Aircraft fault diagnosis method, model training method, device and computer equipment

Abstract

The present application relates to an aircraft fault diagnosis method, a model training method, an apparatus, a computer device and a storage medium. The method comprises: obtaining the measurement information and other auxiliary information of the atmospheric data system and the inertial navigation system in the aircraft system, and constructing an atmospheric parameter model; constructing an unscented Kalman filter model of the atmospheric data system based on the atmospheric parameter model, and estimating the state of the atmospheric data system according to the unscented Kalman filter model; extracting the multidimensional new information and residual sequence of the unscented Kalman filter model; calling the pre-trained aircraft fault diagnosis model, and taking the multidimensional new information and residual sequence as input parameters to obtain the aircraft fault diagnosis result. The combination of model and data is conducive to timely and accurate diagnosis of aircraft atmospheric data system faults, and provides guarantee for the safe flight of the aircraft.

Description

Technical Field

[0001] The present application relates to the field of aircraft technology, and in particular to an aircraft fault diagnosis method, a model training method, an apparatus, a computer device and a storage medium. Background Art

[0002] Among the systems that make up the various parts of an aircraft, the Air Data System (ADS) is mainly responsible for measuring the atmospheric parameters during the flight of the aircraft and calculating the relevant flight parameters through the internal computer. However, the atmospheric data sensor may fail due to harsh flight environments such as impact, electromagnetic interference, and abnormalities of its own components, outputting incorrect flight parameters, causing errors in subsequent control links, and then catastrophic accidents. Therefore, it is of great significance to perform atmospheric data system fault diagnosis and complete detection and identification as soon as possible after a fault occurs, in order to improve the reliability of the aircraft.

[0003] In recent years, data-based fault diagnosis methods have begun to emerge, including support vector machines, neural networks, etc., which can extract fault features from a large amount of data without building an accurate model, and have achieved relatively good results in the field of fault diagnosis. However, since the sensors of the atmospheric data system are located on the surface of the aircraft, the working environment is harsh, and noise and other interference are often large, making it difficult to diagnose small faults. In addition, the aircraft is in an uncertain state such as multiple flight trajectories under different flight missions, and the data of each flight state often changes greatly. Conventional data-based fault diagnosis strategies may not be able to distinguish the changes in atmospheric parameters caused by changes in the flight state itself, and it is not realistic to obtain data in all situations in advance, and the wide application is limited. Summary of the invention

[0004] Based on this, it is necessary to provide an aircraft fault diagnosis method, model training method, device, computer equipment and storage medium that can accurately diagnose aircraft faults through an atmospheric data system to address the above technical problems.

[0005] In a first aspect, the present application provides an aircraft fault diagnosis method. The method comprises:

[0006] Obtain the measurement information and other auxiliary information of the atmospheric data system and inertial navigation system in the aircraft system, and build an atmospheric parameter model;

[0007] An unscented Kalman filter model of the atmospheric data system is constructed based on the atmospheric parameter model, and the state of the atmospheric data system is estimated according to the unscented Kalman filter model;

[0008] Extract multi-dimensional innovation and residual sequences of the unscented Kalman filter model;

[0009] The pre-trained aircraft fault diagnosis model is called, and the multi-dimensional new information and residual sequence are used as input parameters to obtain the aircraft fault diagnosis results.

[0010] In one embodiment, estimating the state of an air data system according to an unscented Kalman filter model includes:

[0011] Obtain the state equation and measurement equation of the unscented Kalman filter model;

[0012] Based on the state equation and Sigma point set, the state variables and error covariance matrix of one-step prediction are obtained;

[0013] The state of the atmospheric data system under normal conditions is estimated based on the one-step prediction value and the measurement equation.

[0014] In one embodiment, the method further comprises:

[0015] obtaining parameter failure signals measured in an atmospheric parameter model;

[0016] The parameter fault signal is input into the measurement equation, and the fault measurement equation under the fault state is updated.

[0017] In one embodiment, extracting a multidimensional innovation and residual sequence of an unscented Kalman filter model includes:

[0018] According to the unscented Kalman filter model, the measurement prediction value, the actual measurement value, the state estimation value and the actual state are obtained;

[0019] Get multi-dimensional new information based on the measurement prediction value and actual measurement value;

[0020] The residual sequence is obtained based on the state estimate and the actual state.

[0021] In a second aspect, the present application also provides an aircraft fault diagnosis model training method. The method comprises:

[0022] Input the multi-dimensional innovation and residual sequence into the Inception V3 network model as training data;

[0023] The weights of each channel of the input feature map in the Inception V3 network model are analyzed by ECA to obtain the weighted feature map.

[0024] The weights are adjusted iteratively according to the training data to make the Inception V3 network model converge or reach the preset number of training times, thus completing the construction of the aircraft fault diagnosis model.

[0025] In one embodiment, the weights of each channel of the input feature map in the Inception V3 network model are analyzed by ECA to obtain a weighted feature map including:

[0026] Perform global flat pooling on the input feature map through ECA and compress the input feature map into a vector;

[0027] Get the adaptive one-dimensional convolution kernel size according to the number of channels of the input feature map;

[0028] Use the one-dimensional convolution kernel size in the one-dimensional convolution to obtain the weight corresponding to each channel of the input feature map;

[0029] The normalized weight of each channel is multiplied by the input feature map channel by channel to obtain a weighted feature map.

[0030] In a third aspect, the present application also provides an aircraft fault diagnosis device. The device comprises:

[0031] The atmospheric model module is used to obtain the measurement information and other auxiliary information of the atmospheric data system and inertial navigation system in the aircraft system and build an atmospheric parameter model;

[0032] An estimation module, used for constructing an unscented Kalman filter model of an atmospheric data system based on an atmospheric parameter model, and estimating a state of the atmospheric data system according to the unscented Kalman filter model;

[0033] An extraction module, used to extract multi-dimensional new information and residual sequences of the unscented Kalman filter model;

[0034] The diagnosis module is used to call the pre-trained aircraft fault diagnosis model, take the multi-dimensional new information and residual sequence as input parameters, and obtain the aircraft fault diagnosis result.

[0035] In a fourth aspect, the present application also provides an aircraft fault diagnosis model training device. The device comprises:

[0036] The input module is used to input the multi-dimensional innovation and residual sequences as training data into the Inception V3 network model;

[0037] The weighting module is used to analyze the weights of each channel of the input feature map in the Inception V3 network model through ECA to obtain a weighted feature map;

[0038] The training module is used to iteratively adjust the weights according to the training data so that the Inception V3 network model converges or reaches the preset number of training times to complete the construction of the aircraft fault diagnosis model.

[0039] In a fifth aspect, the present application further provides a computer device. The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0040] Obtain the measurement information and other auxiliary information of the atmospheric data system and inertial navigation system in the aircraft system, and build an atmospheric parameter model;

[0041] An unscented Kalman filter model of the atmospheric data system is constructed based on the atmospheric parameter model, and the state of the atmospheric data system is estimated according to the unscented Kalman filter model;

[0042] Extract multi-dimensional innovation and residual sequences of the unscented Kalman filter model;

[0043] The pre-trained aircraft fault diagnosis model is called, and the multi-dimensional new information and residual sequence are used as input parameters to obtain the aircraft fault diagnosis results.

[0044] In a sixth aspect, the present application further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0045] Obtain the measurement information and other auxiliary information of the atmospheric data system and inertial navigation system in the aircraft system, and build an atmospheric parameter model;

[0046] An unscented Kalman filter model of the atmospheric data system is constructed based on the atmospheric parameter model, and the state of the atmospheric data system is estimated according to the unscented Kalman filter model;

[0047] Extract multi-dimensional innovation and residual sequences of the unscented Kalman filter model;

[0048] The pre-trained aircraft fault diagnosis model is called, and the multi-dimensional new information and residual sequence are used as input parameters to obtain the aircraft fault diagnosis results.

[0049] The above-mentioned aircraft fault diagnosis method, model training method, device, computer equipment and storage medium construct an atmospheric parameter model based on the measurement information and other auxiliary information of the atmospheric data system and inertial navigation system in the aircraft system; further establish an unscented Kalman filter model of the atmospheric data system to achieve accurate estimation of the normal state of the aircraft, and on this basis extract the multi-dimensional information and residual sequence in the zone-level Kalman filter model, and input them into the pre-trained aircraft fault diagnosis model as fault features, and finally output the fault classification results to achieve fault diagnosis. By combining the model and data, the problem of model-based aircraft fault diagnosis methods relying on the main setting or related experience, and the problem of decreased fault diagnosis effect of data-based aircraft fault diagnosis methods under different conditions are solved, which is conducive to timely and accurate diagnosis of aircraft atmospheric data system faults and provides guarantee for the safe flight of aircraft. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1is an application environment diagram of an aircraft fault diagnosis method in an embodiment;

[0051] Figure 2 is a schematic flow chart of an aircraft fault diagnosis method in one embodiment;

[0052] Figure 3 is a flow chart of an aircraft fault diagnosis method in one embodiment;

[0053] Figure 4a is a curve showing the change of filtered data under the condition of slow-changing airspeed signal fault in one embodiment;

[0054] Figure 4b is a curve showing the change of multi-dimensional new information under the condition of slow-changing airspeed signal fault in one embodiment;

[0055] Figure 4c is a variation curve of the residual sequence under the condition of a slowly varying airspeed signal fault in one embodiment;

[0056] Figure 5 is a flow chart of an aircraft fault diagnosis model training method in one embodiment;

[0057] Figure 6 This is a diagram of the Inception V3 network model structure in one embodiment;

[0058] Figure 7 is an error change curve of the Inception V3 network training process in one embodiment;

[0059] Figure 8 is a confusion matrix result of an aircraft fault diagnosis model in one embodiment;

[0060] Fig. 9 A schematic diagram of the ECA attention mechanism in one embodiment;

[0061] Fig.10 is an overall flow chart of aircraft fault diagnosis in one embodiment;

[0062] Fig.11 is a schematic flow chart of an aircraft fault diagnosis method in another embodiment;

[0063] Fig.12 is a structural block diagram of an aircraft fault diagnosis device in one embodiment;

[0064] Fig.13 is a structural block diagram of an aircraft fault diagnosis model training device in one embodiment;

[0065] Fig.14 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0066] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0067] The applicant noted that in the field of aircraft fault diagnosis, the main methods currently include model-based methods and data-based methods. Model-based fault diagnosis methods generate relevant residuals through parameter estimation methods, state estimation methods, equivalent space methods, etc., and then achieve fault diagnosis through chi-square test methods, optimal parity vector methods, etc. However, compared with combined navigation systems such as INS / GPS (INS, Inertial Navigation System; GPS, Global Positioning System; inertial navigation-satellite integrated navigation system), the model of the atmospheric data system is difficult to describe in an analytical way, and ordinary modeling methods cannot accurately describe the system; and the traditional residual generation method based on the atmospheric data solution equation cannot fully consider the various environmental interference factors in actual situations, making it difficult to distinguish between normal and fault situations, which leads to a decrease in diagnostic accuracy. In addition, due to the high-speed and high-maneuverability characteristics of the aircraft, there are many uncertainties during the flight process, and there are influencing factors such as unstable vortices, which leads to the reduction of the reliability of some hypothesis testing-based methods in the diagnosis stage.

[0068] The aircraft fault diagnosis method provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or it can be placed on the cloud or other network servers. Among them, the terminal 102 can be but is not limited to various personal computers, laptops, smart phones, tablets, Internet of Things devices and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart car-mounted devices, etc. Portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The server 104 can be implemented with an independent server or a server cluster consisting of multiple servers.

[0069] In one embodiment, Figure 2 As shown, a method for diagnosing aircraft faults is provided. Figure 1 The server in the example is used to illustrate the following steps:

[0070] Step 202: Acquire measurement information and other auxiliary information of the atmospheric data system and inertial navigation system in the aircraft system to construct an atmospheric parameter model.

[0071] Step 204: construct an unscented Kalman filter model of the atmosphere data system based on the atmosphere parameter model, and estimate the state of the atmosphere data system according to the unscented Kalman filter model.

[0072] Step 206: extract the multi-dimensional innovation and residual sequence of the unscented Kalman filter model.

[0073] Step 208, calling the pre-trained aircraft fault diagnosis model, taking the multi-dimensional new information and residual sequence as input parameters, and obtaining the aircraft fault diagnosis result.

[0074] Among them, the atmospheric data system is a comprehensive, high-precision atmospheric data information system. As an important airborne equipment, the atmospheric data system is used to measure, calculate and display multiple aircraft flight parameters such as indicated airspeed, true airspeed, flight speed, pressure altitude, Mach number, total atmospheric temperature, and static atmospheric temperature. The inertial navigation system is an autonomous navigation system that does not rely on external information and does not radiate energy to the outside. It is based on Newton's laws of mechanics. By measuring the acceleration of the carrier in the inertial reference system, integrating it over time, and transforming it into the navigation coordinate system, it can obtain information such as speed, attitude angle and position in the navigation coordinate system.

[0075] For details, please refer to Figure 3 , extract the measurement information of the atmospheric data system and inertial navigation system in the aircraft system and other auxiliary information. The extracted aircraft system parameters used to build the atmospheric parameter model include airspeed, angle of attack, sideslip angle, attitude angle (yaw angle, pitch angle, roll angle), acceleration, and angular velocity. According to the conversion relationship between the airflow coordinate system and the aircraft coordinate system, the airspeed V and the components of each axis of the aircraft coordinate system [u, v, w] can be obtained. T The relationship is shown in formula (1):

[0076]

[0077] Differentiating it and transforming it by matrix gives:

[0078]

[0079] Ground speed vector V e 、Airspeed vector V r and wind speed vector v w There is a relationship between them as shown in formula (3):

[0080] V e =V r +V w (3)

[0081] The aircraft dynamics equation in the aircraft coordinate system can be written as shown in formula (4):

[0082]

[0083] Where A is the acceleration vector, T be is the transformation matrix from the ground coordinate system to the missile body coordinate system, g = [0,0,9.81] T , ω is the angular velocity vector, from which we can get formula (5):

[0084]

[0085] Since the wind speed is relatively small during the flight of the aircraft, the wind speed related quantity is recorded as d. Let V r =[u,v,w] T , A=[A x ,A y ,A z ] T ,ω=[p,q,r] T , d=[d u ,d v ,d w ] T , we can get formula (6) by rearranging:

[0086]

[0087] Where φ, θ, ψ are the roll angle, pitch angle, and yaw angle, respectively. The attitude angle can be estimated by the equation of motion (7):

[0088]

[0089] Formula (6) and formula (7) are the atmospheric parameter models of the atmospheric data system.

[0090] The state equation and measurement equation of the atmospheric data system are derived according to the atmospheric parameter model, and the unscented Kalman filter model of the atmospheric data system is established to obtain an accurate estimation of the normal state of the system; the multidimensional new information and residual sequences of the unscented Kalman filter model (UKF, Unscented Kalman Filter) of the atmospheric data system are extracted as fault features. The multidimensional new information and residual sequences are calculated by the unscented Kalman filter model and are continuously iteratively updated with time steps; an aircraft fault diagnosis model is designed, and the fault features are further extracted by training the model, and the fault diagnosis results of the aircraft atmospheric data system are obtained.

[0091] In the above-mentioned aircraft fault diagnosis method, an atmospheric parameter model is constructed based on the measurement information of the atmospheric data system and the inertial navigation system in the aircraft system and other auxiliary information; an unscented Kalman filter model of the atmospheric data system is further established to achieve accurate estimation of the normal state of the aircraft, and on this basis, multi-dimensional information and residual sequences in the unscented Kalman filter model are extracted and input into the pre-trained aircraft fault diagnosis model as fault features, and finally the fault classification results are output to achieve fault diagnosis. By combining the model and data, the problem of model-based aircraft fault diagnosis methods relying on the main setting or related experience, and the problem of decreased fault diagnosis effect of data-based aircraft fault diagnosis methods under different conditions are solved, which is conducive to timely and accurate diagnosis of aircraft atmospheric data system faults and provides guarantee for the safe flight of aircraft.

[0092] In one embodiment, estimating the state of the atmospheric data system according to the unscented Kalman filter model includes: obtaining the state equation and measurement equation of the unscented Kalman filter model; obtaining the one-step predicted state variables and error covariance matrix based on the state equation and the Sigma point set; estimating the state of the atmospheric data system under normal conditions according to the one-step predicted value and the measurement equation.

[0093] Among them, Sigma points are sampling points selected according to a certain rule in the original state distribution, that is, the mean and variance after sampling are equal to the mean and variance of the original state distribution. The probability density distribution of the nonlinear function is approximated through these points, and then the state of the aircraft at the current moment is predicted.

[0094] Specifically, the state equation and measurement equation of the unscented Kalman filter model of the atmospheric data system are obtained, as shown in formula (8):

[0095]

[0096] In the formula, x=[V,α,β,φ,θ,ψ] T is the state vector, is the measurement noise of acceleration and angular velocity measured by the inertial navigation system, input u = [A x ,A y ,A z ,p,q,r] T +w, disturbance d = [d u ,d v ,d w ] T , z=[V m ,α m ,β m ,φ m ,θ m ,ψ m ] Tis the system measurement value, v is the corresponding measurement noise; the nonlinear function f, noise matrix G(x), disturbance matrix E(x), and observation matrix h can all be obtained through the atmospheric parameter equations (6) and (7).

[0097] The Sigma point set is calculated according to formula (8), and the Sigma point set is shown in formula (9):

[0098]

[0099] According to the one-step prediction of the Sigma point set and the state variables, the prediction error covariance matrix is ​​as shown in formula (10):

[0100]

[0101] Where f is the nonlinear state equation function, are the first-order and second-order weight coefficients, w is the process noise, Q is the covariance matrix of the process noise, Q = E[ww T ].

[0102] Generate a new Sigma point set based on the state variables predicted in one step and substitute it into the measurement equation, as shown in formula (11), to calculate the measurement prediction value Autocovariance matrix and the cross-covariance matrix

[0103]

[0104] Where h is the nonlinear measurement equation function, v is the measurement noise, R is the covariance matrix of the measurement noise, R = E[vv T ].

[0105] As shown in formula (12), the Kalman gain K is updated k , state, and covariance:

[0106]

[0107] Among them, x k and P k It is used as the prior information at time k+1 to perform a new round of filtering.

[0108] In this embodiment, the estimation of the state of the atmospheric data system under normal conditions is achieved through unscented Kalman filtering, and the foundation is laid for the subsequent fault information extraction.

[0109] In one embodiment, the method further includes obtaining a parameter fault signal measured in the atmospheric parameter model; inputting the parameter fault signal into the measurement equation, and updating the fault measurement equation under the fault state.

[0110] Specifically, when a fault occurs in the atmospheric data system state, a parameter fault signal measured in the atmospheric parameter model is obtained; the parameter fault signal is input into the measurement equation, and the fault measurement equation under the fault state is updated. The fault measurement equation is shown in formula (13):

[0111] z=hx+v+f (13)

[0112] In the formula, f=[f V ,f α ,f β ,0,0,0] T It represents the failure of the corresponding measurement signals of airspeed V, angle of attack α, and sideslip angle β.

[0113] The fault types existing in the atmospheric data system include sudden change of airspeed fault, slow change of airspeed fault, sudden change of angle of attack fault, slow change of angle of attack fault, sudden change of sideslip angle fault, and slow change of sideslip angle fault.

[0114] The manifestations of sudden faults and slow-changing faults are step faults and ramp faults respectively. Note that the faults of airspeed, angle of attack, and sideslip angle signals can be expressed as:

[0115] Sudden fault, that is, the signal suddenly changes in a step form, as shown in formula (14):

[0116] f(t)=Au(t-t0) (14)

[0117] Where A is the step fault amplitude, u(t-t0) is the unit step function, and t0 is the fault start time.

[0118] Slow-varying faults, that is, the signal deviates over time, as shown in formula (15):

[0119] f(t)=R(t-t0)u(t-t0) (15)

[0120] Where R is the slope of the ramp fault. u(t-t0) is the unit step function, and t0 is the fault start time.

[0121] In this embodiment, the various parts of the aircraft are in a complete closed-loop system, and the impact of the fault will continue to increase over time; the fault measurement equation under the fault state is processed by the unscented Kalman filter model, so that the fault information extraction of the aircraft fault diagnosis model is more accurate.

[0122] In one embodiment, extracting multidimensional new information and residual sequences of an unscented Kalman filter model includes: obtaining measurement prediction values, actual measurement values, state estimation values ​​and actual states according to the unscented Kalman filter model; obtaining multidimensional new information based on the measurement prediction values ​​and the actual measurement values; and obtaining a residual sequence based on the state estimation values ​​and the actual states.

[0123] Specifically, the unscented Kalman filter model of the atmospheric data system is used to extract multidimensional new information and residual sequences. The multidimensional new information is the difference between the actual measured value and the predicted measured value, as shown in formula (16):

[0124]

[0125] In the formula, r k is the multi-dimensional new information, z k is the actual measured value, is the predicted measurement value, the subscript k is the calculation result of the kth time step, the subscript k / k-1 means the result of the kth time step is calculated based on the result of the k-1 time step, and the superscript ∧ represents the estimated value.

[0126] The residual sequence is the difference between the state value and the state estimate, as shown in formula (17):

[0127]

[0128] In the formula, e k is the residual, x k is the status value, is the state estimate, the subscript k is the calculation result of the kth time step, the subscript k / k-1 means the result of the kth time step is calculated based on the result of the k-1 time step, and the superscript ∧ represents the estimate.

[0129] Therefore, the multidimensional innovation and residual sequence is finally selected as shown in formula (18):

[0130]

[0131] The multidimensional innovation and residual sequences both include the 6-dimensional parameters of the atmospheric data system [V, α, β, φ, θ, ψ] T , a total of 12-dimensional data is used as the input of the aircraft fault diagnosis model for subsequent fault diagnosis, location and identification.

[0132] For example, the initial flight speed of the aircraft is set to 240m / s, the take-off altitude is 800m, the navigation points are randomly set, and the range is set to x∈[8000,12000], y∈[-50,50], z∈[1000,2000], all in meters. The simulation duration is set to 50s, and faults are randomly injected within 20 to 35s. At 23.5s, the airspeed signal injection level is 0.05m / s 2 The slow-changing fault lasts for 1.5 seconds. At each sampling moment during the 25-second simulation time, the airspeed signal, airspeed multi-dimensional information, and airspeed residual sequence collected by the aircraft system are Figure 4a , Figure 4b and Figure 4c As shown. It can be seen that for the airspeed signal itself, when the fault degree is small, the data after the fault does not change much relative to the signal itself, and the data itself has certain fluctuations. Compared with before the fault occurs, it is impossible to judge whether it is a normal change or a fault. It is difficult to diagnose from the fault signal level alone. For the multi-dimensional new information, although the changes can also be seen, due to the continuous tracking of the fault by the iterative characteristics of the filter, the subsequent multi-dimensional new information changes to a small extent. When the fault degree is small, it will always fluctuate within a small range. This is contrary to the nature of the damage caused by slow-varying faults increasing over time. For the residual sequence, it can be seen that it gradually increases over time and can effectively reflect the fault situation. Therefore, the residual sequence is added to assist in fault diagnosis.

[0133] In this embodiment, the fault output does not change much at a certain moment due to a slow-varying fault of the aircraft, which makes the multi-dimensional new information change less, and the fault output will further affect the estimated value at the next moment. The multi-dimensional new information that tracks the fault output has been in a small interval, making fault diagnosis difficult. After the residual sequence is introduced, the residual sequence will continue to increase with time when a slow-varying fault occurs, which can effectively make up for the defect that the multi-dimensional new information is insensitive to the slow-varying fault.

[0134] In one embodiment, Figure 5 As shown, a method for training an aircraft fault diagnosis model is provided, comprising the following steps:

[0135] Step 502: Input the multi-dimensional innovation and residual sequence into the Inception V3 network model as training data.

[0136] Step 504: Analyze the weights of each channel of the input feature map in the Inception V3 network model through ECA to obtain a weighted feature map.

[0137] Step 506, repeatedly iteratively adjust the weights according to the training data, so that the Inception V3 network model converges or reaches a preset number of training times, and completes the construction of the aircraft fault diagnosis model.

[0138] Among them, the Inception model is a model that expands the depth and width of the convolutional layer. It constructs a basic neuron structure to build a network structure that maintains network sparsity and high-performance computing. Figure 6 As shown in the figure, Inception V3 improves on Inception by decomposing large convolutions into small convolutions, and then decomposing small convolutions into asymmetric convolutions.

[0139] Specifically, the multidimensional new information and residual sequence are preprocessed, including adding fault labels, normalization and data enhancement, and the preprocessed multidimensional new information and residual sequence are divided into training set, validation set and test set; the training set is input into the Inception V3 network model as training data, and the weights of each channel of the input feature map in the Inception V3 network model are analyzed by ECA (Efficient Channel Attention) to obtain the weighted feature map, and the weights are adjusted iteratively according to the training data to make the Inception V3 network model converge or reach the preset number of training times, and the construction of the aircraft fault diagnosis model is completed. The validation set is used to verify whether the diagnostic error of the aircraft fault diagnosis model meets the requirements. If not, a new round of training is carried out. If it meets the requirements, the construction of the aircraft fault diagnosis model is completed, the model is saved, and finally tested by the test set to output the fault diagnosis results.

[0140] In this embodiment, in terms of the overall network structure design, all convolutional layers in the network are uniformly set to a convolution-batch normalization-activation structure, which can accelerate convergence and improve generalization ability; the aircraft fault diagnosis model introduces a 1×1 convolution kernel for each channel of the input feature map, which can reduce the dimension of the data, reduce the amount of parameter calculation, and improve the nonlinear fitting ability of the model; decomposing large convolutions into small convolutions, and then decomposing small convolutions into asymmetric convolutions, can reduce the amount of parameters while maintaining the receptive field range, and achieve a balance between network performance and computational efficiency; further reflects the effectiveness and advantages of the aircraft fault diagnosis model in atmospheric data system fault diagnosis.

[0141] In one embodiment, the weights of each channel of the input feature map in the Inception V3 network model are analyzed by ECA to obtain a weighted feature map, including: globally flattening the input feature map by ECA to compress the input feature map into a vector; obtaining an adaptive one-dimensional convolution kernel size according to the number of channels of the input feature map; using the one-dimensional convolution kernel size in a one-dimensional convolution to obtain a weight corresponding to each channel of the input feature map; and multiplying the normalized weight of each channel by the input feature map channel by channel to obtain the weighted feature map.

[0142] Specifically, Fig. 9 and Fig.10 As shown in Figure 1, the input feature map in the Inception V3 network model is analyzed through the ECA attention mechanism. The input feature map is subjected to global average pooling, and the input feature map is transformed from a matrix of [h, w, c] to a vector of [1, 1, c]. The adaptive one-dimensional convolution kernel size is calculated according to the number of channels of the feature map, and the one-dimensional convolution kernel size is shown in formula (19):

[0143]

[0144] Where kernel_size is the one-dimensional convolution kernel size, γ=2, and b=1.

[0145] The one-dimensional convolution kernel size kernel_size is used for the one-dimensional convolution to obtain the weight for each channel of the feature map; the normalized weight is multiplied by the original input feature map channel by channel to generate a weighted feature map.

[0146] For example, the decomposition strategy of Inception V3 performs best on medium feature maps (m×m, m between 12-20), so the input time series span is selected as 20 points, and the input parameter is set to 12 according to the analysis in the previous section, then the input dimension is (20, 12), ensuring the efficiency of the Inception V3 network. Since the ECA attention mechanism does not change the data dimension, the result of the splicing of the Inception V3 channels is directly input into ECA, and then the fault diagnosis result is obtained through the fully connected layer and the Softmax classifier (a commonly used classifier in neural networks).

[0147] The number of tracks for each type of aircraft fault is 600. After data enhancement through sliding windows, the number of samples for each situation increases to 3000, including 2500 training set samples and 500 test set samples. The training process of the Inception V3 neural network model is as follows: Figure 7 As shown in the figure, after 50 rounds of training, the error is getting smaller and smaller, and it is stable at about 0.001. Test on the test set and calculate and draw the confusion matrix. The horizontal axis of the confusion matrix is ​​the true value label, the vertical axis is the predicted value label, the diagonal elements are the proportion of correct prediction classification, and the remaining elements represent the proportion of misjudgment, such as Figure 8 As shown. Figure 8 It can be seen that based on the multi-dimensional new information and residual sequence of UKF as input, the diagonal elements of the confusion matrix of the diagnosis results of the Inception V3 network are close to 1, indicating that it can almost identify all fault situations, and only a small number of samples are classified incorrectly. The three indicators of accuracy (Precision), recall (Recall), and F1 score (F1-Score) are used to quantify the diagnosis results. The closer these three indicators are to 1, the better the diagnosis effect. According to statistics, the accuracy of this method is 0.9917, the recall rate is 0.9917, and the F1 score is 0.9917. All indicators are above 99%, and the results can be diagnosed within 200ms. It can be seen that the fault diagnosis strategy through the data-assisted model is more effective.

[0148] Furthermore, Inception V3+ECA and Inception-V3 are compared with the commonly used neural network models CNN (Convolutional Neural Network), FCN (Fully Convolutional Networks), and TCN (Temporal Convolutional Networks). The same test set is used for testing and the fault diagnosis results are statistically analyzed, including accuracy, recall, F1 score, and the number of model parameters. The indicators of the test set under different models are shown in Table 1.

[0149] Table 1

[0150] Evaluation indicators Inception-V3+ECA Inception-V3 CNN FCN TCN Accuracy 0.9917 0.9869 0.9851 0.9774 0.9879 Recall 0.9917 0.9869 0.9851 0.9774 0.9879 F1 score 0.9917 0.9868 0.9850 0.9777 0.9878 Model parameter quantity 58284 58279 97607 94023 117319

[0151] As shown in Table 1, after extracting the UKF multi-dimensional information and residual sequence, the fault features are clearly characterized. All neural network models have achieved certain results. In comparison, the Inception V3 model can extract features from multiple scales and performs better in processing multiple fault types. Inception V3 maintains the sparsity of the network while ensuring high performance. The number of parameters is significantly less than other networks, achieving a balance between network performance and computing resources. In addition, the addition of the ECA attention mechanism can enhance the influence of key features, further improving the effect while almost increasing the number of parameters.

[0152] In this embodiment, the feature extraction capability is enhanced by the ECA attention mechanism. The ECA attention mechanism replaces the fully connected layer after global average pooling with a 1×1 convolution, which can avoid dimensionality reduction operations and does not need to capture the dependencies between all channels. Good results can be achieved with a small number of parameters.

[0153] In one embodiment, Fig.11 As shown, a method for diagnosing aircraft faults is provided, the method comprising the following steps:

[0154] Step 1102, obtaining measurement information and other auxiliary information of the atmospheric data system and inertial navigation system in the aircraft system, and constructing an atmospheric parameter model.

[0155] Step 1104: construct an unscented Kalman filter model of the atmospheric data system based on the atmospheric parameter model.

[0156] Step 1106, obtaining the state equation and measurement equation of the unscented Kalman filter model.

[0157] Step 1108, obtaining the state variables and error covariance matrix of one-step prediction based on the state equation and the Sigma point set.

[0158] Step 1110, estimating the state of the atmospheric data system under normal conditions according to the one-step prediction value and the measurement equation.

[0159] Step 1112, obtaining the measurement prediction value, the actual measurement value, the state estimation value and the actual state according to the unscented Kalman filter model.

[0160] Step 1114, obtaining multi-dimensional new information based on the measurement prediction value and the actual measurement value.

[0161] Step 1116, obtaining a residual sequence based on the state estimate and the actual state.

[0162] Step 1118, obtaining a parameter fault signal measured in the atmospheric parameter model.

[0163] Step 1120, input the parameter fault signal into the measurement equation, and update the fault measurement equation under the fault state.

[0164] Step 1122, input the multi-dimensional innovation and residual sequence as training data into the Inception V3 network model.

[0165] Step 1124, perform global flat pooling on the input feature map through ECA to compress the input feature map into a vector.

[0166] Step 1126, obtaining an adaptive one-dimensional convolution kernel size according to the number of channels of the input feature map.

[0167] Step 1128, use the one-dimensional convolution kernel size in the one-dimensional convolution to obtain the weight corresponding to each channel of the input feature map.

[0168] Step 1130, multiply the normalized weight of each channel and the input feature map channel by channel to obtain a weighted feature map.

[0169] Step 1132, repeatedly iteratively adjust the weights according to the training data, so that the Inception V3 network model converges or reaches a preset number of training times, and completes the construction of the aircraft fault diagnosis model.

[0170] Step 1134, calling the pre-trained aircraft fault diagnosis model, taking the multi-dimensional new information and residual sequence as input parameters, and obtaining the aircraft fault diagnosis result.

[0171] In this embodiment, an atmospheric parameter model is constructed based on the measurement information of the atmospheric data system and the inertial navigation system in the aircraft system and other auxiliary information; an unscented Kalman filter model of the atmospheric data system is further established to achieve accurate estimation of the normal state of the aircraft, and on this basis, multi-dimensional information and residual sequences in the zone-level Kalman filter model are extracted and input into the pre-trained aircraft fault diagnosis model as fault features, and finally the fault classification results are output to achieve fault diagnosis. By combining the model and data, the problem of model-based aircraft fault diagnosis methods relying on the main setting or related experience, and the problem of decreased fault diagnosis effect of data-based aircraft fault diagnosis methods under different conditions are solved, which is conducive to timely and accurate diagnosis of aircraft atmospheric data system faults and provides protection for the safe flight of aircraft.

[0172] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are shown in sequence, these steps are not necessarily executed in sequence in the order indicated. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of the steps or stages in other steps.

[0173] Based on the same inventive concept, the embodiment of the present application also provides an aircraft fault diagnosis device for implementing the above-mentioned aircraft fault diagnosis method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above-mentioned method, so the specific limitations in one or more aircraft fault diagnosis device embodiments provided below can refer to the above-mentioned limitations on the aircraft fault diagnosis method, and will not be repeated here.

[0174] In one embodiment, Fig.12 As shown, an aircraft fault diagnosis device is provided, including: an atmospheric model module 1202, an estimation module 1204, an extraction module 1206 and a diagnosis module 1208, wherein:

[0175] The atmosphere model module 1202 is used to obtain the measurement information and other auxiliary information of the atmosphere data system and the inertial navigation system in the aircraft system to build an atmosphere parameter model.

[0176] The estimation module 1204 is used to construct an unscented Kalman filter model of the atmospheric data system based on the atmospheric parameter model, and estimate the state of the atmospheric data system according to the unscented Kalman filter model.

[0177] The extraction module 1206 is used to extract the multi-dimensional innovation and residual sequence of the unscented Kalman filter model.

[0178] The diagnosis module 1208 is used to call the pre-trained aircraft fault diagnosis model, take the multi-dimensional new information and residual sequence as input parameters, and obtain the aircraft fault diagnosis result.

[0179] In one embodiment, the estimation module 1204 includes:

[0180] The equation module is used to obtain the state equation and measurement equation of the unscented Kalman filter model.

[0181] Point set module is used to obtain the state variables and error covariance matrix of one-step prediction based on the state equation and Sigma point set.

[0182] The state estimation module is used to estimate the state of the atmospheric data system under normal conditions based on the one-step prediction value and the measurement equation.

[0183] In one embodiment, the apparatus further comprises:

[0184] The fault signal module is used to obtain the parameter fault signal measured in the atmospheric parameter model.

[0185] The fault measurement module is used to input the parameter fault signal into the measurement equation and update the fault measurement equation under the fault state.

[0186] In one embodiment, the extraction module 1206 includes:

[0187] Numerical acquisition is used to obtain measurement prediction values, actual measurement values, state estimation values ​​and actual states based on the unscented Kalman filter model.

[0188] The multi-dimensional information module is used to obtain multi-dimensional information based on the measurement prediction value and the actual measurement value.

[0189] The residual sequence module is used to obtain the residual sequence based on the state estimate and the actual state.

[0190] Each module in the above-mentioned aircraft fault diagnosis device can be implemented in whole or in part by software, hardware or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each module.

[0191] Based on the same inventive concept, the embodiment of the present application also provides an aircraft fault diagnosis model training device for implementing the aircraft fault diagnosis model training method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more aircraft fault diagnosis model training device embodiments provided below can refer to the limitations of the aircraft fault diagnosis model training method above, and will not be repeated here.

[0192] In one embodiment, Fig.13 As shown, an aircraft fault diagnosis model training device is provided, comprising: an input module 1302, a weighting module 1304 and a training module 1306, wherein:

[0193] The input module 1302 is used to input the multi-dimensional innovation and residual sequence as training data into the Inception V3 network model.

[0194] The weighting module 1304 is used to analyze the weights of each channel of the input feature map in the Inception V3 network model through ECA to obtain a weighted feature map.

[0195] The training module 1306 is used to iteratively adjust the weights according to the training data so that the Inception V3 network model converges or reaches a preset number of training times to complete the construction of the aircraft fault diagnosis model.

[0196] In one embodiment, the weighted modulus 1304 block includes:

[0197] The flat pooling module is used to perform global flat pooling on the input feature map through ECA and compress the input feature map into a vector.

[0198] The convolution module is used to obtain an adaptive one-dimensional convolution kernel size according to the number of channels of the input feature map.

[0199] The channel weight module is used to use the one-dimensional convolution kernel size in the one-dimensional convolution to obtain the weight corresponding to each channel of the input feature map.

[0200] The weighted feature module is used to multiply the normalized weight of each channel and the input feature map channel by channel to obtain a weighted feature map.

[0201] Each module in the above-mentioned aircraft fault diagnosis model training device can be implemented in whole or in part by software, hardware and their combination. Each of the above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the corresponding operations of each of the above modules.

[0202] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Fig.14 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for diagnosing aircraft faults is implemented.

[0203] Those skilled in the art will understand that Fig.14 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0204] In one embodiment, the system includes a memory and a processor. The memory stores a computer program. When the processor executes the computer program, the steps in the above method embodiments are implemented.

[0205] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0206] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0207] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0208] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0209] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0210] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. An aircraft fault diagnosis method, characterized in that: The method comprises: Obtain measurement information and other auxiliary information from the atmospheric data system and inertial navigation system in the aircraft system to build an atmospheric parameter model; Constructing an unscented Kalman filter model of the atmospheric data system based on the atmospheric parameter model, and estimating the state of the atmospheric data system according to the unscented Kalman filter model; Extracting multidimensional new information and residual sequence of the unscented Kalman filter model; the multidimensional new information and the residual sequence include 6-dimensional parameters of an atmospheric data system, and the 6-dimensional parameters include airspeed, angle of attack, sideslip angle, roll angle, pitch angle and yaw angle; Calling a pre-trained aircraft fault diagnosis model, taking the multi-dimensional new information and the residual sequence as input parameters, and obtaining an aircraft fault diagnosis result; the training method of the aircraft fault diagnosis model includes: Input the multi-dimensional innovation and residual sequence into the Inception V3 network model as training data; Analyzing the weights of each channel of the input feature map in the Inception V3 network model through ECA to obtain a weighted feature map, including performing global flat pooling on the input feature map through ECA to compress the input feature map into a vector; obtaining an adaptive one-dimensional convolution kernel size according to the number of channels of the input feature map; using the one-dimensional convolution kernel size in a one-dimensional convolution to obtain a weight corresponding to each channel of the input feature map; multiplying the normalized weight of each channel by the input feature map channel by channel to obtain a weighted feature map; The weights are adjusted iteratively according to the training data to make the Inception V3 network model converge or reach a preset number of training times, thereby completing the construction of the aircraft fault diagnosis model.

2. The method according to claim 1, characterized in that: The estimating the state of the air data system according to the unscented Kalman filter model comprises: Obtaining the state equation and measurement equation of the unscented Kalman filter model; Based on the state equation and the Sigma point set, a state variable and an error covariance matrix for one-step prediction are obtained; The state of the atmospheric data system under normal conditions is estimated according to the one-step prediction and the measurement equation.

3. The method according to claim 2, characterized in that The method further comprises: obtaining parameter failure signals measured in an atmospheric parameter model; The parameter fault signal is input into the measurement equation to update the fault measurement equation under the fault state.

4. The method according to claim 1, characterized in that The extracting of the multidimensional innovation and residual sequence of the unscented Kalman filter model comprises: Obtaining a measurement prediction value, an actual measurement value, a state estimation value and an actual state according to the unscented Kalman filter model; Obtaining the multi-dimensional new information based on the measurement prediction value and the actual measurement value; The residual sequence is obtained based on the state estimation value and the actual state.

5. An aircraft fault diagnosis device, characterized in that: The device comprises: The atmospheric model module is used to obtain the measurement information and other auxiliary information of the atmospheric data system and inertial navigation system in the aircraft system and build an atmospheric parameter model; An estimation module, configured to construct an unscented Kalman filter model of the atmospheric data system based on the atmospheric parameter model, and estimate the state of the atmospheric data system according to the unscented Kalman filter model; An extraction module is used to extract multidimensional new information and residual sequence of the unscented Kalman filter model; the multidimensional new information and the residual sequence are 6-dimensional parameters including airspeed, angle of attack, sideslip angle, roll angle, pitch angle and yaw angle; The diagnosis module is used to call a pre-trained aircraft fault diagnosis model, take the multi-dimensional new information and the residual sequence as input parameters, and obtain an aircraft fault diagnosis result; the training method of the aircraft fault diagnosis model includes: Input the multi-dimensional innovation and residual sequence into the Inception V3 network model as training data; Analyzing the weights of each channel of the input feature map in the Inception V3 network model through ECA to obtain a weighted feature map, including performing global flat pooling on the input feature map through ECA to compress the input feature map into a vector; obtaining an adaptive one-dimensional convolution kernel size according to the number of channels of the input feature map; using the one-dimensional convolution kernel size in a one-dimensional convolution to obtain a weight corresponding to each channel of the input feature map; multiplying the normalized weight of each channel by the input feature map channel by channel to obtain a weighted feature map; The weights are adjusted iteratively according to the training data to make the Inception V3 network model converge or reach a preset number of training times, thereby completing the construction of the aircraft fault diagnosis model.

6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 4 are implemented.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images