Pneumatic control valve fault monitoring method and device based on microwave measurement

Abstract

The invention relates to the field of pneumatic control valve fault monitoring, in particular to a pneumatic control valve fault monitoring method and device based on microwave measurement, and the method comprises the steps: obtaining working data of a pneumatic control valve in a preset period, and carrying out the preprocessing, and obtaining a state signal at a preset moment; linear features of typical variable dissimilarity are extracted from the state signals through a CVDA method; nonlinear features in the dissimilarity of the typical variables are extracted from the state signals through an MKPCA method; and the linear features and the nonlinear features are converted into statistical magnitudes, the statistical magnitudes are compared with control limits obtained through corresponding offline training, and fault monitoring is carried out on the pneumatic control valve. According to the technical scheme provided by the invention, the CVDA and the MKPCA are combined, the number of monitored fault types is effectively increased, the fault detection rate is improved, the fault discovery time is shortened, the stability and reliability of the pneumatic system are improved, and the accident rate is reduced.

Description

Technical Field

[0001] The present application relates to the technical field of control valve fault monitoring, and in particular to a method and device for monitoring pneumatic control valve fault based on microwave measurement. Background Art

[0002] Control valves are actuators in process industry automatic control systems. With the continuous improvement of automation, they are widely used in various fields. Pneumatic control valves are an important branch of control valves. In process control, pneumatic control valves directly control the pressure, flow rate and flow direction of the fluid. The stability and reliability of their quality will directly affect the entire process system. Therefore, it is very necessary to conduct real-time dynamic process monitoring of pneumatic control valves.

[0003] Among the common multivariate statistical methods, the Canonical Variate Analysis (CVA) algorithm projects high-dimensional state variables into low-dimensional feature space based on the maximum dynamic correlation between the "past" data and the "future" data of the observed variables, which can effectively handle the autocorrelation and cross-correlation of the dynamic data of the industrial process, thereby realizing the characterization of the dynamic characteristics of the system; the standard Principal Component Analysis (PCA) rotates the coordinate axis by finding the orthogonal vector that coincides with the direction of the maximum variance. The new coordinate axis obtained is a linear combination of the coordinate axis after linear transformation. It is a linear dimensionality reduction technology, and the actual industrial process often has certain nonlinearity, time lag and other characteristics. However, due to the complex data characteristics of the pneumatic control valve state signal (such as time series correlation), the traditional method is not ideal when processing this type of data, resulting in a low fault detection rate and a high false detection rate.

[0004] Therefore, it has become an urgent task to develop a pneumatic control valve fault monitoring solution based on microwave measurement to improve the accuracy, number of types, and detection time of pneumatic control valve fault detection. Summary of the invention

[0005] In view of this, it is necessary to provide a pneumatic control valve fault monitoring method and device based on microwave measurement to solve the problem that the existing method is not ideal when processing complex data of pneumatic control valve status signals, resulting in low fault detection rate and high false detection rate.

[0006] A method for monitoring a pneumatic control valve fault based on microwave measurement, comprising:

[0007] The working data of the pneumatic control valve is obtained and preprocessed to obtain a state signal at a preset time; wherein, the obtaining of the working data includes obtaining the real-time valve stem displacement data of the pneumatic control valve by carrying a microwave sensing sensor, measuring the real-time flow signal by an ultrasonic sensor, and collecting the valve inlet and outlet pressure signals by a pressure sensor; and the working data is preprocessed by an edge device to obtain the state signal, and the length of all the state signal data in the state signal is consistent;

[0008] Extracting linear features of typical variable dissimilarity from the state signal by CVDA method;

[0009] Extracting nonlinear features in typical variable dissimilarity from the state signal by using the MKPCA method;

[0010] The linear characteristics and the nonlinear characteristics are respectively converted into statistics, and compared with the corresponding control limits obtained by offline training, so as to perform fault monitoring on the pneumatic control valve.

[0011] Preferably, the extracting of the linear features of the typical variable dissimilarity from the state signal by the CVDA method comprises:

[0012] According to the preset time at which the state signal is located, extracting from the state signal first observation data earlier than the preset time and second observation data later than the preset time;

[0013] The difference regularization variable d is constructed by comparing the difference between the first observation data and the second observation data. n (t), the difference regularization variable d n (t)Satisfy:

[0014] d n (t) = L n X f(t) -S n J n X p(t)

[0015] S n =diag(λ1,λ2,…,λ n )

[0016] Among them, X p(t) and X f(t) are the first observation data and the second observation data respectively; L n and J n is the transformation matrix obtained by performing singular value decomposition calculation based on the first observation data and the second observation data; S n The selection criteria of singular values, λ nrepresents the nth value in the descending eigenvalue sequence;

[0017] According to the difference regularization variable d n (t) Extracting and obtaining the linear features.

[0018] Preferably, the extracting of nonlinear features in the typical variable dissimilarity from the state signal by the MKPCA method comprises:

[0019] Construct the MKPCA model according to the preset hybrid kernel function;

[0020] Extract the difference regularization variable d according to the MKPCA model n (t) is the principal component variable t m , obtain the nonlinear characteristics;

[0021] The preset mixed kernel function Kmix satisfies:

[0022] K mix =γK lap +βK rbf +(1-β-γ)K poly

[0023] Among them, γ (0≤γ≤0.5) and β (0≤β≤0.5) are mixing coefficients that balance interpolation and extrapolation capabilities;

[0024] K lap represents the Laplace kernel function,

[0025] K poly represents the polynomial kernel function,

[0026] K rbf represents the Gaussian kernel function,

[0027] Preferably, the difference regularization variable d is extracted according to the MKPCA model n (t) is the principal component variable t m , the nonlinear characteristics are obtained, including:

[0028] Extract the difference regular variable d n (t) is the principal component variable t m ,satisfy:

[0029]

[0030] Among them, the eigenvector α=[α1,α2,..,α M ] T , i=1,2,…,m.

[0031] Preferably, converting the linear features and the nonlinear features into statistics respectively comprises:

[0032] Converting the linear features into statistics includes: 2 Statistics and SPE statistics satisfy:

[0033]

[0034] Among them, Z k，p represents the p-window state vector extracted from the past data vector, e k represents the residual vector;

[0035] Converting the nonlinear features into statistics includes: 2 m Statistics and SPE m Statistics satisfy:

[0036]

[0037] SPE m (k) = t M t M T -t m (k)t m (k) T

[0038] Among them, ∑ m Indicates t m The sample covariance matrix Σ m = <t m (k) T t m (k)> k ,

[0039] t m (k) represents the kernel principal component t corresponding to the first m eigenvalues i (k) is a sequence consisting of

[0040] t m (k)=[t1(k), t2(k),...,t m (k)],t M =[t1(k), t2(k), ..., t M (k)];

[0041] Preferably, the linear characteristics and the nonlinear characteristics are respectively converted into statistics, and compared with the corresponding control limits obtained by offline training, and the pneumatic control valve is monitored for faults, including:

[0042] The control limits satisfy: P(J<JUCL ] = α;

[0043] Among them, J UCL is the control limit, J is the fault detection index, α is the preset significance level; J∈{T 2 , SPE}, J UCL ∈{T 2 UCL , SPE UCL};

[0044] If J < J UCL , it is determined that the pneumatic control valve is in normal operation, otherwise it is determined that the pneumatic control valve is faulty.

[0045] Preferably, the method further comprises performing a centralization operation on the mixed kernel function Kmix to satisfy:

[0046] K mix，c =K mix -I N K mix -K mix I N +I N K mix I N

[0047]

[0048] Preferably, the p-window state vector Z k，p , f-window state vector Z k，f and the residual vector e k ,satisfy:

[0049] z k，p =J n X p(t)

[0050] z k，f =L n X f(t)

[0051] e k =F n X p(t)

[0052] Among them, the transformation matrix Transformation Matrix Σ pp ,Σ ff ,Σ pf represents the covariance matrix and cross-covariance matrix of the Hankel matrix, V n represents the first n columns of V, U nrepresents the first n columns of U, where n is the inflection point of the singular value curve.

[0053] A pneumatic control valve fault monitoring device based on microwave measurement, comprising:

[0054] A data acquisition module is used to acquire the working data of the pneumatic control valve within a preset period and perform preprocessing to obtain a state signal at a preset time; wherein, the acquisition of working data includes acquiring the real-time valve stem displacement data of the pneumatic control valve through a microwave sensing sensor, measuring the real-time flow signal through an ultrasonic sensor, and collecting the valve inlet and outlet pressure signals through a pressure sensor; and preprocessing the working data through an edge device to obtain the state signal, wherein the length of all state signal data in the state signal is consistent;

[0055] A CVDA processing module, used for extracting linear features of typical variable dissimilarity from the state signal by using a CVDA method;

[0056] An MKPCA processing module, used for extracting nonlinear features in typical variable dissimilarity from the state signal by using an MKPCA method;

[0057] The monitoring module is used to convert the linear characteristics and the nonlinear characteristics into statistics respectively, and compare them with the control limits obtained by the corresponding offline training to perform fault monitoring on the pneumatic control valve.

[0058] Preferably, it also includes:

[0059] The offline training module is used to obtain the control limit based on offline training through the CVDA method and the MKPCA method.

[0060] Compared with the prior art, the beneficial effects of the present application are as follows: the pneumatic control valve fault monitoring scheme proposed in the present application is based on typical variable difference analysis and mixed kernel principal component analysis (CVDA-MKPCA) for fault monitoring, CVDA is used to extract the linear characteristics of the typical variable dissimilarity of the state signal, the mixed kernel function of MKPCA is constructed, and MKPCA is used to extract the nonlinear characteristics in the typical variable dissimilarity; finally, statistics are obtained according to the extracted linear and nonlinear characteristics, and the statistics are compared with the control limits determined offline to achieve fault monitoring. It effectively increases the number of fault types that can be monitored, improves the fault detection rate, and shortens the fault discovery time, and realizes the effective monitoring of pneumatic control valve faults, which is of great significance for improving the stability and reliability of process systems and reducing the accident rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a flow chart of a method for monitoring a pneumatic control valve fault based on microwave measurement provided in an embodiment of the present application.

[0062] Figure 2 It is a flow chart of a method for monitoring a pneumatic control valve fault based on microwave measurement provided in another embodiment of the present application.

[0063] Figure 3a The CVA fault detection method of the fault F10-AM provided in the embodiment of the present application is 2 and schematic diagram of SPE statistics;

[0064] Figure 3b It is a schematic diagram of SMD statistics in CVDA fault detection of fault F10-AM provided in an embodiment of the present application;

[0065] Figure 3c This is the KPCA fault detection method for the fault F10-AM provided in the embodiment of the present application. 2 and schematic diagram of SPE statistics;

[0066] Figure 3d It is a schematic diagram of T2 and SPE statistics in CVDA-MKPCA fault detection of fault F10-AM provided in an embodiment of the present application;

[0067] Figure 4a This is the CVA fault detection method for the fault F17-I provided in the embodiment of the present application. 2 and schematic diagram of SPE statistics;

[0068] Figure 4b It is a schematic diagram of SMD statistics in CVDA fault detection of fault F17-I provided in an embodiment of the present application;

[0069] Figure 4c is the KPCA fault detection method for the fault F17-I provided in the embodiment of the present application. 2 and schematic diagram of SPE statistics;

[0070] Figure 4d This is the CVDA-MKPCA fault detection method for the fault F17-I provided in the embodiment of the present application. 2 And schematic diagram of SPE statistics.

[0071] Figure 5 Schematic diagram of the structure of a pneumatic control valve fault monitoring device 500 based on microwave measurement provided in an embodiment of the present application. DETAILED DESCRIPTION

[0072] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0073] The pneumatic control valve fault monitoring scheme based on microwave measurement proposed in this application is processed based on CVDA-MKPCA. Among them, Canonical Variate Dissimilarity Analysis (CVDA) is a method for extracting linear features of state signals, which is particularly suitable for processing complex industrial process data with time series correlation. The application of CVDA method in pneumatic control valve fault monitoring is to extract linear features from state signals in order to better understand and predict system behavior. Compared with traditional Canonical Variate Analysis (CVA), CVDA focuses on extracting the difference between historical data in a continuous time before a specific observation moment and the predicted or actual data in a continuous time after the same specific observation moment, that is, the linear characteristics of the canonical variable dissimilarity. CVDA not only takes into account the maximum dynamic correlation of the data, but also emphasizes the differences between these data, which helps to more accurately capture the dynamic changes and potential fault signs of the system. CVDA can effectively capture the dynamic characteristics of the system, especially when dealing with complex industrial processes with strong timing correlation. By analyzing these two sets of data, this application can more accurately monitor the state changes of pneumatic control valves and provide early warning of potential problems.

[0074] In this application, the mixed kernel principal component analysis (Mixed Kernel Principal Component Analysis, MKPCA) introduces the concept of mixed kernel function compared to the traditional KPCA (Kernel Principal Component Analysis), that is, using multiple different kernel functions (for example, Gaussian kernel, polynomial kernel and Laplace kernel) at the same time. This hybrid strategy allows the model to better adapt to different types of data structures, and can provide stronger interpolation and extrapolation performance. By adjusting the weights between each kernel function, MKPCA can improve generalization ability while maintaining a good fit. In this application, MKPCA is used to extract nonlinear features in difference regularized variables, which are then used to construct nonlinear statistics (T 2 and SPE) to assist in fault diagnosis.

[0075] This application combines CVDA with MKPCA, and MKPCA effectively supplements the deficiencies of CVDA in linear feature extraction, thereby achieving more comprehensive and accurate monitoring of the operating status of the pneumatic control valve.

[0076] See also Figure 1 , Figure 1The flowchart of a method for monitoring a pneumatic control valve fault based on microwave measurement provided by an embodiment of the present application is shown in FIG. The method for monitoring a pneumatic control valve fault based on microwave measurement comprises the following steps:

[0077] S101: Acquire the working data of the pneumatic control valve and perform preprocessing to obtain a status signal.

[0078] Collect the working data of the pneumatic control valve during operation, that is, the state variables Xnew, X new =[X1, X2, …, X N ] T ∈R N×m , where N is the number of samples and m is the number of state variables. The working data includes control signals, valve inlet and outlet pressure signals, flow signals and valve stem displacement signals. Specifically, obtaining working data includes obtaining real-time valve stem displacement data of the pneumatic control valve by carrying a microwave sensing sensor, measuring real-time flow signals by an ultrasonic sensor, and collecting valve inlet and outlet pressure signals by a pressure sensor; and preprocessing the working data by an edge device to obtain the state signal, preprocessing the working data to form a state signal, and all state signal data in the state signal have the same length.

[0079] According to the preset time state signal, the first observation data Xp earlier than the preset time is generated by the collected working data within the preset period (t) and the second observation data X later than the preset time f(t) , or respectively called "past" data and "future" data, and standardized them into observed variables with zero mean and unit variance, and constructed the corresponding Hankel matrices Xp and X f .

[0080] Furthermore, the preprocessing of the working data of the pneumatic control valve includes standardization processing:

[0081] The p-window length and the f-window length are both n, and after normalization processing, the first observation vector and the second observation vector corresponding to the first observation data and the second observation data are obtained, which are X p(t) , X f(t) .

[0082] X p(t) =[X(t-1),X(t-2),…,X(tn)]∈R nm

[0083] X f(t) =[X(t),X(t+1),…,X(t+n-1)]∈R nm

[0084] S102: extracting linear features of typical variable dissimilarity from the state signal through the CVDA method.

[0085] According to the normalized first observation vector and second observation vector, the corresponding Hankel matrices Xp and X f , get the first observation matrix (past observation matrix) and the second observation matrix (future observation matrix), and calculate the covariance matrix Σ pp ,Σ pf ,∑ ff :

[0086] X p =[X p(p+1) , X p(p+2) , …, X p(p+M) ]∈R mp×M

[0087] X f =[X f(p+1) , X f(p+2) , …, X f(p+M) ]∈R mf×M

[0088] Wherein, M=N-2n+1, and N is the number of samples.

[0089] Covariance matrix∑ pp ,∑ ff and the cross covariance matrix ∑ fp The expressions are:

[0090]

[0091] The first measurement matrix and the second measurement matrix of the Hankel matrix are subjected to singular value decomposition (SVD) to obtain a left singular matrix U and a right singular matrix V. Transformation matrices J, L and F are calculated based on the results of the singular value decomposition.

[0092] Specifically, the result of singular value decomposition is:

[0093]

[0094] Among them, U and V are orthogonal matrices composed of left and right singular vectors, respectively, and the diagonal matrix Λ consists of ordered singular values.

[0095] According to the results of singular value decomposition, the expressions for calculating the transformation matrices J and L are:

[0096]

[0097] Among them, Σ pp and Σ ffRepresent the covariance matrices of the "past" and "future" respectively.

[0098] Construct the CVDA model based on the transformation matrices J and L and extract the p-window state vector Z k，p , f-window state vector Z k，f and the residual vector e k , calculate the difference between the first data predictor variable and the second data regular variable d n (t), to quantify the model residual in the CVA state subspace, satisfying:

[0099] z k，p =J n X p(t)

[0100] z k，f =L n X f(t)

[0101] e k =F n X p(t)

[0102] Among them, the transformation matrix Transformation Matrix Σ pp ,Σ ff ,Σ pf represents the covariance matrix and cross-covariance matrix of the Hankel matrix, V n represents the first n columns of V, U n represents the first n columns of U, where n is the inflection point of the singular value curve.

[0103] Further, the difference regular variable d between the first observed variable and the second observed variable in the past is calculated according to the transformation matrix J, L and F. n (t), satisfying:

[0104] d n (t) = L n X f(t) -S n J n X p(t)

[0105] S n =diag(λ1,λ2,…,λ n )

[0106] Among them, X p(t) and X f(t) are the first observation data and the second observation data respectively; L n and J nis the transformation matrix obtained by performing singular value decomposition calculation based on the first observation data and the second observation data; S n The selection criteria of singular values, λ n Represents the nth value in the descending sequence of eigenvalues.

[0107] According to the difference regularization variable d n (t) Extracting and obtaining the linear features.

[0108] Furthermore, the linear characteristics include T 2 Statistics and SPE statistics:

[0109] According to the extracted state vector Z k，p , Z k，f and the residual vector e k , calculate the linear characteristic state variable index T 2 Statistics and SPE statistics.

[0110]

[0111] Among them, Z k，p represents the p-window state vector extracted from the past data vector during the online monitoring process, e k Represents the residual vector during online process monitoring.

[0112] S103: extracting nonlinear features in typical variable dissimilarity from the state signal through the MKPCA model.

[0113] Design the hybrid kernel function K mix , and based on the mixed kernel function K mix Construct the MKPCA model.

[0114] Furthermore, the embodiment of the present application provides a hybrid kernel function K capable of processing complex data. mix , which combines Gaussian kernel function, polynomial kernel function and Laplace kernel function to satisfy:

[0115] K mix =γK lap +βK rbf +(1-β-γ)K poly

[0116] Among them, γ (0≤γ≤0.5) and β (0≤β≤0.5) are mixing coefficients that balance interpolation and extrapolation capabilities;

[0117] K lap represents the Laplace kernel function, K poly represents the polynomial kernel function, K rbfrepresents the Gaussian kernel function,

[0118] Furthermore, for the mixed kernel function K mix Perform centralized operation to get K mix The expression is:

[0119] K mix =K mix -I N K mix -K mix I N +I N K mix I N

[0120]

[0121] Then, using the above hybrid kernel function K mix , build the MKPCA model and extract the residual variable d n The kernel principal component variable t of the nonlinear characteristics in (t) m .

[0122] Extract the difference regular variable d n (t) is the principal component variable t m ,satisfy:

[0123]

[0124] Among them, the eigenvector α=[α1,α2,..,α M ] T , i=1,2,…,m.

[0125] Construct statistics T of nonlinear features 2 m and SPE m The nonlinear characteristics include the statistic T 2 m and SPE m ,satisfy:

[0126]

[0127] SPE m (k) = t M t M T -t m (k)t m (k) T

[0128] Among them, ∑ m Indicates t m The sample covariance matrix Σm = <t m (k) T t m (k)> k ,

[0129] t m (k) represents the kernel principal component t corresponding to the first m eigenvalues i (k) is a sequence consisting of

[0130] t m (k)=[t1(k), t2(k),…,t m (k)],t M =[t1(k), t2(k), …, t M (k)];

[0131] S104: The linear characteristics and the nonlinear characteristics are converted into statistics respectively, and compared with corresponding control limits to perform fault monitoring on the pneumatic control valve.

[0132] Calculate the linear feature statistic T 2 and SPE and statistics of nonlinear characteristics T 2 UCL and SPE UCL :

[0133] T is estimated by kernel density estimation (KDE). 2 and SPE and T 2 and SPE probability distribution, and select Gaussian kernel function as the kernel function of KDE. T can be determined by normal state variables. 2 And the control limits of the SPE statistic are calculated as:

[0134]

[0135] Among them, J∈{T 2 , SPE}, k = 1, 2, ..., M, h represents the core bandwidth, J UCL ∈{T 2 UCL , SPE UCL}, α represents a given significance level, Gaussian kernel function

[0136] For the linear characteristic state variable index T 2 and SPE and nonlinear characteristic state variable index T 2 m and SPE m Compared with the corresponding control limit, the expression for judging whether a fault occurs is:

[0137] J>J UCL

[0138] Among them, J∈{T 2 , SPE}, J UCL ∈{T 2 UCL , SPE UCL If the statistic does not exceed the control limit, the control valve is considered to be in normal operation, otherwise it is considered to be faulty.

[0139] Further, such as Figure 2 The flow chart of the pneumatic control valve fault monitoring method based on microwave measurement is shown. The control limits mentioned above can be calculated offline, for example, by calculating statistics using the kernel density estimation (KDE) method. At the same time, during offline training, the data of the normal operating state of the pneumatic control valve is collected as a sample set, and training is performed based on the sample set, and then linear and nonlinear monitoring control indicators that can be used as standards or references are obtained.

[0140] Finally, the control limits obtained through offline training are applied to the real-time online monitoring of pneumatic control valves.

[0141] The pneumatic control valve in the embodiment of the present application selects the control signal, valve inlet and outlet pressure signal, flow rate, valve stem displacement and signal as statistical monitoring variables. For example, the first 5000s of the pneumatic control valve operation process can be collected for monitoring, and the fault occurs at the 900th second. By considering the autocorrelation of the sum of squares of the process variables, the p-window length and the f-window length are both 30. For the determination of the number of singular values, it is generally considered that the number of singular values ​​is slightly larger than the singular value inflection point, so the number of singular values ​​n is taken as 80. For the threshold calculation, the control limit confidence α is taken as 0.999. The present application uses the fault detection rate (FDR) evaluation index to evaluate the detection results of the fault categories corresponding to the common faults in the operation process of the pneumatic control valve, and the specific calculation is as follows:

[0142] FDR is the ratio of correctly detected fault samples to the total number of fault samples:

[0143]

[0144] Wherein, FN represents the number of fault samples determined to be faulty, and TN represents the number of fault samples determined to be normal.

[0145] In order to deeply and intuitively evaluate the detection performance of the application technical solution, the following will be compared and explained with visual detection results. The present invention monitors 19 types of faults during the operation of the pneumatic control valve, verifies the fault monitoring performance of this method for the pneumatic control valve, and compares and analyzes with CVA, CVDA and KPCA multivariate statistical analysis methods. The above 19 types of faults can be represented by F1-F19. The F1-F19 fault types refer to the fault types in the DAMADICS (Development and Application of Methods for Actuator Diagnosis in Industrial Control Systems) benchmark platform. Each fault type can be subdivided into several different fault types according to the degree of fault. For example, AL represents a large mutation fault, AM represents a medium mutation fault, AS represents a small mutation fault, and I represents a gradual fault. For example, F1-AM represents a medium mutation fault type of F1 fault.

[0146] Table 1 summarizes the test results of different statistical methods for all faults. The best detection results for each fault are marked in bold. Faults with FDR exceeding 60% are considered to be effectively detected. KPCA only effectively detects 10 fault types corresponding to 5 faults, and its FDR is the lowest compared with other methods. CVA effectively detects 19 fault types corresponding to 10 faults, and its FDR is only higher than KPCA compared with other methods. CVDA uses T 2 ,When SPE is used as a statistic, FDR is low, so SMD is used as a statistic.,CVDA achieves effective detection of 19 fault types corresponding to 10 faults, and achieves the highest FDR for 7 fault types corresponding to 5 faults.

[0147] The present application achieves effective detection of 29 fault types corresponding to 13 faults, and achieves the highest FDR for all detectable fault types. In addition, the present invention achieves effective detection of 10 fault types F1-AS, F4-I, F10-AM, F10-AS, F11-AL, F12-AL, F12-AM, F12-AS, F19-AM and F19-AS that cannot be detected by other methods. Except for F1-AS, F4-I, F10-AS and F12-AS, the FDR exceeds 90%, and the FDR increase is more than 40%.

[0148] The present invention realizes effective detection of all mutation fault types except F8, F14, F16-AM and F16-AS. First, the mutation fault is discussed in detail by taking fault F10-AM as an example. The visual detection results are shown in Figure 2. Figure 3a-3d As shown, the dotted lines represent control limits and the solid lines represent detection statistics.

[0149] Table 1:

[0150]

[0151]

[0152] Combined with the test results in Table 1 and Figure 3a It can be seen that the CVA fault detection rate FDR is 42.06%, its detection statistics fluctuate relatively violently, and a large number of statistics are below the control limit. The CVDA fault detection rate FDR is 55.67%. Figure 3b It can be seen that the detection statistics fluctuate violently, and about half of the statistics are below the control limit. The KPCA fault detection rate FDR is 3.66%. Figure 3c It can be seen that a large number of its statistics are below the control limit and fault detection cannot be achieved. The fault detection rate FDR of the present invention is 99.95%. Figure 3d It can be seen that almost all of its statistics are above the control limit and far away from the control limit, which can indicate that all fault points can be effectively detected. Taking fault F10-AM as an example, the statistics of other statistical methods can achieve a small separation, while the pneumatic control valve fault monitoring method based on microwave measurement proposed in the present invention enhances the separation effect of fault samples by extracting the nonlinear characteristics of state variables based on the CVDA-MKPCA method.

[0153] According to the development time of the fault, the fault of the pneumatic control valve can be divided into a sudden fault and a gradual fault. The monitoring variables of the sudden fault respond quickly to the fault and change greatly, and the monitoring difficulty is relatively low. Compared with the sudden fault, the gradual fault is more difficult to monitor due to factors such as the long development time and the unclear change range of the state variable. The present invention realizes effective detection of the gradual faults F4-I, F13-I and F17-I. Taking the detection process of F17-I as an example, the detection of the gradual fault is described in detail.

[0154] The comparison of the detection results of fault F17-I is shown in the figure below: Figure 4a-4d As shown in Table 1, combined with the detection results, the CVA fault detection rate FDR is 95.66%, the CVDA fault detection rate FDR is 95.54%, and the KPCA fault detection rate FDR is 88.88%. These methods all show good fault detection rates. Figures 4a-4dIt can be seen that the statistics of CVA exceed the control limit at about 1050s, the statistics of CVDA exceed the control limit at about 1000s, the statistics of KPCA exceed the control limit at about 1300s, and the statistics of CVDA-MKPCA of the present invention exceed the control limit at about 950s and find the fault, which is at least 50s earlier than other methods. And the fault detection rate FDR of the SPE statistic of the present invention is 98.56%, T 2 The fault detection rate FDR of the statistic is 97.88%, which is higher than other methods, and can illustrate the timeliness of the invention for gradual fault detection.

[0155] Based on the same inventive concept, the present application also provides a pneumatic control valve fault monitoring device 500 based on microwave measurement, such as Figure 5 The schematic structural diagram of the pneumatic control valve fault monitoring device 500 based on microwave measurement shown includes:

[0156] The data acquisition module 501 is used to acquire the working data of the pneumatic control valve within a preset period and perform preprocessing to obtain a state signal; wherein, the acquisition of working data includes acquiring the real-time valve stem displacement data of the pneumatic control valve through a microwave sensing sensor, measuring the real-time flow signal through an ultrasonic sensor, and collecting the valve inlet and outlet pressure signals through a pressure sensor; and preprocessing the working data through an edge device to obtain the state signal, wherein the length of all state signal data in the state signal is consistent;

[0157] A CVDA processing module 502 is used to extract linear features of typical variable dissimilarity from the state signal by using a CVDA method;

[0158] An MKPCA processing module 503 is used to extract nonlinear features in typical variable dissimilarity from the state signal by using an MKPCA method;

[0159] The monitoring module 504 is used to convert the linear characteristics and the nonlinear characteristics into statistics respectively, and compare them with the corresponding control limits obtained by offline training to perform fault monitoring on the pneumatic control valve.

[0160] Furthermore, it also includes an offline training module 505, which is used to obtain the control limit based on offline training through the CVDA method and the MKPCA method.

[0161] In the above embodiment, the specific implementation method of each functional module of the pneumatic control valve fault monitoring device 500 based on microwave measurement can refer to the introduction of steps S101-S104 in the above-mentioned pneumatic control valve fault monitoring method based on microwave measurement and any optional embodiment thereof, which will not be described in detail in this embodiment.

[0162] In summary, the technical solution provided by the present application can detect faults in a timely and rapid manner, and ensure the safe and stable operation of pneumatic control valves. The present invention first uses CVDA to extract the linear characteristics of the typical variable dissimilarity of the state signal, and then constructs the MKPCA model to extract the nonlinear characteristics in the typical variable dissimilarity, and finally comprehensively extracts the linear and nonlinear characteristics to construct a fault detection index to achieve timely detection of faults. Compared with traditional multivariate statistical methods CVA, CVDA and KPCA, the method of the present invention is superior to other methods in terms of the number of fault types that can be monitored, fault detection rate and fault discovery time, and realizes the effective detection of pneumatic control valve faults and meets the detection requirements.

[0163] It should be understood that the above description of the specific embodiments of the present invention is only for illustrating the design concept and features of the present invention, and its purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly, but the scope of the present invention is not limited to the above specific implementation methods. All modifications made within the scope of the claims of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for monitoring faults of pneumatic control valves based on microwave measurement, characterized in that: include: The working data of the pneumatic control valve within a preset period is obtained and preprocessed to obtain a state signal at a preset time; wherein, the obtaining of the working data includes obtaining the real-time valve stem displacement data of the pneumatic control valve through a microwave sensing sensor, measuring the real-time flow signal through an ultrasonic sensor, and collecting the valve inlet and outlet pressure signals through a pressure sensor; and the working data is preprocessed through an edge device to obtain the state signal, and the data length of all signals in the state signal is consistent; Extracting linear features of typical variable dissimilarity from the state signal by CVDA method; Extracting nonlinear features in typical variable dissimilarity from the state signal by using the MKPCA method; The linear characteristics and the nonlinear characteristics are respectively converted into statistics, and compared with the corresponding control limits obtained by offline training, so as to perform fault monitoring on the pneumatic control valve.

2. The method according to claim 1, characterized in that: The method of extracting the linear features of the typical variable dissimilarity from the state signal by the CVDA method includes: According to the preset time at which the state signal is located, extracting from the state signal first observation data earlier than the preset time and second observation data later than the preset time; Construct a difference regularization variable d according to the difference between the first observation data and the second observation data n (t), the difference regularization variable d n (t)Satisfy: d n (t)=L n X f(t) -S n J n X p(t) S n =diag(λ1,λ2,…,λ n ) Among them, X p(t) and X f(t) are the first observation data and the second observation data respectively; L n and J n is the transformation matrix obtained by performing singular value decomposition calculation based on the first observation data and the second observation data; S n The selection criteria of singular values, λ n represents the nth value in the descending eigenvalue sequence; According to the difference regularization variable d n (t) Extracting and obtaining the linear features.

3. The method according to claim 2, characterized in that The method of extracting nonlinear features in the typical variable dissimilarity from the state signal by the MKPCA method includes: Construct the MKPCA model according to the preset hybrid kernel function; Extract the difference regularization variable d according to the MKPCA model n (t) is the principal component variable t m , obtain the nonlinear characteristics; The preset hybrid kernel function K mix ,satisfy: K mix =γK lap +βK rbf +(1-β-γ)K poly Among them, γ (0≤γ≤0.5) and β (0≤β≤0.5) are mixing coefficients that balance interpolation and extrapolation capabilities; K lap represents the Laplace kernel function, K poly represents the polynomial kernel function, K rbf represents the Gaussian kernel function, 4. The method according to claim 3, characterized in that Extract the difference regularization variable d according to the MKPCA model n (t) is the principal component variable t m , the nonlinear characteristics are obtained, including: Extract the difference regular variable d n (t) is the principal component variable t m ,satisfy: Among them, the eigenvector α=[α1,α2,..,α M ] T , i=1,2,…,m.

5. The method according to claim 4, characterized in that The linear features and the nonlinear features are respectively converted into statistics, including: Converting the linear features into statistics includes: 2 Statistics and SPE statistics satisfy: Among them, Z k,p represents the p-window state vector extracted from the past data vector, e k Represents the residual vector; converting the nonlinear features into statistics includes: T 2 m and SPE m ,satisfy: SPE m (k)=t M t M T -t m (k)t m (k) T Among them, ∑ m Indicates t m The sample covariance matrix Σ m = <t m (k) T t m (k)> k , t m (k) represents the kernel principal component t corresponding to the first m eigenvalues i (k) is a sequence consisting of t m (k)=[t1(k),t2(k),…,t m (k)],t M =[t1(k),t2(k),…,t M (k)]。 6. The method according to claim 5, characterized in that The linear feature and the nonlinear feature are respectively converted into statistics, and compared with the control limits obtained by the corresponding offline training, and the pneumatic control valve is monitored for faults, including: The control limits satisfy: P(J <J UCL )=α; Among them, J UCL is the control limit, J is the fault detection index, α is the preset significance level; J∈{T 2 , SPE}, J UCL ∈{T 2 UCL , SPE UCL }; If J <J UCL , it is determined that the pneumatic control valve is in normal operation, otherwise it is determined that the pneumatic control valve is faulty.

7. The method according to claim 3, characterized in that Also includes the mixing kernel function K mix Centralized operations are performed to meet the following requirements: K mix，c =K mix -I N K mix -K mix I N +I N K mix I N 8. The method according to claim 5, characterized in that The p-window state vector Z k,p , f-window state vector Z k,f and the residual vector e k ,satisfy: z k,p =J n X p(t) z k,f =L n X f(t) e k =F n X p(t) Among them, the transformation matrix Transformation Matrix Σ pp ,Σ ff ,Σ pf represents the covariance matrix and cross-covariance matrix of the Hankel matrix, V n represents the first n columns of V, U n represents the first n columns of U, where n is the inflection point of the singular value curve.

9. A pneumatic control valve fault monitoring device based on microwave measurement, characterized in that: include: A data acquisition module is used to acquire the working data of the pneumatic control valve within a preset period and perform preprocessing to obtain a state signal at a preset time; wherein, the acquisition of working data includes acquiring the real-time valve stem displacement data of the pneumatic control valve through a microwave sensing sensor, measuring the real-time flow signal through an ultrasonic sensor, and collecting the valve inlet and outlet pressure signals through a pressure sensor; and preprocessing the working data through an edge device to obtain the state signal, wherein the length of all state signal data in the state signal is consistent; A CVDA processing module, used for extracting linear features of typical variable dissimilarity from the state signal by using a CVDA method; An MKPCA processing module, used for extracting nonlinear features in typical variable dissimilarity from the state signal by using an MKPCA method; The monitoring module is used to convert the linear characteristics and the nonlinear characteristics into statistics respectively, and compare them with the control limits obtained by the corresponding offline training to perform fault monitoring on the pneumatic control valve.

10. The device according to claim 9, characterized in that Also includes: The offline training module is used to obtain the control limit based on offline training through the CVDA method and the MKPCA method.

Images