A Model-Based Method for Estimating Faults in Atmospheric Data

Abstract

This invention discloses a model-based method for atmospheric data fault estimation, comprising the following steps: acquiring measurement information; constructing the state equation and measurement equation of a dual-model filter; updating the dual-model filter using an unscented Kalman filter; calculating the model probability of the dual-model filter and determining whether an atmospheric data system fault has occurred based on the model probability; identifying the fault information source using the residual component chi-square detection method when a fault occurs, and selectively reinitializing the dual-model filter; calculating and outputting atmospheric data estimates and atmospheric data fault estimates based on the state estimation results of the dual-model filter. This method can promptly detect bias faults in the atmospheric data system and ensure accurate estimation of airspeed, angle of attack, and sideslip angle under fault conditions, further improving the reliability of the atmospheric data system and playing a significant role in ensuring safe flight of aircraft.

Description

Technical Field

[0001] This invention belongs to the field of sensor fault diagnosis technology, and in particular relates to a model-based method for estimating atmospheric data faults. Background Technology

[0002] The atmospheric data system is a key airborne avionics device for modern military and civilian aircraft to perceive surrounding atmospheric information during flight. It utilizes total / static pressure sensors, total temperature probes, and angle-of-attack sensors to measure information such as total pressure, static pressure, total temperature, and angle of attack. An atmospheric data computer processes and interprets the output signals from each sensor, providing crucial atmospheric parameters such as airspeed, angle of attack, and sideslip angle to the airborne navigation and flight control systems. However, in actual flight, atmospheric data sensors exposed to the air are susceptible to wear, weathering, and icing, which can lead to atmospheric data system malfunctions and incorrect flight parameter outputs. The reliability of the atmospheric data system's output data is critical to the safe flight of the aircraft. Therefore, appropriate fault diagnosis and handling measures must be implemented for the atmospheric data system to ensure the safe and stable flight of the aircraft even after a potential failure of the atmospheric data system.

[0003] Currently, the air data systems of major military and civilian aircraft both domestically and internationally generally adopt an architecture based on redundant sensor configuration and voting mechanisms. This involves voting on and integrating multiple air data signals through statistical consistency checks. While sensor fault diagnosis methods based on redundant signal voting are mature, they also have limitations, such as the inability to identify fault types and characteristics, and the risk of difficulty in correctly isolating faults when two or more sensors experience common-mode faults.

[0004] The dual-model-based sensor fault estimation method replaces physical entities with mathematical models, enabling the estimation of key atmospheric parameters such as airspeed, angle of attack, and sideslip angle without increasing hardware redundancy. By comparing the consistency of the estimated parameters with prior information, the consistency between the estimated parameters and the sensor can be determined, and fault signal identification can be achieved through real-time estimation of fault information. This model-based sensor fault estimation method reduces hardware redundancy, lowers operating costs, and ensures operational safety. It represents a key research direction for meeting increasingly stringent future airworthiness requirements and further improving the usability of atmospheric data systems. Summary of the Invention

[0005] To achieve the above objectives, this invention provides a model-based method for atmospheric data fault estimation, comprising the following steps:

[0006] Collect measurement information from inertial navigation systems and atmospheric data systems;

[0007] Based on the measurement information, the state equation and measurement equation of the dual-model filter are constructed.

[0008] An unscented Kalman filter is used to update the dual-model filter.

[0009] Calculate the model probability of the dual-model filter, and use the model probability to determine whether the atmospheric data system has malfunctioned;

[0010] When a fault occurs, the source of the fault information is found by the chi-square test of the residual components, and the dual-model filter is selectively reinitialized.

[0011] The atmospheric data estimates and atmospheric data fault estimates are calculated and output based on the state estimation results of the dual-model filter.

[0012] Preferably, the dual model includes a fault-free model and a fault model;

[0013] The measurement information includes inertial navigation system information and atmospheric data system information:

[0014] The inertial navigation system information includes roll rate, pitch rate, yaw rate, acceleration components on the three axes of the aircraft system, roll angle, and pitch angle;

[0015] The atmospheric data system information includes airspeed, angle of attack, and sideslip angle.

[0016] Preferably, the method for constructing the state equation and measurement equation of the dual-model filter includes:

[0017] The state equations of a dual model are constructed based on the measurement input of the inertial navigation system and the measurement noise of the inertial navigation system.

[0018] The measurement equations of the dual model are constructed based on the airspeed, angle of attack, and sideslip angle measurements of the atmospheric data system, the measurement noise of the airspeed, angle of attack, and sideslip angle of the atmospheric data system, and the observation matrix of the dual model filter.

[0019] Preferably, the process of updating the dual-model filter using unscented Kalman filtering includes:

[0020] The Sigma point set that approximates the nonlinear density function of the dual model is obtained by using UT transformation;

[0021] Calculate the weights of the sampling points of the dual models, calculate the one-step prediction of the state of the dual models, and perform time updates;

[0022] Use UT transformation again to update the Sigma point set of the bimodel;

[0023] Substitute the Sigma point set into the measurement equation and calculate the mean and covariance of the predicted measurement values ​​of the two models using weighted averages.

[0024] The filtering gain of the dual model is calculated based on the mean and covariance.

[0025] Update the state estimates and error covariance matrix of the dual model.

[0026] Preferably, the method for calculating the model probabilities of the dual-model filter includes:

[0027] The innovation and innovation covariance matrix of the dual model are obtained based on the filtering update.

[0028] The conditional probability density function is obtained based on the innovation and the innovation covariance matrix;

[0029] Based on the conditional probability density function, the model probability of the bi-model is recursively calculated using Bayes' theorem.

[0030] Preferably, the method for identifying the source of fault information using the chi-square method of residual components includes:

[0031] The detection threshold is determined by the false detection rate;

[0032] The fault detection value of the j-th innovation component of the model is constructed based on the j-th element of the fault-free filter innovation and the j-th diagonal element of the fault-free filter innovation covariance matrix; j = 1, 2, 3 represent airspeed, angle of attack, and sideslip angle, respectively.

[0033] When the fault detection value is greater than the detection threshold, the j component is faulty; when the fault detection value is less than or equal to the detection threshold, the j component is not faulty. The source of the fault information can be determined by the value of j.

[0034] Preferably, the method for selectively reinitializing the dual-model filter includes:

[0035] When the atmospheric data system malfunctions, both the fault-free model and the faulty model are initialized. When the atmospheric data system is functioning correctly, the faulty model is initialized, while the fault-free model is not initialized.

[0036] Preferably, the method for calculating the estimated values ​​of atmospheric data and atmospheric data faults includes:

[0037] The state estimate of the fault-free model filter is weighted with the first three state variables of the fault model filter to calculate the atmospheric data estimate.

[0038] The last three-dimensional state variables of the fault filter model are weighted and calculated to obtain the fault estimate of the atmospheric data.

[0039] Compared with the prior art, the present invention has the following advantages and technical effects:

[0040] This invention provides a model-based method for estimating atmospheric data faults. Compared with existing technologies, this invention utilizes mathematical models to replace physical sensors, effectively reducing the hardware redundancy requirements of traditional atmospheric data systems. The designed selective re-initialization algorithm enables input interaction between the two models, which can not only detect the existence of faults in key atmospheric data (airspeed, angle of attack, sideslip angle) in a timely manner, but also determine the specific source of fault information and identify the magnitude of the fault. Furthermore, this invention can accurately estimate airspeed, angle of attack, and sideslip angle under atmospheric data system fault conditions, improving the availability of atmospheric data systems in common-mode fault scenarios, which is of great significance for ensuring safe flight of aircraft. Attached Figure Description

[0041] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0042] Figure 1 This is a flowchart of the atmospheric data fault estimation method according to an embodiment of the present invention. Detailed Implementation

[0043] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0044] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0045] like Figure 1 As shown, this application proposes a model-based method for atmospheric data fault estimation, including the following steps:

[0046] Step 1: In this embodiment, the measurement information of the inertial navigation system and the atmospheric data system at time k is obtained through the simulation experiment cycle. The specific process is as follows:

[0047] This embodiment first establishes a flight path for a five-sided approach at Lhasa Gonggar Airport, with a flight duration of 800 seconds, including ascent, turning, descent, acceleration, and deceleration. The initial attitude, velocity, and position information are as follows:

[0048] [0° 0° 180° 0m / s 0m / s 0m / s 90.906802° 29.296007° 3600m]

[0049] Under this flight path, the simulation outputs measurement information from the inertial navigation system and the atmospheric data system. Specifically, the gyroscope drift is set to 0.1° / h, and the accelerometer drift is set to 0.001m / s². 2 The white noise error for airspeed measurement in the air data system is 0.8 m / s; the white noise error for angle of attack measurement is 0.2°; and the white noise error for sideslip angle measurement is 0.2°. The sampling frequency for both the inertial navigation system and the air data system is 50 Hz.

[0050] The output measurement information includes: roll rate p, pitch rate q, yaw rate r, and acceleration components A along the three axes of the inertial navigation system obtained from the inertial navigation system. x A y A z Roll angle φ, pitch angle θ; airspeed obtained from the atmospheric data system Angle of attack α m Sideslip angle β m .

[0051] To verify the effectiveness of the proposed technical solution in detecting faults in atmospheric data systems, and to compare the accuracy of traditional methods with the proposed method in estimating airspeed faults, angle of attack faults, sideslip angle faults, and airspeed, angle of attack, and sideslip angle, this embodiment sets up the fault scenarios shown in Table 1, that is, injecting corresponding faults into the actual measurement information of the atmospheric data system.

[0052] Table 1

[0053]

[0054] Step 2: Construct the state equations and measurement equations for the fault-free model filter and the fault model filter. The specific method is as follows:

[0055] The state of the fault-free model filter is: x nf =[V a α β] T The state of the fault model filter is: The state equations for the two models are as follows:

[0056]

[0057]

[0058] In the formula, V a α and β are the state estimates of airspeed, angle of attack and sideslip angle, respectively; f α ,f β These are faults related to airspeed, angle of attack, and sideslip angle, respectively; f(x) nf ,u) and f(x) f,u) are the nonlinear functions of the fault-free model and the fault model, respectively; u=[A x A y A z pqr φ θ] T It is the measurement input of the inertial navigation system in step 1; w = [w ax w ay w az w p w q w r w φ w θ ] T It is system measurement noise; G nf and G f These are the system noise matrices for the fault-free model filter and the faulty model filter, respectively. The relevant functions and matrix calculation formulas are as follows:

[0059]

[0060]

[0061]

[0062]

[0063]

[0064] The measurement equations for the two models are as follows:

[0065]

[0066]

[0067] In the formula: V am ,α m ,β m These are the measurements from the atmospheric data system in step 1; It is measurement noise; H nf =I 3×3 and H f =[H nf I 3×3 ] are the observation matrices of the fault-free model filter and the fault model filter, respectively.

[0068] Step 3: Use unscented Kalman filtering to update the filtering of the fault-free model and the fault model respectively, specifically including the following steps:

[0069] Step 3.1: Use UT transformation to obtain the Sigma point set that approximates the nonlinear density function of the fault-free model and the fault model.

[0070] In this embodiment, the initial state of the fault-free filter Take the initial output value of the atmospheric data system and the initial estimation error covariance matrix. Initial state of the fault filter Initial estimation error covariance matrix

[0071] The sampling points for the two models are calculated as follows:

[0072]

[0073]

[0074]

[0075] In the formula, the superscript md = nf, f represents the fault-free model and the fault model, respectively; This represents the state value of the model at time k-1; This represents the value of the i-th sampling point of the model at time k-1; This represents the error covariance matrix of the model at time k-1; n is the dimension of the state. Represents the i-th column of the matrix; λ = α 2 (n+k)-n is the scaling parameter; α is the coefficient that controls the distribution of sampling points, which is 0.6 in this embodiment; k is the quantity that ensures the positive semidefiniteness of the variance matrix, which is 0 in this embodiment.

[0076] Step 3.2, calculate the weights of the sampling points of the two models:

[0077]

[0078]

[0079] Wherein, β is a non-negative weight coefficient, which is related to the distribution of the states; in this embodiment, it is set to 2.

[0080] Step 3.3: Calculate the one-step state predictions for both models. The time update equations are as follows:

[0081]

[0082]

[0083]

[0084] In the formula, T is the filter update period, and in this embodiment, T = 0.2s; This represents the one-step prediction value for the i-th sampling point of the corresponding model; This is a one-step prediction of the corresponding model state variable; Let Q be the one-step prediction error covariance matrix of the corresponding model; Q = E{ww T} is the system noise variance matrix.

[0085] Step 3.4, apply UT transformation again to update the Sigma point set of both models. The sampling points are calculated as follows:

[0086]

[0087]

[0088]

[0089] Step 3.5: Substitute the Sigma point set into the measurement equation and calculate the weighted mean of the predicted measurement values ​​of the two models. and covariance

[0090]

[0091]

[0092] In the formula, R = E{vv T} is the measurement noise variance matrix.

[0093] Step 3.6, calculate the filtering gain of the two models.

[0094]

[0095]

[0096] Step 3.7, update the state estimates of the two models at time k. and error covariance matrix

[0097]

[0098]

[0099] Step 4: Calculate the model probabilities for the fault-free model and the fault model. The specific method is as follows:

[0100] The information at time k for both models can be obtained from the filtering update in step 3. With the new covariance matrix

[0101]

[0102]

[0103] If the innovations of both models follow a zero-mean Gaussian distribution, then the conditional probability density function is... Satisfy the following formula:

[0104]

[0105] The probabilities of the two models are recursively calculated using Bayes' theorem.

[0106]

[0107] The model probability is used to determine whether the atmospheric data system has malfunctioned. If the result is positive, it indicates that the atmospheric data system has malfunctioned, and you should continue to step 5; otherwise, it indicates that the atmospheric data system has not malfunctioned, and you should proceed to step 6.

[0108] Step 5: Use the residual component chi-square test method to determine the specific source of fault information. The specific method is as follows:

[0109] For a fault-free model filter, when the system has no faults, its innovation satisfies a mean of 0 and a variance of . Gaussian white noise. Therefore, the fault detection value of the j-th innovation component of this model is constructed.

[0110]

[0111] In the formula, This represents the j-th element of the fault-free filter information; This represents the j-th diagonal element of the information covariance matrix of the fault-free filter; j = 1, 2, 3 represent airspeed, angle of attack, and sideslip angle, respectively.

[0112] As can be seen from the chi-square test principle, under fault-free conditions... Follows the χ² pattern with 1 degree of freedom 2 If the distribution is such that component j has a fault, the criterion for determining whether component j has a fault is as follows:

[0113] j-component fault

[0114] The j-component is fault-free.

[0115] In the formula, T o The detection threshold is determined by the false detection rate μ and satisfies the following relationship:

[0116] P{χ 2 (1) > T o}=μ

[0117] Step 6: Selectively reinitialize the state values ​​and estimation error covariance matrix of the dual-model filter, as follows:

[0118] If the fault detection results in step 4 indicate that the atmospheric data system is not faulty, then the fault-free model filter will not be initialized, and its own filter estimate at time k will be used. and estimation error covariance matrix This serves as the initial value for the filter iteration update at time k+1. The fault model filter is initialized as follows:

[0119]

[0120]

[0121] In the formula, P0 represents the initial state value of the fault model filter; P0 is the initial estimation error covariance matrix of the fault model filter.

[0122] If the fault detection results in step 4 indicate that the atmospheric data system has malfunctioned, and step 5 identifies the specific fault information source as component j, then the fault model filter is initialized as follows:

[0123]

[0124] And i≠j

[0125] In the formula, The i-th element of the initial state value of the fault model filter; The filter estimate of the fault model filter at time k. The i-th dimension state variable; P 0,ii The element in the i-th row and i-th column of the initial estimation error covariance matrix of the fault model filter; The covariance matrix of the estimation error of the filter at time k in the fault model represents the value of the filter. The element in the i-th row and i-th column.

[0126] The fault-free model filter is initialized as follows:

[0127]

[0128]

[0129] In the formula, The first three state variables represent the fault model filter. The first 3 rows and 3 columns represent the elements of the covariance matrix of the error estimation error of the fault model filter.

[0130] The initial state values ​​and estimation error covariance matrix of the dual-model filter are used as the initial values ​​for filter iteration update at time k+1 in step 3.

[0131] Step 7: Calculate and output the estimated atmospheric data and atmospheric data fault values ​​at time k. The calculation method is as follows:

[0132] By weighting the state estimate of the fault-free model filter with the first three state variables of the faulty model filter, the atmospheric data estimate at time k can be obtained.

[0133]

[0134] In the formula, These represent the estimated airspeed, angle of attack, and sideslip angle at time k, respectively.

[0135] By weighting the last three-dimensional state variables of the fault filter model, the estimated value of the atmospheric data fault at time k can be obtained.

[0136]

[0137] In the formula, These represent the estimated values ​​for airspeed fault, angle of attack fault, and sideslip angle fault at time k, respectively.

[0138] The root mean square error between the estimated fault value and the actual fault value of the atmospheric data, and the root mean square error between the estimated atmospheric data and the actual atmospheric data output value were statistically analyzed. The experimental results are shown in Table 2.

[0139] Table 2

[0140]

[0141] The experimental results above demonstrate that this invention provides a model-based method for estimating atmospheric data faults. The model probability calculation results of this method match the actual model, enabling timely and accurate detection of faults in the atmospheric data system. Compared to traditional dual-model estimation methods, this invention can also identify specific fault information sources, avoiding incorrect estimations of other fault-free information. Furthermore, the estimated atmospheric data faults and the accuracy of the atmospheric data are significantly improved, ensuring the reliability of the output information when an atmospheric data system experiences a fault.

[0142] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A model-based atmospheric data failure estimation method, characterized by, The method comprises the following steps: collecting measurement information of an inertial navigation system and an air data system; constructing a state equation and a measurement equation of a dual-model filter based on the measurement information; the dual model comprises a fault-free model and a fault model; the measurement information comprises inertial navigation system information and air data system information; the inertial navigation system information comprises a roll rate, a pitch rate, a heading rate, acceleration components in three axes of a machine body, a roll angle and a pitch angle; the air data system information comprises an air speed, an attack angle and a sideslip angle; filtering and updating the dual-model filter by using an unscented Kalman filter; calculating a model probability of the dual-model filter, and determining whether the air data system has a fault by using the model probability; when a fault occurs, finding a fault information source by using a residual component chi-square detection method, and selectively reinitializing the dual-model filter; the method for selectively reinitializing the dual-model filter comprises: when the air data system has a fault, initializing the fault-free model and the fault model, and when the air data system has no fault, initializing the fault model and not initializing the fault-free model; calculating and outputting an air data estimation value and an air data fault estimation value based on a state estimation result of the dual-model filter; the method for calculating the air data estimation value and the air data fault estimation value comprises: weighting a state estimation value of the fault-free model filter and a first three-dimensional state quantity of the fault model filter to obtain the air data estimation value; and weighting a second three-dimensional state quantity of the fault model filter to obtain the air data fault estimation value.

2. The model-based atmospheric data failure estimation method according to claim 1, characterized in that, the method for constructing the state equation and the measurement equation of the dual-model filter comprises: constructing the state equation of the dual model based on inertial navigation system measurement input and inertial navigation system measurement noise; constructing the measurement equation of the dual model based on air speed, attack angle and sideslip angle measurement values of the air data system, air speed, attack angle and sideslip angle measurement noise of the air data system and an observation matrix of the dual-model filter.

3. The model-based atmospheric data fault estimation method according to claim 1, characterized in that, the process of filtering and updating the dual-model filter by using the unscented Kalman filter comprises: obtaining a Sigma point set approximating a nonlinear density function of the dual model by using UT variation; calculating a weight value of a dual-model sampling point, calculating a one-step prediction value of the state of the dual model, and performing time updating; updating the Sigma point set of the dual model again by using UT variation; bringing the Sigma point set into the measurement equation, and weighting and calculating a mean value and a covariance of a predicted measurement value of the dual model; calculating a filtering gain of the dual model based on the mean value and the covariance; updating a state estimation value and an error covariance matrix of the dual model.

4. The model-based atmospheric data failure estimation method according to claim 1, characterized in that, the method for calculating the model probability of the dual-model filter comprises: obtaining a new information and a new information covariance matrix of the dual model based on filtering and updating; obtaining a conditional probability density function based on the new information and the new information covariance matrix; recursively calculating the model probability of the dual model by using a Bayesian formula based on the conditional probability density function.

5. The model-based atmospheric data failure estimation method of claim 1, wherein, the method for finding the fault information source by using the residual component chi-square detection method comprises: determining a detection threshold value by using a false detection rate; The fault detection value of the jth innovation component of the model is constructed based on the jth element of the fault-free filter innovation and the jth diagonal element of the fault-free filter innovation covariance matrix; respectively represent airspeed, angle of attack, and sideslip angle. When the fault detection value is greater than the detection threshold, the j component is faulty, and when the fault detection value is less than or equal to the detection threshold, the j component is not faulty, and the fault information source is determined by the value of j.

Images