Controller reference power supply fault diagnosis and correction method based on compressed sensing

Abstract

The present invention relates to a controller reference power supply fault diagnosis and correction method based on compressed sensing, comprising the following steps: systematically analyzing the internal reference power supply circuit of an aircraft engine controller, establishing a correlation model, and obtaining a correlation graphical model and a mathematical model matrix; establishing a linearized fault equation based on the correlation model; reconstructing the fault mode from the measured values and the fault feature matrix based on an orthogonal matching pursuit algorithm to achieve fault diagnosis of the reference power supply; and evaluating the offset ratio of all components by calculating the projection length of the measured values in different fault subspaces, thereby achieving correction of the reference power supply and correction of the measured values. The present invention improves the fault detection rate of the system, reduces the demand for test points, and can simultaneously locate single-point faults and multi-point combination faults, providing an effective means for system health management, and providing a reference power supply deviation correction method to improve the measurement accuracy of the controller.

Description

Technical Field

[0001] The present invention belongs to the field of overall design of aero-engine control systems, and in particular relates to a controller reference power supply fault diagnosis and correction method based on compressed sensing. Background Art

[0002] Aircraft engine control systems are the engine's central control center. As a key component of the control system, the reliability of the electronic controller is crucial to the engine's safe operation. A voltage reference is a high-precision, high-stability power supply within the controller's internal circuits, typically providing a reference for other operating voltages. The accuracy of the reference power supply is closely related to the operating environment's temperature, humidity, and long-term operational stability. External or internal factors can cause the reference power supply to attenuate, drift, or even fail, leading to deviations in numerous related measurement parameters and severely impacting the control accuracy and performance of the entire system.

[0003] Currently, the controller can monitor, locate, and isolate some power supply faults through resources such as BITs and power supply alarm circuits. However, there are still two deficiencies: First, the system only judges and processes the normal and fault failure modes of the controller's internal power supply, but lacks system-level impact analysis, anomaly detection, and correction measures for abnormal and decaying states other than normal and faulty. Second, the controller contains some unmeasurable reference power supplies, requiring positive test design and verification based on internal power supply correlations. This fully utilizes the test resources within the controller to achieve 100% coverage of critical power supply fault detection and improve system testability and reliability. Summary of the Invention

[0004] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a controller reference power supply fault diagnosis and correction method based on compressed sensing that can improve the testability and reliability of the control system.

[0005] According to the technical solution provided by the present invention, the controller reference power supply fault diagnosis and correction method based on compressed sensing includes the following steps:

[0006] Step 1: Conduct a systematic analysis of the internal reference power supply circuit of the aircraft engine controller, establish a correlation model, and obtain a correlation graphical model and a mathematical model D matrix;

[0007] Step 2: Based on the correlation model, establish the linearized fault equation:

[0008] y=Ax+w

[0009] Where y is the M-dimensional measurement value, A∈C M×N is the fault feature matrix, w is the measurement noise;

[0010] Step 3: Based on the orthogonal matching pursuit algorithm, the fault mode x is reconstructed from the measured value y and the fault feature matrix A to achieve fault diagnosis of the reference power supply;

[0011] Step 4: By calculating the projection length l of the measured value in different fault subspaces, the offset ratio k of all components is evaluated, and then the reference power supply correction is implemented to correct the measured value.

[0012] Preferably, the step 1 is specifically as follows:

[0013] 1-1) Through system analysis, determine all possible fault components F in the reference power supply circuit n and available test points T m , where n = 1, 2, ..., N, m = 1, 2, ..., M, N represents the total number of components, and M represents the total number of test points;

[0014] 1-2) Analyze the correlation between components and test points and establish a correlation diagram model; in the correlation diagram model, a box is used to represent the component F n , the circle represents the test point T m ,The arrows indicate the flow of functional information;

[0015] 1-3) According to the functional information flow in the correlation diagram model, establish component F n and test point T m The D matrix model D M×N The D matrix model is a correlation mathematical model, using d mn Denotes the matrix D M×N The element in row m and column n, d mn The value of is:

[0016]

[0017] Preferably, the step 2 is specifically as follows:

[0018] For a complex system, there are usually multiple types of failures; an N-dimensional vector x is used to represent the failure mode, x n represents the nth fault, n=1,2,…,N, assuming the fault is binary, then x n The value of can be expressed as,

[0019]

[0020] For failure mode x i ∈{0,1}, the linearized fault equation is,

[0021] y=Ax+w

[0022] Where y is the M-dimensional measurement value, A∈C M×N is the fault feature matrix, and w is the measurement noise.

[0023] The principle of compressed sensing refers to a method that can reconstruct the original signal from a small number of measurements when the original signal is sparse; the sparsity means that most of the elements in the vector are 0;

[0024] The above fault equation satisfies the following conditions:

[0025] Since the probability of each fault occurring is very low, only a few elements of vector x are non-zero, making it a sparse vector.

[0026] Due to limitations such as sensor settings, the object parameters that can be measured are very limited, so the dimension of the measured value is much smaller than the dimension of the fault mode, that is, M<<N;

[0027] The fault equation meets the conditions of compressed sensing. The compressed sensing algorithm can be used to reconstruct the fault mode through the measurement value y and the fault feature matrix A to achieve anomaly detection and fault diagnosis.

[0028] The specific process of establishing the fault equation from the correlation model is as follows:

[0029] 2-1) The measured value y is obtained from all the test points T in step 1) m , m=1,2,…,M measurement data relative deviation composition, that is

[0030]

[0031] Among them, t m Indicates the test point T m The test data, t 0m Indicates the test point T m The baseline value;

[0032] 2-2) Failure mode x represents component F in step 1) n , n=1,2,…,N whether a fault occurs, that is

[0033]

[0034] 2-3) The fault feature matrix A is obtained by the matrix model D in step 1) M×N Each column vector is multiplied by the proportion of the corresponding fault on the measured value y, and then normalized to obtain, that is,

[0035]

[0036] Where n = 1, 2, ..., N, A n and D n Represents A and D respectivelyM×N The nth column vector, K n is an M-dimensional column vector, representing component F n The influence ratio of each element on the measurement value y.

[0037] Preferably, the step 3 is specifically as follows:

[0038] 3-1) Set the initial value: residual r 0 =y, failure mode x=0, index set Iteration counter q = 0;

[0039] 3-2) Update counter: q = q +1 ;

[0040] 3-3) Find the index value λ q : Solve the optimization problem λ q =argmax n＝1,…,N | n ,r q >|, if there are multiple solutions, choose one;

[0041] 3-4) Update parameters: Ω q =Ω q-1 ∪{λ q};

[0042] 3-5) Update failure mode: solve optimization problem x q =argmin supp(x)＝Ωq ||Ax-y||2;

[0043] 3-6) Update residual: r q =y-Ax q ;

[0044] 3-7) When q=Q, the algorithm ends, x Q This is the desired failure mode. Otherwise, repeat steps 3-2) to 3-7).

[0045] Preferably, the step 4 is specifically as follows:

[0046] 4-1) Calculate the projection length l and record the failure mode x Q The index vector corresponding to the faulty component is f, A Q =A f The fault feature vector corresponding to the faulty component is represented by the projection length l of the subspace corresponding to the column vector of A.

[0047]

[0048] Where l is an N-dimensional column vector, and the subspace projection length of the fault-free component is 0; ​

[0049] 4-2) Evaluate component offset ratio, |k n |The larger the component F n The larger the deviation, the n By l n Calculated,

[0050]

[0051] 4-3) This step mainly calibrates the reference power supply and calculates the test point T according to the offset ratio. m The measurement data t m The correction value of is,

[0052]

[0053] Wherein, c represents the component index value corresponding to the reference power supply.

[0054] The advantages of the present invention are as follows:

[0055] 1. The present invention establishes a correlation model between components and test points by performing correlation analysis on the controller reference power supply, making components in the system that were originally not directly measurable explicit, and then realizing fault detection of components by establishing a fault strategy, thereby improving the fault detection rate of the system.

[0056] 2. The present invention proposes a controller reference power supply fault diagnosis based on a compressed sensing algorithm. By establishing a fault feature matrix, the mapping between all relevant components of the system and the test points is completed. By utilizing the sparsity of the fault mode vector, the fault positions of multiple components can be located using a small amount of test point data. Compared with traditional threshold methods, fault tree methods, etc., on the one hand, it can reduce the demand for test points, and on the other hand, in addition to locating single-point faults, it can also locate multi-point combined faults.

[0057] 3. The present invention calculates the deviation ratio of each component and evaluates the degree of component deviation, thereby realizing the monitoring of abnormalities such as attenuation and drift of all components, providing an effective means for system health management.

[0058] 4. Based on anomaly detection and fault diagnosis, the present invention proposes a reference power supply correction method. By calculating the reference power supply offset ratio, the measurement data can be corrected, which can avoid measurement deviation caused by reference power supply attenuation in the system and improve the measurement accuracy of the controller. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a flowchart of the overall process of the method of the present invention.

[0060] Figure 2 This is a controller reference power model according to an embodiment of the present invention.

[0061] Figure 3 This is a correlation diagram model according to an embodiment of the present invention.

[0062] Figure 4 This is a comparison chart of the calculated value and actual value of the relative deviation of the reference power supply for test case 8 of the present invention.

[0063] Figure 5 This is a comparison chart of the calculated value and actual value of the relative deviation of the reference power supply for test case 9 of the present invention.

[0064] Figure 6-1 This is a curve diagram of the original voltage measurement values of U3, U4, and U5.

[0065] Figure 6-2 The figure shows the comparison between the original voltage measurement value of U3 and the measurement value after calibration of U3.

[0066] Figure 6-3 The figure shows the comparison between the original voltage measurement value of U4 and the measurement value after calibration of U4.

[0067] Figure 6-4 The figure shows the comparison between the original voltage measurement value of U5 and the measurement value after calibration of U5. DETAILED DESCRIPTION

[0068] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0069] by Figure 2 The controller power supply model in the example is an object, and the model features are:

[0070] The power supply model includes regulated power supplies U1 through U5, whose voltage values are dimensionless and are marked in the figure. The measurement noise w is white noise. The A / D acquisition module converts analog quantities into digital quantities. The digital value converted from reference power supply U1 serves as a reference. The digital values converted from other analog quantities are proportional to this reference value and converted to corresponding measured values. Therefore, reference power supply U1 itself cannot be measured. However, its failure or degradation can cause measurement deviations in all measurements of the A / D acquisition module, making it a critical component of the system. The measured value deviations of regulated power supplies U3, U4, and U5 can be used to monitor U1 for faults. However, the measured values of U3, U4, and U5 are also related to the actual values of power supplies U2 through U5. Therefore, it is necessary to comprehensively consider the correlation of the model to locate the fault and correct the deviation of reference power supply U1.

[0071] The specific embodiment includes the following steps:

[0072] 1) Perform a systematic analysis of the internal reference power supply circuit of the aircraft engine controller, establish a correlation model, and obtain a correlation graphical model and a mathematical model D matrix. The specific steps are as follows:

[0073] 1-1) Through system analysis, determine all possible fault components F in the reference power supply circuit n and available test points T m as follows,

[0074] F1: Reference power supply U1

[0075] F2: regulated power supply U2

[0076] F3: regulated power supply U3

[0077] F4: regulated power supply U4

[0078] F5: regulated power supply U5

[0079] T1: U3 measurement value

[0080] T2: U4 measurement value

[0081] T3: U5 measurement value

[0082] Among them, the total number of components is 5 and the total number of test points is 3.

[0083] 1-2) Establish a correlation diagram model based on the circuit connection relationship and the principle of the A / D acquisition module, using a box to represent the component F n , the circle represents the test point T m ,The arrows indicate the flow of functional information, e.g. Figure 3 shown.

[0084] 1-3) According to the functional information flow in the correlation diagram model, establish component F n and test point T m The D matrix model D M×N .

[0085]

[0086] 2) According to the correlation model and based on the principle of compressed sensing, a linearized fault equation is established.

[0087] 2-1) The measured value y is obtained from all the test points T in step 1) m , m=1,2,3 measurement data relative deviation composition, that is

[0088]

[0089] Among them, t m Indicates the test point T m The test data, t 0m They are 3, 1, and 2 respectively.

[0090] 2-2) Use failure mode x to represent component F in step 1)n , n=1,2,…,5 whether a fault occurs, that is

[0091]

[0092] 2-3) The fault feature matrix A is obtained by the matrix model D in step 1) M×N Each column vector is multiplied by the proportion of the corresponding fault's impact on the measured value y, and then normalized. That is,

[0093] A1=[1 1 1] T / |[1 1 1] T |=[0.5774 0.5774 0.5774] T

[0094] A2=[1 1 0] T |[1 1 0] T |=[0.7071 0.7071 0] T

[0095] A3=[1 0 0] T

[0096] A4=[0 1 0] T

[0097] A5=[0 0 1] T

[0098] Then we get the measurement matrix

[0099] 3) Using the OMP (Orthogonal Matching Pursuit) algorithm, the fault mode x is reconstructed from the measured value y and the fault feature matrix A. Where Q = 2, meaning that at most two components are considered to have simultaneous faults. Faults are injected into different components. The fault diagnosis results of the embodiment of the present invention are shown in Table 1. For test cases involving single-point faults, two-point combined faults, attenuation, and disconnection faults of different components, the reconstructed fault mode x is consistent with the actual fault injection location, demonstrating that the present invention, based on the principle of compressed sensing, achieves fault diagnosis of the control reference power supply and improves the system's fault detection rate.

[0100] Table 1 Fault diagnosis results

[0101]

[0102] 4) For the test cases in Table 1, the projection length l of the measured value in different fault subspaces is calculated, and the offset ratio k of all components is calculated. The results are shown in the last column of Table 1. Compared with the injected deviation, the calculated offset ratio is basically consistent with the actual deviation value. In test cases 8 and 9, the reference power supply injects a deviation that decays over time. The calculated values of k1 over time are respectively Figure 4 and Figure 5 The above results show that the component offset ratio k calculated by the present invention can accurately reflect the actual deviation of the component.

[0103] For test case 9 in Table 1, i.e., the combined failure of the reference power supply U1 attenuation and the regulated power supply U4 deviation, the reference power supply and the measured value are calibrated. The test results of the embodiment of the present invention are as follows: Figure 6-1 to Figure 6-4 As shown in the figure, due to the attenuation of the reference power supply U1, the A / D conversion voltage acquisition value deviates. By applying the correction method of the present invention, the corrected acquisition value is consistent with the true value, eliminating the influence of the attenuation of the reference power supply U1 on the acquisition value and improving the acquisition accuracy of the controller.

Claims

1. A controller reference power supply fault diagnosis and correction method based on compressed sensing, characterized by The method comprises the following steps: Step 1: Perform a systematic analysis of the internal reference power supply circuit of the aircraft engine controller, establish a correlation model, and obtain a correlation graphical model and a mathematical model D matrix; specifically: 1-1) Through system analysis, determine all possible fault components F in the reference power supply circuit n and available test points T m , where n = 1, 2, ..., N, m = 1, 2, ..., M, N represents the total number of components, and M represents the total number of test points; 1-2) Analyze the correlation between components and test points and establish a correlation diagram model; in the correlation diagram model, a box is used to represent the component F n , the circle represents the test point T m ,The arrows indicate the flow of functional information; 1-3) According to the functional information flow in the correlation diagram model, establish component F n and test point T m The D matrix model D M×N The D matrix model is a correlation mathematical model, using d mn Denotes the matrix D M×N The element in row m and column n, d mn The value of is: Step 2: Based on the correlation model, establish the linearized fault equation: y=Ax+w Where y is the M-dimensional measurement value, A∈C M×N is the fault feature matrix, x represents the fault mode, and w is the measurement noise; specifically: 2-1) The measured value y is obtained from all the test points T in step 1) m , m=1,2,…,M measurement data relative deviation composition, that is Among them, t m Indicates the test point T m The test data, t 0m Indicates the test point T m The baseline value; 2-2) Failure mode x represents component F in step 1) n , n=1,2,…,N whether a fault occurs, that is 2-3) The fault feature matrix A is obtained by the matrix model D in step 1) M×N Each column vector is multiplied by the proportion of the corresponding fault on the measured value y, and then normalized to obtain, that is, Where n = 1, 2, ..., N, A n and D n Represents A and D respectively M×N The nth column vector, K n is an M-dimensional column vector, representing component F n The influence ratio of each element on the measured value y; Step 3: Based on the orthogonal matching pursuit algorithm, the fault mode x is reconstructed from the measured value y and the fault feature matrix A to realize the fault diagnosis of the reference power supply; specifically: 3-1) Set the initial value: residual r 0 =y, failure mode x=0, index set Iteration counter q = 0; 3-2) Update counter: q = q + 1; 3-3) Find the index value λ q : Solve the optimization problem λ q =argmax n＝1,…,N | n ,r q >|, if there are multiple solutions, choose one;​ 3-4) Update parameters: Ω q =Ω q-1 ∪{λ q }; 3-5) Update failure mode: Solve the optimization problem 3-6) Update residual: r q =y-Ax q ; 3-7) When q=Q, the algorithm ends, x Q This is the desired fault mode, otherwise repeat steps 3-2) to 3-7); Step 4: By calculating the projection length l of the measured value in different fault subspaces, the offset ratio k of all components is evaluated, and then the reference power supply correction is performed to correct the measured value; specifically: 4-1) Calculate the projection length l and record the failure mode x Q The index vector corresponding to the faulty component is f, A Q =A f The fault feature vector corresponding to the faulty component is represented by the projection length l of the subspace corresponding to the column vector of A. Where l is an N-dimensional column vector, and the subspace projection length of the fault-free component is 0; 4-2) Evaluate component offset ratio, |k n |The larger the component F n The larger the deviation, the n By l n Calculated, 4-3) Calibrate the reference power supply and calculate the test point T according to the offset ratio m The measurement data t m The correction value of is, Wherein, c represents the component index value corresponding to the reference power supply.

Images