes a processing server; and the data storage and publishing unit includes a database server. The method is: the data analysis unit simultaneously performs pantograph slide wear monitoring, pantograph centerline deviation monitoring, and horn loss monitoring processing on the pantograph image collected on site, and packages the processing results of vehicle number recognition and the monitoring results to the data storage and publishing unit. This patent application uses online non-contact measurement, which is simple and convenient to implement. It needs to judge the fault according to wear, centerline deviation, and horn loss. For vehicles that do not have specific fault phenomena such as wear, centerline deviation, and horn loss, the fault cannot be identified. Summary of the invention

[0007] The technical problem to be solved by the present invention is that the existing rail transit vehicle pantograph fault detection method realizes vehicle fault monitoring according to specific fault phenomena, and cannot identify the fault when no specific fault phenomenon occurs, and the accuracy of fault identification is not high.

[0008] The present invention solves the above technical problems through the following technical means: a distributed fault detection method for a pantograph of a rail transit vehicle, the method comprising:

[0009] Step 1: The pantograph of the rail transit vehicle is divided into multiple nodes, and each node is equipped with a sensor to detect corresponding information and collect the original data on each pantograph device;

[0010] Step 2: Perform dimensionality reduction processing on the original data of the local node and the adjacent nodes to obtain the reduced-dimensional data;

[0011] Step 3: Use the reduced-dimensional data to construct a system matrix and use the system matrix to estimate the covariance matrix, and perform singular value decomposition on the covariance matrix to obtain a singular matrix;

[0012] Step 4: Use the singular matrix to design the residual generator and construct the test statistic T 2 Conduct online monitoring;

[0013] Step 5: Use the fault detection threshold setting method for non-Gaussian processes to solve the threshold J in real time th , if T 2 ≤J th , then the pantograph of the rail transit vehicle has no fault. If T 2>J th , then the pantograph of the rail transit vehicle is faulty.

[0014] The present invention sets a sensor on the vehicle pantograph, collects relevant data of the local node and the adjacent nodes, performs dimensionality reduction processing, and obtains the reduced data. Therefore, only the data with the strongest correlation is transmitted, reducing the uncertainty between data transmissions. In addition, by constructing a test statistic, the relationship between it and a preset threshold is judged, thereby identifying the fault, rather than realizing vehicle fault monitoring based on a specific fault phenomenon. Fault identification is not limited to a specific fault phenomenon, has a high accuracy, and the threshold is updated in real time during the identification process, further improving the accuracy of fault identification.

[0015] Furthermore, the step 2 comprises:

[0016] Assume that the sample set is X n*m , one sample per row, the number of samples is n, and the feature dimension is m; the covariance matrix between m features is:

[0017] Find the eigenvalue λ of this covariance matrix i and the eigenvector ρ i , and arrange the eigenvalues ​​in descending order: λ1≥λ2≥…≥λ k ≥λ k+1 ≥…≥λ m , rearrange the eigenvectors according to their corresponding eigenvalues, then Z m*m =[ρ1,ρ2,…,ρ m ]; Let the first k eigenvalues ​​from large to small form a diagonal matrix S k*k And S k*k =diag(λ1,λ2,…λ k ), k corresponding eigenvectors form the dimension reduction matrix P m*k And P m*k =[ρ1,ρ2,…,ρ k ].

[0018] Furthermore, the step three includes:

[0019] Assume that the measurement vector u comes from the local node, y i are nodes from the neighborhood, where

[0020] Node i collects m i Dimensional signal y i The measured value of y iGenerate y in m dimensions; let U be represented by an l×N measurement matrix, whose columns contain N measurement values ​​of u. Similarly, Y is represented by an m×N measurement matrix. Assuming that the number of measurements is large enough, the system matrix ∑u, ∑y, ∑uy can be approximated by the following formula: Then the covariance matrix The superscript symbol indicates taking the square root of a matrix and then its inverse matrix;

[0021] Performing singular value decomposition on the matrix K yields K = τ∑β T , where τ=(μ1, ...μ l )∈R l×l , β=(v1,...ν m )∈R m×m , μ1, ..., μ l and v1, ... ν m are the corresponding singular vectors, R l*m represents a matrix of dimension l*m, a singular matrix ∑k=diag(p1, p2, ... p k )∈R k×k , where p1, p2, …p k It represents the singular value. The relationship between the singular value and the eigenvalue is shown as follows: The eigenvalues ​​are arranged in descending order: λ1≥λ2≥…≥λ k .

[0022] Furthermore, the step 4 includes:

[0023] By formula Design a residual generator, where ∑k=diag(p1,p2,…p k )∈R k×k ,τ'∈R l×k , β'∈R m×k ;

[0024] According to the residual generator, using the formula Construct a test statistic where

[0025] Furthermore, the step five includes:

[0026] Step 501: Assume the probability of false alarm P (FAR) ∈(0,1) and there is δ∈(0,1), let ∈>0 and satisfy the condition P (FAR) -∈>0, iteration tolerance Δ>0, where P (FAR) =prob(T 2 >Jth |f=0);

[0027] Step 502: Setting the initial threshold J th0 , and J th0 >0;

[0028] Step 503: The number of random samples can be determined to be an integer N by the Chernoff bound theorem;

[0029] Step 504: Set initial P (FAR) The value of P (FAR0) And P (FAR0) =α, α is the given significance level;

[0030] Step 505: If the estimated value of the false alarm obtained under the current threshold satisfies P′ (FAR) ≤P (FAR) -∈ condition, then output the current threshold and end, where P′ (FAR) is the estimate of false alarms;

[0031] Step 506: If the condition of step 505 is not met, then set J th =J th0 +Δ,P′ (FAR) =P (FAR0) -Δ, and return to step 505 until the number of iterations is less than N-1;

[0032] After the threshold is determined, if T 2 ≤J th , then the pantograph of the rail transit vehicle has no fault;

[0033] If T 2 >J th , then the pantograph of the rail transit vehicle is faulty.

[0034] The present invention also provides a distributed fault detection device for a rail transit vehicle pantograph, the device comprising:

[0035] The data acquisition module is used to divide the pantograph of the rail transit vehicle into multiple nodes, each node is equipped with a sensor to detect the corresponding information and collect the original data on each pantograph device;

[0036] The data dimension reduction module is used to perform dimension reduction processing on the original data of the local node and the adjacent nodes to obtain the reduced-dimensional data;

[0037] A singular decomposition module is used to construct a system matrix using the reduced-dimensional data and estimate the covariance matrix using the system matrix, and perform singular value decomposition on the covariance matrix to obtain a singular matrix;

[0038] Test statistics module, used to design residual generators using singular matrices and construct test statistics T 2 Conduct online monitoring;

[0039] The threshold update module is used to solve the threshold J in real time using the fault detection threshold setting method for non-Gaussian processes. th , if T 2 ≤J th , then the pantograph of the rail transit vehicle has no fault. If T 2 >J th , then the pantograph of the rail transit vehicle is faulty.

[0040] Furthermore, the data dimension reduction module is also used for:

[0041] Assume that the sample set is X n*m , one sample per row, the number of samples is n, and the feature dimension is m; the covariance matrix between m features is:

[0042] Find the eigenvalue λ of this covariance matrix i and the eigenvector ρ i , and arrange the eigenvalues ​​in descending order: λ1≥λ2≥…≥λ k ≥λ k+1 ≥…≥λ m , rearrange the eigenvectors according to their corresponding eigenvalues, then Z m*m =[ρ1,ρ2,…,ρ m ]; Let the first k eigenvalues ​​from large to small form a diagonal matrix S k*k And S k+k =diag(λ1,λ2,…λ k ), k corresponding eigenvectors form the dimension reduction matrix P m*k And P m*k =[ρ1,ρ2,…,ρ k ].

[0043] Furthermore, the singular decomposition module is also used for:

[0044] Assume that the measurement vector u comes from the local node, y i are nodes from the neighborhood, where

[0045] Node i collects m i Dimensional signal y i The measured value of y iGenerate y in m dimensions; let U be represented by an l×N measurement matrix, whose columns contain N measurement values ​​of u. Similarly, Y is represented by an m×N measurement matrix. Assuming that the number of measurements is large enough, the system matrix ∑u, ∑y, ∑uy can be approximated by the following formula: Then the covariance matrix The superscript symbol indicates taking the square root of a matrix and then its inverse matrix;

[0046] Performing singular value decomposition on the matrix K yields K = τ∑β T , where τ=(μ1, ...μ l )∈R l×l , β=(v1,...ν m )∈R m×m , μ1, ..., μ l and v1, ... ν m are the corresponding singular vectors, Singular matrix ∑k=diag(p1,p2,…p k )∈R k×k , where p1, p2, ... pk represent singular values. The relationship between singular values ​​and eigenvalues ​​is shown in the following formula: The eigenvalues ​​are arranged in descending order: λ1≥λ2≥…≥λ k .

[0047] Furthermore, the test statistics module is also used for:

[0048] By formula Design a residual generator, where ∑k=diag(p1,p2,…p k )∈R k×k ,τ'∈R l×k , β'∈R m×k ;

[0049] According to the residual generator, using the formula Construct a test statistic where

[0050] Furthermore, the threshold updating module is further used for:

[0051] Step 501: Assume the probability of false alarm P (FAR) ∈(0,1) and there is δ∈(0,1), let ∈>0 and satisfy the condition P (FAR) -∈>0, iteration tolerance Δ>0, where P (FAR) =prob(T 2 >J th|f=0);

[0052] Step 502: Setting the initial threshold J th0 , and J th0 >0;

[0053] Step 503: The number of random samples can be determined to be an integer N by the Chernoff bound theorem;

[0054] Step 504: Set initial P (FAR) The value of P (FAR0) And P (FAR0) =α, α is the given significance level;

[0055] Step 505: If the estimated value of the false alarm obtained under the current threshold satisfies P′ (FAR) ≤P (FAR) -∈ condition, then output the current threshold and end, where P′ (FAR) is the estimate of false alarms;

[0056] Step 506: If the condition of step 505 is not met, then set J th =J th0 +Δ,P′ (FAR) =P (FAR0) -Δ, and return to step 505 until the number of iterations is less than N-1;

[0057] After the threshold is determined, if T 2 ≤J th , then the pantograph of the rail transit vehicle has no fault;

[0058] If T 2 >J th , then the pantograph of the rail transit vehicle is faulty.

[0059] The advantages of the present invention are:

[0060] (1) The present invention sets a sensor on the vehicle pantograph, collects relevant data of the local node and the adjacent nodes, performs dimensionality reduction processing, and obtains the reduced-dimensional data. Therefore, only the data with the strongest correlation is transmitted, reducing the uncertainty between data transmissions. In addition, by constructing a test statistic and judging the relationship between it and a preset threshold, the fault is identified, rather than realizing vehicle fault monitoring based on a specific fault phenomenon. Fault identification is not limited to a specific fault phenomenon, and has a high accuracy. In addition, the threshold is updated in real time during the identification process, further improving the accuracy of fault identification.

[0061] (2) The present invention performs fault diagnosis by collecting and processing sensor data and applying algorithms, and the process does not add any additional costs, so the overall cost is low. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 This is a flow chart of a distributed fault detection method for a rail transit vehicle pantograph disclosed in Embodiment 1 of the present invention;

[0063] Figure 2 A schematic structural diagram of a pantograph in a distributed fault detection method for a pantograph of a rail transit vehicle disclosed in Embodiment 1 of the present invention;

[0064] Figure 3 A schematic diagram of information interaction between a local node and a neighboring node in a distributed fault detection method for a rail transit vehicle pantograph disclosed in Embodiment 1 of the present invention;

[0065] Figure 4 This is a fault decision logic diagram in a distributed fault detection method for a pantograph of a rail transit vehicle disclosed in Example 1 of the present invention. DETAILED DESCRIPTION

[0066] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in combination with the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0067] Example 1

[0068] like Figure 1 As shown, a distributed fault detection method for pantographs of rail transit vehicles is provided, and the basic idea of ​​the method is: first, learning (identifying) the correlation of changes in real time; second, using correlation-based estimates and measurements from adjacent nodes to reduce the uncertainty of local sensor signals; then, developing and statistically analyzing test statistics at local nodes; then, updating the iterative threshold through a preset algorithm; finally, through decision logic, comparing the monitoring statistics with the threshold to draw a conclusion on whether a fault has occurred. The method includes:

[0069] S1: The pantograph of a rail transit vehicle is divided into multiple nodes, each of which is equipped with a sensor to detect the corresponding information and collect the original data on each pantograph device; the specific process is: the entire vehicle is regarded as a whole, the pantograph on the upper part of the car is divided into multiple nodes, and each node is equipped with a sensor to detect the corresponding information. Figure 2 shown.

[0070] S2: Perform dimensionality reduction processing on the original data of the local node and the adjacent nodes to obtain the reduced-dimensional data;

[0071] Fault detection is to determine whether a fault occurs during the driving process based on the characteristics of the monitored data information. Online monitoring refers to the use of partial test data to simulate the actual situation of the driving process, and fault monitoring is performed once a new measurement variable is collected. For online monitoring, data transmission is required for detection. In order to reduce the transmission cost, only low-dimensional strongly correlated data is transmitted between network nodes. During the high-speed operation of rail transit vehicles, friction between the pantograph and the contact wire always exists. A lot of information can be saved and transmitted through sensors, such as the operating temperature of the pantograph slide, the extension angle of the lower arm, and the stability performance of the insulation equipment. The collected data information is statistically analyzed, and the principal component analysis and dimensionality reduction methods are used to select the data with the strongest correlation after dimensionality reduction processing.

[0072] The dimensionality reduction method in principal component analysis is: Assume that the sample set is X n*m , one sample per row, the number of samples is n, and the feature dimension is m. The covariance matrix between m features is calculated as follows: Find the eigenvalue λ of this covariance matrix i and the eigenvector ρ i , and arrange the eigenvalues ​​in descending order: λ1≥λ2≥…≥λ k ≥λ k+1 ≥…≥λ m . Rearrange each eigenvector according to the corresponding eigenvalue. Assume that the sorted matrix is: Z m*m =[ρ1,ρ2,…,ρ m ]. According to the cumulative and proportion of eigenvalues, select the first k eigenvalues ​​for principal component analysis and dimensionality reduction. Let the first k eigenvalues ​​from large to small form a diagonal matrix S k*k , k corresponding eigenvectors form the dimension reduction matrix P m*k The expressions of the two matrices are as follows: S k*k =diag(λ1,λ2,…λ k ), P m*k =[ρ1,ρ2,…,ρ k ]. After principal component analysis, the number of samples in the matrix is ​​still n, and the number of features becomes k. The dimension reduction formula is as follows: n*k =XP (i.e. Λ n*k is the matrix obtained after dimensionality reduction).

[0073] The connection between the local node and the adjacent nodes is as follows Figure 3 As shown in the figure, node one is regarded as a local node, and the other two nodes are regarded as adjacent nodes. In the process of dividing the local node and the adjacent nodes, only the data information with the strongest correlation is transmitted from the adjacent nodes to the local node through the principal component analysis method. The whole process reduces the uncertainty between data transmission and improves the accuracy.

[0074] S3: Use the reduced-dimensional data to construct a system matrix and use the system matrix to estimate the covariance matrix, and perform singular value decomposition on the covariance matrix to obtain a singular matrix; the specific process is:

[0075] Assume that the measurement vector u comes from the local node, y i is a node from the neighborhood, where i = 1, ..., n-1. u ~ N (0, ∑u), y i ~N(0,∑y i ), u∈R l ,

[0076] Node i collects m i Dimensional signal y i The measured value of y i Generate y in m dimensions. Let U be represented by a l×N measurement matrix, whose columns contain N measurement values ​​of u. Similarly, Y is represented by an m×N measurement matrix. Assuming that the number of measurements is large enough, the system matrices ∑u, ∑y, ∑uy can be approximated by the following formula:

[0077] Then the covariance matrix The superscript symbol indicates that the square root of the matrix is ​​taken and then its inverse matrix is ​​taken. Performing singular value decomposition on the matrix K yields K = τ∑β T , where τ=(μ1, ...μ l )∈R l×l , β=(v1,...ν m )∈R m×m , μ1, ..., μ l and v1, ... ν m are the corresponding singular vectors, ∑k=diag(p1,p2,…p k )∈R k×k , where p1, p2, …p k It represents the singular value. The relationship between the singular value and the eigenvalue is shown as follows: The eigenvalues ​​are arranged in descending order: λ1≥λ2≥…≥λ k .

[0078] S4: Designing residual generator using singular matrix and constructing test statistic T 2 Carry out online monitoring; the specific process is:

[0079] Online monitoring is performed at the local node, and a residual generator is developed, which is expressed as: in ∑k=diag(p1,p2,…p k )∈R k×k ,τ'∈R l×k , β'∈R m×k .

[0080] The most widely used statistic is X 2 , which is defined as Assuming the number is large enough, then X 2 The test statistic becomes T 2 Test statistic. Combining the above formula, we have In the formula,

[0081] S5: Real-time solution of threshold J using the fault detection threshold setting method for non-Gaussian processes th , if T 2 ≤J th , then the pantograph of the rail transit vehicle has no fault. If T 2 >J th , then the pantograph of the rail transit vehicle is faulty; the specific process is:

[0082] In the case where the residual vector belongs to Gaussian distribution, fault detection is 2 The threshold value of the test statistic J th The setting can be determined by the standard distribution table with a significance level of α. The significance level is a concept in hypothesis testing, which refers to the probability or risk of rejecting the null hypothesis when it is correct. It is a recognized probability value of a low-probability event and must be determined before each statistical test. Usually α = 0.05 or α = 0.01 is used. The decision logic of fault detection is as follows Figure 4 shown.

[0083] The data collected by the sensors on the pantograph equipment and the data analysis show that the residual signal has non-Gaussian characteristics. In this case, fault detection is very important for T 2 The threshold value of the test statistic J thThe setting cannot be determined by the standard distribution table. Therefore, the present invention proposes a fault detection threshold setting method for non-Gaussian processes. The solution to non-Gaussian distribution is to use a method based on distribution estimation. Here, it is necessary to introduce the related concepts of two parameters, FAR and FDR. Among them, FAR (false alarm rate) refers to the false alarm rate, and FDR (fault detection rate) refers to the fault detection rate. The purpose of introducing the threshold is to reduce the false alarm rate to an acceptable level. Ideally, it is best to set the threshold to zero false alarm rate, but in this case, the fault detection rate will be greatly reduced. Given the time when the fault starts, the false alarm is understood as an event that triggers an alarm when there is no fault, and the detection alarm is regarded as an event that the alarm occurs when a fault actually occurs. From a probability perspective, FAR and FDR can be defined as the probabilities of false alarms and detection alarms, respectively. There are: P (FAR) =prob(T 2 >J th |f=0),P (FDR) =prob(T 2 >J th |f≠0). The method flow for updating the threshold is:

[0084] Step 1: Assume P (FAR) ∈(0,1) and there is δ∈(0,1), let ∈>0 and satisfy the condition P (FAR) -∈>0, iteration tolerance Δ>0.

[0085] Step 2: Set the initial threshold J th0 , and J th0 >0;

[0086] Step 3: The number of random samples can be determined to be an integer N by the Chernoff bound theorem;

[0087] Step 4: Set the initial P (FAR) The value of P (FAR0) And P (FAR0) =α, α is a given significance level, usually a constant between 0 and 1;

[0088] Step 5: If the estimated value of the false alarm obtained under the current threshold satisfies P′ (FAR) ≤P (FAR) -∈ condition, then output the current threshold and end, where P′ (FAR) is the estimated value of FAR;

[0089] Step 6: If the conditions of step 5 are not met, then there is J th =J th0 +Δ,P′ (FAR) =P (FAR0)-Δ, and return to step 5 until the number of iterations is less than N-1.

[0090] For the third step: For δ∈(0,1) and ∈∈(0,1), the random sample satisfies the one-sided Chernoff bound, the number of random samples is N, and there is Through the one-sided Chernoff boundary formula, we can obtain: p<p′+ε, that is, prob(p<p′+ε)>1-δ.

[0091] For the fourth step: Assume P (FAR) It is constant at any time, so the distribution of false alarm time has nothing to do with whether a false alarm occurred before. The average false alarm time is the expectation of the false alarm time. From the formula: Wherein η represents the false alarm time, prob(η=j) represents the probability density function of the false alarm time, and the value range of j is j=1, 2…, ∞. Simplifying the above formula by taking the first n terms and summing them, we can get Then we can get Ideally, the average false alarm time is (α is a given significance level), that is, the initial FAR estimate is α, which is: P (FAR0) =α.

[0092] The threshold J returned by the above algorithm th It can be obtained that prob(J>J th |f=0)<P FAR The initial threshold value must be greater than 1-δ, that is, prob{prob(J>J th )<P FAR}>1-δ. It is obvious that for a smaller Δ, the threshold of the algorithm returning the iteration is very small, and the estimated value of the false alarm rate can fluctuate within a small range during the iteration process, making the calculation result more accurate.

[0093] J th After confirmation, the fault is determined by the following logic:

[0094] If T 2 ≤J th , then the pantograph of the rail transit vehicle has no fault, J th is the preset threshold;

[0095] If T 2 >J th , then the pantograph of the rail transit vehicle is faulty.

[0096] Through the above technical scheme, the present invention sets a sensor on the vehicle pantograph, collects relevant data of local nodes and adjacent nodes, performs dimensionality reduction processing, and obtains the reduced data. Therefore, only the data with the strongest correlation is transmitted, reducing the uncertainty between data transmission. In addition, by constructing a test statistic, the relationship between it and a preset threshold is judged to identify the fault, rather than realizing vehicle fault monitoring based on a specific fault phenomenon. Fault identification is not limited to a specific fault phenomenon, has a high accuracy, and the threshold is updated in real time during the identification process, further improving the accuracy of fault identification.

[0097] Example 2

[0098] Based on Embodiment 1, Embodiment 2 of the present invention further provides a distributed fault detection device for a pantograph of a rail transit vehicle, the device comprising:

[0099] The data acquisition module is used to divide the pantograph of the rail transit vehicle into multiple nodes, each node is equipped with a sensor to detect the corresponding information and collect the original data on each pantograph device;

[0100] The data dimension reduction module is used to perform dimension reduction processing on the original data of the local node and the adjacent nodes to obtain the reduced-dimensional data;

[0101] A singular decomposition module is used to construct a system matrix using the reduced-dimensional data and estimate the covariance matrix using the system matrix, and perform singular value decomposition on the covariance matrix to obtain a singular matrix;

[0102] Test statistics module, used to design residual generators using singular matrices and construct test statistics T 2 Conduct online monitoring;

[0103] The threshold update module is used to solve the threshold J in real time using the fault detection threshold setting method for non-Gaussian processes. th , if T 2 ≤J th , then the pantograph of the rail transit vehicle has no fault. If T 2 >J th , then the pantograph of the rail transit vehicle is faulty.

[0104] Specifically, the data dimension reduction module is also used for:

[0105] Assume that the sample set is X n*m , one sample per row, the number of samples is n, and the feature dimension is m; the covariance matrix between m features is:

[0106] Find the eigenvalue λ of this covariance matrix i and the eigenvector ρi , and arrange the eigenvalues ​​in descending order: λ1≥λ2≥…≥λ k ≥λ k+1 ≥…≥λ m , rearrange the eigenvectors according to their corresponding eigenvalues, then Z m*m =[ρ1,ρ2,…,ρ m ]; Let the first k eigenvalues ​​from large to small form a diagonal matrix S k*k And S k*k =diag(λ1,λ2,…λ k ), k corresponding eigenvectors form the dimension reduction matrix P m*k And P m*k =[ρ1,ρ2,…,ρ k ].

[0107] More specifically, the singular decomposition module is also used for:

[0108] Assume that the measurement vector u comes from the local node, y i are nodes from the neighborhood, where

[0109] Node i collects m i Dimensional signal y i The measured value of y i Generate y in m dimensions; let U be represented by an l×N measurement matrix, whose columns contain N measurement values ​​of u. Similarly, Y is represented by an m×N measurement matrix. Assuming that the number of measurements is large enough, the system matrix ∑u, ∑y, ∑uy can be approximated by the following formula: Then the covariance matrix The superscript symbol indicates taking the square root of a matrix and then its inverse matrix;

[0110] Performing singular value decomposition on the matrix K yields K = τ∑β T , where τ=(μ1, ...μ l )∈R l×l , β=(v1,...ν m )∈R m×m , μ1, ..., μ l and v1, ... ν m are the corresponding singular vectors, Singular matrix ∑k=diag(p1,p2,…p k )∈R k×k , where p1, p2, …p k It represents the singular value. The relationship between the singular value and the eigenvalue is shown as follows: The eigenvalues ​​are arranged in descending order: λ1≥λ2≥…≥λ k .

[0111] More specifically, the test statistics module is also used for:

[0112] By formula Design a residual generator, where ∑k=diag(p1,p2,…p k )∈R k×k ,τ'∈R l×k , β'∈R m×k ;

[0113] According to the residual generator, using the formula Construct a test statistic where

[0114] More specifically, the threshold updating module is further used for:

[0115] Step 501: Assume the probability of false alarm P (FAR) ∈(0,1) and there is δ∈(0,1), let ∈>0 and satisfy the condition P (FAR) -∈>0, iteration tolerance Δ>0, where P (FAR) =prob(T 2 >J th |f=0);

[0116] Step 502: Setting the initial threshold J th0 , and J th0 >0;

[0117] Step 503: The number of random samples can be determined to be an integer N by the Chernoff bound theorem;

[0118] Step 504: Set initial P (FAR) The value of P (FAR0) And P (FAR0) =α, α is the given significance level;

[0119] Step 505: If the estimated value of the false alarm obtained under the current threshold satisfies P′ (FAR) ≤P (FAR) -∈ condition, then output the current threshold and end, where P′ (FAR) is the estimate of false alarms;

[0120] Step 506: If the condition of step 505 is not met, then set J th =J th0 +Δ,P′ (FAR) =P (FAR0)-Δ, and return to step 505 until the number of iterations is less than N-1;

[0121] After the threshold is determined, if T 2 ≤J th , then the pantograph of the rail transit vehicle has no fault;

[0122] If T 2 >J th , then the pantograph of the rail transit vehicle is faulty.

[0123] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A distributed fault detection method for a rail transit vehicle pantograph, characterized in that: The method comprises: Step 1: The pantograph of the rail transit vehicle is divided into multiple nodes, and each node is equipped with a sensor to detect corresponding information and collect the original data on each pantograph device; Step 2: Perform dimensionality reduction processing on the original data of the local node and the adjacent nodes to obtain the reduced-dimensional data; Step 3: Use the reduced-dimensional data to construct a system matrix and use the system matrix to estimate the covariance matrix, and perform singular value decomposition on the covariance matrix to obtain a singular matrix; Step 4: Use the singular matrix to design the residual generator and construct the test statistic T 2 Conduct online monitoring; Step 5: Use the fault detection threshold setting method for non-Gaussian processes to solve the threshold J in real time th , if T 2 ≤J th , then the pantograph of the rail transit vehicle has no fault. If T 2 >J th , then the pantograph of the rail transit vehicle is faulty.

2. A distributed fault detection method for a rail transit vehicle pantograph according to claim 1, characterized in that: The second step comprises: Assume that the sample set is X n*m , one sample per row, the number of samples is n, and the feature dimension is m; the covariance matrix between m features is: Find the eigenvalue λ of this covariance matrix i and the eigenvector ρ i , and arrange the eigenvalues ​​in descending order: λ1≥λ2≥…≥λ k ≥λ k+1 ≥…≥λ m , rearrange the eigenvectors according to their corresponding eigenvalues, then Z m*m =[ρ1,ρ2,…,ρ m ]; Let the first k eigenvalues ​​from large to small form a diagonal matrix S k*k And S k*k =diag(λ1,λ2,…λ k ), k corresponding eigenvectors form the dimension reduction matrix P m*k And P m*k =[ρ1,ρ2,…,ρ k ].

3. A distributed fault detection method for a rail transit vehicle pantograph according to claim 2, characterized in that: The step three comprises: Assume that the measurement vector u comes from the local node, y i are nodes from the neighborhood, where i=1,...,n-1, Node i collects m i Dimensional signal y i The measured value of y i Generate y in m dimensions; let U be represented by an l×N measurement matrix, whose columns contain N measurement values ​​of u. Similarly, Y is represented by an m×N measurement matrix. Assuming that the number of measurements is large enough, the system matrices Σu, Σy, Σuy can be approximated by the following formula: Then the covariance matrix The superscript symbol indicates taking the square root of a matrix and then its inverse matrix; Performing singular value decomposition on the matrix K yields K = τΣβ T , where τ=(μ1,…μ l )∈R l×l , β=(v1,…v m )∈R m×m ,μ1,…,μ l and ν1,…ν m are the corresponding singular vectors, Singular matrix ∑k=diag(p1,p2,…p k )∈R k×k , where p1, p2, … p k It represents the singular value. The relationship between the singular value and the eigenvalue is shown as follows: i∈1,…,k, the eigenvalues ​​are arranged in descending order: λ1≥λ2≥…≥λ k .

4. A distributed fault detection method for a rail transit vehicle pantograph according to claim 3, characterized in that: The fourth step comprises: By formula Design a residual generator, where ∑k=diag(p1,p2,…p k )∈R k×k ,τ'∈R l×k , β'∈R m×k ; According to the residual generator, using the formula Construct a test statistic where 5. A distributed fault detection method for a rail transit vehicle pantograph according to claim 4, characterized in that: The step five comprises: Step 501: Assume the probability of false alarm P (FAR) ∈(0,1) and there is δ∈(0,1), let ∈>0 and satisfy the condition P (FAR) -∈>0, iteration tolerance Δ>0, where P (FAR) =prob(T 2 >J th |f=0); Step 502: Setting the initial threshold J th0 , and J th0 >0; Step 503: The number of random samples can be determined to be an integer N by the Chernoff bound theorem; Step 504: Set initial P (FAR) The value of P (FAR0) And P (FAR0) =α, α is the given significance level; Step 505: If the estimated value of the false alarm obtained under the current threshold satisfies P′ (FAR) ≤P (FAR) -∈ condition, then output the current threshold and end, where P′ (FAR) is the estimate of false alarms; Step 506: If the condition of step 505 is not met, then set J th =J th0 +Δ,P′ (FAR) =P (FAR0) -Δ, and return to step 505 until the number of iterations is less than N-1; After the threshold is determined, if T 2 ≤J th , then the pantograph of the rail transit vehicle has no fault; If T 2 >J th , then the pantograph of the rail transit vehicle is faulty.

6. A distributed fault detection device for a pantograph of a rail transit vehicle, characterized in that: The device comprises: The data acquisition module is used to divide the pantograph of the rail transit vehicle into multiple nodes, each node is equipped with a sensor to detect the corresponding information and collect the original data on each pantograph device; The data dimension reduction module is used to perform dimension reduction processing on the original data of the local node and the adjacent nodes to obtain the reduced-dimensional data; A singular decomposition module is used to construct a system matrix using the reduced-dimensional data and estimate the covariance matrix using the system matrix, and perform singular value decomposition on the covariance matrix to obtain a singular matrix; Test statistics module, used to design residual generators using singular matrices and construct test statistics T 2 Conduct online monitoring; The threshold update module is used to solve the threshold J in real time using the fault detection threshold setting method for non-Gaussian processes. th , if T 2 ≤J th , then the pantograph of the rail transit vehicle has no fault. If T 2 >J th , then the pantograph of the rail transit vehicle is faulty.

7. A distributed fault detection device for a pantograph of a rail transit vehicle according to claim 6, characterized in that: The data dimension reduction module is also used for: Assume that the sample set is X n*m , one sample per row, the number of samples is n, and the feature dimension is m; the covariance matrix between m features is: Find the eigenvalue λ of this covariance matrix i and the eigenvector ρ i , and arrange the eigenvalues ​​in descending order: λ1≥λ2≥…≥λ k ≥λ k+1 ≥…≥λ m , rearrange the eigenvectors according to their corresponding eigenvalues, then Z m*m =[ρ1,ρ2,…,ρ m ]; Let the first k eigenvalues ​​from large to small form a diagonal matrix S k*k And S k*k =diag(λ1,λ2,…λ k ), k corresponding eigenvectors form the dimension reduction matrix P m*k And P m*k =[ρ1,ρ2,…,ρ k ].

8. A distributed fault detection device for a pantograph of a rail transit vehicle according to claim 7, characterized in that: The singular decomposition module is also used for: Assume that the measurement vector u comes from the local node, y i are nodes from the neighborhood, where i=1,...,n-1, Node i collects m i Dimensional signal y i The measured value of y i Generate y in m dimensions; let U be represented by an l×N measurement matrix, whose columns contain N measurement values ​​of u. Similarly, Y is represented by an m×N measurement matrix. Assuming that the number of measurements is large enough, the system matrices Σu, Σy, Σuy can be approximated by the following formula: Then the covariance matrix The superscript symbol indicates taking the square root of a matrix and then its inverse matrix; Performing singular value decomposition on the matrix K yields K = τΣβ T , where τ=(μ1,…μ l )∈R l×l , β=(v1,…v m )∈R m×m ,μ1,…,μ l and v1,…v m are the corresponding singular vectors, Singular matrix ∑k=diag(p1,p2,…p k )∈R k×k , where p1, p2, … p k It represents the singular value. The relationship between the singular value and the eigenvalue is shown as follows: i∈1,…,k, the eigenvalues ​​are arranged in descending order: λ1≥λ2≥…≥λ k .

9. A distributed fault detection device for a pantograph of a rail transit vehicle according to claim 8, characterized in that: The test statistics module is also used for: By formula Design a residual generator, where ∑k=diag(p1,p2,…p k )∈R k×k ,τ'∈R l×k , β'∈R m×k ; According to the residual generator, using the formula Construct a test statistic where 10. A distributed fault detection device for a pantograph of a rail transit vehicle according to claim 9, characterized in that: The threshold updating module is also used for: Step 501: Assume the probability of false alarm P (FAR) ∈(0,1) and there is δ∈(0,1), let ∈>0 and satisfy the condition P (FAR) -∈>0, iteration tolerance Δ>0, where P (FAR) =prob(T 2 >J th |f=0); Step 502: Setting the initial threshold J th0 , and J th0 >0; Step 503: The number of random samples can be determined to be an integer N by the Chernoff bound theorem; Step 504: Set initial P (FAR) The value of P (FAR0) And P (FAR0) =α, α is the given significance level; Step 505: If the estimated value of the false alarm obtained under the current threshold satisfies P′ (FAR) ≤P (FAR) -∈ condition, then output the current threshold and end, where P′ (FAR) is the estimate of false alarms; Step 506: If the condition of step 505 is not met, then set J th =J th0 +Δ,P′ (FAR) =P (FAR0) -Δ, and return to step 505 until the number of iterations is less than N-1; After the threshold is determined, if T 2 ≤J th , then the pantograph of the rail transit vehicle has no fault; If T 2 >J th , then the pantograph of the rail transit vehicle is faulty.

Images