CN114186578A - Gas turbine electric actuating mechanism fault diagnosis method based on signal decomposition - Google Patents

Gas turbine electric actuating mechanism fault diagnosis method based on signal decomposition Download PDF

Info

Publication number
CN114186578A
CN114186578A CN202111001150.0A CN202111001150A CN114186578A CN 114186578 A CN114186578 A CN 114186578A CN 202111001150 A CN202111001150 A CN 202111001150A CN 114186578 A CN114186578 A CN 114186578A
Authority
CN
China
Prior art keywords
formula
dimensional
decomposition
low
signal
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202111001150.0A
Other languages
Chinese (zh)
Inventor
孙建平
高文捷
田乐乐
李朝雅
张文广
牛玉广
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
North China Electric Power University
Original Assignee
North China Electric Power University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by North China Electric Power University filed Critical North China Electric Power University
Priority to CN202111001150.0A priority Critical patent/CN114186578A/en
Publication of CN114186578A publication Critical patent/CN114186578A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/02—Preprocessing
    • G06F2218/04—Denoising
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching

Abstract

The invention provides a fault diagnosis method for an electric actuating mechanism of a gas turbine based on signal decomposition, and an online diagnosis platform is constructed. Determining the type of possible faults of the electric actuating mechanism, connecting the electric actuating mechanism with a real object part by using a dSPACE system and acquiring data; eliminating noise interference by using a Variational Modal Decomposition (VMD) and Hilbert (Hilbert) transformation combined improved variational modal decomposition (VHT) method, and extracting fault characteristics; the secondary dimensionality reduction visualization is carried out by using a t-distribution random neighborhood embedding algorithm (t-SNE), so that the fault characteristics are more obvious; and finishing classification diagnosis by using k-means clustering. The invention can make up the defects of low diagnosis accuracy rate caused by mode aliasing, multiple decomposition layers, long time consumption and unobvious fault characteristics of the traditional method, and can improve the diagnosis efficiency and the diagnosis accuracy rate.

Description

Gas turbine electric actuating mechanism fault diagnosis method based on signal decomposition
Technical Field
The invention belongs to the technical field of online fault diagnosis of electric actuators, and particularly relates to a fault diagnosis method of a gas turbine electric actuator based on signal decomposition.
Background
In recent years, with the complication and intellectualization of a new type of industrialization process, industrial control systems are often characterized by nonlinearity and uncertainty. The actuator is the most important component of the control system and can precisely move the valve of the regulating valve to any position. The regulating valve is used as an execution terminal part of the whole control system, and has great influence on the quality of the control system; however, because of the harsh use environment, the actuating mechanism, especially the regulating valve, is prone to various faults. According to statistics, the fault of the regulating valve accounts for more than one third of the total fault of the control system. Therefore, the fault diagnosis technology of the executing mechanism is researched, the accuracy of fault diagnosis is improved, and the method has important significance for guaranteeing safe and stable operation of the control process.
In order to effectively extract the fault information and to make the diagnosis precision higher, it is necessary to decompose the original signal. The conventional decomposition method such as Empirical Mode Decomposition (EMD) has problems of a large number of decomposition layers, low decomposition efficiency, mode confusion and the like, and the problems can occur because hilbert-yellow transform (HHT) is based on the empirical mode decomposition. The Variational Modal Decomposition (VMD) can effectively improve the decomposition efficiency and inhibit the modal confusion problem due to the advantages of the solid mathematical basis and the self-adaptive decomposition. Combining the variable component with Hilbert transform to form an improved variable component modal decomposition (VHT) method, then utilizing a t-distribution random neighborhood embedding algorithm (t-SNE) to reduce dimension for the second time and visualize, and finally finishing diagnosis through k-means clustering. The method can improve the diagnosis accuracy and enable the diagnosis result to be displayed more visually. At present, the method is not widely applied to fault diagnosis of the electric actuating mechanism.
Disclosure of Invention
In order to solve the problems existing in the background art, the invention provides a fault diagnosis method for an electric actuator of a gas turbine based on signal decomposition, which is characterized by comprising the following steps of:
step 1, collecting normal and fault data of equipment as original data, and calculating the current flow before and after the valve:
the method comprises the steps that the output flow Q of a servo valve cylinder and the valve port flow Q measured under a given opening degree are calculated to obtain the flow difference delta Q between the front and the back of a valve as an input signal;
step 2, carrying out variation modal decomposition on the original flow data, namely sequentially carrying out variation problem construction and variation problem solving; obtaining k modal components;
and step 3: performing Hilbert transform on each modal component obtained by the variation modal decomposition for the second time, setting parameters of the Hilbert transform for the second time according to experience, and performing operation; extracting frequency domain characteristic quantities of all modal components to be combined to obtain characteristic quantities of original flow signals;
and 4, step 4: constructing a high-dimensional characteristic matrix, carrying out VHT (very high-speed) transformation on data groups, dividing the data groups into 10 groups under the same condition, and sequentially paralleling the 10 groups of results into a high-dimensional output matrix as the input of the next step;
and 5: sending the result of the step 4 into a t-distribution random neighborhood embedding algorithm for dimension reduction visualization, and carrying out secondary processing on the data; obtaining a visual image of a classification result, and outputting a two-dimensional characteristic matrix for final classification; the t-distribution random neighborhood embedding algorithm is specifically to set algorithm parameters, and perform secondary dimensionality reduction and extraction on the high-dimensional feature matrix obtained in the step 4;
step 6: and (5) sending the two-dimensional feature matrix obtained in the step (5) into a k-means classifier for clustering, and finishing classification and identification of the fault signals subjected to the t-SNE secondary dimensionality reduction extraction to obtain a fault diagnosis result.
The step 1 specifically comprises the following steps:
step 11, calculating the flow of the servo valve cylinder body:
Figure BDA0003235689380000021
wherein Q is the output flow of the cylinder body and the unit is m3/s；a1Is the piston diameter in m; a is2Is the diameter of the piston rod, and the unit is m; Δ T is the relative velocity of the piston in m/s;delta is an empirical error value in m3/s；
Step 12, calculating the flow of a valve port of the servo valve:
Figure BDA0003235689380000022
wherein q is the flow rate of the valve port of the servo valve and the unit is m3S; tau is an empirical coefficient; y is the motion displacement of the valve core, and the unit is m; delta P is the relative pressure of the valve core, and the unit is MPa; rho is the fluid density in kg/m3；
And step 13, calculating the flow difference between the front and the rear of the valve as follows:
ΔQ＝Q-q (3)
in the formula: Δ Q is the flow difference before and after the valve, in m3S; q is the output flow of the cylinder body and is m3S; q is the flow rate of the valve port of the servo valve and is m3/s。
The step 2 specifically comprises:
step 21: the variation modal decomposition redefines the modal function into an amplitude modulation-frequency modulation signal;
step 22: for each mode function, solving an analysis signal of each mode function by utilizing first Hilbert transform to obtain a single-side frequency spectrum;
step 23: modulating the frequency spectrum to a corresponding base frequency band according to the mixed pre-estimated central frequency;
step 24: calculating the time gradient L of the demodulated signal2Squaring the norm, estimating the bandwidth of the modal component, introducing constraint conditions, and constructing a constrained variation model;
the expression of the constraint variational model is as follows:
Figure BDA0003235689380000031
in the formula: δ (t) is the dirichlet function; is a convolution operation; { phikThe variable mode decomposition is a set of K Intrinsic Mode Functions (IMF) after the variable mode decomposition;
Figure BDA0003235689380000032
a set of K modal component center frequencies; k is the number of decomposition modes set by the variational mode decomposition; f represents an input signal; u. ofkRepresenting the respective center frequencies of the K modal components;
step 25: when solving the variation problem, a penalty factor alpha is introduced to ensure the signal reconstruction accuracy under the condition of Gaussian noise, and a Lagrange multiplier initially selected according to experience is introduced to keep the constraint condition strict
Figure BDA0003235689380000033
A and
Figure BDA0003235689380000034
the introduction of (2) converts the constrained variation problem into an unconstrained variation problem;
step 26: solving the variation problem in step 25 by using an alternative direction multiplicative operator to obtain k modal components; specifically, the method comprises the following steps:
by alternating updates
Figure BDA0003235689380000035
Finding a 'saddle point' of an extended Lagrange table formula, and alternately updating the specific formula as follows:
Figure BDA0003235689380000036
in the formula:
Figure BDA0003235689380000037
is the phi obtained after the alternate updatek、
Figure BDA0003235689380000038
A value of (d);
Figure BDA0003235689380000039
lagrange functions constructed for solution using the ADMM algorithm; A. b, c is the constraint of equation (4) < >k、
Figure BDA00032356893800000310
The coefficient obtained by the previous calculation is 3 constants; n is an algebra, the formula needs to be updated circularly for a plurality of iterations in an alternating mode, the initial value is taken as the first generation, namely n is 1; ρ is a penalty factor.
In the step 3, performing a second hilbert transform on the component signal of each inherent mode function to obtain a corresponding instantaneous frequency; the hilbert transform formula is:
Figure BDA00032356893800000311
in the formula: h (t) represents a time-frequency function of the Hilbert transform; p represents a Cauchy principal value; m iskRepresenting each natural mode function (IMF);
after the second hilbert transform is performed, the original flow rate signal Δ q (t) can be expressed as follows:
Figure BDA0003235689380000041
in the formula, H (t, ω) represents a function of time and instantaneous frequency in the hilbert transform space, i.e., a time-frequency function in formula (6); biRepresents a switching factor; a. thei(t) represents a magnitude function; omegai(t) represents an instantaneous frequency.
The t-distribution random neighborhood embedding algorithm in the step 5 comprises the following steps:
step 51: taking the high dimensional data sequence H (x) { x }1,x2,...,xnSolving the high-dimensional data point x according to the following formulaiAnd xjConditional probability distribution p betweenj|i
Figure BDA0003235689380000042
In the formula: x is the number ofi、xjIs inputting a high dimensional datasetTwo data points of (a); p is a radical ofj|iIs a high dimensional data point xiAnd xjA conditional probability distribution therebetween; xiiIs xiA gaussian distribution variance for the center point, determined by a given confusion and binary search; m is the mth high-dimensional space sample data point from i but not equal to i;
step 52: computing a joint probability density p of high dimensional samplesij
Figure BDA0003235689380000043
In the formula: p is a radical ofj|iAnd pi|jIs a high dimensional data point xiAnd xjA conditional probability distribution therebetween; p is a radical ofijIs the joint probability density of the high dimensional samples; n is the number of data points;
step 53: initializing sample data Y of low dimensional space(0)
Y(0)＝{y1,y2,...,yn} (10)
In the formula: y is(0)A low-dimensional space sample data set is obtained; y is1To ynA low-dimensional spatial sample data point;
step 54: calculating a joint probability density f of low-dimensional spatial sample points based on a t-distribution with a degree of freedom of 1ijAnd gradient
Figure BDA0003235689380000044
Figure BDA0003235689380000045
In the formula: y isiAnd yjA low-dimensional space sample data point reduced to a low dimension; u is the u-th low-dimensional spatial sample data point from i but not equal to i; f. ofijIs the joint probability density of the low-dimensional space sample points;
Figure BDA0003235689380000046
in the formula, C is a cost function; y is a low-dimensional space sample data set; y isiAnd yjA low-dimensional space sample data point reduced to a low dimension; p is a radical ofijA joint probability density for the high dimensional samples; f. ofijIs the joint probability density of the low-dimensional space sample points;
Figure BDA0003235689380000051
is the gradient of the objective function;
step 55: update output
Figure BDA0003235689380000052
In the formula, r is iteration times; eta is the learning rate; t is a momentum factor;
Figure BDA0003235689380000053
is the gradient of the objective function; y is a low-dimensional space sample data set;
step 56: if the iteration number r is satisfied, the iteration is stopped, otherwise, the process returns to the step 54.
The invention has the beneficial effects that:
1. the method has the advantages of high diagnosis accuracy and suitability for fault diagnosis of the actuating mechanism of the gas turbine control system, the problem that the fault features are weak and difficult to identify can be greatly improved by the unique advantage of VHT on weak feature extraction, misjudgment and late judgment are reduced, and the diagnosis accuracy is improved.
2. The Hilbert-Huang transform (HHT) extracts fault features by using Empirical Mode Decomposition (EMD) decomposition, nonlinear data are converted into linear signals, so that the problem of severe mode confusion exists, and the Empirical Mode Decomposition (EMD) method has no strict mathematical basis and has certain loopholes in use, so that the Hilbert-Huang transform (HHT) has the defects of mode confusion, low decomposition efficiency and unsatisfactory fault feature extraction effect. The problem encountered by the Variable Mode Decomposition (VMD) can be effectively solved by replacing the Empirical Mode Decomposition (EMD), and the quality of signal decomposition is greatly improved, so that the fault diagnosis accuracy is improved.
Drawings
FIG. 1 is a flow chart of an embodiment of a method for diagnosing a fault of an electric actuator of a gas turbine based on signal decomposition according to the present invention;
FIG. 2 is a schematic diagram of a modified variational modal decomposition (VHT) architecture used in the present invention;
FIG. 3 is a block diagram of a method for diagnosing a fault of an electric actuator of a gas turbine based on signal decomposition according to the present invention;
FIG. 4 is a diagram showing the result of fault diagnosis of a terminal device according to the method for fault diagnosis of an electric actuator of a gas turbine based on signal decomposition;
FIG. 5 is a comparison graph of the diagnosis effect of the method for diagnosing the fault of the electric actuator of the gas turbine based on signal decomposition according to the invention.
Detailed Description
The embodiments are described in detail below with reference to the accompanying drawings.
The invention carries out fault diagnosis by analyzing and processing flow signals of an electric actuating mechanism of a gas turbine, wherein the electric actuating mechanism comprises a servo motor, a servo valve, a speed reducer, a flow pipeline, a water tank, a position generator and the like, and the electric actuating mechanism controls an adjusting valve to change the opening of the valve through electric signals so as to control the flow.
The fault diagnosis method used by the gas turbine electric actuator shown in fig. 1 includes:
step 1, collecting normal and fault data of equipment as original data, and randomly dividing the data into 10 groups; and then calculating the current flow of the front valve and the rear valve of each group:
the method comprises the steps that the output flow Q of a servo valve cylinder and the valve port flow Q measured under a given opening degree are calculated to obtain the flow difference delta Q between the front and the back of a valve as an input signal; the method is specifically divided into the following steps:
step 11, calculating the flow of the servo valve cylinder body:
Figure BDA0003235689380000061
wherein Q is the output flow of the cylinder body and the unit is m3/s；a1Is the piston diameter in m; a is2Is the diameter of the piston rod, and the unit is m; Δ T is the relative velocity of the piston in m/s; delta is an empirical error value in m3/s。
Step 12, calculating the flow of a valve port of the servo valve:
Figure BDA0003235689380000062
wherein q is the flow rate of the valve port of the servo valve and the unit is m3S; tau is an empirical coefficient; y is the motion displacement of the valve core, and the unit is m; delta P is the relative pressure of the valve core, and the unit is MPa; rho is the fluid density in kg/m3。
And step 13, calculating the flow difference between the front and the rear of the valve as follows:
ΔQ＝Q-q (3)
in the formula: Δ Q is the flow difference before and after the valve, in m3S; q is the output flow of the cylinder body and is m3S; q is the flow rate of the valve port of the servo valve and is m3/s。
Step 2, carrying out variation modal decomposition on each group, namely sequentially carrying out variation problem construction and variation problem solving; obtaining k modal components; the method specifically comprises the following steps:
step 21: redefining the IMF as an amplitude modulation-frequency modulation signal by the variation modal decomposition;
step 22: for each mode function, performing a first Hilbert transform (Hilbert transform) to obtain an analysis signal so as to obtain a single-side frequency spectrum;
step 23: modulating the frequency spectrum to a corresponding base frequency band according to the mixed pre-estimated central frequency;
step 24: calculating the time gradient L of the demodulated signal2Squaring the norm, estimating the bandwidth of the modal component, introducing constraint conditions, and constructing a constrained variation model;
the expression of the constraint variational model is as follows:
Figure BDA0003235689380000071
in the formula: δ (t) is the dirichlet function; is a convolution operation; { phikThe variable mode decomposition is a set of K Intrinsic Mode Functions (IMF) after the variable mode decomposition;
Figure BDA0003235689380000072
a set of K modal component center frequencies; k is the number of decomposition modes set by the variational mode decomposition; f represents an input signal; u. ofkRepresenting the center frequencies of each of the K modal components.
Step 25: when solving the variation problem, a penalty factor alpha is introduced to ensure the signal reconstruction accuracy under the condition of Gaussian noise, and a Lagrange multiplier initially selected according to experience is introduced to keep the constraint condition strict
Figure BDA0003235689380000073
A and
Figure BDA0003235689380000074
the introduction of (2) converts the constrained variation problem into an unconstrained variation problem;
step 26: solving the variation problem in step 25 by using an alternating direction multiplicative operator (ADMM) to obtain k modal components; specifically, the method comprises the following steps:
by alternating updates
Figure BDA0003235689380000075
Finding a 'saddle point' of an extended Lagrange table formula, and alternately updating the specific formula as follows:
Figure BDA0003235689380000076
in the formula:
Figure BDA0003235689380000077
is the phi obtained after the alternate updatek、
Figure BDA0003235689380000078
A value of (d);
Figure BDA0003235689380000079
lagrange functions constructed for solution using the ADMM algorithm; A. b, c is the constraint of equation (4) < >k、
Figure BDA00032356893800000710
The coefficient obtained by the previous calculation is 3 constants; n is an algebra, the formula needs to be updated circularly for a plurality of iterations in an alternating mode, the initial value is taken as the first generation, namely n is 1; alpha is a penalty factor.
And step 3: performing Hilbert transform on each modal component obtained by the variation modal decomposition for the second time, setting parameters of the Hilbert transform for the second time according to experience, and performing operation; extracting frequency domain characteristic quantities of all modal components to be combined to obtain characteristic quantities of original flow signals; specifically, the method comprises the following steps:
performing Hilbert transform on the component signal of each inherent mode function for the second time to obtain corresponding instantaneous frequency; the hilbert transform formula (the same formula is used for both hilbert transforms) is:
Figure BDA0003235689380000081
in the formula: h (t) represents a time-frequency function of the Hilbert transform; p represents a Cauchy principal value; m iskEach IMF is represented.
After the second hilbert transform, the original signals Δ q (t) can be expressed as follows:
Figure BDA0003235689380000082
in the formula, H (t, ω) represents a function of time and instantaneous frequency in the hilbert transform space, i.e., a time-frequency function in formula (6); biRepresents a switching factor; a. thei(t) represents a magnitude function; omegai(t) represents an instantaneous frequency.
Aiming at the problems that a signal decomposition part in the traditional fault diagnosis method cannot adapt to weak fault characteristics, the number of decomposition layers is complex, mode confusion and the like, the invention utilizes an improved variational mode decomposition algorithm (VHT) consisting of the step 3 and the step 4, namely, Variational Mode Decomposition (VMD) is combined with Hilbert transformation to form a signal decomposition method with stronger adaptability; the flow of the VHT algorithm structure of the present invention is shown in fig. 2.
And 4, step 4: constructing a high-dimensional characteristic matrix, and paralleling 10 groups of results into a high-dimensional output matrix in sequence to be used as input of the next step;
and 5: sending the result of the step 4 into a t-distribution random neighborhood embedding algorithm (t-SNE) for dimension reduction visualization, and carrying out secondary processing on the data; obtaining a visual image of a classification result, and outputting a two-dimensional characteristic matrix for final classification; the t-distribution random neighborhood embedding algorithm is specifically to set algorithm parameters, and perform secondary dimensionality reduction and extraction on a high-dimensional feature matrix constructed (step 4) after the improved secondary Hilbert transformation; specifically, the t-distribution random neighborhood embedding algorithm comprises the following steps:
step 51: taking the high dimensional data sequence H (x) { x }1,x2,...,xnSolving the high-dimensional data point x according to the following formulaiAnd xjConditional probability distribution p betweenj|i
Figure BDA0003235689380000083
In the formula: x is the number ofi、xjAre two data points in the input high-dimensional dataset; p is a radical ofj|iIs a high dimensional data point xiAnd xjA conditional probability distribution therebetween; xiiIs xiA gaussian distribution variance for the center point, determined by a given confusion and binary search; m is the mth high-dimensional spatial sample data point from i but not equal to i.
Step 52: computing a joint probability density p of high dimensional samplesij
Figure BDA0003235689380000091
In the formula: p is a radical ofj|iAnd pi|jIs a high dimensional data point xiAnd xjA conditional probability distribution therebetween; p is a radical ofijIs the joint probability density of the high dimensional samples; n is the number of data points.
Step 53: initializing sample data Y of low dimensional space(0)
Y(0)＝{y1,y2,...,yn} (10)
In the formula: y is(0)A low-dimensional space sample data set is obtained; y is1To ynAre low-dimensional spatial sample data points.
Step 54: calculating a joint probability density f of low-dimensional spatial sample points based on a t-distribution with a degree of freedom of 1ijAnd gradient
Figure BDA0003235689380000092
Figure BDA0003235689380000093
In the formula: y isiAnd yjA low-dimensional space sample data point reduced to a low dimension; u is the u-th low-dimensional spatial sample data point from i but not equal to i; f. ofijIs the joint probability density of the low-dimensional spatial sample points.
Figure BDA0003235689380000094
In the formula, C is a cost function defined by an example; y is a low-dimensional space sample data set; y isiAnd yjA low-dimensional space sample data point reduced to a low dimension; p is a radical ofijA joint probability density for the high dimensional samples; f. ofijIs the joint probability density of the low-dimensional space sample points;
Figure BDA0003235689380000095
is the gradient of the objective function.
Step 55: update output
Figure BDA0003235689380000096
In the formula, r is iteration times; eta is the learning rate; t is a momentum factor;
Figure BDA0003235689380000097
is the gradient of the objective function; and Y is a low-dimensional space sample data set.
Step 56: if the iteration times r are met, stopping iteration, otherwise, returning to the step 54;
step 6: and (5) sending the two-dimensional feature matrix obtained in the step (5) into a k-means classifier for clustering, and finishing classification and identification of the fault signals subjected to the t-SNE secondary dimensionality reduction extraction to obtain a fault diagnosis result.
And 7: and connecting the final output of the algorithm module to terminal equipment, and displaying a diagnosis result in real time to realize online diagnosis.
Fig. 3 is a block diagram of a fault diagnosis system based on signal decomposition for an electric actuator, which includes an electric actuator module, a controller module, a fault type module, a signal receiving and processing module, a fault diagnosis algorithm module, and a terminal device. The electric actuating mechanism module comprises a servo valve module and a servo motor module.
The electric actuator module is an established electric actuator model and comprises a servo valve module and a servo motor module, and the control system is used for controlling the flow of the model.
The controller module is a PID controller and is used for tracking errors, adjusting and controlling the flow of the hydraulic actuator module, and performing closed-loop control on the electric actuator module to enable the actual flow to be closer to the ideal flow.
The fault type module is formed by connecting a dSPACE system and a physical part and is used for determining the type of the fault which can occur.
The signal receiving and processing module adopts an algorithm of improved variational modal decomposition (VHT), performs variational modal decomposition on input data, and performs Hilbert transform on the decomposed data to obtain the partial output. The good performance of the improved algorithm is utilized in the part, the noise of the original signal is eliminated, the weak fault signal is better extracted, and a high-quality input signal is provided for the subsequent diagnostic algorithm.
And the fault diagnosis algorithm module is used for constructing the data output by the signal receiving and processing module into a high-dimensional characteristic matrix, sending the high-dimensional characteristic matrix into a t-distribution random neighborhood embedding algorithm (t-SNE) for secondary dimensionality reduction and visualization, and finally sending the high-dimensional characteristic matrix into k-means clustering to complete classification. Because of the excellent data processing capability of the t-distribution random neighborhood embedding algorithm (t-SNE), the output of the t-distribution random neighborhood embedding algorithm is sent into k-means clusters, and higher diagnosis accuracy can be achieved.
And the terminal equipment is used for visually representing the diagnosis result obtained by the fault diagnosis algorithm module and realizing online real-time data monitoring.
The terminal equipment fault diagnosis result display interface of the fault diagnosis online control system based on signal decomposition of the electric actuating mechanism is shown in fig. 4. As shown in fig. 4, it can be seen through the terminal image recorder that there is no overlap or boundary ambiguity between the faults, the results of fault diagnosis are well shown, and at the same time, the records thereof can be archived.
The diagnosis effect graph of the gas turbine electric actuator fault diagnosis algorithm based on the signal decomposition is shown in FIG. 5. The improved variational modal decomposition (VHT) and the t-distribution random neighborhood embedding algorithm (t-SNE) are combined to obtain an excellent diagnosis effect, all faults are not overlapped or fuzzy in boundary, noise is well inhibited, weak fault features are effectively extracted, and high diagnosis accuracy is achieved.
The present invention is not limited to the above embodiments, and any changes or substitutions that can be easily made by those skilled in the art within the technical scope of the present invention are also within the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims (5)

1. A fault diagnosis method for an electric actuator of a gas turbine based on signal decomposition is characterized by comprising the following steps:
step 1, collecting normal and fault data of equipment as original data, and calculating the current flow before and after the valve:
the method comprises the steps that the output flow Q of a servo valve cylinder and the valve port flow Q measured under a given opening degree are calculated to obtain the flow difference delta Q between the front and the back of a valve as an input signal;
step 2, carrying out variation modal decomposition on the original flow data, namely sequentially carrying out variation problem construction and variation problem solving; obtaining k modal components;
and step 3: performing Hilbert transform on each modal component obtained by the variation modal decomposition for the second time, setting parameters of the Hilbert transform for the second time according to experience, and performing operation; extracting frequency domain characteristic quantities of all modal components to be combined to obtain characteristic quantities of original flow signals;
and 4, step 4: constructing a high-dimensional characteristic matrix, carrying out VHT (very high-speed) transformation on data groups, dividing the data groups into 10 groups under the same condition, and sequentially paralleling the 10 groups of results into a high-dimensional output matrix as the input of the next step;
and 5: sending the result of the step 4 into a t-distribution random neighborhood embedding algorithm for dimension reduction visualization, and carrying out secondary processing on the data; obtaining a visual image of a classification result, and outputting a two-dimensional characteristic matrix for final classification; the t-distribution random neighborhood embedding algorithm is specifically to set algorithm parameters, and perform secondary dimensionality reduction and extraction on the high-dimensional feature matrix obtained in the step 4;
step 6: and (5) sending the two-dimensional feature matrix obtained in the step (5) into a k-means classifier for clustering, and finishing classification and identification of the fault signals subjected to the t-SNE secondary dimensionality reduction extraction to obtain a fault diagnosis result.
2. The method for diagnosing the fault of the electric actuator of the gas turbine based on the signal decomposition as claimed in claim 1, wherein the step 1 is specifically as follows:
step 11, calculating the flow of the servo valve cylinder body:
Figure FDA0003235689370000011
wherein Q is the output flow of the cylinder body and the unit is m3/s；a1Is the piston diameter in m; a is2Is the diameter of the piston rod, and the unit is m; Δ T is the relative velocity of the piston in m/s; delta is an empirical error value in m3/s；
Step 12, calculating the flow of a valve port of the servo valve:
Figure FDA0003235689370000012
wherein q is the flow rate of the valve port of the servo valve and the unit is m3S; tau is an empirical coefficient; y is the motion displacement of the valve core, and the unit is m; delta P is the relative pressure of the valve core, and the unit is MPa; rho is the fluid density in kg/m3；
And step 13, calculating the flow difference between the front and the rear of the valve as follows:
ΔQ＝Q-q (3)
in the formula: Δ Q is the flow difference before and after the valve, in m3S; q is the output flow of the cylinder body and is m3S; q is the flow rate of the valve port of the servo valve and is m3/s。
3. The method for diagnosing the fault of the electric actuator of the gas turbine based on the signal decomposition as claimed in claim 1, wherein the step 2 specifically comprises:
step 21: the variation modal decomposition redefines the modal function into an amplitude modulation-frequency modulation signal;
step 22: for each mode function, solving an analysis signal of each mode function by utilizing first Hilbert transform to obtain a single-side frequency spectrum;
step 23: modulating the frequency spectrum to a corresponding base frequency band according to the mixed pre-estimated central frequency;
step 24: calculating the time gradient L of the demodulated signal2Squaring the norm, estimating the bandwidth of the modal component, introducing constraint conditions, and constructing a constrained variation model;
the expression of the constraint variational model is as follows:
Figure FDA0003235689370000021
in the formula: δ (t) is the dirichlet function; is a convolution operation; { phikThe variable mode decomposition is a set of K Intrinsic Mode Functions (IMF) after the variable mode decomposition;
Figure FDA0003235689370000026
a set of K modal component center frequencies; k is the number of decomposition modes set by the variational mode decomposition; f represents an input signal; u. ofkRepresenting the respective center frequencies of the K modal components;
step 25: when solving the variation problem, a penalty factor alpha is introduced to ensure the signal reconstruction accuracy under the condition of Gaussian noise, and a Lagrange multiplier initially selected according to experience is introduced to keep the constraint condition strict
Figure FDA0003235689370000022
A and
Figure FDA0003235689370000023
the introduction of (2) converts the constrained variation problem into an unconstrained variation problem;
step 26: solving the variation problem in step 25 by using an alternative direction multiplicative operator to obtain k modal components; specifically, the method comprises the following steps:
by alternating updates
Figure FDA0003235689370000024
Finding a 'saddle point' of an extended Lagrange table formula, and alternately updating the specific formula as follows:
Figure FDA0003235689370000025
in the formula:
Figure FDA0003235689370000031
is the phi obtained after the alternate updatek、
Figure FDA0003235689370000032
A value of (d);
Figure FDA0003235689370000033
lagrange functions constructed for solution using the ADMM algorithm; A. b, c is the constraint of equation (4) < >k、
Figure FDA0003235689370000034
The coefficient obtained by the previous calculation is 3 constants; n is an algebra, the formula needs to be updated circularly for a plurality of iterations in an alternating mode, the initial value is taken as the first generation, namely n is 1; ρ is a penalty factor.
4. The method for diagnosing the fault of the electric actuator of the gas turbine based on the signal decomposition as claimed in claim 1, wherein in the step 3, the component signal of each natural mode function is subjected to a second hilbert transform to obtain a corresponding instantaneous frequency; the hilbert transform formula is:
Figure FDA0003235689370000035
in the formula: h (t) represents a time-frequency function of the Hilbert transform; p represents a Cauchy principal value; m iskRepresenting each natural mode function (IMF);
after the second hilbert transform is performed, the original flow rate signal Δ q (t) can be expressed as follows:
Figure FDA0003235689370000036
in the formula, H (t, ω) represents a function of time and instantaneous frequency in the hilbert transform space, i.e., a time-frequency function in formula (6); biRepresents a switching factor; a. thei(t) represents a magnitude function; omegai(t) represents an instantaneous frequency.
5. The method for diagnosing the fault of the electric actuator of the gas turbine based on the signal decomposition as claimed in claim 1, wherein the t distribution random neighborhood embedding algorithm in the step 5 comprises the following steps:
step 51: taking the high dimensional data sequence H (x) { x }1,x2,...,xnSolving the high-dimensional data point x according to the following formulaiAnd xjConditional probability distribution p betweenj|i
Figure FDA0003235689370000037
In the formula: x is the number ofi、xjAre two data points in the input high-dimensional dataset; p is a radical ofj|iIs a high dimensional data point xiAnd xjA conditional probability distribution therebetween; xiiIs xiA gaussian distribution variance for the center point, determined by a given confusion and binary search; m is the mth high-dimensional space sample data point from i but not equal to i;
step 52: computing a joint probability density p of high dimensional samplesij
Figure FDA0003235689370000038
In the formula: p is a radical ofj|iAnd pi|jIs a high dimensional data point xiAnd xjA conditional probability distribution therebetween; p is a radical ofijIs the joint probability density of the high dimensional samples; n is the number of data points;
step 53: initializing sample data Y of low dimensional space(0)
Y(0)＝{y1,y2,...,yn} (10)
In the formula: y is(0)A low-dimensional space sample data set is obtained; y is1To ynA low-dimensional spatial sample data point;
step 54: calculating a joint probability density f of low-dimensional spatial sample points based on a t-distribution with a degree of freedom of 1ijAnd gradient
Figure FDA0003235689370000043
Figure FDA0003235689370000041
In the formula: y isiAnd yjA low-dimensional space sample data point reduced to a low dimension; u is the u-th low-dimensional spatial sample data point from i but not equal to i; f. ofijIs the joint probability density of the low-dimensional space sample points;
Figure FDA0003235689370000044
in the formula, C is a cost function; y is a low-dimensional space sample data set; y isiAnd yjA low-dimensional space sample data point reduced to a low dimension; p is a radical ofijA joint probability density for the high dimensional samples; f. ofijIs the joint probability density of the low-dimensional space sample points;
Figure FDA0003235689370000045
is the gradient of the objective function;
step 55: update output
Figure FDA0003235689370000042
In the formula, r isThe number of iterations; eta is the learning rate; t is a momentum factor;
Figure FDA0003235689370000046
is the gradient of the objective function; y is a low-dimensional space sample data set;
step 56: if the iteration number r is satisfied, the iteration is stopped, otherwise, the process returns to the step 54.
CN202111001150.0A 2021-08-30 2021-08-30 Gas turbine electric actuating mechanism fault diagnosis method based on signal decomposition Pending CN114186578A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111001150.0A CN114186578A (en) 2021-08-30 2021-08-30 Gas turbine electric actuating mechanism fault diagnosis method based on signal decomposition

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111001150.0A CN114186578A (en) 2021-08-30 2021-08-30 Gas turbine electric actuating mechanism fault diagnosis method based on signal decomposition

Publications (1)

Publication Number Publication Date
CN114186578A true CN114186578A (en) 2022-03-15

Family

ID=80601044

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111001150.0A Pending CN114186578A (en) 2021-08-30 2021-08-30 Gas turbine electric actuating mechanism fault diagnosis method based on signal decomposition

Country Status (1)

Country Link
CN (1) CN114186578A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114994517A (en) * 2022-07-04 2022-09-02 哈尔滨理工大学 Soft fault diagnosis method for analog circuit
CN115356631A (en) * 2022-10-24 2022-11-18 新黎明科技股份有限公司 Motor state monitoring method and system under high-dimensional variable

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114994517A (en) * 2022-07-04 2022-09-02 哈尔滨理工大学 Soft fault diagnosis method for analog circuit
CN115356631A (en) * 2022-10-24 2022-11-18 新黎明科技股份有限公司 Motor state monitoring method and system under high-dimensional variable

Similar Documents

Publication Publication Date Title
Grezmak et al. Interpretable convolutional neural network through layer-wise relevance propagation for machine fault diagnosis
Tang et al. A novel adaptive convolutional neural network for fault diagnosis of hydraulic piston pump with acoustic images
Dai et al. Signal-based intelligent hydraulic fault diagnosis methods: Review and prospects
CN111830408A (en) Motor fault diagnosis system and method based on edge calculation and deep learning
CN114186578A (en) Gas turbine electric actuating mechanism fault diagnosis method based on signal decomposition
CN110096810B (en) Industrial process soft measurement method based on layer-by-layer data expansion deep learning
Xu et al. An imbalanced fault diagnosis method for rolling bearing based on semi-supervised conditional generative adversarial network with spectral normalization
Li et al. Fault diagnosis method based on encoding time series and convolutional neural network
CN116167668B (en) BIM-based green energy-saving building construction quality evaluation method and system
Ayodeji et al. Causal augmented ConvNet: A temporal memory dilated convolution model for long-sequence time series prediction
CN103714045A (en) Information fusion estimation method for asynchronous multi-rate non-uniform sampled observation data
CN114169110B (en) Motor bearing fault diagnosis method based on feature optimization and GWAA-XGboost
Hu et al. Data simulation by resampling—A practical data augmentation algorithm for periodical signal analysis-based fault diagnosis
GB2581524A (en) Facial shape representation and generation system and method
Ma et al. Multimodal convolutional neural network model with information fusion for intelligent fault diagnosis in rotating machinery
Li A comprehensive survey of sparse regularization: Fundamental, state-of-the-art methodologies and applications on fault diagnosis
Wang et al. A novel data augmentation method for intelligent fault diagnosis under speed fluctuation condition
CN114117954B (en) Dynamic real-time visualization method for three-dimensional reaction field in reactor
CN106054167A (en) Intensity filter-based multi-extended target tracking method
Wang et al. A RUL prediction of bearing using fusion network through feature cross weighting
Xue et al. A novel framework for motor bearing fault diagnosis based on multi-transformation domain and multi-source data
CN116468767B (en) Airplane surface reconstruction method based on local geometric features and implicit distance field
CN115824519B (en) Comprehensive diagnosis method for valve leakage faults based on multi-sensor information fusion
CN110701487A (en) KPCA and Cas-SVDD-based multi-working-condition pipeline leakage detection method
CN116070126A (en) Aviation plunger pump oil distribution disc abrasion detection method and system based on countermeasure self-supervision

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination