A method, device and storage medium for fault diagnosis of flight control system

Abstract

The present invention provides a method, device and storage medium for fault diagnosis of a flight control system, the method comprising: step 1, determining the composition and connection structure of the flight control system; step 2, determining the distributed structure of the flight control computer; step 3, determining the unmanned aerial vehicle object model; step 4, designing the fault detection filter of the actuator and the airborne sensor; step 5, training a class of support vector machine models by an ant colony algorithm to obtain the optimal model parameters; step 6, according to the distributed flight control computer structure, a detection filter and a class of support vector machine models are built into each functional unit; step 7, according to the hardware redundancy and the class of support vector machine models, the fault diagnosis of the flight control system is performed; step 8, testing the fault diagnosis effect and optimizing the system structure. The method can improve the accuracy of fault judgment under the output coupling condition of a complex flight control system.

Description

Technical Field

[0001] The present invention belongs to the technical field of fault diagnosis of flight control systems, and in particular relates to a fault diagnosis method, device and storage medium for a flight control system. Background Art

[0002] High reliability and safety flight control system has always been an important research direction for aircraft. The flight control system includes flight control computers, actuators, onboard sensors and other recording computers. The flight control computer is the core of the flight control system. Its job is to complete the flight control tasks of the UAV and manage the scheduling of onboard equipment. More importantly, it can realize resource sharing, integrate data and information, and is also the connection hub of the air and ground command systems. However, the number and complexity of UAV missions are increasing. At the same time, the multi-data fusion of sensors, the increase in flight envelopes and performance boundaries have put forward higher requirements on the reliability and safety of flight control computers.

[0003] Redundancy technology is the main means for flight control computers to achieve fault tolerance. Under the current complex computer architecture, as functions become more powerful, the complexity of the system increases greatly. Therefore, complete computer hardware and software do not exist. The emergence of redundancy technology greatly reduces the probability of the same components failing at the same time, thereby increasing the reliability of the system.

[0004] At present, flight control computers under distributed architecture mainly adopt a series of fault-tolerant control methods such as hardware redundancy, fault detection, system reconstruction, and fault recovery. As the current UAV missions become increasingly complex and the mission environment becomes increasingly severe, the number of resources required by distributed flight control computers has also increased dramatically, and the coupling between nodes has also increased. Therefore, a more reliable and effective fault-tolerant control method must be designed to ensure the reliability of distributed flight control computers. Summary of the invention

[0005] Purpose of the invention: In view of the current situation where complex UAVs have serious output coupling and insufficient utilization of fault diagnosis information, the present invention proposes a flight control system fault diagnosis method, which can effectively improve the fault diagnosis accuracy under the condition of high coupling of UAV output channels, and by combining the method with hardware redundancy, expand the fault diagnosis coverage to the flight control computer, actuators and airborne sensors in the flight control system, thereby improving the fault diagnosis accuracy and coverage of the flight control system.

[0006] The method comprises the following steps:

[0007] Step 1, (computer equipment) establishes a UAV object model, and uses the UAV object model to generate a fault detection filter (of the actuator and the airborne sensor), wherein the UAV object model includes: a triple-redundant central processing unit, a dual-redundant serial interface unit, and a quad-redundant bus unit;

[0008] Step 2: collecting residual data output by the fault detection filter of the airborne sensor, using the residual data and training a support vector machine model through an ant colony algorithm, and recording the optimal model parameters obtained by the training;

[0009] Step 3: The flight control computer loads the fault detection filter and the first-class support vector machine model;

[0010] Step 4: diagnose the fault of the flight control system by using the hardware redundancy and the first-class support vector machine model;

[0011] Step 5: Update the system structure using the results of the fault diagnosis.

[0012] Before step 1, also include:

[0013] (Computer device) constructing topological structure information of a flight control system, wherein the topological structure information is used to represent components of the flight control system, wherein the components include: a flight control computer, an actuator, an airborne sensor system, and an airframe;

[0014] The topology structure information also includes: distributed structure information of the flight control computer. Specifically, the flight control computer adopts a distributed structure, including a triple-redundant central processing unit, a dual-redundant serial communication unit and a quad-redundant bus unit. The triple-redundant central processing unit and the dual-redundant serial communication unit are bidirectionally connected through the quad-redundant bus unit.

[0015] Among them, the triple-redundant central processing unit is used for control law solution, logical operation and equipment management of the flight control system; the dual-redundant serial communication unit is responsible for communication with external devices; and the quadruple-redundant bus unit is responsible for two-way communication within the flight control computer.

[0016] Step 1 includes: assuming that the linear time-invariant flight control system dynamic equation is:

[0017]

[0018] Where u(t) is the system input vector, y(t) is the system output vector, A, B, C are coefficient matrices, A∈R n×n , B∈R n×m , C∈R p×n , R n×n represents an n×n matrix, Rn×m represents an n×m matrix, R p×n represents a p×n matrix, u(t) is the input, x(t) is the state matrix, is the state matrix after the derivative of x(t);

[0019] set up is the state estimate vector, is the output estimation vector, K is the system gain matrix, and the system state error e(t) is given as:

[0020]

[0021] Taking the derivative of e(t) we get The state error equation is further obtained as:

[0022]

[0023] yes is the result of the derivation;

[0024] The output error ε(t) is:

[0025]

[0026] For actuator failure: In the fault detection filter, if actuator i fails at time t, the system input u(t) is:

[0027] u(t)=u d (t)+e i n(t) (5)

[0028] Among them, u d (t) is the desired control input under normal conditions; e i =[0…1…0] is the control information direction of actuator i, n(t) is the current time;

[0029] Substituting equation (2) and equation (5) into equation (4), we get the following updated state error equation:

[0030]

[0031] Where f is the sensor fault vector:

[0032] f=Be i (7)

[0033] According to formula (6) and formula (4), we can directly solve:

[0034]

[0035] Where e(0) is the initial error of the system, n(τ) is an arbitrary scalar time function, and dτ is the differential term; both e(t) and ε(t) are divided into two independent terms, the former is the transient solution of the system, and the latter is the stable solution of the system. The transient problem is not considered, and only the system state in the stable state is considered:

[0036]

[0037] Where e s (t) is the steady-state system state error, ε s (t) is the steady-state output error;

[0038] For airborne sensor failure, in the fault detection filter, if sensor j fails at time t, the system output is:

[0039] y(t)=Cx(t)+e j n(t) (10)

[0040] In the formula, e j =[0…0 1 0…0] T is the vector direction of sensor j, n(t) is the current moment;

[0041] Substituting equation (2) and equation (10) into equation (3), we get the following updated state error equation:

[0042]

[0043] Where f is the sensor fault vector:

[0044] f=Ke j (12)

[0045] Substituting formula (10) into formula (4), we get:

[0046]

[0047] According to formula (11) and formula (13), we can directly solve:

[0048]

[0049] In the formula, e(t) and ε(t) have the same structure and are divided into two independent terms. The first term is the transient solution of the system, and the second term is the stable solution of the system. The transient problem is not considered, and only the system state in the stable state is considered:

[0050]

[0051] Step 2 includes:

[0052] Step 2-1: The fault detection filter outputs n independent and identically distributed residual vectors Es = {E1, E2, ... E i ,…,E n}, where E i ={e1,e2,…,e j} is the residual vector of the i-th output channel in period i, e i is the residual value of the output channel i saved periodically from the fault detection filter, i=1…j;

[0053] The Es in the fault-free state is used as the training data set of a support vector machine model, and a nonlinear transformation φ from the input feature space to the high-dimensional space χ is given, where φ satisfies φ(E i )∈χ,φ(E i ) is the mapping function, and a hyperplane ω is established in χ T φ(E)-ρ=0, where the superscript T represents the matrix transpose, ω is the normal vector of the hyperplane, and ρ is the distance from the origin to the hyperplane;

[0054] The first-class support vector machine model objective solves the following optimization problem:

[0055]

[0056] Among them, ξ i >0 is the relaxation factor; v∈(0,1] represents the credibility of the classification performance of the hyperplane;

[0057] When solving the quadratic programming problem in equation (16), the Lagrangian function L is introduced. p :

[0058]

[0059] Among them, α i ≥0, β i ≥0 is the Lagrange multiplier, ω is the normal vector of the hyperplane;

[0060] By i And ρ derivative, do minimization:

[0061]

[0062]

[0063]

[0064] Substituting equations (18), (19), and (20) into equation (17), the Gaussian kernel function K(E i ,E j ) instead of the inner product (E i,E j ), we get the following optimization problem:

[0065]

[0066]

[0067] Among them, st stands for subject to, which means subject to; exp is the exponential operation, σ is the kernel parameter, It means that α takes the minimum value, α i and α j are the Lagrange multipliers when the sequence i and sequence j change respectively;

[0068] The decision function f(e) corresponding to the optimization problem is:

[0069] f(e)=sgn(ω T ·φ(E)-ρ) (23)

[0070] sgn() is the following symbolic function:

[0071]

[0072] Step 2-2, establish the ant colony algorithm to optimize the model parameters:

[0073] The penalty parameter v and the kernel parameter σ in the model parameters are taken as the parameters to be optimized, and the following parameter optimization method is performed:

[0074] Step 2-2-1, with the parameter (v, σ) as the position of each ant, initialize the number of iterations M, the total number of ants N, the pheromone transfer coefficient Rho and the transition probability constant in the ant colony algorithm; randomly generate the initial position (v, σ) of each ant, and then learn the corresponding accuracy model through a class of support vector machine models under the normal training set:

[0075] T0(i)=precision(i) (25)

[0076] Where T0(i) is the accuracy of the current ant's position, and precision is the accuracy calculation formula after abstraction;

[0077] Step 2-2-2, in the main loop of the ant colony algorithm, first find the ant with the largest pheromone and save it as T0 (bestindex), and then determine the next transfer probability p of each ant based on its own pheromone. (i) :

[0078]

[0079] Let the current number of iterations be k, mark (vold , σ old ), as the ant position before iteration, where v old , σ(old) are the penalty parameter v and the kernel parameter σ before iteration respectively, marking (v new , σ new ), as the ant position after iteration, where v new , σ(new) are the penalty parameter v and the kernel parameter σ after iteration respectively; if the transition probability p (i) < p0, then local search is performed, v new = v old + χ·α, σ new = σ old + χ·α; if p (i) ≥ p0, then global search is performed, v new = v old + χ·β, σ new = σ old + χ·β; where p0 is the state transition factor; α = 1 / k is a constant value; χ is the search range; β is the global search step size;

[0080] Step 2-2-3, if all ants have completed local or global search iterations, then update the pheromone of each ant according to the rules of the ant colony algorithm, where the pheromone update rule is:

[0081]

[0082] where is the pheromone updated according to the pheromone at time k at time k + 1;

[0083] Take the updated maximum pheromone as the new round of T0(bestindex), repeat Step 2-2-2 until all iterations are completed, and finally take the ant position (v best , σ best ) as the optimal model parameters of the one-class support vector machine model OCSVM (one-class support vector machine).

[0084] In Step 2, the training data of the one-class support vector machine model is the output parameter data of the fault detection filter under normal conditions, and the optimal model parameters are obtained through the iteration of the ant colony algorithm. The optimal model parameters include the penalty coefficient and the kernel parameter.

[0085] In Step 2, the training of the one-class support vector machine model by the ant colony algorithm includes the parameter tuning of the ant colony algorithm and the parameter tuning of the one-class support vector machine. The parameter tuning of the ant colony algorithm and the parameter tuning of the one-class support vector machine are in a sequential relationship and do not have coupling.

[0086] In step 3, the flight control computer loads the fault detection filter and the type of support vector machine model through a functional unit, and the functional unit is a dual-redundancy serial interface unit.

[0087] Step 3 includes: the input of the fault detection filter loaded in the dual-redundancy serial interface unit is the actuator control data solved by the control law of the triple-redundancy central processing unit and the airborne sensor data collected by the dual-redundancy serial communication unit, and the output of the fault detection filter is the residual data containing the fault information;

[0088] The input of a type of support vector machine model configured in the dual-redundancy serial interface unit is the residual data containing fault information output by the fault detection filter, and the output is the fault information of the dual-redundancy serial communication unit;

[0089] Fault diagnosis is performed using sensor information collected by a dual-redundancy serial communication unit and the output results of a type of support vector machine model. The fault diagnosis covers input / output functional interface faults of the dual-redundancy serial interface unit, actuator faults, and human-onboard sensor system faults.

[0090] On the other hand, the present invention also provides a fault diagnosis device for a flight control system, comprising:

[0091] A UAV object model building module is used to build a UAV object model and generate fault detection filters of actuators and airborne sensors using the UAV object model, wherein the UAV object model includes: a triple-redundant central processing unit, a dual-redundant serial interface unit and a quad-redundant bus unit;

[0092] A support vector machine model building module, used to collect residual data output by the fault detection filter of the airborne sensor, train a support vector machine model using the residual data and an ant colony algorithm, and record the optimal model parameters obtained by the training;

[0093] The flight control computer loads the fault detection filter and the one-class support vector machine model;

[0094] The fault diagnosis module is used to perform fault diagnosis of the flight control system by utilizing hardware redundancy and the support vector machine model; and to update the system structure by utilizing the result of the fault diagnosis.

[0095] On the other hand, the present invention further provides a storage medium storing a computer program or instruction, which, when executed, implements the fault diagnosis method for a flight control system described in the present invention.

[0096] The beneficial effects of the present invention are as follows:

[0097] 1. A class of support vector machines is used as a fault diagnosis tool to improve the utilization of fault information and reduce the complexity caused by high coupling of the system.

[0098] 2. The ant colony algorithm is used to find the optimal model parameters, which reduces human influence and improves the accuracy of the model.

[0099] 3. Improve the fault diagnosis coverage, which can effectively diagnose the faults of the serial communication unit and airborne sensors in the flight control computer. BRIEF DESCRIPTION OF THE DRAWINGS

[0100] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments, and the above and / or other advantages of the present invention will become more clear.

[0101] Figure 1 It is a schematic diagram of the flight control system structure.

[0102] Figure 2 It is a schematic diagram of the distributed flight control computer structure.

[0103] Figure 3 It is a schematic diagram of the information flow structure of the flight control system.

[0104] Figure 4 Figure 5 is a schematic diagram of the detection filter tracking structure.

[0105] Figure 5 This is a typical fault type diagram.

[0106] Figure 6 It is a schematic diagram of model parameter optimization. DETAILED DESCRIPTION

[0107] The present invention discloses a comprehensive fault diagnosis method which optimizes the parameters of a support vector machine model by using an ant colony algorithm and combines the hardware redundancy of a flight control system. Figure 1 As shown in the figure, it mainly includes flight control computers, actuators, airborne sensor systems and fuselage. The actuators receive control signals from the servo controller and feed back the actuator response results to the airborne sensors through the control fuselage. After receiving the sensor information, the flight control computer outputs the control results to the servo controller according to the control law. Figure 2 As shown, in the distributed flight control computer, the central processing unit adopts a triple-redundancy structure, the serial communication unit adopts a double-redundancy structure, and the bus unit adopts a quadruple-redundancy structure. Figure 2In the figure, RX represents the receiving interface between the serial communication unit and the external device, TX represents the sending interface between the serial communication unit and the external device, CU is the central processing unit, and IU is the interface unit. The central processing unit is the core component of the flight control computer, responsible for the control law solution and logical operation functions; the serial interface unit is the channel for the flight control computer to communicate with the outside world, and its function is to provide an interface for the flight control computer to communicate with the outside world; the bus unit is a two-way interactive channel for communication between the functional components inside the flight control computer.

[0108] The information flow structure of the flight control system is as follows: Figure 3 As shown, from Figure 3 It can be learned that the information exchange of the flight control system mainly consists of onboard sensors, flight control computers, actuators, drone bodies and ground stations. When the drone starts to perform a mission, the flight control computer will communicate with the onboard sensors, actuators and ground stations at a fixed frequency to achieve a closed-loop control of the drone.

[0109] The detection filter tracking structure is as follows Figure 4 As shown, A, B, and C are coefficient matrices, u(t) is the input, and y(t) is the output. y(t) is the derivative, K is the system gain matrix. Typical fault types are Figure 5 As shown, it is divided into sudden fault, lateral deviation fault and gradual fault.

[0110] The following is a detailed description of the design of the flight control system fault diagnosis process implementation.

[0111] 1. Fault detection filter

[0112] Assume that the linear time-invariant flight control system dynamic equation is:

[0113]

[0114] Where u(t) is the system input vector, y(t) is the system output vector, A, B, C are coefficient matrices, A∈R n×n , B∈R n×m , C∈R p×n , R n×n represents an n×n matrix, R n×m represents an n×m matrix, R p×n represents a p×n matrix, u(t) is the input, x(t) is the state matrix, is the state matrix after the derivative of x(t).

[0115] exist Figure 4In the above, the difference between the observation vector and the output vector of the detection filter is added to the closed-loop control of the detection filter after passing through the designed system gain matrix. Under the action of this gain matrix, the detection filter will gradually eliminate the detection filter tracking error caused by the initial error.

[0116] according to Figure 4 The observation form of the detection filter is given to prove the fault detectability of the detection filter. In the figure, is the state estimate vector, is the output estimation vector and K is the system gain matrix.

[0117] The system state error e(t) is given as:

[0118]

[0119] Furthermore, taking the derivative of e(t) we get The state error equation is further obtained as:

[0120]

[0121] yes is the result of the derivation;

[0122] The output error ε(t) is:

[0123]

[0124] Two types of faults are considered below: actuator failure and onboard sensor failure.

[0125] (1) Actuator failure

[0126] In the fault detection filter, consider that the actuator i fails at time t. At this time, the system input u(t) is:

[0127] u(t)=u d (t)+e i n(t) (5)

[0128] Among them, u d (t) is the desired control input under normal conditions; e i =[0…1…0] is the control information direction of actuator i, n(t) is the current time;

[0129] Substituting equation (2) and equation (5) into equation (4), we get the following updated state error equation:

[0130]

[0131] Where f is the sensor fault vector:

[0132] f=Be i (7)

[0133] According to formula (6) and formula (4), we can directly solve:

[0134]

[0135] Where e(0) is the initial error of the system, n(τ) is an arbitrary scalar time function, and dτ is the differential term. Both e(t) and ε(t) are divided into two independent terms, the first term is the transient solution of the system, and the second term is the stable solution of the system. The transient problem is not considered here, only the system state in the stable state is considered:

[0136]

[0137] Where e s (t) is the steady-state system state error, ε s (t) is the steady-state output error;

[0138] (2) Airborne sensor failure

[0139] In the fault detection filter, it is assumed that sensor j fails at time t. At this time, the system output is:

[0140] y(t)=Cx(t)+e j n(t) (10)

[0141] In the formula, e j =[0…0 1 0…0] T is the vector direction of sensor j, and n(t) is the current moment.

[0142] Substituting equation (2) and equation (10) into equation (3), we get the following updated state error equation:

[0143]

[0144] Where f is the sensor fault vector:

[0145] f=Ke j (12)

[0146] Substituting formula (10) into formula (4), we get:

[0147]

[0148] According to formula (11) and formula (13), we can directly solve:

[0149]

[0150] In the formula, e(t) and ε(t) have the same structure and are divided into two independent terms. The first term is the transient solution of the system, and the second term is the stable solution of the system. The transient problem is not considered here, only the system state in the stable state is considered:

[0151]

[0152] From equations (9) and (15), we can draw the following conclusions: whether it is an actuator failure or a sensor failure, it has the same form, and the system failure caused by the external airborne equipment failure is only related to its own fault condition and has nothing to do with external disturbances.

[0153] It can be seen from formula (6) and formula (11) that Figure 4 In a linear system, the detectability of the fault is determined by the system gain matrix K. The sufficient condition for fault detectability is: the stable output error Ce s The output direction of (t) is fixed; (A-KC) is full rank.

[0154] Under the above conditions, if the actuator fault and sensor fault are measurable, it is only necessary to output the error Ce s The direction of (t) can locate the fault position.

[0155] 2. A class of support vector machine models

[0156] The fault detection filter outputs n independent and identically distributed residual vectors Es = {E1, E2, ... E i ,…,E n}, where E i ={e1,e2,…,e j} is the residual vector of the i-th output channel in period i, e i (i=1...j) are the residual values ​​under output channel i saved periodically from the fault detection filter.

[0157] The Es in the fault-free state is used as the training data set of a support vector machine model, and a nonlinear transformation φ from the input feature space to the high-dimensional space χ is given, where φ satisfies φ(E i )∈χ,φ(E i ) is the mapping function, and a hyperplane ω is established in χ T ·φ(E)-ρ=0, where the superscript T represents the matrix transpose, ω is the normal vector of the hyperplane, and ρ is the distance from the origin to the hyperplane.

[0158] The first-class support vector machine model objective solves the following optimization problem:

[0159]

[0160] Among them, ξi >0 is a relaxation factor, which makes the analytical hyperplane robust; v∈(0,1] plays a role similar to the penalty coefficient in the standard support vector machine in the optimization problem, indicating the credibility of the classification performance of the hyperplane.

[0161] When solving the quadratic programming problem in equation (16), the Lagrangian function L is introduced. p :

[0162]

[0163] Among them, α i ≥0, β i ≥0 is the Lagrange multiplier, and ω is the normal vector of the hyperplane.

[0164] By i And ρ derivative, do minimization:

[0165]

[0166]

[0167]

[0168] Substituting equations (18), (19), and (20) into equation (17), the Gaussian kernel function K(E i ,E j ) instead of the inner product (E i ,E j ), we get the following optimization problem:

[0169]

[0170] K(E i ,E j )=φ(E i )·φ(E j )=exp(-||E i -E j || 2 / σ 2 ) (twenty two)

[0171] Among them, st stands for subject to, which means subject to; exp is the exponential operation, σ is the kernel parameter, It means that α takes the minimum value, α i and α j are the Lagrange multipliers when the sequence i and sequence j change respectively.

[0172] The decision function f(e) corresponding to the optimization problem is:

[0173] f(e) = sgn(ω T ·φ(E) - ρ) (23)

[0174] sgn() is the following sign function:

[0175]

[0176] 3. Optimize the model parameters using the ant colony algorithm

[0177] Take the penalty parameter v and the kernel parameter σ in the model parameters as the parameters to be optimized, and perform the Figure 6 parameter optimization method shown below:

[0178] 1) First, take the parameters (v, σ) as the position of each ant, and initialize the number of iterations M, the total number of ant colonies N, the pheromone transmission coefficient Rho, and the transition probability constant in the ant colony algorithm. Randomly generate the initial positions (v, σ) of each ant, and then learn the corresponding accuracy model through a one-class support vector machine model under the training set in the normal state:

[0179] T0(i) = precission(i) (25)

[0180] In the formula, T0(i) is the accuracy of the current ant's position, and precission is the abstract accuracy calculation formula;

[0181] 2) In the main loop of the ant colony algorithm, first find the ant with the largest pheromone and save it as T0(bestindex), and then determine its next transition probability p according to the pheromone of each ant itself (i) :

[0182]

[0183] Let the current number of iterations be k, and mark (v old , σ old ) as the ant position before iteration, where v old , σ(old) are the penalty parameter v and the kernel parameter σ before iteration respectively. Mark (v new , σ new ) as the ant position after iteration, where v new , σ(new) are the penalty parameter v and the kernel parameter σ after iteration respectively. If the transition probability p (i) < p0, then perform local search, v new = v old + χ·α, similarly σ new = σ old + χ·α; if p (i) ≥ p0, then perform global search, v new = vold +χ·β. Similar to σ new =σ old +χ·β. Where p0 is the state transfer factor, and the transfer probability of the ant with the median T0(i) is generally taken as the value of p0; α=1 / k is a constant value; χ is the search range; β is the global search step.

[0184] 3) If all ants have completed local or global search iterations, each ant’s pheromone is updated according to the rules of the ant colony algorithm, where the pheromone update rule is:

[0185]

[0186] in The pheromone at time k+1 is updated based on the pheromone at time k;

[0187] The updated maximum pheromone is used as the new round of T0 (bestindex), and step 2) is repeated until all iterations are completed. Finally, the position of the ant with the largest pheromone (v best ,σ best ) as the optimal model parameters of OCSVM.

[0188] 4. Serial communication unit external interface fault diagnosis

[0189] The serial communication unit uses hardware dual-redundancy input and a type of support vector machine model to form a redundant fault diagnosis structure, as shown in Table 1. The serial interface unit is equipped with an onboard filter Fil_s(U,Y). After the decision is made, the data sent to the bus is Y sM , the airborne sensor signal input to the serial interface unit is Y s , input via serial interface as Y sA , Y sB The two form a master-slave backup mode, and use the onboard filter and a type of support vector machine model fault judgment to form a four-redundancy structure in the serial communication unit, and according to Y sA , Y sB And a class of support vector machine models are used to diagnose the faults of the serial interface unit, as shown in Table 1.

[0190] Table 1

[0191]

[0192] Numbers 1 to 4 in the table represent various output conditions of the fault detection filter when the serial interface unit receives the same data. Among them, if the output results of the two groups of detection filters are different, the node results with the output result of no fault are taken as reliable data; if the results of the two groups of detection filters are the same and both are judged as faults, the sensor is faulty and the sensor is self-checked internally. If both are judged as no faults, the system is normal. Numbers 4-8 represent different data received by the serial interface unit. Among them, if the output results of the two groups of detection filters are different, the node results with the output result of no fault are taken as reliable data; if the results of the two groups of detection filters are the same and both are judged as faults, it is considered that all serial interface units are faulty. If both are judged as no faults, it cannot be judged and the faulty unit is unknown. From the results, it is clear that unless the voting results of the two detection filters are correct and the output results are inconsistent, this type of fault cannot be judged.

[0193] This embodiment also provides a fault diagnosis device for a flight control system, comprising:

[0194] A UAV object model building module is used to build a UAV object model and generate fault detection filters of actuators and airborne sensors using the UAV object model, wherein the UAV object model includes: a triple-redundant central processing unit, a dual-redundant serial interface unit and a quad-redundant bus unit;

[0195] A support vector machine model building module, used to collect residual data output by the fault detection filter of the airborne sensor, train a support vector machine model using the residual data and an ant colony algorithm, and record the optimal model parameters obtained by the training;

[0196] The flight control computer loads the fault detection filter and the one-class support vector machine model;

[0197] The fault diagnosis module is used to perform fault diagnosis of the flight control system by utilizing hardware redundancy and the support vector machine model; and to update the system structure by utilizing the result of the fault diagnosis.

[0198] As described above, the apparatus according to the embodiment of the present application can be implemented in various terminal devices, such as a server of a distributed computing system. In one example, the apparatus according to the embodiment of the present application can be integrated into the terminal device as a software module and / or a hardware module. For example, the apparatus can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the apparatus can also be one of the many hardware modules of the terminal device.

[0199] Alternatively, in another example, the apparatus and the terminal device may also be separate terminal devices, and the apparatus may be connected to the terminal device via a wired and / or wireless network, and transmit the interactive information in accordance with an agreed data format.

[0200] This embodiment also provides a computer-readable storage medium having computer program instructions stored thereon. When the computer program instructions are executed by a computing device, the computer program instructions can be operated to execute the fault diagnosis method of a flight control system as described above.

[0201] The present invention provides a method, device and storage medium for diagnosing a fault of a flight control system. There are many methods and ways to implement the technical solution. The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be considered as the protection scope of the present invention. All components not specified in this embodiment can be implemented by existing technologies.

Claims

1. A method for diagnosing a fault in a flight control system, characterized in that: The following steps are involved: Step 1: Establish a UAV object model, and use the UAV object model to generate a fault detection filter, wherein the UAV object model includes: a triple-redundant central processing unit, a dual-redundant serial interface unit, and a quad-redundant bus unit; Step 2: collecting residual data output by the fault detection filter of the airborne sensor, using the residual data and training a support vector machine model through an ant colony algorithm, and recording the optimal model parameters obtained by the training; Step 3: The flight control computer loads the fault detection filter and the first-class support vector machine model; Step 4: diagnose the fault of the flight control system by using the hardware redundancy and the first-class support vector machine model; Step 5: Update the system structure using the results of the fault diagnosis; Before step 1, also include: Constructing topological structure information of a flight control system, wherein the topological structure information is used to represent components of the flight control system, the components including: a flight control computer, an actuator, an airborne sensor system, and an airframe; The topology structure information also includes: distributed structure information of the flight control computer. Specifically, the flight control computer adopts a distributed structure, including a triple-redundant central processing unit, a dual-redundant serial communication unit and a quad-redundant bus unit. The triple-redundant central processing unit and the dual-redundant serial communication unit are bidirectionally connected through the quad-redundant bus unit. The triple-redundant central processing unit is used for control law solution, logic operation and equipment management of the flight control system; the dual-redundant serial communication unit is responsible for communication with external devices; the quadruple-redundant bus unit is responsible for two-way communication within the flight control computer; Step 1 includes: assuming that the linear time-invariant flight control system dynamic equation is: Where u(t) is the system input vector, y(t) is the system output vector, A, B, C are coefficient matrices, A∈R n×n , B∈R n ×m , C∈R p×n , R n×n represents an n×n matrix, R n×m represents an n×m matrix, R p×n represents a p×n matrix, u(t) is the input, x(t) is the state matrix, is the state matrix after the derivative of x(t); set up is the state estimate vector, is the output estimation vector, K is the system gain matrix, and the system state error e(t) is given as: Taking the derivative of e(t) we get The state error equation is further obtained as: yes is the result of the derivation; The output error ε(t) is: For actuator failure: In the fault detection filter, if actuator i fails at time t, the system input u(t) is: u(t)=u d (t)+e i n(t) (5) Among them, u d (t) is the desired control input under normal conditions; e i =[0…1…0] is the control information direction of actuator i, n(t) is the current time; Substituting equation (2) and equation (5) into equation (4), we get the following updated state error equation: Where f is the sensor fault vector: f=Be i (7) According to formula (6) and formula (4), we can directly solve: Where e(0) is the initial error of the system, n(τ) is an arbitrary scalar time function, and dτ is the differential term; both e(t) and ε(t) are divided into two independent terms, the former is the transient solution of the system, and the latter is the stable solution of the system, and we get: Where e s (t) is the steady-state system state error, ε s (t) is the steady-state output error; For airborne sensor failure, in the fault detection filter, if sensor j fails at time t, the system output is: y(t)=Cx(t)+e j n(t) (10) In the formula, e j =[0 … 0 1 0 … 0] T is the vector direction of sensor j, n(t) is the current moment; Substituting equation (2) and equation (10) into equation (3), we get the following updated state error equation: Where f is the sensor fault vector: f=Ke j (12) Substituting formula (10) into formula (4), we get: According to formula (11) and formula (13), we can directly solve: In the formula, e(t) and ε(t) have the same structure and are divided into two independent terms. The first term is the transient solution of the system and the second term is the stable solution of the system. We can get: Step 2 includes: Step 2-1: The fault detection filter outputs n independent and identically distributed residual vectors Es = {E1, E2, ... E i ,…,E n }, where E i ={e1,e2,…,e j } is the residual vector of the i-th output channel in period i, ei is the residual value of the output channel i saved periodically from the fault detection filter, i = 1…j; The Es in the fault-free state is used as the training data set of a support vector machine model, and a nonlinear transformation φ from the input feature space to the high-dimensional space χ is given, where φ satisfies φ(E i )∈χ,φ(E i ) is the mapping function, and a hyperplane ω is established in χ T φ(E)-ρ=0, where superscript T denotes matrix transpose, ω is the normal vector of the hyperplane, and ρ is the distance from the origin to the hyperplane; The first-class support vector machine model objective solves the following optimization problem: Among them, ξ i >0 is the relaxation factor; v∈(0,1] represents the credibility of the classification performance of the hyperplane; When solving the quadratic programming problem in equation (16), the Lagrangian function L is introduced. p : Among them, α i ≥0, β i ≥0 is the Lagrange multiplier, ω is the normal vector of the hyperplane; By i And ρ derivative, do minimization: Substituting equations (18), (19), and (20) into equation (17), the Gaussian kernel function K(E i ,E j ) instead of the inner product (E i ,E j ), we get the following optimization problem: Among them, st means restricted; exp is the exponential operation, σ is the kernel parameter, It means that α takes the minimum value, α i and α j are the Lagrange multipliers when the sequence i and sequence j change respectively; The decision function f(e) corresponding to the optimization problem is: f(e)=sgn(ω T ·φ(E)-ρ) (23) sgn() is the following symbolic function: Step 2-2, establish the ant colony algorithm to optimize the model parameters: The penalty parameter v and the kernel parameter σ in the model parameters are taken as the parameters to be optimized, and the following parameter optimization method is performed: Step 2-2-1, with the parameter (v, σ) as the position of each ant, initialize the number of iterations M, the total number of ants N, the pheromone transfer coefficient Rho and the transition probability constant in the ant colony algorithm; randomly generate the initial position (v, σ) of each ant, and then learn the corresponding accuracy model through a class of support vector machine models under the normal training set: T0(i)=precision(i) (25) Where T0(i) is the accuracy of the current ant's position, and precision is the accuracy calculation formula after abstraction; Step 2-2-2, in the main loop of the ant colony algorithm, first find the ant with the largest pheromone and save it as T0 (bestindex), and then determine the next transfer probability p of each ant based on its own pheromone. (i) : Let the current iteration number be k, and mark (v old , σ old ) as the ant position before iteration, where v old , σ(old) are the penalty parameter v and the kernel parameter σ before iteration respectively. Mark (v new , σ new ) as the ant position after iteration, where v new , σ(new) are the penalty parameter v and the kernel parameter σ after iteration respectively. If the transition probability p (i) < p0, then local search is performed, v new = v old + χ·α, σ new = σ old + χ·α; if p (i) ≥ p0, then global search is performed, v new = v old + χ·β, σ new = σ old + χ·β; where p0 is the state transition factor; α = 1 / k is a constant value; χ is the search range; β is the global search step size; Step 2-2-3, if all ants have completed the local or global search iteration, then update each ant's pheromone according to the rules of the ant colony algorithm, where the pheromone update rule is: in The pheromone at time k+1 is updated based on the pheromone at time k; The updated maximum pheromone is used as the new round of T0 (bestindex), and steps 2-2-2 are repeated until all iterations are completed. Finally, the position of the ant with the largest pheromone (v best ,σ best ) as the optimal model parameters for a class of support vector machine models.

2. The method according to claim 1, characterized in that In step 2, the training data of the support vector machine model is the output parameter data of the fault detection filter under normal conditions, and the optimal model parameters are obtained through iteration of the ant colony algorithm, and the optimal model parameters include a penalty coefficient and a kernel parameter.

3. The method according to claim 2 is characterized in that, in step 2, the training of a class of support vector machine model by using the ant colony algorithm includes parameter tuning of the ant colony algorithm and parameter tuning of the class of support vector machine, and the parameter tuning of the ant colony algorithm and the parameter tuning of the class of support vector machine are in a succession relationship and are not coupled.

4. The method according to claim 3, characterized in that In step 3, the flight control computer loads the fault detection filter and the type of support vector machine model through a functional unit, and the functional unit is a dual-redundancy serial interface unit.

5. The method according to claim 4, characterized in that Step 3 includes: the input of the fault detection filter loaded in the dual-redundancy serial interface unit is the actuator control data solved by the control law of the triple-redundancy central processing unit and the airborne sensor data collected by the dual-redundancy serial communication unit, and the output of the fault detection filter is the residual data containing the fault information; The input of a type of support vector machine model configured in the dual-redundancy serial interface unit is the residual data containing fault information output by the fault detection filter, and the output is the fault information of the dual-redundancy serial communication unit; Fault diagnosis is performed using sensor information collected by a dual-redundancy serial communication unit and the output results of a type of support vector machine model. The fault diagnosis covers input / output functional interface faults of the dual-redundancy serial interface unit, actuator faults, and human-onboard sensor system faults.

6. A fault diagnosis device for a flight control system, used to implement the method according to claim 1, characterized in that: include: A UAV object model building module is used to build a UAV object model and generate fault detection filters of actuators and airborne sensors using the UAV object model, wherein the UAV object model includes: a triple-redundant central processing unit, a dual-redundant serial interface unit and a quad-redundant bus unit; A support vector machine model building module, used to collect residual data output by the fault detection filter of the airborne sensor, train a support vector machine model using the residual data and an ant colony algorithm, and record the optimal model parameters obtained by the training; The flight control computer loads the fault detection filter and the one-class support vector machine model; The fault diagnosis module is used to perform fault diagnosis of the flight control system by utilizing hardware redundancy and the support vector machine model; and to update the system structure by utilizing the result of the fault diagnosis.

7. A storage medium, characterized in that: A computer program or instruction is stored, and when the computer program or instruction is executed, the method according to any one of claims 1 to 5 is implemented.

Images