Aircraft fault intelligent diagnosis method, device and equipment and readable storage medium

Abstract

The invention discloses an aircraft fault intelligent diagnosis method, device and equipment and a readable storage medium, and relates to the field of aircraft fault diagnosis, and the method comprises the steps: carrying out the spatial feature extraction of target flight state data corresponding to an aircraft based on a CNN network in a target fault diagnosis model, and obtaining a target feature vector; performing time sequence processing on the target feature vector through a long short-term memory network in the target fault diagnosis model to obtain a target time sequence feature; carrying out feature integration on the target time sequence features according to a full connection layer in the target fault diagnosis model, and outputting a fault gain value used for representing the fault severity of the aircraft execution mechanism; wherein a loss function of the target fault diagnosis model is determined based on a mean square error and an attitude residual error, and the attitude residual error is determined based on an aircraft six-degree-of-freedom model containing a fault gain value parameter. According to the invention, intelligent diagnosis of the fault of the aircraft execution mechanism can be realized, so that the diagnosis reliability, accuracy and adaptability are effectively improved, and the diagnosis difficulty is reduced.

Description

Technical Field

[0001] This application relates to the field of aircraft fault diagnosis technology, specifically to an intelligent fault diagnosis method, apparatus, device, and readable storage medium for aircraft. Background Technology

[0002] Aircraft play a vital role in modern society, both in military and civilian applications. As key components of the flight control system, the reliability of an aircraft's actuators directly impacts its attitude stability and flight performance. However, during actual flight, aircraft actuators may malfunction for various reasons, such as mechanical wear, electrical failures, and shock wave interference. A failure in an actuator can lead to loss of attitude control and ultimately, a flight accident.

[0003] Currently, traditional methods for fault diagnosis of aircraft actuators mainly include physical model-based methods, signal processing-based methods, and expert system-based methods. Physical model-based methods require establishing accurate aircraft dynamics models and actuator physical fault models (i.e., mathematical models). However, due to the complexity of aircraft flight and the diversity of actuator fault modes, accurate modeling is difficult, and the models have poor adaptability and versatility. Signal processing-based methods have high requirements for fault signal feature extraction, and for some complex fault modes, it is difficult to accurately extract effective fault features. Expert system-based methods rely on expert experience and knowledge, making knowledge acquisition difficult, and they struggle to handle newly emerging fault modes, lacking self-learning and adaptive capabilities.

[0004] It is evident that how to achieve intelligent diagnosis of aircraft actuator failures, so as to improve the reliability, accuracy and adaptability of diagnosis and reduce the difficulty of diagnosis, is an urgent problem to be solved. Summary of the Invention

[0005] This application provides a method, apparatus, device, and readable storage medium for intelligent diagnosis of aircraft faults, which can realize intelligent diagnosis of aircraft actuator faults, thereby effectively improving the reliability, accuracy, and adaptability of diagnosis, and reducing the difficulty of diagnosis.

[0006] In a first aspect, embodiments of this application provide an intelligent fault diagnosis method for aircraft, comprising the following steps: Acquire target flight status data corresponding to the aircraft; Based on the CNN network in the pre-defined target fault diagnosis model, spatial features are extracted from the target flight state data to obtain the target feature vector. The target feature vector is processed by time series analysis using the long short-term memory network in the target fault diagnosis model to obtain the target time series features; Based on the fully connected layer in the target fault diagnosis model, the target temporal features are integrated to output the fault gain value of the aircraft actuator, which is used to characterize the fault severity of the aircraft actuator. The loss function of the target fault diagnosis model is determined based on the mean square error and attitude residuals, and the attitude residuals are determined based on a preset six-degree-of-freedom aircraft model that includes fault gain parameters.

[0007] In conjunction with the first aspect, in one implementation, when the aircraft is a quadcopter, the six-degree-of-freedom model of the aircraft is as follows:

[0008]

[0009] In the formula, , and These represent the accelerations in the X, Y, and Z directions, respectively. , and These represent pitch acceleration, roll acceleration, and yaw acceleration, respectively. , and These represent pitch angle, roll angle, and yaw angle, respectively. , and These represent the pitch rate, roll rate, and yaw rate, respectively; m represents the mass of the aircraft; and g represents the acceleration due to gravity. , and This represents the three-axis inertia of the aircraft body along the body coordinate system. , , and These represent the fault gain values ​​of the four rotors in the aircraft. , , and These represent the control values ​​for the elevator, pitch, roll, and yaw channels, respectively. , , and These represent the rotational speeds of the four motors in the aircraft, and d represents the length of the aircraft's fuselage arm. Indicates the lift coefficient of the aircraft's propeller blades. This represents the moment coefficient of the aircraft propeller blades.

[0010] In conjunction with the first aspect, in one implementation, the loss function is:

[0011]

[0012] In the formula, Represents the target loss function. Indicates weight, This represents the mean square error. Let n represent the pose residual, and n represent the number of samples. and Let represent the predicted and actual values ​​of the i-th pitch rate, respectively. and Let represent the predicted and actual values ​​of the i-th roll rate, respectively. and Let represent the predicted and actual values ​​of the i-th yaw rate, respectively.

[0013] In conjunction with the first aspect, in one implementation, the CNN network includes convolutional layers, activation layers, and pooling layers. The CNN network based on the preset target fault diagnosis model extracts spatial features from the target flight state data to obtain a target feature vector, including: The first feature matrix is ​​obtained by performing one-dimensional convolution operation on the target flight state data through a convolutional layer. The second feature matrix is ​​obtained by performing a nonlinear transformation on the first feature matrix based on the activation layer; The target feature vector is obtained by performing max pooling on the second feature matrix through a pooling layer.

[0014] In conjunction with the first aspect, in one embodiment, prior to the step of acquiring target flight status data corresponding to the aircraft, the method further includes: Collect the original attitude angles corresponding to the aircraft; The original attitude angles are then subjected to denoising and normalization processes to obtain the target attitude angles. The target flight state data is calculated based on the target attitude angle, and the target flight state data includes attitude angle tracking error and attitude angle rate.

[0015] Secondly, embodiments of this application provide an intelligent diagnostic device for aircraft faults, including: a data acquisition module and a target fault diagnosis model, wherein the target fault diagnosis model includes a CNN network, a long short-term memory network and a fully connected layer; The data acquisition module is used to acquire target flight status data corresponding to the aircraft; The CNN network is used to extract spatial features from the target flight state data to obtain the target feature vector. The long short-term memory network is used to perform time series processing on the target feature vector to obtain the target time series features; The fully connected layer is used to integrate the target temporal features to output the fault gain value of the aircraft actuator, which is used to characterize the fault severity of the aircraft actuator. The loss function of the target fault diagnosis model is determined based on the mean square error and attitude residuals, and the attitude residuals are determined based on a preset six-degree-of-freedom aircraft model that includes fault gain parameters.

[0016] In conjunction with the second aspect, in one implementation, when the aircraft is a quadcopter, the six-degree-of-freedom model of the aircraft is as follows:

[0017]

[0018] In the formula, , and These represent the accelerations in the X, Y, and Z directions, respectively. , and These represent pitch acceleration, roll acceleration, and yaw acceleration, respectively. , and These represent pitch angle, roll angle, and yaw angle, respectively. , and These represent the pitch rate, roll rate, and yaw rate, respectively; m represents the mass of the aircraft; and g represents the acceleration due to gravity. , and This represents the three-axis inertia of the aircraft body along the body coordinate system. , , and These represent the fault gain values ​​of the four rotors in the aircraft. , , and These represent the control values ​​for the elevator, pitch, roll, and yaw channels, respectively. , , and These represent the rotational speeds of the four motors in the aircraft, and d represents the length of the aircraft's fuselage arm. Indicates the lift coefficient of the aircraft's propeller blades. This represents the moment coefficient of the aircraft propeller blades.

[0019] In conjunction with the second aspect, in one implementation, the loss function is:

[0020]

[0021] In the formula, Represents the target loss function. Indicates weight, This represents the mean square error. Let n represent the pose residual, and n represent the number of samples. and Let represent the predicted and actual values ​​of the i-th pitch rate, respectively. and Let represent the predicted and actual values ​​of the i-th roll rate, respectively. and Let represent the predicted and actual values ​​of the i-th yaw rate, respectively.

[0022] In conjunction with the second aspect, in one embodiment, the CNN network includes a convolutional layer, an activation layer, and a pooling layer. The convolutional layer is used to perform one-dimensional convolution operations on the target flight state data to obtain a first feature matrix; the activation layer is used to perform nonlinear transformations on the first feature matrix to obtain a second feature matrix; and the pooling layer is used to perform max pooling operations on the second feature matrix to obtain a target feature vector.

[0023] In conjunction with the second aspect, in one embodiment, the intelligent diagnostic device for aircraft faults further includes: a data preprocessing module; a data acquisition module for acquiring the original attitude angles corresponding to the aircraft; the data preprocessing module for sequentially performing noise reduction and normalization processing on the original attitude angles to obtain the target attitude angles; and calculating the target flight state data based on the target attitude angles, wherein the target flight state data includes attitude angle tracking error and attitude angle rate.

[0024] Thirdly, embodiments of this application provide an intelligent diagnostic device for aircraft faults. The intelligent diagnostic device for aircraft faults includes a processor, a memory, and an intelligent diagnostic program for aircraft faults stored in the memory and executable by the processor. When the intelligent diagnostic program for aircraft faults is executed by the processor, it implements the steps of the aforementioned intelligent diagnostic method for aircraft faults.

[0025] Fourthly, embodiments of this application provide a computer-readable storage medium storing an intelligent diagnostic program for aircraft faults, wherein when the intelligent diagnostic program for aircraft faults is executed by a processor, it implements the steps of the aforementioned intelligent diagnostic method for aircraft faults.

[0026] The beneficial effects of the technical solutions provided in this application include: This application employs a CNN (Convolutional Neural Network) in the target fault diagnosis model to extract spatial features from the target flight state data corresponding to the aircraft, obtaining a target feature vector. Then, based on the long short-term memory network in the same model, temporal features are extracted from the target feature vector to obtain target temporal features. Finally, the target temporal features are integrated using fully connected layers in the model to output a fault gain value characterizing the severity of the aircraft's actuator faults. The loss function of the target fault diagnosis model is determined based on the mean squared error and attitude residuals, while the attitude residuals are determined based on a six-degree-of-freedom model of the aircraft that includes the fault gain value parameter. Therefore, this application autonomously learns and extracts the internal features of actuator fault data through a CNN network, and utilizes the memory capacity of the long short-term memory network to more effectively utilize the actuator fault data information, effectively solving problems such as gradient vanishing and gradient exploding. Furthermore, physical equation constraints are added to the loss function, ensuring the model follows physical laws, enabling accurate fault diagnosis of complex dynamic systems without relying on expert experience or precise physical modeling. This effectively improves diagnostic reliability, accuracy, and adaptability, while reducing diagnostic difficulty. Attached Figure Description

[0027] Figure 1 This is a flowchart illustrating an embodiment of the intelligent fault diagnosis method for aircraft of this application; Figure 2 This is a schematic diagram of the structure of the target fault diagnosis model involved in the embodiments of this application; Figure 3 For this application Figure 1 A detailed flowchart of step S20; Figure 4 This is a schematic diagram of the functional modules of an embodiment of the intelligent diagnostic device for aircraft faults in this application; Figure 5 This is a schematic diagram of the hardware structure of the intelligent diagnostic device for aircraft faults involved in the embodiments of this application. Detailed Implementation

[0028] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0029] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0030] In one aspect, embodiments of this application provide an intelligent diagnostic method for aircraft faults.

[0031] In one embodiment, reference is made to Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the intelligent fault diagnosis method for aircraft according to this application. Figure 1 As shown, the intelligent fault diagnosis method for aircraft includes: Step S10: Obtain the target flight status data corresponding to the aircraft.

[0032] Exemplary and understandable, aircraft include, but are not limited to, fixed-wing aircraft and rotorcraft, with rotorcraft further including single-rotor aircraft and multi-rotor aircraft (such as tri-rotor and quadcopter aircraft). The actuating structure of the aircraft includes, but is not limited to, the aircraft's rotor. Target flight state data includes, but is not limited to, the aircraft's attitude angle tracking error and attitude angular rate. Attitude angle tracking error refers to the difference between the actual attitude angle and the desired attitude angle, while attitude angular rate refers to the rate of change of the attitude angle over time. The attitude angle includes pitch angle. Roll angle and yaw angle In other words, attitude angle tracking error includes pitch angle tracking error, roll angle tracking error, and yaw angle tracking error. Similarly, attitude angle rate includes pitch angle rate, roll angle rate, and yaw angle rate.

[0033] Furthermore, in one embodiment, before the step of acquiring the target flight status data corresponding to the aircraft, the method further includes: Collect the original attitude angles corresponding to the aircraft; The original attitude angles are then subjected to denoising and normalization processes to obtain the target attitude angles. The target flight state data is calculated based on the target attitude angle, and the target flight state data includes attitude angle tracking error and attitude angle rate.

[0034] Exemplary and understandable, the raw attitude angle refers to the actual flight attitude angle of the aircraft during flight, which can be measured by aircraft sensors (such as inertial measurement units, gyroscopes, accelerometers, etc.). In this embodiment, the collected raw attitude angle data is first preprocessed by denoising and normalization to obtain the target attitude angle. Specifically, filtering algorithms (such as Kalman filtering) are used to denoise the raw attitude angle to remove noise interference and improve data quality; and Min-Max normalization is used. Scaled to the range [0,1] to eliminate dimensional differences between different features and avoid any single feature in the data having an excessive impact on the fault diagnosis model; among which, This represents the normalized attitude angle (i.e., the target attitude angle). This represents the attitude angle after denoising. This represents the minimum attitude angle after noise reduction. This represents the maximum attitude angle after denoising. Then, the attitude angle tracking error is calculated based on the target attitude angle and the desired attitude angle. At the same time, the target attitude angle is differentiated to obtain the attitude angular rate, so as to obtain the target flight state data including the attitude angle tracking error and the attitude angular rate.

[0035] Step S20: Based on the CNN network in the preset target fault diagnosis model, spatial features are extracted from the target flight state data to obtain the target feature vector.

[0036] Exemplary, see Figure 2 As shown, this embodiment constructs a fault diagnosis model consisting of an output layer, a CNN network, an LSTM (Long Short-Term Memory) network model, a fully connected layer, and an output layer. The number of neurons in the input layer can be determined based on the dimension of the input data, while the number of neurons in the output layer is determined by the number of actuators in the aircraft; that is, the output is the fault gain corresponding to each actuator. The CNN network is used to extract spatially relevant information from the flight data, the LSTM network model is used to extract temporally relevant information from the flight data, and the fully connected layer is used to convert the hidden states output by the LSTM network model into the final fault gain value. It is understood that after constructing the fault diagnosis model, it will be trained and tested until the requirements are met, thus obtaining the final target fault diagnosis model.

[0037] The following examples will explain the construction process of the target fault diagnosis model.

[0038] First, the aircraft is controlled to simulate flight along a predetermined trajectory. During this process, flight data (i.e., actual attitude angles) is collected in real time based on the aircraft's sensors. Then, the collected flight data is denoised and normalized, and the attitude angle tracking error is calculated. and attitude angular rate Based on this, the attitude angle tracking error of each simulated flight is... and attitude angular rate As input, and the corresponding fault gain (the fault gain value is between 0 and 1, where 0 represents complete actuator failure and 1 represents no actuator failure) is used as output to construct a data sample. Then, all data samples are divided in an 8:2 ratio to obtain the training dataset and the test dataset. At this point, the dataset collection for the fault diagnosis model is complete. It should be noted that during the dataset collection and construction process, the flight time and simulation step size for each flight simulation can be set, for example, setting the flight simulation time to 100s and the step size to 0.05s. A fault gain value is introduced every second during the simulation. However, considering the needs of actual flight, only the case of a single actuator failure can be considered, i.e., only the case of a single actuator's reduced performance is considered.

[0039] Taking a quadcopter as an example, the input to the quadcopter fault diagnosis model is a 6-dimensional flight state vector consisting of pitch tracking error, roll tracking error, yaw tracking error, pitch rate, roll rate, and yaw rate, flying along a predetermined trajectory. The output is a 4-dimensional fault gain vector containing the fault gain values ​​of the four rotors, and there is a one-to-one correspondence between the input flight state vector and the output fault gain vector. The flight time for each flight simulation of the controllable aircraft is 20 seconds, and the simulation step size is set to 0.05. 100 data samples are sampled at equal intervals each time, thus obtaining 300 simulation data points per flight simulation using sliding window sampling. To ensure the dataset includes all scenarios, a fault gain value is introduced every second. However, considering actual flight needs, a fault is only introduced in one rotor, i.e., considering the case where only a single rotor's control effect is reduced. Understandably, in order to collect the aircraft's attitude data under all fault conditions, when only one of the four rotors experiences a fault gain at a certain moment, the aircraft needs to undergo 1000 flight simulations. The flight state vector and fault gain vector from each simulation are used to construct a training sample. At this point, the dataset for the fault diagnosis model is complete, containing 30,000 data points in total, including training and testing sets. Each data point is a 100×10 dimension data, where 100 represents the time-series dimension and 10 represents the dimensions of the flight state vector and fault gain vector.

[0040] Next, an initial fault diagnosis model is constructed and trained and tested using training and testing datasets. The input to the training samples consists of attitude angle tracking error and attitude angular rate, which are simulation data. The output of the training samples is the corresponding fault gain of the actuator. Training continues until the model converges, generating the target fault diagnosis model. During training, backpropagation and Adam optimization algorithms can be used. It should be noted that the specific training and testing methods, processes, and principles are common knowledge in the field and will not be elaborated upon here for the sake of brevity. As can be seen, this embodiment trains the fault diagnosis model with a large amount of training data, enabling the model to automatically learn the fault gain of the actuator under different flight conditions, exhibiting good adaptability and the ability to adapt to complex situations and changes during aircraft flight.

[0041] Based on this, see Figure 2 As shown, after obtaining the target flight status data, the target flight status data is directly input into the CNN network so that the CNN network can extract spatial features from the target flight status data to extract high-order features related to fault diagnosis, that is, to obtain the target feature vector.

[0042] Furthermore, the CNN network includes convolutional layers, activation layers, and pooling layers, see [link to documentation]. Figure 3 As shown, the CNN network in the preset target fault diagnosis model extracts spatial features from the target flight state data to obtain a target feature vector, including: Step S201: Perform one-dimensional convolution operation on the target flight state data through a convolutional layer to obtain the first feature matrix; Step S202: Perform a nonlinear transformation on the first feature matrix based on the activation layer to obtain the second feature matrix; Step S203: Perform max pooling on the second feature matrix through a pooling layer to obtain the target feature vector.

[0043] Exemplary, see Figure 2As shown, the CNN network includes convolutional layers, activation layers, and pooling layers. In this embodiment, the convolutional layers are used to process temporal data, the activation layers are used to introduce nonlinearity to enhance the network's expressive power, and the pooling layers are used to reduce feature dimensionality while retaining important features. Specifically, the convolutional layers perform one-dimensional convolution operations on the target flight state data to obtain a first feature matrix; the activation layer performs a nonlinear transformation on the first feature matrix using an activation function (such as the ReLU function) to obtain a second feature matrix; the pooling layer performs max pooling on the second feature matrix to obtain the target feature vector, that is, to extract the features of the aircraft's attitude angle tracking error and attitude angular rate. It can be seen that the CNN network in this embodiment relies on powerful feature extraction capabilities and, through operations such as convolution and pooling, can autonomously learn and extract the internal features of actuator fault data, thereby achieving accurate fault diagnosis.

[0044] Step S30: Perform time series processing on the target feature vector through the long short-term memory network in the target fault diagnosis model to obtain the target time series features.

[0045] As an example, in this embodiment, a two-layer LSTM network model will be introduced to perform time series analysis on the target feature vector to obtain the target temporal features; the following will combine... Figure 2 The working principle of the LSTM network model is explained as follows: First, the forget gate in the LSTM network model uses the feature values ​​of the current spacecraft state variables. and the hidden state of the previous moment Generate a value between 0 and 1 (i.e., the output of the forget gate), which represents the cell state at the previous time step. The proportion of information to be retained, where 1 represents complete retention and 0 represents complete discarding, is obtained from the previous time step using the sigmoid function. Whether it can be passed or the proportion of passing, The calculation formula is as follows:

[0046] In the formula, This represents the sigmoid function, which is a non-linear function that compresses the input to a range between 0 and 1; This represents the forget gate weight matrix, which is used to store the hidden state from the previous time step. and the current moment The concatenated vector is mapped to the output space of the forget gate to learn the non-linear relationship between input features and retention ratio; This represents the forget gate bias term, which is used to provide an offset for forget gate calculation.

[0047] Then update the information within the node: the input at the current moment. and the hidden state of the previous moment A candidate cell state is generated by a tanh function. It should be understood that the candidate unit state generated by the current function It may be used to update the cell state at the current time step. Simultaneously, the input gate adjusts according to the input at the current moment. and the hidden state of the previous moment Generate a value between 0 and 1 (i.e., input gate output), which represents the proportion of the input information that needs to be added to the cell state at the current moment, determined by the sigmoid function. It is possible; among which, and The calculation formulas are as follows:

[0048] In the formula, This represents the input gate weight matrix, which is used to map the concatenated input to the input gate output space to learn the nonlinear relationship between input features and the proportion of new information written. represents the input gate bias term, which is used to provide an offset for input gate computation; tanh represents the tanh function (i.e., the hyperbolic tangent activation function), which is used to map input values ​​to the interval [-1, 1]. Represents the candidate state weight matrix; This represents the candidate state bias term.

[0049] It should be understood that, in relation to the unit state When updating, the values ​​generated from the two parts above need to be combined first, and then... Value multiplied by By forgetting the unnecessary information in the cell state from the previous moment, we can obtain the updated cell state value. :

[0050] In the formula, This indicates element-wise multiplication.

[0051] Finally, the information in the nodes is output to obtain the latest hidden state. First, based on the input at the current moment... and the hidden state of the previous moment Generate a value between 0 and 1 (i.e., output gate output), which represents the proportion of information that needs to be output from the current cell state. In other words, the sigmoid function determines how much information can be output from this node. After passing through a tanh layer, the two parts are multiplied to obtain the node's output, as shown in the following formula:

[0052] In the formula, represents the output gate weight matrix, which is used to map the concatenated input to the output gate output space to learn the nonlinear relationship between input features and the proportion of hidden state generation; This represents the output gate bias term, which is used to provide the offset for output gate calculation.

[0053] Step S40: Integrate the target temporal features according to the fully connected layer in the target fault diagnosis model to output the fault gain value of the aircraft actuator. The fault gain value is used to characterize the fault severity of the aircraft actuator. The loss function of the target fault diagnosis model is determined based on the mean square error and attitude residual. The attitude residual is determined based on a preset six-degree-of-freedom aircraft model that includes fault gain value parameters.

[0054] Exemplary, see Figure 2 As shown, this embodiment uses a fully connected layer (FC) to integrate the target temporal features and map out N×M dimensional data to locate the faulty actuator, thereby obtaining the fault gain value of each actuator of the aircraft. And the time of the failure; it should be noted that N represents the dimension corresponding to the flight time sequence and For example, if the aircraft's flight time is 100 seconds, the update step size is 0.05, and the sliding window length is 100, then... M represents the number of actuators in the aircraft. For example, for a quadcopter, M=4, then the output fault gain value is... include to ,and to These represent the fault gain values ​​for each of the four rotors. The fault gain value is used to... The severity of the aircraft actuator failure (i.e. the degree of performance reduction) can be known, and the failure gain value is between 0 and 1, where 0 represents complete failure of the actuator and 1 represents no failure of the actuator. For actuators with low failure gain values, this embodiment can provide early warning so that staff can quickly take corresponding solutions.

[0055] It is worth noting that, in order to ensure that the predictions of the fault diagnosis model do not violate the laws of aircraft dynamics, this embodiment will integrate a Physical Information Neural Network (PINN) into the target fault diagnosis model, that is, construct a six-degree-of-freedom model of the aircraft (i.e., the physical model of the aircraft), and incorporate the fault gain value as a parameter into the control quantity of the six-degree-of-freedom model of the aircraft, so that the attitude residual constructed based on the attitude angular rate predicted by the six-degree-of-freedom model of the aircraft can truly reflect the degree of fault. In summary, the LSTM network model can effectively handle long-term dependencies in time series data, and the integration of the Physical Information Neural Network makes the model follow physical laws, which has unique advantages in fault diagnosis. Based on this, the high-precision and high-reliability aircraft fault diagnosis method based on a hybrid network of CNN, LSTM, and PINN proposed in this embodiment has important practical significance and research value.

[0056] It is understandable that the six-degree-of-freedom (6DOF) models corresponding to different types of aircraft may differ slightly, but the overall architecture is similar. Only minor adjustments to the control variables in the 6DOF model are needed to generate the final aircraft 6DOF model; that is, adding the fault gain parameter to the control variables of the 6DOF model. This embodiment uses a quadcopter as an example, and its corresponding 6DOF model is as follows:

[0057]

[0058] In the formula, , and These represent the accelerations in the X, Y, and Z directions, respectively. , and These represent pitch acceleration, roll acceleration, and yaw acceleration, respectively. , and These represent pitch angle, roll angle, and yaw angle, respectively. , and These represent the pitch rate, roll rate, and yaw rate, respectively; m represents the mass of the aircraft; and g represents the acceleration due to gravity. , and This represents the three-axis inertia of the aircraft body along the body coordinate system. , , and These represent the fault gain values ​​of the four rotors in the aircraft. , , and These represent the control values ​​for the elevator, pitch, roll, and yaw channels, respectively. , , and These represent the rotational speeds of the four motors in the aircraft, and d represents the length of the aircraft's fuselage arm. Indicates the lift coefficient of the aircraft's propeller blades. This represents the moment coefficient of the aircraft propeller blades.

[0059] Based on this, the predicted attitude angular rate under the current fault factor is calculated using historical information and rotor control information, and based on the aircraft's six-degree-of-freedom model. Then, attitude residuals are generated based on the attitude angular rate, and these residuals are added as constraints to the physical information equation in the loss function of the target fault diagnosis model. This ensures that the loss function includes not only the mean squared error but also the physical information equation constraints, thereby effectively improving the accuracy, interpretability, and robustness of fault diagnosis. The loss function of the target fault diagnosis model is as follows:

[0060]

[0061]

[0062] In the formula, Represents the target loss function. This represents the weights, and their specific values ​​can be determined through model training. This represents the mean square error. Let n represent the pose residual, and n represent the number of samples. and Let represent the predicted and actual values ​​of the i-th pitch rate, respectively. and Let represent the predicted and actual values ​​of the i-th roll rate, respectively. and Let represent the predicted and actual values ​​of the i-th yaw rate, respectively; This represents the fault gain value output by the target fault diagnosis model. This represents the actual fault gain value.

[0063] As can be seen, the CNN network in this embodiment can autonomously learn and extract the internal features of actuator fault data through operations such as convolution and pooling, thereby achieving fault diagnosis. The LSTM network model, on the other hand, obtains the state of the next time step by comprehensively considering past and current time steps, possessing strong memory capabilities. It can more effectively utilize the fault data information of the actuator and effectively solve problems such as gradient vanishing and gradient exploding. Furthermore, physical equation constraints are added to the loss function, ensuring the model follows physical laws. Moreover, the LSTM network model has significant advantages in processing time series data. The introduction of the "gate" concept allows some errors to pass directly through the "gate" during propagation. This structure prevents gradient vanishing regardless of the number of network layers. Simultaneously, by integrating physical laws with time series modeling capabilities, it can accurately predict complex dynamic systems. Therefore, by adding the LSTM network model and physical information equation constraints to the powerful feature extraction capabilities of the CNN network, the fault diagnosis effect can be significantly improved.

[0064] In summary, this embodiment proposes an aircraft fault diagnosis method based on a Physical Information Long Short-Term Memory (PIS) network. First, a CNN network is used to extract spatially relevant information from flight data. Then, an LSTM network model is constructed, and physical laws are combined with the LSTM and CNN networks. This ensures that the model follows physical laws when extracting time-series features, improving the accuracy and real-time performance of fault diagnosis, ensuring flight safety, and solving the problems of poor diagnostic reliability, insufficient adaptability, and lack of versatility in existing aircraft fault diagnosis methods. It is worth noting that this embodiment is not dependent on specific aircraft models or actuator structures. Only appropriate adjustments to the model are needed based on the sensor data characteristics and the number of actuators of different aircraft, demonstrating strong versatility. Furthermore, this embodiment can collect flight status data in real time during flight and analyze and reason about the time-series information of the flight status data through a target fault diagnosis model, achieving rapid and accurate diagnosis of aircraft actuators, timely detection of actuator faults, improving flight safety and reliability, and providing real-time assurance for safe flight.

[0065] Secondly, embodiments of this application also provide an intelligent diagnostic device for aircraft faults.

[0066] In one embodiment, reference is made to Figure 4 , Figure 4 This is a functional module diagram of an embodiment of the intelligent diagnostic device for aircraft faults according to this application. Figure 4 As shown, the intelligent diagnostic device for aircraft faults includes: a data acquisition module and a target fault diagnosis model, wherein the target fault diagnosis model includes a CNN network, a long short-term memory network and a fully connected layer; The data acquisition module is used to acquire target flight status data corresponding to the aircraft; The CNN network is used to extract spatial features from the target flight state data to obtain the target feature vector. The long short-term memory network is used to perform time series processing on the target feature vector to obtain the target time series features; The fully connected layer is used to integrate the target temporal features to output the fault gain value of the aircraft actuator, which is used to characterize the fault severity of the aircraft actuator. The loss function of the target fault diagnosis model is determined based on the mean square error and attitude residuals, and the attitude residuals are determined based on a preset six-degree-of-freedom aircraft model that includes fault gain parameters.

[0067] Furthermore, in one embodiment, when the aircraft is a quadcopter, the six-degree-of-freedom model of the aircraft is as follows:

[0068]

[0069] In the formula, , and These represent the accelerations in the X, Y, and Z directions, respectively. , and These represent pitch acceleration, roll acceleration, and yaw acceleration, respectively. , and These represent pitch angle, roll angle, and yaw angle, respectively. , and These represent the pitch rate, roll rate, and yaw rate, respectively; m represents the mass of the aircraft; and g represents the acceleration due to gravity. , and This represents the three-axis inertia of the aircraft body along the body coordinate system. , , and These represent the fault gain values ​​of the four rotors in the aircraft. , , and These represent the control values ​​for the elevator, pitch, roll, and yaw channels, respectively. , , and These represent the rotational speeds of the four motors in the aircraft, and d represents the length of the aircraft's fuselage arm. Indicates the lift coefficient of the aircraft's propeller blades. This represents the moment coefficient of the aircraft propeller blades.

[0070] Furthermore, in one embodiment, the loss function is:

[0071]

[0072] In the formula, Represents the target loss function. Indicates weight, This represents the mean square error. Let n represent the pose residual, and n represent the number of samples. and Let represent the predicted and actual values ​​of the i-th pitch rate, respectively. and Let represent the predicted and actual values ​​of the i-th roll rate, respectively. and Let represent the predicted and actual values ​​of the i-th yaw rate, respectively.

[0073] Furthermore, in one embodiment, the CNN network includes a convolutional layer, an activation layer, and a pooling layer. The convolutional layer is used to perform one-dimensional convolution operations on the target flight state data to obtain a first feature matrix; the activation layer is used to perform nonlinear transformations on the first feature matrix to obtain a second feature matrix; and the pooling layer is used to perform max pooling operations on the second feature matrix to obtain a target feature vector.

[0074] Furthermore, in one embodiment, the intelligent diagnostic device for aircraft faults further includes: a data preprocessing module; a data acquisition module for acquiring the original attitude angles corresponding to the aircraft; the data preprocessing module for sequentially performing noise reduction and normalization processing on the original attitude angles to obtain the target attitude angles; and calculating the target flight state data based on the target attitude angles, wherein the target flight state data includes attitude angle tracking error and attitude angle rate.

[0075] The functions of each part of the above-mentioned intelligent aircraft fault diagnosis device correspond to the steps in the above-mentioned intelligent aircraft fault diagnosis method embodiment, and their functions and implementation processes will not be described in detail here.

[0076] Thirdly, embodiments of this application provide an intelligent diagnostic device for aircraft faults. The intelligent diagnostic device for aircraft faults can be a personal computer (PC), a laptop computer, a server, or other device with data processing capabilities.

[0077] Reference Figure 5 , Figure 5 This is a schematic diagram of the hardware structure of the intelligent aircraft fault diagnosis device involved in the embodiments of this application. In this embodiment, the intelligent aircraft fault diagnosis device may include a processor, a memory, a communication interface, and a communication bus.

[0078] The communication bus can be of any type and is used to interconnect the processor, memory, and communication interface.

[0079] The communication interface includes input / output (I / O) interfaces, physical interfaces, and logical interfaces used for interconnecting internal components of the intelligent aircraft fault diagnosis equipment, as well as interfaces used for interconnecting the intelligent aircraft fault diagnosis equipment with other devices (such as other computing devices or user equipment). Physical interfaces can be Ethernet interfaces, fiber optic interfaces, ATM interfaces, etc.; user equipment can be displays, keyboards, etc.

[0080] Memory can be various types of storage media, such as random access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), flash memory, optical storage, hard disk, programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), etc.

[0081] The processor can be a general-purpose processor, which can call the intelligent aircraft fault diagnosis program stored in the memory and execute the intelligent aircraft fault diagnosis method provided in the embodiments of this application. For example, the general-purpose processor can be a central processing unit (CPU). The method executed when the intelligent aircraft fault diagnosis program is called can be referred to in the various embodiments of the intelligent aircraft fault diagnosis method of this application, and will not be repeated here.

[0082] Those skilled in the art will understand that Figure 5 The hardware structure shown does not constitute a limitation of this application and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0083] Fourthly, embodiments of this application also provide a computer-readable storage medium.

[0084] The present application has a readable storage medium storing an intelligent diagnostic program for aircraft faults, wherein when the intelligent diagnostic program for aircraft faults is executed by a processor, it implements the steps of the intelligent diagnostic method for aircraft faults as described above.

[0085] The method implemented when the intelligent diagnostic program for aircraft faults is executed can be referred to in various embodiments of the intelligent diagnostic method for aircraft faults in this application, and will not be repeated here.

[0086] It should be noted that the sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0087] The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus. The terms "first," "second," and "third," etc., are used to distinguish different objects, etc., and do not indicate a sequence, nor do they limit "first," "second," and "third" to different types.

[0088] In the description of the embodiments of this application, terms such as "exemplary," "for example," or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "exemplary," "for example," or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary," "for example," or "for instance" is intended to present the relevant concepts in a concrete manner.

[0089] In the description of the embodiments of this application, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. The "and / or" in the text is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist simultaneously, and B exists alone. In addition, in the description of the embodiments of this application, "multiple" means two or more.

[0090] In some processes described in the embodiments of this application, multiple operations or steps are included in a specific order. However, it should be understood that these operations or steps may not be executed in the order they appear in the embodiments of this application, or they may be executed in parallel. The sequence number of the operation is only used to distinguish the different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed sequentially or in parallel, and these operations or steps may be combined.

[0091] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device to execute the methods described in the various embodiments of this application.

[0092] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for intelligent diagnosis of aircraft faults, characterized in that, Includes the following steps: Acquire target flight status data corresponding to the aircraft; Based on the CNN network in the pre-defined target fault diagnosis model, spatial features are extracted from the target flight state data to obtain the target feature vector. The target feature vector is processed by time series analysis using the long short-term memory network in the target fault diagnosis model to obtain the target time series features. Based on the fully connected layer in the target fault diagnosis model, the target temporal features are integrated to output the fault gain value of the aircraft actuator, which is used to characterize the fault severity of the aircraft actuator. The loss function of the target fault diagnosis model is determined based on the mean square error and attitude residuals, and the attitude residuals are determined based on a preset six-degree-of-freedom aircraft model that includes fault gain parameters.

2. The intelligent fault diagnosis method for aircraft as described in claim 1, characterized in that, When the aircraft is a quadcopter, the six-degree-of-freedom model of the aircraft is as follows: In the formula, , and These represent the accelerations in the X, Y, and Z directions, respectively. , and These represent pitch acceleration, roll acceleration, and yaw acceleration, respectively. , and These represent pitch angle, roll angle, and yaw angle, respectively. , and These represent the pitch rate, roll rate, and yaw rate, respectively; m represents the mass of the aircraft; and g represents the acceleration due to gravity. , and This represents the three-axis inertia of the aircraft body along the body coordinate system. , , and These represent the fault gain values ​​of the four rotors in the aircraft. , , and These represent the control values ​​for the elevator, pitch, roll, and yaw channels, respectively. , , and These represent the rotational speeds of the four motors in the aircraft, and d represents the length of the aircraft's fuselage arm. Indicates the lift coefficient of the aircraft's propeller blades. This represents the moment coefficient of the aircraft propeller blades.

3. The intelligent fault diagnosis method for aircraft as described in claim 1, characterized in that, The loss function is: In the formula, Represents the target loss function. Indicates weight, This represents the mean square error. Let n represent the pose residual, and n represent the number of samples. and Let represent the predicted and actual values ​​of the i-th pitch rate, respectively. and Let represent the predicted and actual values ​​of the i-th roll rate, respectively. and Let represent the predicted and actual values ​​of the i-th yaw rate, respectively.

4. The intelligent fault diagnosis method for aircraft as described in claim 1, characterized in that, The CNN network includes convolutional layers, activation layers, and pooling layers. The CNN network based on the preset target fault diagnosis model extracts spatial features from the target flight state data to obtain a target feature vector, including: The first feature matrix is ​​obtained by performing one-dimensional convolution operation on the target flight state data through a convolutional layer. The second feature matrix is ​​obtained by performing a nonlinear transformation on the first feature matrix based on the activation layer; The target feature vector is obtained by performing max pooling on the second feature matrix through a pooling layer.

5. The intelligent fault diagnosis method for aircraft as described in claim 1, characterized in that, Before the step of acquiring the target flight status data corresponding to the aircraft, the method further includes: Collect the original attitude angles corresponding to the aircraft; The original attitude angles are then subjected to denoising and normalization processes to obtain the target attitude angles. The target flight state data is calculated based on the target attitude angle, and the target flight state data includes attitude angle tracking error and attitude angle rate.

6. An intelligent diagnostic device for aircraft faults, characterized in that, include: The data acquisition module and the target fault diagnosis model include a CNN network, a long short-term memory network, and a fully connected layer. The data acquisition module is used to acquire target flight status data corresponding to the aircraft; The CNN network is used to extract spatial features from the target flight state data to obtain the target feature vector. The long short-term memory network is used to perform time series processing on the target feature vector to obtain the target time series features; The fully connected layer is used to integrate the target temporal features to output the fault gain value of the aircraft actuator, which is used to characterize the fault severity of the aircraft actuator. The loss function of the target fault diagnosis model is determined based on the mean square error and attitude residuals, and the attitude residuals are determined based on a preset six-degree-of-freedom aircraft model that includes fault gain parameters.

7. The intelligent diagnostic device for aircraft faults as described in claim 6, characterized in that, When the aircraft is a quadcopter, the six-degree-of-freedom model of the aircraft is as follows: In the formula, , and These represent the accelerations in the X, Y, and Z directions, respectively. , and These represent pitch acceleration, roll acceleration, and yaw acceleration, respectively. , and These represent pitch angle, roll angle, and yaw angle, respectively. , and These represent the pitch rate, roll rate, and yaw rate, respectively; m represents the mass of the aircraft; and g represents the acceleration due to gravity. , and This represents the three-axis inertia of the aircraft body along the body coordinate system. , , and These represent the fault gain values ​​of the four rotors in the aircraft. , , and These represent the control values ​​for the elevator, pitch, roll, and yaw channels, respectively. , , and These represent the rotational speeds of the four motors in the aircraft, and d represents the length of the aircraft's fuselage arm. Indicates the lift coefficient of the aircraft's propeller blades. This represents the moment coefficient of the aircraft propeller blades.

8. The intelligent diagnostic device for aircraft faults as described in claim 6, characterized in that, The loss function is: In the formula, Represents the target loss function. Indicates weight, This represents the mean square error. Let n represent the pose residual, and n represent the number of samples. and Let represent the predicted and actual values ​​of the i-th pitch rate, respectively. and Let represent the predicted and actual values ​​of the i-th roll rate, respectively. and Let represent the predicted and actual values ​​of the i-th yaw rate, respectively.

9. An intelligent diagnostic device for aircraft faults, characterized in that, The intelligent diagnostic device for aircraft faults includes a processor, a memory, and an intelligent diagnostic program for aircraft faults stored in the memory and executable by the processor, wherein when the intelligent diagnostic program for aircraft faults is executed by the processor, it implements the steps of the intelligent diagnostic method for aircraft faults as described in any one of claims 1 to 5.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores an intelligent diagnostic program for aircraft faults, wherein when the intelligent diagnostic program for aircraft faults is executed by a processor, it implements the steps of the intelligent diagnostic method for aircraft faults as described in any one of claims 1 to 5.

Images