A method for diagnosing faults of carrier-based aircraft based on time causal diagram in the scene of aircraft carrier landing

Abstract

This invention provides a time-causal graph-based fault diagnosis method for carrier-based aircraft in landing scenarios, addressing the technical problem that data-driven carrier-based aircraft fault diagnosis methods cannot meet the high-precision requirements of practical fault diagnosis. First, the basic structure and variable parameter table of the triplet time-causal graph model are defined, transforming the nonlinear mathematical model into a small-deviation form. The time-causal graph model of the carrier-based aircraft is determined using an equation method. Typical failure modes are classified according to the residual characteristics of the failure injection parameters, ultimately obtaining a failure feature vector table for subsequent fault diagnosis. Then, the residuals of the observable parameters of the carrier-based aircraft under fault conditions are calculated. Finally, the original failure feature vectors inferred from the time-causal graph model are corrected using the failure feature vectors trained from simulation data, resulting in a feature vector table of failure modes trained from data, thus completing the quantitative description of fault features. This method achieves carrier-based aircraft fault diagnosis based on the time-causal graph model in landing scenarios that meets the high-precision requirements of practical fault diagnosis.

Description

Technical Field

[0001] This invention relates to a fault diagnosis method for prediction and health management, and particularly to a fault diagnosis method for carrier-based aircraft based on time-cause-effect graphs in a carrier landing scenario. Background Technology

[0002] With the continuous development and performance improvement of carrier-based aircraft flight technology, the risks associated with manually operating carrier-based aircraft for landing are gradually increasing, and the skill requirements for pilots are also rising. To effectively improve the safety of carrier-based aircraft landings and reduce the burden on pilots, automatic landing systems have emerged. However, due to the highly integrated and complex nature of this system, the diagnosis of related faults becomes crucial.

[0003] Carrier-based aircraft landing technology is developing towards a high degree of automation. Aircraft fault diagnosis involves monitoring the aircraft's status, processing flight data, and acquiring corresponding fault feature vectors to locate, isolate, and diagnose the type of faults that have occurred. Research on aircraft fault diagnosis technology both domestically and internationally is gradually evolving from experience-based to data-driven approaches. The rise of artificial intelligence has promoted the widespread application of machine learning, neural networks, and other methods in the field of fault diagnosis.

[0004] However, current data-driven fault diagnosis methods typically rely on reliable datasets, while real-world datasets for carrier-based aircraft automatic landing are difficult to obtain. Therefore, it is necessary to study how to obtain data under different types of faults through simulation in order to construct data-driven fault diagnosis methods. Furthermore, single fault diagnosis methods sometimes fall short of achieving accurate fault diagnosis. Therefore, combining carrier-based aircraft landing models to study the fault generation mechanisms and establish time-cause-effect graphs can achieve more precise diagnosis.

[0005] To address the current state and problems of carrier-based aircraft landing fault diagnosis methods, this invention proposes a fault diagnosis method based on a time-causal graph physical model. This method aims to determine the relationships between various parameters during carrier-based aircraft landing, characterize the system's failure propagation path, overcome the limitations of static systems, and simplify the complex nonlinear mathematical model of the system into a description of the relationships between various state variables, thus providing a foundation for subsequent fault diagnosis. This innovation makes the failure diagnosis research of carrier-based aircraft systems more consistent with actual operating conditions, providing an effective means to improve the safety and reliability of carrier-based aircraft landings. Summary of the Invention

[0006] To address the challenge that data-driven carrier-based aircraft fault diagnosis methods cannot meet the high-precision requirements of practical fault diagnosis, this invention discloses a carrier-based aircraft fault diagnosis method based on a time-causal graph model. This method, starting from a physical model perspective, determines the relationships between various parameters during carrier-based aircraft landing, overcoming the limitations of static systems. It simplifies the complex nonlinear mathematical model of the system into a description of the relationships between various state variables, thereby achieving more accurate fault diagnosis. This is of great significance for improving the safety and reliability of carrier-based aircraft landings.

[0007] This invention proposes a carrier-based aircraft fault diagnosis method based on a time-causal graph dynamic model. This method aims to determine the relationships between various parameters during the carrier-based aircraft's landing process, characterize the system's failure propagation path, overcome the shortcomings of static systems, and simplify the complex nonlinear mathematical model of the system into a description of the relationships between various state variables, providing a foundation for subsequent fault diagnosis. Its core idea is to establish a carrier-based aircraft causal graph model using the equation method based on the mathematical model of the carrier-based aircraft system, determine the fault feature vector table, and further refine the current fault feature vector table by combining observable parameters with residual analysis. Finally, based on the refined fault vector table, automatic fault diagnosis is achieved using a fuzzy matching algorithm. The proposed carrier-based aircraft fault diagnosis method is as follows: Figure 1 As shown, the specific steps are as follows:

[0008] S1: Define the basic structure and variable parameter table of the triplet time cause-effect graph model; define the triplet time cause-effect graph model as G =<V,E,A> Here, V is the set of nodes in the TCG graph, corresponding to the variables in the system; E is the set of directed edges in the TCG; and A represents the relationships between nodes, which can be considered as the attributes of the directed edges. Next, the variable set V is divided into an input variable set U, a state variable set X, an output variable set Y, a constant parameter set Θ, and an auxiliary variable set P. Specifically, the input variable u∈U corresponds to the known input quantities of the system; the state variable x∈X corresponds to the variables describing the future behavior of the system; the output variable y∈Y corresponds to the measurable response output of the system; the constant parameter θ∈Θ corresponds to the immutable constant parameters used for model constraints in the system, used to calculate the system state quantities; and the auxiliary variable p∈P corresponds to variables that are not strictly necessary but help describe the relationships between other nodes, used to assist in model construction and analysis.

[0009] S2: Determine the carrier-based aircraft time-cause graph model using the equation method; based on the definition of the cause-effect graph model, analyze the signal behavior of the carrier-based aircraft system in the scenario by module. Establish the carrier-based aircraft TCG model for the landing scenario using the equation method. For the eight main modules of the carrier-based aircraft (track controller, trajectory angular rate controller, angle controller, angular rate controller, speed controller, control distribution module, engine and control surfaces, and airframe), establish their corresponding mathematical model equations. These equations are dimensionless small-deviation equations in the carrier-based aircraft system, requiring the nonlinear mathematical model to be converted into a small-deviation form.

[0010] S3: Determine the initial failure feature vector table for carrier-based aircraft; After establishing the TCG model of the carrier-based aircraft system, forward reasoning needs to be performed in the TCG model according to the possible failure modes of the carrier-based aircraft system to obtain the signal characteristics of each failure mode, and the typical failure modes are classified according to the residual characteristics of the failure injection parameters, so as to finally obtain the failure feature vector table for subsequent fault diagnosis.

[0011] S4: Calculate the residuals of observable parameters of the carrier-based aircraft under fault conditions; using the carrier-based aircraft landing simulation model after failure injection, simulate the landing process of the carrier-based aircraft under normal conditions and under various typical failure modes of the control surfaces and flight control system, and export the carrier-based aircraft state information of each landing process as a dataset for fault diagnosis. Then, perform residual analysis using training data: extract the time series of output variables from normal data and fault data, determine a set of average values ​​using a sliding window of equal length, and use the difference between each variable under fault data and normal data as the residual value of each variable.

[0012] S5: Correct the feature vector table based on the residual of observable parameters; use the failure feature vector obtained by training with simulation data to correct the original failure feature vector obtained by reasoning from the time cause-effect graph model, and finally obtain the feature vector table of failure modes trained by data, thus completing the quantitative description of fault features.

[0013] The features of this invention are:

[0014] (1) This invention explores the relationship between parameters starting from the model, which can make the research on fault diagnosis of carrier-based aircraft more in line with actual working conditions;

[0015] (2) This invention describes dynamic structures and can utilize derivative information and discontinuous information in the system;

[0016] (3) The present invention can simplify the complex nonlinear mathematical model of the system into a description of the relationship between various state variables, and clearly reflect the failure transmission path in the carrier-based aircraft system. Attached Figure Description

[0017] Figure 1A flowchart for fault diagnosis of carrier-based aircraft landing process based on a time-cause-effect graph model.

[0018] Figure 2 The image shows a partial TCG diagram of a trajectory controller, where single-circle nodes represent system state variables; square-framed nodes represent input variables; double-circle nodes represent measurable state variables, i.e., output variables; and unframed nodes represent auxiliary variables.

[0019] Figure 3 This is the complete TCG diagram of the trajectory controller obtained based on the mathematical model of the trajectory controller.

[0020] Figure 4 The diagram shows the TCG (Trajectory Angular Rate Controller) diagram.

[0021] Figure 5 This is a TCG diagram for the angle controller.

[0022] Figure 6 This is a TCG diagram for an angular velocity controller.

[0023] Figure 7 This is a TCG diagram for a speed controller.

[0024] Figure 8 TCG diagram for control allocation module.

[0025] Figure 9 This is a TCG diagram of the engine and control surfaces.

[0026] Figure 10 This is a TCG diagram of a carrier-based aircraft.

[0027] Figure 11 The diagram illustrates the failure propagation process of the controller output signal C_T, taking a constant positive deviation as an example.

[0028] Figure 12 This is a graph of the p-variable residuals obtained from a set of data for aileron hardware failure (upper limit). In the graph, the actual residual curve refers to the actual curve of the corresponding variable residual within 10 windows after the failure, and the fitted residual curve refers to the second-order fitted curve of the residual data within 8 windows after the failure.

[0029] Figure 13 The residual curve of the q-variable after the aileron hardware failure.

[0030] Figure 14 This is a graph showing the residual curve of the r variable after the aileron hardware failure. Detailed Implementation

[0031] The following detailed description, with reference to the accompanying drawings, illustrates a fault diagnosis method based on a temporal causal graph (TCG) model provided by this invention. This paper studies a diagnostic method for carrier-based aircraft landing processes based on a TCG model, the specific process of which is as follows: Figure 1 As shown, the data detected during the dynamic process of carrier-based aircraft landing is compared with the normal model. After residual analysis, a time-cause-effect graph model derived from the normal model through causal relationships is used to obtain a qualitative description of the fault characteristics, ultimately achieving fuzzy matching fault isolation. This method establishes a carrier-based aircraft cause-effect graph model under the landing scenario based on the mathematical model of the carrier-based aircraft landing system, and extracts features of typical failure modes of carrier-based aircraft, laying the foundation for subsequent fault diagnosis.

[0032] S1: Define the basic structure and variable parameter table of the triplet time cause-effect graph model.

[0033] First, this method defines the time-cause-effect graph model of triples as G =<V,E,A> Where V is the set of nodes in the TCG graph, corresponding to the variables in the system; E is the set of directed edges in the TCG; and A is the relationship between nodes, which can be regarded as the attribute of the directed edges, that is, it corresponds to the structural parameters of the system.

[0034] Given the complexity and numerous variables in carrier-based aircraft systems, this method further categorizes system variables to distinguish different types and more clearly identify the more important variables. Therefore, the variable set V is divided into an input variable set U, a state variable set X, an output variable set Y, a constant parameter set Θ, and an auxiliary variable set P. Specifically, input variables u∈U correspond to the known input quantities of the system; state variables x∈X correspond to variables describing the future behavior of the system; output variables y∈Y correspond to the measurable response output quantities of the system; constant parameters θ∈Θ correspond to invariant constant parameters used for model constraints and for calculating system state quantities; and auxiliary variables p∈P correspond to variables that are not strictly necessary but help describe the relationships between other nodes, assisting in model construction and analysis.

[0035] Next, based on the definition of the cause-effect graph model, the signal behavior of the carrier-based aircraft system in the scenario is analyzed module by module. First, the variables of the carrier-based aircraft system are classified and organized based on the carrier-based aircraft landing mathematical model. The classification results are shown in Table 1 below.

[0036] Table 1 Summary of Variables for Carrier-based Aircraft Systems

[0037]

[0038] In the table, p, q, and r represent the angular velocities of the carrier-based aircraft, and V... kx represents the aircraft's flight speed, x, y, and z represent the aircraft's position information, α and β represent the pitch angle and sideslip angle, respectively, and μ represents the roll angle. s y s z s Let d be the coordinates of the aircraft carrier's position. p d q d r To mitigate the impact of the wake model on the angular velocity of carrier-based aircraft, d χ d γ To mitigate the interference of the wake model on the trajectory angle, d Vk The influence of the wake model on velocity is represented by σ, where σ is the engine mounting angle, and δ is the velocity. a δ e δ r δ lef δ tef These are the aileron deflection, rudder deflection, elevator deflection, leading-edge flap deflection, and trailing-edge flap deflection, δ p Let y be the throttle opening, D, C, and Y be the aerodynamic angles, L, M, and N be the aerodynamic torques, and χ and γ be the heading angle and glide slope angle, respectively.

[0039] S2: Determining the Time Cause-and-Effect Graph Model of Carrier-based Aircraft Using Equation Method

[0040] This project uses the equation method to establish a carrier-based aircraft TCG model for landing scenarios. The equations are dimensionless small-deviation equations in the carrier-based aircraft system, which need to be converted from nonlinear mathematical models into small-deviation forms. Due to the complexity of the carrier-based aircraft system, its mathematical model is first divided into: trajectory controller, track angular rate controller, angle controller, angular rate controller, speed controller, control distribution module, engine and control surfaces, and airframe.

[0041] Taking the trajectory controller as an example, a TCG model is established. Based on the mathematical model of the trajectory controller, the equations can be obtained:

[0042]

[0043] In the formula, y path The y-coordinate of the ideal landing point in the ground coordinate system.

[0044] Representing it in small deviation form, we get:

[0045]

[0046] make b1=sinγ 0 sinχ 0 b2=-cosγ 0 sinχ 0 +1、 b4=2ξ n ωn b5=ω n 2 Among them, b1 to b5 do not contain variable system structure parameters, which can be represented by "1" in the TCG diagram. Meanwhile, to make the TCG diagram neater and more aesthetically pleasing, an auxiliary variable v1 = δy is introduced. g1 , v3=∫δy path v4=∫δy a .

[0047] In summary, partial trajectory controller TCG graphs can be created using Python and Graphviz, such as... Figure 2 In this diagram, single-circle nodes represent system state variables; double-circle nodes represent measurable state variables, i.e., output variables; and unselected nodes represent auxiliary variables. Nodes with square boxes in the diagram represent input variables. The specific meanings of each variable are shown in Table 2 below.

[0048] Table 2 Summary of variable naming in time-cause-effect diagram

[0049]

[0050]

[0051] Similarly, for the solution C in the trajectory controller γ The analysis focuses on parts of the carrier-based aircraft system and other modules. Using a carrier-based aircraft landing model, the propagation coefficients between parameters are determined by sign judgment, approximate numerical calculations, and selection. This yields the TCG diagrams for each module, as shown below. Figures 3-10 As shown in Table 2, the meanings of the important variables are summarized therein. Finally, by ignoring some minor causal relationships based on the magnitude of the parameters, the TCG models of the eight modules are integrated to obtain the carrier-based aircraft TCG model for the landing scenario.

[0052] S3: Determine the initial failure feature vector table for carrier-based aircraft

[0053] After establishing the TCG model of the carrier-based aircraft system, forward inference needs to be performed within the TCG model based on the possible failure modes of the system to predict system behavior and derive failure mode feature vectors that can be used for subsequent fault diagnosis. Because the carrier-based aircraft TCG model contains negative feedback loops, which may create loops, contradictory inference results may occur for some variables in the system during the forward inference process. To address this issue, this method assumes the following: negative feedback loops in the system will prevent changes in variables, but they will not change the direction of those changes.

[0054] Based on the above assumptions, the propagation and signal characteristics of typical failure modes of carrier-based aircraft are analyzed: typical failure modes are listed, and for each failure f, its fault characteristic set S is given. f ={S1,S2,...,S} n}, where S i =(y:s magnitude ,s d1 ,s d2 ) represents the signal analysis tuple for the TCG output variable y in the current failure mode, and s magnitude ,s d1 ,s d2 The values ​​∈{0,+,-} are used to qualitatively describe the amplitude, first derivative, and expected direction of change of the second derivative of the response signal when the system detects a residual.

[0055] The controller output signal C T Taking a constant positive deviation as an example, the failure propagation process is as follows: Figure 11 As shown: C T It is represented as "+,0,0"; after propagation to δ p δ p It is expressed as "+,0,0", and then through a 16 The propagation reaches T, where T is represented as "+, 0, 0"; changes in T then affect v. 74 , They are represented as "+,0,0", "+,0,0", and "-,0,0" respectively; then v 74 Propagation to V via integral relation k V k It is represented as "0,+,0"; Changes affect v 71 v 73 This, in turn, affects α and μ. Changes affect v 72 This, in turn, affects β, so α, β, and μ are represented as "0,-,0", "0,-,0", and "0,+,0" respectively; similarly, V k It will continue to affect x a y a z a , and V k α and μ will affect the second derivatives of χ and γ through integral relations. Figure 11 In the diagram, + and - represent changes in magnitude; ↑ and ↓ represent changes in the first derivative; and ↑↑ and ↓↓ represent changes in the second derivative.

[0056] Similarly, the signal characteristics of each failure mode can be deduced from the cause-effect graph model of the carrier-based aircraft system, as shown in Table 3:

[0057] Table 3. Failure Mode Characteristics Table for Cause-Effect Graph Model of Carrier-based Aircraft System (Definition)

[0058]

[0059]

[0060] Since time-cause-effect diagrams can only qualitatively describe the causal relationships between variables, classifying the typical failure modes of the control surfaces and flight control systems according to the residual characteristics of the failure injection parameters yields 12 sets of failure feature vectors, as shown in Table 4 below:

[0061] Table 4. Failure Feature Vectors under Typical Failure Modes of Control Surfaces and Flight Control Systems

[0062] Group Failure characteristics p q r <![CDATA[V k ]]> <![CDATA[x a ]]> <![CDATA[y a ]]> <![CDATA[z a ]]> α β μ χ γ 1 <![CDATA[δ a +、C N +]]> 0+- / 0++ / / / / 00- 00- 00+ / / 2 <![CDATA[δ a -、C N -]]> 0-+ / 0-- / / / / 00+ 00+ 00- / / 3 <![CDATA[δ e +、C Y +]]> 00+ 0-+ 00- 0-+ 00- 00- 00- 0-- 0-- 0+- 0-- 0+- 4 <![CDATA[δ e -、C Y -]]> 00- 0+- 00+ 0+- 00+ 00+ 00+ 0++ 0++ 0-+ 0++ 0-+ 5 <![CDATA[δ r +、C L +]]> 0+- / 0-+ / / / / 00- 00+ 00+ / / 6 <![CDATA[δ r -、C L -]]> 0-+ / 0+- / / / / 00+ 00- 00- / / 7 <![CDATA[δ lef +]]> 00+ 00+ 00- 0-+ 00- 00- 00- 0-+ 0-- 0+- 0-- 0+- 8 <![CDATA[δ lef -]]> 00- 00- 00+ 0+- 00+ 00+ 00+ 0+- 0++ 0-+ 0++ 0-+ 9 <![CDATA[δ tef +、C M +]]> 00+ 0+？ 00- 0-+ 00- 00- 00- 0-+ 0-- 0+- 0-- 0+- 10 <![CDATA[δ tef -、C M -]]> 00- 0-？ 00+ 0+- 00+ 00+ 00+ 0+- 0++ 0-+ 0++ 0-+ 11 <![CDATA[δ p +、C T +、T+]]> 00- 00- 00+ 0+- 00+ 00+ 00+ 0-- 0-+ 0+- 0-+ 0++ 12 <![CDATA[δ p +、C T -、T-]]> 00+ 00+ 00- 0-+ 00- 00- 00- 0++ 0+- 0-+ 0+- 0--

[0063] In the table above, the "+" and "-" signs for failure characteristics represent the theoretical residual signs of the corresponding parameters after failure injection. Taking aileron hardware failure (upper limit) as an example, after this failure is injected into the carrier-based aircraft system, δ a The signal rapidly increases to its upper limit, exceeding the value at the corresponding moment during normal landing. Therefore, the residual is positive, exhibiting a failure characteristic of δ. a +. In Table 4, “ / ” indicates that when performing system behavior reasoning in this failure mode, the magnitude, first derivative, and second derivative of the corresponding variable are not affected, equivalent to “000”; “?” indicates that in the process of system behavior reasoning, different branches yield contradictory results, and it is impossible to determine which branch has a greater impact based on the magnitude of the causal relationship between variables, so it is treated as “0” in the code.

[0064] Due to the extreme complexity of carrier-based aircraft systems, while the time-causal graphical model is based on the mathematical model of the system, it inevitably overlooks many minor dynamic processes. Furthermore, higher-order features may influence changes in lower-order features during these dynamic processes. In addition, during the system behavior reasoning process, contradictory results from different branches frequently occurred. Although a simple judgment was made based on the magnitude of the causal relationships between variables, some uncertainties remain due to qualitative analysis. These issues render the failure feature vector obtained from the time-causal graphical model unreliable, necessitating correction based on experimental simulation data.

[0065] S4: Calculate the residuals of observable parameters of carrier-based aircraft under fault conditions.

[0066] Using a carrier-based aircraft landing simulation model after failure injection, the landing process of carrier-based aircraft under normal conditions and with control surfaces and flight control systems in various typical failure modes were simulated. The carrier-based aircraft state information for each landing process was exported as a dataset for fault diagnosis. During dataset generation, a total of 200 simulations were performed for each landing process under each condition, with 100 sets used as training data and the other 100 sets used as test data. Furthermore, the carrier-based aircraft's initial landing position, the aircraft carrier's initial position, the aircraft carrier's speed, and the failure injection time in each simulation all exhibit a certain degree of randomness, and once a failure occurs, it cannot be recovered.

[0067] Then, residual analysis was performed using the training data: the output variables p, q, r, and V were extracted from the normal and fault data. k x a y a z a The time series of α, β, and μ are averaged by taking the data at 15 time points as a window. The variable values ​​of the fault data are subtracted from the variable values ​​of the normal data at the corresponding time points. Based on the difference results before the failure injection, the influence of simulation random quantities such as the initial landing position of the carrier-based aircraft is eliminated, thus obtaining the residuals of each variable corresponding to each set of fault data. Figures 12-14 The graph shows the residuals of p, q, and r obtained after the above operations for a set of data related to aileron hardware failure (upper limit). In the graph, the actual residual curve refers to the actual curve of the corresponding variable residual within 10 windows after the failure, and the residual fitted curve refers to the second-order fitted curve of the residual data within 8 windows after the failure. Based on the graph, it can be roughly determined that the p variable characteristic is "0++", the q variable characteristic is "00-", and the r variable characteristic is "0++" for a set of data related to aileron hardware failure (upper limit).

[0068] S5: Eigenvector table corrected based on observable parameter residuals

[0069] To unify the qualitative standards for the features of each set of data and to more accurately describe the features of the variables, it is necessary to use the training data to obtain the true calculation results of 36 features. By analyzing the results, a threshold is set. If the value is less than the threshold, the feature is judged as "0" and recorded as 0. If the value is greater than the threshold, the feature is judged as "+" or "-" and recorded as 1 or -1 according to the sign of the feature calculation result.

[0070] Using the threshold set in the previous step as the basis for qualitative feature description, the training data is iterated again. Each data set will yield a corresponding feature vector. Based on the actual failure modes injected into the training data, all the obtained feature vectors are divided into 12 groups as shown in Table 3. The mode of each feature in each group is taken as the qualitative description result of that feature. The failure mode feature table trained from the data is shown in Table 5 below:

[0071] Table 5. Failure Mode Feature Vector Table Based on Data-Driven Approach

[0072] Failure characteristics p q r <![CDATA[V k ]]> <![CDATA[x a ]]> <![CDATA[y a ]]> <![CDATA[z a ]]> α β μ χ γ <![CDATA[δ a +、C N +]]> 0++ 00- 0++ 00+ 00- 00+ 00- 00- 0++ 0++ 0++ 00- <![CDATA[δ a -、C N -]]> 0-- 00- 0-- 00+ 00+ 00- 00- 00- 0-- 0-- 0-- 00- <![CDATA[δ e +、C Y +]]> 00- 0-- 00- 0-+ 00+ 00- 00- 0-+ 00- 0-+ 0-- 0++ <![CDATA[δ e -、C Y -]]> 00- 0+- 00- 0+- 00- 00- 00+ 0++ 00+ 0+- 0-+ 0-- <![CDATA[δ r +、C L +]]> 0++ 00- 0-- 00- 00+ 00+ 00- 00- 0++ 0++ 0++ 00- <![CDATA[δ r -、C L -]]> 0-- 00- 0++ 00+ 00+ 00- 00+ 00- 0-- 0-- 0-- 00- <![CDATA[δ lef +]]> 00+ 0++ 00+ 0-+ 00+ 00- 00- 0-- 00+ 0-+ 0-- 0++ <![CDATA[δ lef -]]> 00- 0-- 00- 0+- 00- 00- 00+ 0++ 00+ 0-- 0-- 0-- <![CDATA[δ tef +、C M +]]> 00- 0++ 00- 0-+ 00+ 00+ 00- 0++ 00- 0-- 0++ 00+ <![CDATA[δ tef -、C M -]]> 00- 0-- 00- 0+- 00- 00- 00+ 0++ 00+ 0+- 0-- 0-- <![CDATA[δ p +、C T +、T+]]> 00- 00- 00- 0-- 00- 00+ 00- 00+ 00- 0-- 0-- 00- <![CDATA[δ p +、C T -、T-]]> 00+ 00- 00+ 0-- 00- 00- 00- 0-+ 00+ 0++ 0++ 00+

[0073] Finally, the original failure feature vector obtained from the time-cause-effect graph model inference is corrected using the failure feature vector obtained from data training. The proportion of each feature in the table above in all training results of its group is calculated. If the proportion is greater than 50%, the feature at the corresponding position in the original failure feature table is corrected. The final corrected failure feature table is shown in Table 6 below:

[0074] Table 6. Corrected Failure Mode Feature Vector Table

[0075]

[0076]

[0077] In the table above, the symbols marked with an underline are features that changed during the correction process, while the symbols without an underline are features that did not change.

Claims

1. A carrier-based aircraft fault diagnosis method based on time-cause-effect graph in a carrier landing scenario, characterized in that, Includes the following steps: This paper defines the basic structure and variable parameter classification method of a triplet time-cause graph model; its key feature is that the definition of the basic structure and variable parameter classification method of the triplet time-cause graph model specifically includes: defining the triplet time-cause graph model as... Here, V is the set of nodes in the temporal cause-effect graph model, corresponding to the variables in the system; E is the set of directed edges in the temporal cause-effect graph model; A represents the relationships between nodes, which can be considered as the attributes of the directed edges; secondly, the variable set V is divided into the input variable set U, the state variable set X, the output variable set Y, the constant parameter set Θ, and the auxiliary variable set P; among which, the input variables... The known inputs of the corresponding system; state variables Variables that describe the future behavior of the system in the corresponding system; output variables The corresponding system's measurable response output; constant parameters The invariant constant parameters that correspond to the model constraints in the system are used to calculate the system state variables; auxiliary variables Variables that are not strictly required in the system but help describe the relationships between other nodes are used to assist in the construction and analysis of the model. Next, according to the definition of the causal graph model, the signal behavior of the carrier-based aircraft system in the magic carpet scenario is sorted out in modules. This paper describes a method for determining the time-cause-effect graph model of a carrier-based aircraft using dimensionless small-deviation equations. Specifically, this method involves establishing a time-cause-effect graph model of a carrier-based aircraft under a magic carpet landing scenario using an equation-based approach. The equations are dimensionless small-deviation equations within the carrier-based aircraft system, requiring the conversion of the nonlinear mathematical model into a small-deviation form. Due to the complexity of the carrier-based aircraft system, its mathematical model is first divided into: trajectory controller, trajectory angular rate controller, angle controller, angular rate controller, speed controller, control distribution module, engine and control surfaces, and airframe. Taking the trajectory controller as an example, a time-cause-effect graph model is established. Based on the mathematical model equation of the trajectory controller, the corresponding small-deviation form can be obtained. （1） make ;in, ~ The system does not contain variable structural parameters, which can be represented by "1" in the time-cause-effect graph model diagram; at the same time, to make the time-cause-effect graph model diagram cleaner and more aesthetically pleasing, auxiliary variables are introduced. Similarly, the solution in the trajectory controller The analysis focuses on the parts of the carrier-based aircraft system and other modules. Using the carrier-based aircraft magic carpet landing model, the propagation coefficients between parameters are judged by sign, roughly calculated and discarded to obtain the time causal graph model of each module. Finally, based on the order of magnitude of the parameters, some causal relationships with minimal impact are ignored, and the time causal graph models of the eight modules are integrated together to obtain the carrier-based aircraft time causal graph model under the magic carpet landing scenario. Based on the forward thrust generated by the time-causal graph model, the initial failure feature vector table of the carrier-based aircraft is determined. A residual vector table using the differences between variables under fault data and normal data as observable parameters; The initial failure feature vector table is corrected based on the residual vector table of observable parameters, thus completing the quantitative description of the fault characteristics.

2. The carrier-based aircraft fault diagnosis method based on time-cause-effect graph in the landing scenario according to claim 1, characterized in that, The aforementioned forward inference based on the time-causal graph model, obtaining the signal characteristics of each failure mode, and determining the initial failure feature vector table of the carrier-based aircraft, specifically includes: Based on the possible failure modes of the carrier-based aircraft system, forward reasoning is performed in the time-cause-effect graph model to predict system behavior and derive failure mode feature vectors that can be used for subsequent fault diagnosis. Since there are negative feedback loops in the carrier-based aircraft time-cause-effect graph model, which may generate loops, some variables in the system may produce contradictory reasoning results during the forward reasoning process. To solve this problem, this method assumes the following: negative feedback loops in the system will prevent changes in variables, but they will not change the direction of change of variables. Based on the above assumptions, the propagation and signal characteristics of typical failure modes of carrier-based aircraft are analyzed: typical failure modes are listed, and for each failure f, its set of fault characteristics is given. ,in This provides the signal analysis tuple for the output variable y of the time-cause-effect graphical model under the current failure mode. It is used to qualitatively describe the amplitude, first derivative, and expected direction of change of the second derivative of the response signal when the system detects a residual; The signal characteristics of each failure mode can be deduced from the causal graph model of the carrier-based aircraft system. Since the time causal graph can only qualitatively describe the causal relationship between variables, the typical failure modes of the control surface and flight control system are classified according to the residual characteristics of the failure injection parameters, and the initial failure feature vector table can be finally obtained.

3. The carrier-based aircraft fault diagnosis method based on time-cause-effect graph in the landing scenario according to claim 1, characterized in that, The simulation data generated by injecting typical faults uses the differences between each variable under fault data and normal data as the residual vector table of observable parameters, as detailed below: Multiple simulations were conducted for the landing process under each condition, divided into training and testing groups. At the same time, the starting position of the carrier-based aircraft, the starting position of the carrier, the speed of the carrier, and the time of failure injection in each simulation all had a certain degree of randomness, and once a failure occurred, it could not be recovered. Then, residual analysis was performed using the training data: output variables p, q, r, ... were extracted from the normal and fault data. , The time series data is obtained by extracting the average value using a fixed sliding window, subtracting the normal data variable value at the corresponding time from the variable value of the fault data, and eliminating the influence of simulation random quantities such as the initial landing position of the carrier-based aircraft based on the difference results before failure injection. This yields the residual vector table of each variable corresponding to each set of fault data.

4. The carrier-based aircraft fault diagnosis method based on time-cause-effect graph in the landing scenario according to claim 1, characterized in that, The residual vector table based on observable parameters is used to correct the initial failure feature vector table, thus completing the quantitative description of fault characteristics; specifically as follows: The actual calculation results of the residual vector table corresponding to the feature are obtained by using the training data. By analyzing and setting a threshold, if the value is less than the threshold, the feature is judged as "0" and recorded as 0. If the value is greater than the threshold, the feature is judged as "+" or "-" and recorded as 1 or -1 according to the sign of the feature calculation result. Using the threshold defined above as the basis for quantitative feature division, the training data is traversed again, and each group of data will obtain a corresponding feature vector. According to the grouping method of the failure mode feature table of the carrier-based aircraft system causal graph model, the specific grouping of all feature vectors obtained according to the actual failure modes injected into the training data is determined. The mode of each feature in each group is used as the quantitative description result of that feature, and the failure mode feature vector table generated based on the data training can be obtained. Finally, the original failure feature vector table obtained by the time-cause-effect graph model is corrected using the failure feature vector table obtained by the data training. The proportion of each feature in the table is calculated in the total training results of its group. If the proportion is greater than 50%, the feature at the corresponding position in the original failure feature table is corrected. The corrected failure mode feature vector table can be obtained in the end. Automatic fault diagnosis is then achieved by using the fuzzy matching algorithm.

Images