An aircraft airfoil fault diagnosis method based on unscented Kalman filter estimation

Abstract

The application discloses an aircraft wing fault diagnosis method based on unscented Kalman filtering estimation, and the method comprises the following steps: determining an aircraft wing fault diagnosis model set; constructing a group of parallel unscented Kalman filters according to the aircraft wing fault diagnosis model set; determining a group of parallel filtering results by using the group of parallel unscented Kalman filters according to aircraft control quantities and aircraft state quantities; and obtaining an aircraft wing fault diagnosis result by using a Bayesian rule according to the group of parallel filtering results. The method is simple to operate, can directly perform fault diagnosis on a nonlinear model of an aircraft wing fault, improves real-time performance of fault diagnosis, and effectively improves accuracy of fault diagnosis.

Description

Technical Field

[0001] This invention relates to the field of aircraft flight control systems, and more specifically to an aircraft wing surface fault diagnosis method based on unscented Kalman filter estimation. Background Technology

[0002] As a crucial component of an aircraft, the flight control system plays a vital role in stabilizing aircraft attitude, maintaining flight path, and improving flight performance. Since the 1970s, with the rapid development of direct force control, relaxed static stability technology, and active control technologies aimed at improving flight performance, the importance of the flight control system to aircraft has gradually increased. Traditional control systems have been gradually replaced by fly-by-wire systems, thus modern flight control systems widely adopt fly-by-wire technology. While single-channel fly-by-wire systems offer advantages such as high battlefield survivability, simple structure, and light weight, their reliability is lower than that of mechanical control systems. Therefore, redundancy techniques are generally used to improve system reliability, which significantly increases the complexity and scale of the flight control system. A failure in such a system can lead to incalculable losses. Therefore, when a flight control system malfunctions, the ability to quickly detect and estimate the magnitude and location of the fault, ensuring the aircraft can continue its mission or return safely, is of paramount importance.

[0003] To enhance aircraft maintainability and the reliability and safety of flight control, domestic and international scientists and scholars have developed numerous methods for diagnosing aircraft wing surface faults. Among existing methods, filter methods and state observer methods are typical examples. These methods establish a specific system model, generate residuals by comparing the model's output estimates with the actual system's measurements, extract fault features, and then determine the fault according to verification rules. However, real-world fault diagnosis targets are large, complex, and nonlinear systems. Therefore, existing methods often require linearizing these nonlinear systems, which severely impacts the robustness of fault state estimation. Summary of the Invention

[0004] To address the aforementioned shortcomings in existing technologies, this invention provides an aircraft wing surface fault diagnosis method based on unscented Kalman filter estimation. This method is simple to operate and can directly perform fault diagnosis on the nonlinear model of aircraft wing surface faults, improving the real-time performance of fault diagnosis and effectively enhancing the accuracy of fault diagnosis.

[0005] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows:

[0006] An aircraft wing surface fault diagnosis method based on unscented Kalman filter estimation includes the following steps:

[0007] S1. Determine the set of aircraft wing surface fault diagnosis models;

[0008] S2. Based on the aircraft wing surface fault diagnosis model set in step S1, construct a set of parallel unscented Kalman filters.

[0009] S3. Based on the aircraft control variables and aircraft state variables, determine a set of parallel filtering results using a set of parallel unscented Kalman filters from step S2.

[0010] S4. Based on a set of parallel filtering results from step S3, obtain the aircraft wing surface fault diagnosis results using the Bayesian criterion.

[0011] Furthermore, in step S1, the aircraft wing surface fault diagnosis model set includes a fault-free aircraft model, an aircraft model with wing damage, an aircraft model with aileron failure, an aircraft model with elevator failure, and an aircraft model with rudder failure.

[0012] Furthermore, step S2 includes the following sub-steps:

[0013] S21. Based on the aircraft wing surface fault diagnosis model set in step S1, construct a fault-free, unscented Kalman filter.

[0014] S22. Based on the aircraft wing surface fault diagnosis model set in step S1, construct an unscented Kalman filter for wing damage.

[0015] S23. Based on the aircraft wing surface fault diagnosis model set in step S1, construct a stray Kalman filter for control surface faults.

[0016] Furthermore, step S21 includes the following sub-steps:

[0017] S211. Based on the aircraft wing surface fault diagnosis model set in step S1, determine the fault-free aircraft model, represented as:

[0018]

[0019] Where: X k Let f(X) be the aircraft state vector at time k. k-1 U k-1 ) is about vector X k-1 sum vector U k-1 The one-step transfer function of the nonlinear state of a fault-free aircraft, X k-1 Let U be the aircraft state vector at time (k+1). k-1 Let W be the aircraft control vector at time (k+1). k-1 Z is the system noise vector at time (k+1). k Let h(X) be the aircraft measurement vector at time k. k ) is about vector X k The nonlinear measurement function, V kLet be the measurement noise vector at time k;

[0020] S212. Combine the fault-free aircraft model from step S211 with the unscented Kalman filter to construct a fault-free unscented Kalman filter.

[0021] Furthermore, step S22 includes the following sub-steps:

[0022] S221. Based on the aircraft wing surface fault diagnosis model set in step S1, determine the aircraft model with wing damage, represented as:

[0023]

[0024] Where: X k Let f be the aircraft state vector at time k. wing_dam (X k-1 U k-1 ) is about vector X k-1 sum vector U k-1 The one-step transfer function of the nonlinear state of the aircraft with wing damage, W k-1 Z is the system noise vector at time (k+1). k Let h(X) be the aircraft measurement vector at time k. k ) is about vector X k The nonlinear measurement function, V k Let be the measurement noise vector at time k;

[0025] S222. Combine the wing damage aircraft model from step S221 with an unscented Kalman filter to construct an initial wing damage unscented Kalman filter.

[0026] S223. Construct the fading factor matrix and introduce it into the initial wing damage unscented Kalman filter in sub-step S222 to obtain the wing damage unscented Kalman filter.

[0027] Furthermore, in step S223, constructing the fading factor matrix includes the following steps:

[0028] A1. Calculate the first intermediate variable, expressed as:

[0029]

[0030] Where: M k Let i be the first intermediate variable at time k, where i is the index of the sampling point and L is the total number of sampling points. The weights are the mean square errors of the sampling points. for, Here, T is the transpose symbol;

[0031] A2. Calculate the second intermediate variable, expressed as:

[0032] N k =Q k -R k

[0033] Where: M k Let i be the first intermediate variable at time k, where i is the index of the sampling point and L is the total number of sampling points. The weights are the mean square errors of the sampling points. This is a one-step prediction of the observation at time k. Let T be the mean of the system prediction measurements at time k, and let T be the transpose.

[0034] A3. Based on the first intermediate variable in step A1 and the second intermediate variable in step A2, calculate the fading factor, expressed as:

[0035]

[0036] Where: ρ k Let be the fading factor at time k, and tr be the trace symbol;

[0037] A4. Based on the fading factor in step A3, calculate the fading factor conversion coefficient, expressed as:

[0038]

[0039] Wherein: κ i Here, is the fading factor conversion coefficient, max is the sign of the maximum value, n is the total number of columns of the mean square error, and p i Let p be the element in the i-th row of the mean squared error matrix, where i is the row number of the mean squared error matrix, j is the column number of the mean squared error matrix, and p is the element in the i-th row of the mean squared error matrix. j Let j be the element in the j-th column of the mean square error matrix;

[0040] A5. Based on the fading factor in step A3 and the fading factor transformation coefficient in step A4, calculate the diagonal elements of the fading factor matrix, represented as:

[0041] λ i,i =ρ k κ i

[0042] Where: λ i,i The i-th diagonal element of the fading factor matrix;

[0043] A6. Based on the diagonal elements of the fading factor matrix in A5, construct the fading factor matrix as follows:

[0044] λ k =diag(λ 1,1 ,λ 2,2 ,…λi,i )

[0045] Where: λ k is the fading factor matrix, and diag is the symbol for the diagonal matrix.

[0046] Furthermore, step S23 includes the following sub-steps:

[0047] S231. Determine the initial control surface fault aircraft model based on the aircraft wing surface fault diagnosis model set in step S1.

[0048] S232. Based on the initial control surface failure aircraft model in step S231, the deflection estimate of the control surface is augmented to the state matrix using the state matrix augmentation method to determine the control surface failure aircraft model, which is expressed as:

[0049]

[0050] in: X k Let be the aircraft state vector at time k. Here, T represents the deflection estimate of the control surface, and f is the transpose sign. u_fault (X' k-1 U k-1 ) is about vector X k-1 sum vector U k-1 The one-step transfer function of the nonlinear state of an aircraft with control surface failure, Z' k Let the aircraft state vector be X' k The aircraft measurement vector at time k under the given condition, h(X') k ) is about vector X' k The nonlinear measurement function, V k Let be the measurement noise vector at time k;

[0051] S233. Combine the aircraft model with the control surface fault in step S232 with the unscented Kalman filter to construct the initial unscented Kalman filter for the control surface fault.

[0052] S234. Introduce the fading factor matrix into the initial unscented Kalman filter for rudder surface faults in sub-step S232 to obtain the unscented Kalman filter for rudder surface faults.

[0053] Furthermore, step S4 includes the following sub-steps:

[0054] S41. Based on a set of parallel filtering results from step S3, calculate the probability density of the measured values, expressed as:

[0055]

[0056] Where: L i (k) represents the i-th measurement value Z.k The probability density function, where n is the total number of unscented Kalman filters, and C i (k) is the measurement one-step prediction mean square error matrix P. ZZ,k|k-1 , ε i (k) represents the new information ε output by the i-th Kalman filter. k T is the transpose symbol;

[0057] S42. Based on the probability density of the measured values ​​in step S41, calculate the probability of each model in the aircraft wing surface fault diagnosis model set using the Bayesian criterion, expressed as:

[0058]

[0059] Where: μ i (k) represents the probability of each model in the aircraft wing surface fault diagnosis model set. Let L be the prior probability of the i-th model in the aircraft wing surface fault diagnosis model set, j be the index of each model in the aircraft wing surface fault diagnosis model set, m be the number of model types in the aircraft wing surface fault diagnosis model set, and L be the probability of the i-th model in the aircraft wing surface fault diagnosis model set. j (k) is the probability density function of the j-th measurement value z(k). Let be the prior probability of the j-th model in the set of aircraft wing surface fault diagnosis models;

[0060] S43. Based on the probability of each model in the aircraft wing surface fault diagnosis model set and the fault probability threshold of each model in step S42, obtain the aircraft wing surface fault diagnosis results.

[0061] S44. Determine whether the aircraft wing surface fault diagnosis result in sub-step S43 includes control surface faults; if yes, proceed to sub-step S45, otherwise end the operation.

[0062] S45. Use the sequential probability ratio test to determine the type of rudder surface failure in step S44.

[0063] Furthermore, step S45 includes the following sub-steps:

[0064] S451. Determine the lower bound of the threshold based on the false alarm rate and the false negative rate, expressed as:

[0065]

[0066] Where: T(H0) is the lower limit of the threshold, H0 is the control surface damage fault, β is the false alarm rate, and α is the missed alarm rate;

[0067] S452. Determine the upper limit of the threshold based on the false alarm rate and the false negative rate, expressed as:

[0068]

[0069] Where: T(H1) is the upper limit of the threshold, and H1 is the control surface jamming fault;

[0070] S453. Calculate the conditional probability density of control surface damage failure, expressed as:

[0071]

[0072] Where: H0:p[R M [|H0] represents the conditional probability density of control surface damage faults, σ ​​represents the variance of sampled faults, exp represents the exponential function, and R... M R represents the mean of the sampled faults. damage The magnitude of the damage or fault identified in the unscented Kalman filter;

[0073] S454. Calculate the conditional probability density of a control surface jamming fault, expressed as:

[0074]

[0075] Where: H1:p[R M [|H1] represents the conditional probability density of a control surface jamming fault, R stuck The size of the stuck fault identified in the unscented Kalman filter;

[0076] S455. Based on the conditional probability density of the control surface damage fault in step S453 and the conditional probability density of the control surface jamming fault in step S454, calculate the log-likelihood ratio, expressed as:

[0077]

[0078] Where: λ M (R) is the likelihood ratio, k is the accumulated cycle value of the sampled sequence, and M is the total number of times the sampled data sequence is sampled;

[0079] S456. Determine the type of rudder surface fault in step S44 based on the lower limit of the threshold in step S451, the upper limit of the threshold in step S452, and the log-likelihood ratio in step S455.

[0080] The beneficial effects of this invention are as follows:

[0081] (1) By constructing an unscented Kalman filter for the aircraft wing surface fault diagnosis model, this invention can directly perform fault diagnosis on the nonlinear model of the aircraft wing surface fault, thereby improving the real-time performance of fault diagnosis and effectively enhancing the accuracy of fault diagnosis.

[0082] (2) By introducing the fading factor matrix into the wing damage unscented Kalman filter and the control surface fault unscented Kalman filter, this invention can increase the one-step prediction mean square error matrix, thereby increasing the filter gain matrix, so as to improve the utilization weight of recent measurement data and increase the accuracy of prediction.

[0083] (3) This invention utilizes the state matrix augmentation method to augment the deflection estimate of the control surface as a state vector to the state matrix, outputs the control surface deflection estimate through the unscented Kalman filter for control surface faults, and uses the sequential probability ratio test to determine the type of control surface fault. Attached Figure Description

[0084] Figure 1 This is a schematic diagram of a fault diagnosis method for aircraft wing surfaces based on unscented Kalman filter estimation. Detailed Implementation

[0085] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0086] like Figure 1 As shown, an aircraft wing surface fault diagnosis method based on unscented Kalman filter estimation includes the following steps:

[0087] S1. Determine the set of aircraft wing surface fault diagnosis models.

[0088] In an optional embodiment of the present invention, the aircraft wing surface fault diagnosis model set includes a fault-free aircraft model, an aircraft model with wing damage, an aircraft model with aileron failure, an aircraft model with elevator failure, and an aircraft model with rudder failure.

[0089] S2. Based on the set of aircraft wing surface fault diagnosis models in step S1, construct a set of parallel unscented Kalman filters.

[0090] In an optional embodiment of the present invention, a set of parallel unscented Kalman filters is constructed based on the aircraft wing surface fault diagnosis model set in step S1. The unscented Kalman filters can estimate the system's state signals, which cannot be directly measured, based on the system's measurable output signals. When a flight control system malfunctions, the changes in internal parameters are unknown, but the changes in output are known; therefore, the constructed filters can be used to identify the system.

[0091] This invention constructs a set of parallel unscented Kalman filters based on the models in the aircraft wing surface fault diagnosis model set in step S1, and obtains the current observation value based on the estimated value at the previous time step and the estimated value at the current time step. The set of parallel unscented Kalman filters includes a fault-free unscented Kalman filter, a wing damage unscented Kalman filter, and a control surface fault unscented Kalman filter.

[0092] Specifically, this invention constructs a fault-free, traceless Kalman filter based on a fault-free aircraft model, a wing-damaged, traceless Kalman filter based on a wing-damaged aircraft model, and a control surface-damaged, traceless Kalman filter based on an aileron-damaged aircraft model, an elevator-damaged aircraft model, and a rudder-damaged aircraft model.

[0093] Step S2 includes the following sub-steps:

[0094] S21. Based on the aircraft wing surface fault diagnosis model set in step S1, construct a fault-free, traceless Kalman filter.

[0095] Step S21 includes the following sub-steps:

[0096] S211. Based on the aircraft wing surface fault diagnosis model set in step S1, determine the fault-free aircraft model, represented as:

[0097]

[0098] Where: X k Let f(X) be the aircraft state vector at time k. k-1 U k-1 ) is about vector X k-1 sum vector U k-1 The one-step transfer function of the nonlinear state of a fault-free aircraft, X k-1 Let U be the aircraft state vector at time (k+1). k-1 Let W be the aircraft control vector at time (k+1). k-1 Z is the system noise vector at time (k+1). k Let h(X) be the aircraft measurement vector at time k. k ) is about vector X k The nonlinear measurement function, V k Let be the measurement noise vector at time k.

[0099] S212. Combine the fault-free aircraft model from step S211 with the unscented Kalman filter to construct a fault-free unscented Kalman filter.

[0100] Specifically, n is the number of aircraft state variables in the fault-free aircraft model of this invention. x=3, k=0. This invention determines the proportional correction parameter α0=0.001, used to introduce the aircraft state vector X at time k. k The prior distribution's higher-order information β0 = 2. This invention combines a fault-free aircraft model and unscented Kalman filtering to construct a fault-free unscented Kalman filter.

[0101] S22. Based on the aircraft wing surface fault diagnosis model set in step S1, construct an unscented Kalman filter for wing damage.

[0102] Step S22 includes the following sub-steps:

[0103] S221. Based on the aircraft wing surface fault diagnosis model set in step S1, determine the aircraft model with wing damage, represented as:

[0104]

[0105] Where: X k Let f be the aircraft state vector at time k. wing_dam (X k-1 U k-1 ) is about vector X k-1 sum vector U k-1 The one-step transfer function of the nonlinear state of the aircraft with wing damage, W k-1 Z is the system noise vector at time (k+1). k Let h(X) be the aircraft measurement vector at time k. k ) is about vector X k The nonlinear measurement function, V k Let be the measurement noise vector at time k.

[0106] S222. Combine the wing damage aircraft model from step S221 with an unscented Kalman filter to construct an initial wing damage unscented Kalman filter.

[0107] Specifically, n is the number of aircraft state variables in the fault-free aircraft model of this invention. x =3, k=0. This invention determines the proportional correction parameter α0=0.001, used to introduce the aircraft state vector X at time k. k The prior distribution's higher-order information β0 = 2. This invention combines a wing-damaged aircraft model with unscented Kalman filtering to construct an initial wing-damaged unscented Kalman filter.

[0108] S223. Construct the fading factor matrix and introduce it into the initial wing damage unscented Kalman filter in sub-step S222 to obtain the wing damage unscented Kalman filter.

[0109] The present invention constructs the fading factor matrix by comprising the following steps:

[0110] A1. Calculate the first intermediate variable, expressed as:

[0111]

[0112] Where: M k Let i be the first intermediate variable at time k, where i is the index of the sampling point and L is the total number of sampling points. The weights are the mean square errors of the sampling points. This is a one-step prediction of the observation at time k. Let T be the mean of the system prediction measurements at time k, and let T be the transpose.

[0113] A2. Calculate the second intermediate variable, expressed as:

[0114] N k =Q k -R k

[0115] Where: N k Q is the second intermediate variable at time k. k Let R be the covariance matrix of the system noise vector at time k. k Let be the covariance matrix of the measurement noise vector at time k.

[0116] A3. Based on the first intermediate variable in step A1 and the second intermediate variable in step A2, calculate the fading factor, expressed as:

[0117]

[0118] Where: ρ k Let t be the fading factor at time k, and tr be the trace symbol.

[0119] A4. Based on the fading factor in step A3, calculate the fading factor conversion coefficient, expressed as:

[0120]

[0121] Wherein: κ i Here, is the fading factor conversion coefficient, max is the sign of the maximum value, n is the total number of columns of the mean square error, and p i Let p be the element in the i-th row of the mean squared error matrix, where i is the row number of the mean squared error matrix, j is the column number of the mean squared error matrix, and p is the element in the i-th row of the mean squared error matrix. j Let be the element in the j-th column of the mean square error matrix.

[0122] A5. Based on the fading factor in step A3 and the fading factor transformation coefficient in step A4, calculate the diagonal elements of the fading factor matrix, represented as:

[0123] λi,i =ρ k κ i

[0124] Where: λ i,i It is the i-th diagonal element of the fading factor matrix.

[0125] A6. Based on the diagonal elements of the fading factor matrix in A5, construct the fading factor matrix as follows:

[0126] λ k =diag(λ 1,1 ,λ 2,2 ,…λ i,i )

[0127] Where: λ k is the fading factor matrix, and diag is the symbol for the diagonal matrix.

[0128] S23. Based on the aircraft wing surface fault diagnosis model set in step S1, construct a stray Kalman filter for control surface faults.

[0129] Step S23 includes the following sub-steps:

[0130] S231. Based on the aircraft wing surface fault diagnosis model set in step S1, determine the initial control surface fault aircraft model.

[0131] S232. Based on the initial control surface failure aircraft model in step S231, the deflection estimate of the control surface is augmented to the state matrix using the state matrix augmentation method to determine the control surface failure aircraft model, which is expressed as:

[0132]

[0133] in: X k Let be the aircraft state vector at time k. Here, T represents the deflection estimate of the control surface, and f is the transpose sign. u _ fault (X' k-1 U k-1 ) is about vector X k-1 sum vector U k-1 The one-step transfer function of the nonlinear state of an aircraft with control surface failure, Z' k Let the aircraft state vector be X' k The aircraft measurement vector at time k under the given condition, h(X') k ) is about vector X' k The nonlinear measurement function, V k Let be the measurement noise vector at time k.

[0134] S233. Combine the aircraft model with the control surface fault from step S232 with an unscented Kalman filter to construct an initial unscented Kalman filter for the control surface fault.

[0135] Specifically, n is the number of aircraft state variables in the control surface failure aircraft model of this invention. x =4, k=0. This invention determines the proportional correction parameter α0=0.001, used to introduce the aircraft state vector X at time k. k The prior distribution's higher-order terms β0 = 2. This invention combines a control surface fault aircraft model with unscented Kalman filtering to construct an initial control surface fault unscented Kalman filter.

[0136] S234. Introduce the fading factor matrix into the initial unscented Kalman filter for rudder surface faults in sub-step S232 to obtain the unscented Kalman filter for rudder surface faults.

[0137] S3. Based on the aircraft control variables and aircraft state variables, determine a set of parallel filtering results using a set of parallel unscented Kalman filters from step S2.

[0138] In an optional embodiment of the present invention, the inputs to the set of parallel unscented Kalman filters in step S2 are both aircraft control quantities and state measurements. For the fault-free unscented Kalman filter and the wing-damaged unscented Kalman filter, the output is the measurement one-step prediction mean square error matrix P. ZZ,k|k-1 and the new information vector ε k For the unscented Kalman filter for rudder surface faults, the output is the measurement one-step prediction mean square error matrix P. ZZ,k|k-1 ε, the new information vector k and the rudder deflection estimate δ k Therefore, based on the aircraft control variables and aircraft state variables, the present invention can determine a set of parallel filtering results using a set of parallel unscented Kalman filters in step S2.

[0139] S4. Based on a set of parallel filtering results from step S3, obtain the aircraft wing surface fault diagnosis results using the Bayesian criterion.

[0140] In an optional embodiment of the present invention, based on a set of parallel filtering results in step S3, the probability of each model in the aircraft wing surface fault diagnosis model set is calculated using the Bayesian criterion. A probability threshold is set for each model, and the probability threshold is compared with the probability of each model calculated by the Bayesian criterion. If the probability value of a certain model is not less than the probability threshold, it is determined that a corresponding fault has occurred in the flight control system. Furthermore, the present invention determines whether the aircraft wing surface fault diagnosis results include control surface faults; if so, the specific type of control surface fault is determined using a sequential probability ratio test; otherwise, the operation ends.

[0141] Step S4 includes the following sub-steps:

[0142] S41. Based on a set of parallel filtering results from step S3, calculate the probability density of the measured values, expressed as:

[0143]

[0144] Where: L i (k) represents the i-th measurement value Z. k The probability density function, where n is the total number of unscented Kalman filters, and C i (k) is the measurement one-step prediction mean square error matrix P. ZZ,k|k-1 , ε i (k) represents the new information ε output by the i-th Kalman filter. k , where T is the transpose symbol.

[0145] S42. Based on the probability density of the measured values ​​in step S41, calculate the probability of each model in the aircraft wing surface fault diagnosis model set using the Bayesian criterion, expressed as:

[0146]

[0147] Where: μ i (k) represents the probability of each model in the aircraft wing surface fault diagnosis model set. Let L be the prior probability of the i-th model in the aircraft wing surface fault diagnosis model set, j be the index of each model in the aircraft wing surface fault diagnosis model set, m be the number of model types in the aircraft wing surface fault diagnosis model set, and L be the probability of the i-th model in the aircraft wing surface fault diagnosis model set. j (k) is the probability density function of the j-th measurement value z(k). Let be the prior probability of the j-th model in the set of aircraft wing surface fault diagnosis models.

[0148] S43. Based on the probability of each model in the aircraft wing surface fault diagnosis model set and the fault probability threshold of each model in step S42, obtain the aircraft wing surface fault diagnosis results.

[0149] S44. Determine whether the aircraft wing surface fault diagnosis result in sub-step S43 includes control surface faults; if yes, proceed to sub-step S45, otherwise end the operation.

[0150] S45. Use the sequential probability ratio test to determine the type of rudder surface failure in step S44.

[0151] Step S45 includes the following sub-steps:

[0152] S451. Determine the lower bound of the threshold based on the false alarm rate and the false negative rate, expressed as:

[0153]

[0154] Where: T(H0) is the lower limit of the threshold, H0 is the control surface damage fault, β is the false alarm rate, and α is the missed alarm rate.

[0155] S452. Determine the upper limit of the threshold based on the false alarm rate and the false negative rate, expressed as:

[0156]

[0157] Where: T(H1) is the upper limit of the threshold, and H1 is the control surface jamming fault.

[0158] S453. Calculate the conditional probability density of control surface damage failure, expressed as:

[0159]

[0160] Where: H0:p[R M [|H0] represents the conditional probability density of control surface damage faults, σ ​​represents the variance of sampled faults, exp represents the exponential function, and R... M R represents the mean of the sampled faults. damage This represents the magnitude of the damage or fault identified in the unscented Kalman filter.

[0161] S454. Calculate the conditional probability density of a control surface jamming fault, expressed as:

[0162]

[0163] Where: H1:p[R M [|H1] represents the conditional probability density of a control surface jamming fault, R stuck This represents the size of the stuck fault identified in the unscented Kalman filter.

[0164] S455. Based on the conditional probability density of the control surface damage fault in step S453 and the conditional probability density of the control surface jamming fault in step S454, calculate the log-likelihood ratio, expressed as:

[0165]

[0166] Where: λ M (R) is the likelihood ratio, k is the accumulated cycle value of the sampled sequence, and M is the total number of times the sampled data sequence is sampled.

[0167] S456. Determine the type of rudder surface fault in step S44 based on the lower limit of the threshold in step S451, the upper limit of the threshold in step S452, and the log-likelihood ratio in step S455.

[0168] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.

Claims

1. An aircraft wing surface fault diagnosis method based on unscented Kalman filter estimation, characterized in that, The method comprises the following steps: S1, determining an aircraft wing surface fault diagnosis model set; S2, constructing a group of parallel unscented Kalman filters according to the aircraft wing surface fault diagnosis model set in step S1; S3, determining a group of parallel filtering results by using the group of parallel unscented Kalman filters in step S2 according to the aircraft control quantity and the aircraft state quantity; S4, obtaining the aircraft wing surface fault diagnosis result by using the Bayesian rule according to the group of parallel filtering results in step S3; Step S2 comprises the following steps: S21, constructing a fault-free unscented Kalman filter according to the aircraft wing surface fault diagnosis model set in step S1; S22, constructing a wing damage unscented Kalman filter according to the aircraft wing surface fault diagnosis model set in step S1; S23, constructing a rudder surface fault unscented Kalman filter according to the aircraft wing surface fault diagnosis model set in step S1; Step S22 comprises the following steps: S221, determining a wing damage aircraft model according to the aircraft wing surface fault diagnosis model set in step S1, which is expressed as: where: is the aircraft state vector at time k is the aircraft state vector at time is the aircraft state vector at time is the aircraft state vector at time is the aircraft state vector at time is the aircraft state vector at time k is the system noise vector at time is the aircraft state vector at time k is the aircraft state vector at time is the aircraft state vector at time is the aircraft state vector at time is the measurement noise vector at time k is the measurement noise vector at time S222, combining the wing damage aircraft model in sub-step S221 with the unscented Kalman filter to construct an initial wing damage unscented Kalman filter; S223, constructing a fading factor matrix and introducing the fading factor matrix into the initial wing damage unscented Kalman filter in sub-step S222 to obtain the wing damage unscented Kalman filter; In sub-step S223, constructing the fading factor matrix comprises the following steps: A1, calculating a first intermediate variable, which is expressed as: wherein: is the first intermediate variable at the time instant k is the second intermediate variable at the time instant is the index of the sampling point, is the total number of sampling points, is the weight of the mean square error of the sampling point, is the first intermediate variable at the time instant k is the one-step prediction of the observation at the time instant is the second intermediate variable at the time instant k is the mean of the system predicted measurement at the time instant is the transposition symbol; A2, calculating a second intermediate variable, which is expressed as: wherein: is a first intermediate variable, k is a second intermediate variable, is a third intermediate variable, k is a fourth intermediate variable, is a fifth intermediate variable, k is a sixth intermediate variable. A3, calculating a fading factor according to the first intermediate variable in step A1 and the second intermediate variable in step A2, which is expressed as: wherein: is the first k is the fading factor at time is the trace sign; A4, calculating a fading factor conversion coefficient according to the fading factor in step A3, which is expressed as: in: The fading factor conversion coefficient, To determine the sign of the maximum value, This represents the total number of columns representing the mean square error. The mean square error matrix is ​​the first row element, Let be the number of rows of the mean square error matrix. Let be the number of columns in the mean square error matrix. The mean square error matrix is ​​the first Column elements; A5, calculating a diagonal element of the fading factor matrix according to the fading factor in step A3 and the fading factor conversion coefficient in A4, which is expressed as: wherein: is the jth diagonal element of the evanescent factor matrix; and is the jth diagonal element of the evanescent factor matrix. A6, constructing the fading factor matrix according to the diagonal element of the fading factor matrix in A5, which is expressed as: wherein: is an evanescent factor matrix, is a diagonal matrix of signs; Step S23 comprises the following steps: S231, determining an initial rudder surface fault aircraft model according to the aircraft wing surface fault diagnosis model set in step S1; S232, determining a rudder surface fault aircraft model by using the state matrix augmentation method to augment the deflection estimation of the rudder surface to the state matrix according to the initial rudder surface fault aircraft model in sub-step S231, which is expressed as: wherein: , is the aircraft state vector at time k , is the deflection estimate of the control surface, is the transpose symbol, is the control surface faulted aircraft nonlinear state one-step transition function with respect to the vector and the vector , is the aircraft measurement vector at time given the aircraft state vector k , is the nonlinear measurement function with respect to the vector , is the measurement noise vector at time k ; S233, combining the rudder surface fault aircraft model in sub-step S232 with the unscented Kalman filter to construct an initial rudder surface fault unscented Kalman filter; S234, introducing a fading factor matrix into the initial rudder surface fault unscented Kalman filter in sub-step S232 to obtain the rudder surface fault unscented Kalman filter.

2. The method of claim 1, wherein, In step S1, the aircraft wing surface fault diagnosis model set comprises a fault-free aircraft model, a wing damage aircraft model, an aileron fault aircraft model, an elevator fault aircraft model and a rudder fault aircraft model.

3. The method of claim 1, wherein, Step S21 comprises the following steps: S211, determining a fault-free aircraft model according to the aircraft wing surface fault diagnosis model set in step S1, which is expressed as: in: For the first k The aircraft state vector at time t. For about vectors sum vector The one-step transfer function of the nonlinear state of a fault-free aircraft. For the ( k -1) The aircraft state vector at time t(-1). For the ( k -1) The aircraft control vector at time 1. For the ( k The system noise vector at time -1) For the first k Aircraft measurement vector at time [time] For about vectors Nonlinear measurement function, For the first k The measurement noise vector at time step; S212, combine the fault-free aircraft model in step S211 with the unscented Kalman filter to construct a fault-free unscented Kalman filter.

4. The method of claim 1, wherein, Step S4 includes the following sub-steps: S41, calculate the probability density of the measurement value according to the set of parallel filter results in step S3, denoted as: wherein: is the probability density function of the i th measurement , is the total number of unscented Kalman filters, is the measurement step prediction error covariance matrix , is the innovation of the th Kalman filter output , is the transpose symbol; S42, calculate the probability of each model in the aircraft wing surface fault diagnosis model set according to the probability density of the measurement value in sub-step S41 using the Bayesian rule, denoted as: in: The probabilities of each model in the aircraft wing surface fault diagnosis model set. The first in the set of aircraft wing surface fault diagnosis models i The prior probabilities of the model, This refers to the serial number of each model in the aircraft wing surface fault diagnosis model set. The number of model types in the aircraft wing surface fault diagnosis model set. For the first j Measured values The probability density function, The first in the set of aircraft wing surface fault diagnosis models j The prior probability of the model; S43, obtain the aircraft wing surface fault diagnosis result according to the probability of each model in the aircraft wing surface fault diagnosis model set in sub-step S42 and the fault probability threshold of each model; S44, determine whether the aircraft wing surface fault diagnosis result in sub-step S43 includes a control surface fault; if yes, proceed to sub-step S45, otherwise end the operation; S45, determine the type of the control surface fault in sub-step S44 using the sequential probability ratio test.

5. The method of claim 4, wherein, Step S45 includes the following sub-steps: S451, determine the threshold lower limit according to the false alarm rate and the false dismissal rate, denoted as: wherein: is a threshold lower bound, is a rudder surface damage fault, is a false alarm rate, is a missed alarm rate; S452, determine the threshold upper limit according to the false alarm rate and the false dismissal rate, denoted as: wherein: is a threshold upper bound, is a rudder stuck failure; S453, calculate the conditional probability density of the control surface damage fault, denoted as: wherein: is the conditional probability density of a rudder surface damage fault, is the variance of the sampling fault, is the exponential function, is the mean of the sampling fault, is the identified damage fault size in the unscented Kalman filter; S454, calculate the conditional probability density of the control surface jam fault, denoted as: wherein: is the conditional probability density of a stuck failure of the control surface, is the stuck failure size identified in the unscented Kalman filter. S455, calculate the log-likelihood ratio according to the conditional probability density of the control surface damage fault in sub-step S453 and the conditional probability density of the control surface jam fault in S454, denoted as: wherein: is a likelihood ratio, is a cyclic value accumulated for the sample sequence, is the total number of sample data sequences; S456, determine the type of the control surface fault in sub-step S44 according to the threshold lower limit in sub-step S451, the threshold upper limit in sub-step S452, and the log-likelihood ratio in sub-step S455.

Images