Fault diagnosis method for aircraft executing mechanism

Abstract

The invention discloses a fault diagnosis method for an aircraft executing mechanism. The fault diagnosis method comprises the following steps: collecting a motor signal of the aircraft executing mechanism; sequentially carrying out wavelet decomposition, threshold processing, wavelet reconstruction and denoising preprocessing to obtain denoised motor current one-dimensional data; constructing a fault diagnosis model, wherein the fault diagnosis model comprises a convolutional neural network CNN layer, a long short-term memory network LSTM layer, a channel attention mechanism and a full connection layer; inputting the motor current one-dimensional data into a convolutional neural network (CNN) layer and a long short-term memory (LSTM) layer, applying a channel attention mechanism on the LSTM layer, and performing weight scoring on the output of the LSTM layer to obtain a trained fault diagnosis model; and carrying out aircraft fault detection on the test set by using the trained fault diagnosis model to obtain a fault diagnosis result. According to the method, online learning can be realized, new and unknown fault modes can be immediately adapted, and dependence on historical fault data is reduced.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of motor platform fault diagnosis, and particularly relates to a fault diagnosis method for an aircraft actuator. BACKGROUND

[0002] In the field of aerospace, the reliability of aircraft actuators is crucial for flight safety and mission success. These actuators, including but not limited to motors and rudders, are key components for precise control of aircraft. Traditional aircraft actuator fault diagnosis methods mainly rely on regular maintenance checks and manual experience judgments. In recent years, machine learning-based fault diagnosis techniques have also received widespread attention. These techniques analyze historical data to identify fault patterns, enabling fault prediction and diagnosis.

[0003] However, the existing technology has many shortcomings. On the one hand, traditional fault diagnosis methods have slow response speed and cannot achieve real-time monitoring and rapid response, and the diagnosis results are largely dependent on the experience and skills of maintenance personnel, which limits the accuracy and consistency of diagnosis. On the other hand, existing machine learning models usually require a large amount of labeled data for training, and the generalization ability and adaptability of the model are limited when facing new or unknown fault types. In addition, these models often require a long time to train and adjust in actual application, making it difficult to meet the demand for immediate diagnosis. With the increasing complexity of aircraft systems, traditional fault diagnosis methods are increasingly difficult to adapt to changing fault patterns and complex system structures, making it difficult to meet the requirements of modern aircraft for high reliability and high safety. SUMMARY

[0004] To solve the above technical problems, the present application provides a fault diagnosis method for an aircraft actuator to solve the problems existing in the prior art.

[0005] To achieve the above purpose, in a first aspect, the present application provides a fault diagnosis method for an aircraft actuator, comprising:

[0006] Collecting motor signals of the aircraft actuator, the motor signals including real-time current information and real-time voltage information, the motor signals containing normal data, short-circuit data, overload data, open-circuit data, and shaft current data, and being divided into a training set and a test set;

[0007] Performing wavelet decomposition, threshold processing, wavelet reconstruction, and denoising preprocessing on the training set in sequence to obtain denoised motor current one-dimensional data;

[0008] Building a fault diagnosis model, the fault diagnosis model including a convolutional neural network (CNN) layer, a long short-term memory (LSTM) layer, a channel attention mechanism, and a fully connected layer;

[0009] The motor current one-dimensional data is input to a convolutional neural network (CNN) layer for feature extraction, and key feature data of the motor signal is extracted through convolution and pooling operations;

[0010] The key feature data is fed to a long short-term memory (LSTM) layer to process and memorize long-term dependencies of the input sequence, and a channel attention mechanism is applied to the LSTM layer to score the output of the LSTM layer to focus on and process important parts of the data;

[0011] The fault detection result is output through a fully connected layer, and an Adadelta optimizer is used to adjust the weights and parameters of the fault diagnosis model to obtain a trained fault diagnosis model;

[0012] The trained fault diagnosis model is used to detect aircraft faults in the test set to obtain a fault diagnosis result.

[0013] Preferably, the steps of wavelet decomposition, threshold processing, wavelet reconstruction, and denoising preprocessing of the collected motor current signal in sequence include:

[0014] A wavelet basis function is selected, and the decomposition layers are defined, and the motor signal is decomposed into multiple layers to obtain wavelet decomposition coefficients of each layer;

[0015] A preset threshold is set, and the threshold processing is performed on the wavelet decomposition coefficients of each layer;

[0016] The wavelet coefficients of each layer after threshold processing are recombined through inverse wavelet transform to obtain a denoised motor signal.

[0017] Preferably, in the threshold processing process, an adjustable threshold function is used for processing, when the threshold value is close to 0, the adjustment factor of the improved threshold function tends to be a soft threshold algorithm, and when the threshold value is close to 1, the adjustment factor of the improved threshold function tends to be a hard threshold method.

[0018] Preferably, the CNN layer extracts key features of the motor signal through convolution and pooling operations, the convolution operation uses multiple convolution kernels of different sizes, and the pooling operation uses maximum pooling or average pooling.

[0019] Preferably, the number of units of the LSTM layer is adjusted according to the dimension and sequence length of the input feature data.

[0020] Preferably, the Adadelta optimizer adjusts the learning rate adaptively to minimize the cross-entropy loss function.

[0021] Preferably, the output layer of the fully connected layer adopts a Softmax activation function, which maps the output of the fault diagnosis model to a category probability distribution of fault diagnosis.

[0022] In a second aspect, the present application also discloses a computer device comprising a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method of the first aspect.

[0023] In a third aspect, the present application also discloses a computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the steps of the method of the first aspect.

[0024] In a fourth aspect, the present application also discloses a computer program product comprising a computer program, wherein the computer program is executed by a processor to implement the steps of the method of the first aspect.

[0025] Compared with the prior art, the present application has the following advantages and technical effects:

[0026] The present application provides a fault diagnosis method for aircraft actuators, comprising: first, collecting motor signals of the aircraft actuators, the motor signals including real-time current information and real-time voltage information, the motor signals containing normal data, short-circuit data, overload data, open-circuit data and shaft current data, and being divided into a training set and a test set; second, sequentially performing wavelet decomposition, threshold processing, wavelet reconstruction and denoising preprocessing on the training set to obtain denoised motor current one-dimensional data; third, constructing a fault diagnosis model, the fault diagnosis model comprising a convolutional neural network (CNN) layer, a long short-term memory (LSTM) layer, a channel attention mechanism and a fully connected layer; fourth, inputting the motor current one-dimensional data into the CNN layer for feature extraction, and extracting key feature data of the motor signals through convolution and pooling operations; fifth, feeding the key feature data to the LSTM layer to process and remember long-term dependencies of the input sequence, applying the channel attention mechanism on the LSTM layer to score the output of the LSTM layer, so as to focus on and process important parts of the data; sixth, outputting fault detection results through the fully connected layer, using an Adadelta optimizer to adjust weights and parameters of the fault diagnosis model to obtain a trained fault diagnosis model; and seventh, using the trained fault diagnosis model to detect aircraft faults of the test set to obtain fault diagnosis results.

[0027] In view of the technical problem that traditional fault diagnosis methods have slow response speed, cannot realize real-time monitoring and rapid response, and the diagnosis results are largely dependent on the experience and skills of maintenance personnel, which limits the accuracy and consistency of diagnosis, the present application can quickly process and analyze motor signals to realize real-time fault diagnosis by constructing a fault diagnosis model combining convolutional neural network (CNN) and long short-term memory network (LSTM) and introducing a channel attention mechanism. This model structure based on deep learning enables the system to have the ability of online learning. During actual operation, the model can continuously receive new motor signal data and update its internal parameters and feature extraction method in real time. When new fault patterns appear, the model can quickly capture these new features and include them in the scope of fault diagnosis, thereby realizing immediate adaptation to new and unknown fault patterns without the need for offline training with a large amount of historical fault data, greatly improving the flexibility and adaptability of the fault diagnosis system.

[0028] In view of the technical problem that existing machine learning models usually need a large amount of labeled data for training, and have limited generalization ability and adaptability when facing new or unknown fault types, the present application adopts preprocessing steps such as wavelet decomposition, threshold processing and wavelet reconstruction to denoise the collected motor signals, which can effectively remove noise signals and retain key features in the motor signals. Through these preprocessing means, more representative and effective features can be extracted from the original motor signals, which can better reflect the actual operating state and potential fault information of the motor; at the same time, the attention mechanism can automatically focus on important parts of the data, improving the model's adaptability to new fault patterns. Therefore, when training the fault diagnosis model, even without a large amount of historical fault data, the model can learn the rules and patterns of motor faults by relying only on these key feature data after preprocessing, thereby reducing the dependence on historical fault data, reducing the cost of data collection and labeling, and improving the practicality and economy of the fault diagnosis system.

[0029] In view of the technical problem that existing models often need a long time to train and adjust in actual application, making it difficult to meet the demand for immediate diagnosis, the present application uses the Adadelta optimizer to adjust the weights and parameters of the model, which can converge more quickly by adaptively adjusting the learning rate, improving the training efficiency of the model and meeting the demand for immediate diagnosis. BRIEF DESCRIPTION OF DRAWINGS

[0030] The accompanying drawings, which form a part of this application, are included to provide a further understanding of the application and are incorporated in and constitute a part of this application. The embodiments of the present application illustrated in the drawings, and their description, are used to explain the present application and are not intended to limit the present application unduly.

[0031] Figure 1A method flowchart of an embodiment of the present application;

[0032] Figure 2 A motor fault signal diagram of various types of embodiments of the present application;

[0033] Figure 3 A wavelet threshold denoising flowchart of an embodiment of the present application;

[0034] Figure 4 A CAMLSTM-CNN overall framework diagram of an embodiment of the present application. DETAILED DESCRIPTION

[0035] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in combination with embodiments.

[0036] It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a group of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in a different order.

[0037] Embodiment one

[0038] As shown in Figure 1 , the present embodiment provides a fault diagnosis method for aircraft actuators, comprising:

[0039] S1, collecting motor signals of aircraft actuators, the motor signals including real-time current information and real-time voltage information, the motor signals containing normal data, short-circuit data, overload data, open-circuit data, and shaft current data, and being divided into a training set and a test set;

[0040] As shown in Figure 2 , the motor fault signal diagram of various types is used as the motor platform fault data.

[0041] S2, sequentially performing wavelet decomposition, threshold processing, wavelet reconstruction, and denoising preprocessing on the training set to obtain denoised motor current one-dimensional data;

[0042] Through the preprocessing method of step S2, the signal quality can be improved and the interference of environmental noise can be reduced, and the preprocessed data can be better used for training neural networks. As shown in Figure 3 , the specific process is as follows:

[0043] S201, wavelet decomposition: selecting a suitable wavelet basis function and defining the decomposition layer number, performing multi-layer decomposition on the original motor signal to obtain the wavelet decomposition coefficients of each layer;

[0044] S202, threshold processing: set a suitable threshold, and perform threshold processing on the obtained wavelet decomposition coefficients of each layer;

[0045] The key step of wavelet threshold denoising is threshold processing. After wavelet decomposition, the original motor signal amplitude is much larger than the noise signal amplitude. The designed threshold denoising function is as follows:

[0046]

[0047] In the formula, the parameter a is an adjustment factor of the threshold function, which can be flexibly set in the range of (0, 1]. When the threshold value is close to 0, the improved threshold function tends to be a soft threshold algorithm, and when the threshold value is close to 1, the improved threshold function tends to be a hard threshold method. The adjustable threshold function overcomes the inherent defects of hard threshold and soft threshold, while retaining the advantages of hard threshold and soft threshold, so that the reconstruction function retains a large amount of effective information and has good continuity.

[0048] At the same time, in order to avoid the signal being filtered out as noise when the signal amplitude is close to or less than the noise, affecting the denoising effect, the traditional threshold estimation method is adjusted. For the wavelet transform value under the m scale, the correlation corr(m, n) of the adjacent scale is calculated, and the normalized result k(m, n) is obtained after normalization. Under the wavelet transform of different scales, the change trend of the noise wavelet signal coefficient and the original signal wavelet coefficient is opposite, so when k(m, n) < 1, the obtained is a noise signal. In order to better filter out noise interference, the traditional threshold λ is adjusted as follows:

[0049]

[0050] S203, wavelet reconstruction: the wavelet coefficients of each layer after threshold processing are recombined through wavelet inverse transform, so as to obtain the denoised motor signal.

[0051] S3, construct a fault diagnosis model, the fault diagnosis model comprising: a convolutional neural network (CNN) layer, a long short-term memory (LSTM) layer, a channel attention mechanism, and a fully connected layer;

[0052] As shown in Figure 4 , the fault diagnosis model is a CAMLSTM-CNN model.

[0053] S4, input the one-dimensional motor current data to the convolutional neural network (CNN) layer for feature extraction, and extract the key feature data of the motor signal through convolution and pooling operations;

[0054] S5, feeding the key feature data to a long short-term memory network (LSTM) layer to process and remember long-term dependencies of the input sequence, and applying a channel attention mechanism on the LSTM layer to weight score the output of the LSTM layer to realize attention and processing of important parts of the data;

[0055] Specifically, the output of the LSTM layer is weighted scored by the channel attention mechanism, which can realize more accurate fault detection.

[0056] S6, outputting the fault detection result through a fully connected layer to obtain a trained fault diagnosis model;

[0057] Specifically, the fault diagnosis result is output through the fully connected layer to realize real-time diagnosis of the actuator fault of the aircraft, and the Adadelta optimizer is used to adjust the weights and parameters of the model to minimize the cross-entropy loss function, thereby improving the detection accuracy of the model for the fault signal.

[0058] S7, using the trained fault diagnosis model to perform aircraft fault detection on the test set to obtain a fault diagnosis result.

[0059] The embodiment uses the data of the built dual-redundancy electric actuator system platform to test the fault diagnosis performance of the embodiment from the aspects of accuracy, sensitivity and specificity, and the motor fault diagnosis performance evaluation results are shown in Table 1.

[0060] Table 1

[0061]

[0062] According to the above table, it can be seen that the fault diagnosis algorithm designed in the embodiment has excellent performance, and for short circuit, overload, open circuit, shaft current and other types of classic abnormalities, the diagnosis accuracy is more than 99%.

[0063] Embodiment Two

[0064] The embodiment develops a software system based on a graphical user interface (GUI), which can display the collection state of the motor signal, the fault diagnosis result and the parameters of the system in the above embodiment one in real time. The system provides an intuitive interactive interface for the operator, so that the operator can easily access the diagnosis result and the system state, and improves the practicability and user friendliness of the system.

[0065] Embodiment Three

[0066] The embodiment also discloses a computer device, which comprises a memory, a processor and a computer program stored in the memory, and the processor executes the computer program to realize the steps of the method in embodiment one.

[0067] Example 4

[0068] This embodiment further discloses a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the method described in the first embodiment are implemented.

[0069] Example 5

[0070] This embodiment further discloses a computer program product, including a computer program, which implements the steps of the method described in the first embodiment when executed by a processor.

[0071] The above are merely preferred embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A fault diagnosis method for an aircraft actuator, characterized in that: The following steps are involved: Collecting motor signals from the aircraft actuator, the motor signals including real-time current information and real-time voltage information, the motor signals containing normal data, short-circuit data, overload data, open circuit data, and shaft current data, and dividing them into a training set and a test set; The training set is sequentially subjected to wavelet decomposition, threshold processing, wavelet reconstruction, and denoising preprocessing to obtain denoised one-dimensional motor current data; Constructing a fault diagnosis model, the fault diagnosis model including: a convolutional neural network (CNN) layer, a long short-term memory (LSTM) network layer, a channel attention mechanism, and a fully connected layer; The one-dimensional data of the motor current is input into the convolutional neural network (CNN) layer for feature extraction, and the key feature data of the motor signal is extracted through convolution and pooling operations; Feeding the key feature data into the long short-term memory network (LSTM) layer to process and memorize the long-term dependencies of the input sequence, and applying a channel attention mechanism on the LSTM layer to weight the output of the LSTM layer to achieve attention and processing of important parts of the data; Outputting the fault detection results through the fully connected layer, and using the Adadelta optimizer to adjust the weights and parameters of the fault diagnosis model to obtain a trained fault diagnosis model; The trained fault diagnosis model is used to perform aircraft fault detection on the test set to obtain a fault diagnosis result.

2. The method according to claim 1, characterized in that The steps of sequentially performing wavelet decomposition, threshold processing, wavelet reconstruction, and denoising preprocessing on the collected motor current signal include: Select a wavelet basis function and customize the number of decomposition layers to perform multi-layer decomposition on the motor signal to obtain wavelet decomposition coefficients of each layer; Set a preset threshold and perform threshold processing on the wavelet decomposition coefficients of each layer; The wavelet coefficients of each layer after threshold processing are recombined through inverse wavelet transform to obtain the denoised motor signal.

3. The method according to claim 2, characterized in that In the threshold processing process, an adjustable threshold function is used for processing. When the threshold is close to 0, the adjustment factor of the improved threshold function tends to the soft threshold algorithm. When the threshold is close to 1, the adjustment factor of the improved threshold function tends to the hard threshold method.

4. The method according to claim 1, wherein The convolutional neural network (CNN) layer extracts key features of motor signals through convolution and pooling operations. The convolution operation uses multiple convolution kernels of different sizes, and the pooling operation uses maximum pooling or average pooling.

5. The method according to claim 1, wherein The number of units in the LSTM layer is adjusted according to the dimension and sequence length of the input feature data.

6. The method according to claim 1, characterized in that The Adadelta optimizer adaptively adjusts the learning rate to minimize the cross-entropy loss function.

7. The method according to claim 1, characterized in that The output layer of the fully connected layer uses a Softmax activation function to map the output of the fault diagnosis model to the category probability distribution of the fault diagnosis.

8. A computer device comprising a memory, a processor, and a computer program stored in the memory, wherein: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

9. 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 7 are implemented.

10. A computer program product comprising a computer program, 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 7 are implemented.

Images