Fault diagnosis method and device for electro-hydraulic servo valve of aircraft brake system

Abstract

The invention provides a fault diagnosis method and device for an electro-hydraulic servo valve of an aircraft brake system, and relates to the technical field of fault diagnosis of electro-hydraulic systems. The method comprises the following steps: collecting a vibration signal of the electro-hydraulic servo valve, and carrying out fast Fourier transform and variational mode decomposition processing on the vibration signal to generate a multi-scale time-frequency characteristic; performing feature extraction, fusion and fault classification on the multi-scale time-frequency features by using a time sequence convolutional network model; and training the time sequence convolutional network model by using a self-adaptive loss function with sample screening to enhance the anti-interference capability of the model to noise labels. According to the method, the problems of poor universality and low diagnosis accuracy of a fault diagnosis method based on deep learning caused by complex fault modes and label noise interference can be solved, and the timeliness and precision of fault diagnosis are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis of electro-mechanical hydraulic systems, and particularly to a method and device for fault diagnosis of an electro-hydraulic servo valve in an aircraft braking system. Background Art

[0002] The aircraft braking system realizes the braking and control of the aircraft during takeoff, landing, taxiing, and turning by bearing the static weight, dynamic impact load of the aircraft, and absorbing the kinetic energy of the aircraft during landing. Once a fault occurs in the aircraft braking system, it will affect the braking efficiency at least, and at worst, tire blowout or even running off the runway will occur, resulting in huge casualties and property losses. As the core control component of the aircraft braking system, the electro-hydraulic servo valve is responsible for converting the electrical signal into hydraulic power and precisely adjusting the braking pressure, directly affecting the braking response speed, stability, and safety. Its performance determines the control reliability of key stages such as aircraft landing and taxiing.

[0003] Aero electro-hydraulic servo valves have the advantages of fast response speed, large control power, and high control accuracy. However, due to the structural characteristics of the servo valve with a compact structure, precise parts, and high integration, it has high requirements for oil and temperature. The aircraft braking system often works in a complex and changeable working environment, and needs to adapt to different runway conditions and harsh weather conditions such as rain and snow. The electro-hydraulic servo valve is continuously affected by various factors such as mechanical wear, variable loads, and the environment, resulting in diverse and complex fault forms, which are difficult to identify and diagnose in a timely manner. Therefore, in order to ensure the normal operation of the aircraft and prevent accidents, it is necessary to develop a safe and efficient fault diagnosis method for aero electro-hydraulic servo valves.

[0004] Currently, the fault diagnosis methods for electro-hydraulic servo valves mainly include model-driven fault diagnosis methods, signal processing-based fault diagnosis methods, and data-driven fault diagnosis methods. Among them, the model-driven fault diagnosis method constructs a mathematical model of the servo valve, designs an observer to estimate the system state, and detects faults through residual analysis. This type of method mainly includes parameter estimation methods and state estimation methods, such as methods based on state observers, Kalman filters, extended Kalman filters, particle filters, and the equivalent space method.

[0005] However, since hydraulic components work in a closed oil circuit or oil chamber, there are limited measurable parameters during operation. And there are many factors affecting the performance of the hydraulic system. Complex systems often exhibit hysteresis effects, strong coupling, and nonlinear behavior with parameters changing over time, making it difficult to construct an accurate mathematical model. Therefore, the signal processing-based method analyzes various signals collected by sensors, including the vibration frequency, pressure, and flow rate of hydraulic components, and applies technologies such as wavelet transform, correlation analysis, and higher-order statistics to identify fault characteristics and achieve accurate diagnosis of system faults. This method can reduce the dependence on an accurate model.

[0006] With the development of data analysis technologies such as machine learning and deep learning, data-driven methods have gradually been applied to the fault diagnosis of electro-hydraulic servo valves. Such methods learn the characteristics of servo valve operation data and construct a mapping relationship between sensor data and the health status of the servo valve. In the early stage, machine learning methods such as expert systems and support vector machines (SVM) were used to model common faults in hydraulic systems. Related research used optimization algorithms to improve the BP neural network for the fault diagnosis of servo valves. In recent years, deep learning methods such as CNN and RNN have gradually been used to improve the accuracy and stability of servo valve fault classification models. The data-driven fault diagnosis method does not need to rely on a complex mechanism model and can directly learn the unique patterns of faults from the data to achieve efficient and accurate fault diagnosis.

[0007] However, there is currently little research on the fault diagnosis of electro-hydraulic servo valves based on deep learning. Most model parameters are relatively large, the generalization ability of the models is weak, and there is a lack of methods for analyzing from the perspective of combining the time-frequency domain of data. And most fault diagnosis methods are based on complete sensor data and corresponding fault categories, lacking research on noisy label data. Summary of the Invention

[0008] In view of the above problems, the present invention combines the Fast Fourier Transform (FFT), Variational Modal Decomposition (VMD), and Temporal Convolutional Network (TCN) to provide a fault diagnosis method and device for electro-hydraulic servo valves in an aircraft braking system, so as to solve the problems of poor generality and low diagnostic accuracy of the deep learning-based fault diagnosis method caused by complex fault modes and label noise interference, and improve the timeliness and accuracy of fault diagnosis.

[0009] To solve the above technical problems, the present invention provides the following technical solutions:

[0010] On the one hand, a fault diagnosis method for electro-hydraulic servo valves in an aircraft braking system is provided. The method includes the following steps:

[0011] S1. Collect the vibration signal of the electro-hydraulic servo valve, and perform fast Fourier transform and variational modal decomposition on the vibration signal to generate multi-scale time-frequency features;

[0012] S2. Use the temporal convolutional network model to perform feature extraction, fusion, and fault classification on the multi-scale time-frequency features;

[0013] S3. Use an adaptive loss function with sample screening to train the temporal convolutional network model, enhancing the model's anti-interference ability against noisy labels.

[0014] Optionally, step S1 specifically includes:

[0015] S11. Use the Fast Fourier Transform (FFT) to extract the frequency-domain information of the vibration signal. Assuming that the discrete signal of length N is represented as: x[n], n = 0, 1, 2,..., N - 1, then the discrete Fourier transform is:

[0016]

[0017] S12. Perform variational mode decomposition (VMD) on the vibration signal. The optimization objective function of VMD is expressed as:

[0018] In the formula, f(t) is the input vibration signal, u k (t) is the k-th intrinsic mode function (IMF) of the decomposition, ω k is the central frequency of the k-th IMF, is the derivative with respect to time, δ(t) is the Dirac function, j is the imaginary unit, and K is the total number of modes.

[0019] Optionally, the value of IMF is 4.

[0020] Optionally, in step S2, the temporal convolutional network model is a multi-scale temporal convolutional network model.

[0021] Optionally, step S2 specifically includes:

[0022] S21. Send the sequence processed by FFT and VMD to the embedding layer, where M represents the number of variables and L represents the length of the input sequence; perform an embedding operation on the input sequence :

[0023]

[0024] In the formula, is the time series after the embedding operation, D is the size of the data feature dimension, and N is the size of the temporal dimension;

[0025] S22. Use convolution operations to extract features from X emb .

[0026] Among them, large-kernel convolution captures the long-term dependencies of the sequence by enhancing the effective receptive field of the model, and small-kernel convolution is used to capture fine-grained features; both convolutions adopt grouped convolution, that is, each input channel is independently convolved with the corresponding convolution kernel:

[0027]

[0028] In the formula, the m-th convolution kernel calculates the output feature map with the feature map F of the corresponding channel ; where m represents the number of channels of the convolution kernel and the feature map, l represents the spatial position of the input and output feature maps, and k represents the position index of the convolution kernel;

[0029] S23. Send the feature map after feature extraction into two feature fusion modules for feature fusion;

[0030] Each feature fusion module consists of two pointwise convolutions and a GeLU activation function. The pointwise convolution performs a linear combination in the variable and feature dimensions, and the GeLU activation function adds a non-linear factor to the feature fusion; among them, the calculation of the pointwise convolution is as follows:

[0031]

[0032] In the formula, W m,n is the mapping weight from the m-th input channel to the n-th output channel; M is the number of input channels;

[0033] The first feature fusion module is used to recombine the features in the variable dimension, and the second feature fusion module is used to capture the temporal dependencies across variables. Both feature fusion modules adopt separable convolution to ensure the independence of the variable dimension and the feature dimension.

[0034] Optionally, in step S2, the fault classification results include: normal operation, nozzle clogging degradation, nozzle clogging fault, spool wear degradation, spool wear fault, and electromagnet degradation.

[0035] Optionally, step S3 specifically includes:

[0036] S31. Use cross-entropy CE and normalized cross-entropy NCE as the active loss and passive loss to improve the robustness and generalization ability of the model. The representation of NCE is as follows:

[0037]

[0038] In the formula, is the dataset sample, is the corresponding noise label, is the output probability distribution of the neural network model, p(y | x) is the probability distribution when the predicted label is the same as the actual label, q(k |x) is the label probability distribution of x, q(y = j | x) is the probability distribution with the label value of j, and it satisfies , ;

[0039] Calculate the variance V of the predicted label probability distribution through the following formula, and dynamically adjust the weight of the CE term:

[0040]

[0041] In the formula, is 's mean value, then the adaptive loss function is:

[0042] In the formula, α,β>0 are weight parameters, and f(x) is the function expression of the neural network;

[0043] S32. At the early stage of training, adopt the adaptive loss without sample screening. At the later stage of training, use the prediction variance V as the performance index of classification effectiveness to screen the training samples;

[0044] The adaptive loss function with sample screening is expressed as:

[0045] In the formula, refers to the screened training samples, refers to based on Calculated sample variance.

[0046] On the other hand, an electro-hydraulic servo valve fault diagnosis device for an aircraft braking system is provided, which is used to implement the method described in any one of the above, and the device includes:

[0047] A signal acquisition and processing module, which is used to collect the vibration signal of the electro-hydraulic servo valve, perform fast Fourier transform and variational mode decomposition processing on the vibration signal, and generate multi-scale time-frequency features;

[0048] A temporal convolutional network model, which is used to extract, fuse and classify faults for the multi-scale time-frequency features;

[0049] A training module, which is used to train the temporal convolutional network model by using the adaptive loss function with sample screening to enhance the anti-interference ability of the model to noise labels.

[0050] On the other hand, an electronic device is provided, and the electronic device includes:

[0051] A processor;

[0052] A memory stores computer-readable instructions. When the computer-readable instructions are loaded and executed by the processor, the steps of the above-mentioned aircraft brake system electro-hydraulic servo valve fault diagnosis method are implemented.

[0053] On the other hand, a computer-readable storage medium is provided. Program code is stored in the computer-readable storage medium, and the program code can be called by the processor to execute the steps of the above-mentioned aircraft brake system electro-hydraulic servo valve fault diagnosis method.

[0054] The beneficial effects brought by the technical solution provided by the present invention at least include:

[0055] The aircraft brake system electro-hydraulic servo valve fault diagnosis method based on the temporal convolutional network provided by the present invention provides new ideas and approaches for data-driven electro-hydraulic servo valve fault diagnosis, and makes up for the problems of poor generality, low diagnostic accuracy, and susceptibility to noise label interference that are difficult to solve by existing fault diagnosis methods. It can improve the real-time performance and accuracy of electro-hydraulic servo valve fault diagnosis, and has important practical application and promotion value. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0057] Figure 1 is a flowchart of an aircraft brake system electro-hydraulic servo valve fault diagnosis method provided by an embodiment of the present invention;

[0058] Figure 2 is a schematic structural diagram of a temporal convolutional network provided by an embodiment of the present invention;

[0059] Figure 3 is a schematic diagram of the electro-hydraulic servo valve fault diagnosis process based on MTCN-ALSS provided by an embodiment of the present invention;

[0060] Figure 4 is a training curve of MTCN-ALSS with a noise ratio of 0.6 provided by an embodiment of the present invention;

[0061] Figure 5 is a confusion matrix diagram of the fault diagnosis results provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0062] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the described embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0063] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to represent examples, illustrations, or explanations. Any embodiment or design solution described as an "example" in the present invention should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Rather, the use of the word "example" is intended to present concepts in a specific manner.

[0064] The embodiments of the present invention provide a method for diagnosing faults in an electro-hydraulic servo valve of an aircraft braking system. This method can be implemented by an electronic device, which can be a terminal or a server. As Figure 1 shown, the processing flow of this method can include the following steps:

[0065] S1. Collect the vibration signal of the electro-hydraulic servo valve, perform fast Fourier transform and variational mode decomposition processing on the vibration signal to generate multi-scale time-frequency features.

[0066] The specific steps of step S1 include:

[0067] S11. Use the fast Fourier transform (FFT) to extract the frequency-domain information of the vibration signal. Assume that the discrete signal of length N is represented as: x[n], n = 0, 1, 2,..., N - 1, then the discrete Fourier transform is:

[0068]

[0069] S12. Perform variational mode decomposition (VMD) on the collected vibration signal. The optimization objective function of VMD is expressed as:

[0070] where f(t) is the input vibration signal, u k (t) is the kth intrinsic mode function (IMF) of the decomposition, ω k is the central frequency of the kth IMF, is the derivative with respect to time, δ(t) is the Dirac function, j is the imaginary unit, and K is the total number of modes.

[0071] Among them, the value of IMF is determined by observing the central frequencies of different numbers of modes. In the embodiments of the present invention, through repeated experiments and verification, the value of IMF is 4.

[0072] The Fast Fourier Transform (FFT) is an effective method for calculating the Discrete Fourier Transform (DFT). It can convert a time-domain signal into a frequency-domain representation, thereby revealing the frequency components of the signal. Moreover, the FFT can effectively reduce the computational complexity. Variational Mode Decomposition (VMD) is a signal processing technique that can decompose a complex signal into several Intrinsic Mode Functions (IMFs) with physical meanings. This is an adaptive decomposition method based on a variational framework, which can effectively overcome some inherent problems in traditional decomposition methods, such as mode mixing, endpoint effects, and sensitivity to noise.

[0073] S2. Use a temporal convolutional network model to extract, fuse, and classify faults for the multi-scale time-frequency features.

[0074] A Temporal Convolutional Network (TCN) is a deep learning architecture designed for modeling time series data. Based on the basic model of a Convolutional Neural Network (CNN), by combining causal convolution and dilated convolution, it solves problems such as the vanishing gradient and low computational efficiency in the learning of long sequence dependencies by traditional Recurrent Neural Networks (RNNs). The TCN can capture temporal dynamic features using convolutional operations while maintaining strict time causality.

[0075] In the embodiments of the present invention, the temporal convolutional network adopted is a multi-scale temporal convolutional network MTCN, and the structure is as Figure 2 shown. The specific steps of step S2 include:

[0076] S21. Send the sequence processed by FFT and VMD into the embedding layer, where M represents the number of variables and L represents the length of the input sequence. To maintain the independence of variable dimensions during feature transformation, an embedding operation is performed on the input sequence :

[0077]

[0078] In the formula, is the time series after the embedding operation, D is the size of the data feature dimension, and N is the size of the time series dimension.

[0079] S22. Use convolutional operations to extract features from X emb .

[0080] Among them, the large-kernel convolution captures the long-term dependencies of the sequence by enhancing the effective receptive field (ERF) of the model, and the small-kernel convolution is used to capture fine-grained features. To capture temporal dependencies in both the variable dimension and the feature dimension without destroying the independence of these two dimensions, both convolutions adopt group convolution, that is, each input channel is independently convolved with the corresponding convolution kernel:

[0081]

[0082] In the formula, the m-th convolution kernel and the feature map F of the corresponding channel are used to calculate the output feature map ; where m represents the number of channels of the convolution kernel and the feature map, l represents the spatial position of the input and output feature maps, and k represents the position index of the convolution kernel.

[0083] The combined use of the two convolution kernel sizes enables the model to perform multi-scale feature extraction on the input data, which helps to improve the computational efficiency and prevent overfitting.

[0084] S23. Send the feature maps after feature extraction into two feature fusion modules for feature fusion.

[0085] Each feature fusion module consists of two pointwise convolutions and a GeLU activation function. The pointwise convolution performs a linear combination in the variable and feature dimensions, and the GeLU activation function adds a non-linear factor to the feature fusion. Among them, the calculation of the pointwise convolution is as follows:

[0086]

[0087] In the formula, W m,n is the mapping weight from the m-th input channel to the n-th output channel, and M is the number of input channels.

[0088] Among them, the first feature fusion module is used to recombine the features in the variable dimension, and the second feature fusion module is used to capture the temporal dependencies across variables. Both feature fusion modules adopt depthwise separable convolution to ensure the independence of the variable dimension and the feature dimension.

[0089] The fused features are classified after downsampling, and the final output fault classification results are shown in Table 1, including: normal operation, nozzle blockage degradation, nozzle blockage fault, spool wear degradation, spool wear fault, and solenoid degradation.

[0090] Table 1 Fault Types of Electro-Hydraulic Servo Valves

[0091]

[0092] S3. Train the temporal convolutional network model using an Adaptive Loss with Sample Screening (ALSS) to enhance the model's anti-interference ability against noisy labels.

[0093] The fault diagnosis process of the electro-hydraulic servo valve based on MTCN-ALSS provided by the embodiments of the present invention is as Figure 3 shown. The specific steps of step S3 include:

[0094] S31. Use Cross Entropy (CE) and Normalized Cross Entropy (NCE) as the active loss and passive loss to improve the robustness and generalization ability of the model. The expression of NCE is as follows:

[0095]

[0096] In the formula, is the dataset sample, is the corresponding noisy label, is the output probability distribution of the neural network model, p(y | x) is the probability distribution when the predicted label is consistent with the actual label, q(k |x) is the label probability distribution of x, q(y = j | x) is the probability distribution with the label value of j, and it satisfies , .

[0097] Calculate the variance V of the predicted label probability distribution through the following formula, and dynamically adjust the weight of the CE term:

[0098]

[0099] In the formula, is the mean value, then the adaptive loss function is:

[0100] In the formula, α, β > 0 are weight parameters, and f(x) is the function expression of the neural network.

[0101] The early characteristics of deep learning enable the accuracy to increase rapidly in the initial stage of training. Therefore, in the early stage of training, the adaptive loss without sample screening is adopted, and in the later stage of training, the prediction variance V is used as the performance index of classification effectiveness to screen the training samples;

[0102] The adaptive loss function with sample screening is expressed as:

[0103] In the formula, Refers to the selected training samples, which is based on the calculated sample variance.

[0104] Compared with the prior art, the present invention has the following advantages:

[0105] (1) Multi-scale feature extraction based on MTCN;

[0106] On the premise of ensuring the independence of variable dimension and feature dimension, two sizes of convolutional kernels are used to extract the temporal dependence of the sequence from different granularities; then the feature fusion module uses pointwise convolution to recombine the features in terms of variable and feature dimensions.

[0107] (2) Training a classification model resistant to noise interference based on ALSS;

[0108] The adaptive loss function combining NCE and CE improves the robustness and generalization ability of the model; in the later stage of training, the variance of the predicted labels is used to screen the training data to ensure the correct update direction of the parameters, further enhancing the anti-interference ability of the model to noise labels.

[0109] In terms of model verification, the vibration signals of the electro-hydraulic servo valve are collected as the sample training set. The training curve of MTCN-ALSS under a noise ratio of 0.6 is as Figure 4 shown, and the confusion matrix of the fault diagnosis results is as Figure 5 shown. It can be seen that the electro-hydraulic servo valve fault diagnosis method based on MTCN-ALSS of the present invention has high prediction accuracy.

[0110] Correspondingly, an embodiment of the present invention further provides an electro-hydraulic servo valve fault diagnosis device for an aircraft braking system, which is used to implement the above-mentioned electro-hydraulic servo valve fault diagnosis method for an aircraft braking system. The device includes:

[0111] A signal acquisition and processing module, which is used to collect the vibration signals of the electro-hydraulic servo valve, perform fast Fourier transform and variational mode decomposition processing on the vibration signals, and generate multi-scale time-frequency features;

[0112] A temporal convolutional network model, which is used to perform feature extraction, fusion, and fault classification on the multi-scale time-frequency features;

[0113] A training module, which is used to train the temporal convolutional network model by using an adaptive loss function with sample screening to enhance the anti-interference ability of the model to noise labels.

[0114] The device of this embodiment can be used to execute Figure 1 the technical solutions of the method embodiments shown, and its implementation principle and technical effects are similar, so they will not be elaborated here.

[0115] In an exemplary embodiment, the present invention further provides an electronic device, which includes:

[0116] A processor;

[0117] A memory, on which computer-readable instructions are stored. When the computer-readable instructions are loaded and executed by the processor, the steps of the above-mentioned aircraft brake system electro-hydraulic servo valve fault diagnosis method are implemented.

[0118] In an exemplary embodiment, the present invention further provides a computer-readable storage medium, in which at least one instruction is stored. The at least one instruction is loaded and executed by the processor to implement the steps of the above-mentioned aircraft brake system electro-hydraulic servo valve fault diagnosis method. For example, the computer-readable storage medium may be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.

[0119] It should be noted that in this article, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such a process, method, article or terminal device. Without more limitations, the element defined by the statement "including one..." does not exclude the existence of another identical element in the process, method, article or terminal device including the element.

[0120] When "an embodiment", "an embodiment", "an exemplary embodiment", "some embodiments", etc. are mentioned in the specification, it indicates that the described embodiment may include a specific feature, structure or characteristic, but not necessarily every embodiment includes this specific feature, structure or characteristic. Additionally, when combining an embodiment to describe a specific feature, structure or characteristic, implementing such a feature, structure or characteristic in combination with other embodiments (whether explicitly described or not) should be within the knowledge scope of those skilled in the relevant art.

[0121] It should be understood that the term "and / or" in this article is merely a description of the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B may represent: A exists alone, A and B exist simultaneously, and B exists alone. Here, A and B may be singular or plural. Additionally, the character " / " in this article generally represents an "or" relationship between the associated objects before and after, but it may also represent an "and / or" relationship, which can be specifically understood with reference to the context before and after.

[0122] In the present invention, "at least one" means one or more, and "a plurality" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single item or plural items. For example, at least one of a, b, or c can represent: a, b, c, a - b, a - c, b - c, or a - b - c, where a, b, and c can be single or plural.

[0123] It should be understood that in various embodiments of the present invention, the sequence numbers of the above - mentioned processes do not imply the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

[0124] In several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there can be other division methods. For example, multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections between each other can be through some interfaces. The indirect couplings or communication connections of the devices or units can be in electrical, mechanical, or other forms.

[0125] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0126] In addition, in each embodiment of the present invention, the functional units can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit.

[0127] When the above-mentioned functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs.

[0128] The present invention covers any alternatives, modifications, equivalent methods, and solutions made within the spirit and scope of the present invention. For the public to have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention. However, those skilled in the art can fully understand the present invention without these detailed descriptions. Additionally, to avoid unnecessary confusion to the essence of the present invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0129] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for diagnosing faults of an electro-hydraulic servo valve in an aircraft brake system, characterized in that: The following steps are involved: S1, collecting the vibration signal of the electro-hydraulic servo valve, performing fast Fourier transform and variational mode decomposition processing on the vibration signal, and generating multi-scale time-frequency features; S2. Using a temporal convolutional network model to extract, fuse and classify the multi-scale time-frequency features; S3. Use an adaptive loss function with sample screening to train the temporal convolutional network model to enhance the model's ability to resist interference from noise labels.

2. The method for diagnosing faults of an electro-hydraulic servo valve of an aircraft brake system according to claim 1, characterized in that: The step S1 specifically includes: S11. Use fast Fourier transform FFT to extract the frequency domain information of the vibration signal. Assuming that the discrete signal with a length of N is represented by: x[n], n = 0, 1, 2, ..., N-1, then the discrete Fourier transform is: ； S12, performing variational mode decomposition (VMD) on the vibration signal, and the optimization objective function of VMD is expressed as: ； Where f(t) is the input vibration signal, u k (t) is the kth intrinsic mode function IMF decomposed, ω k is the center frequency of the kth IMF, is the time derivative, δ(t) is the Dirac function, j is the imaginary unit, and K is the total number of modes.

3. The method for diagnosing faults of an electro-hydraulic servo valve of an aircraft brake system according to claim 2, characterized in that: The value of IMF is 4.

4. The method for diagnosing faults of an electro-hydraulic servo valve of an aircraft brake system according to claim 1, characterized in that: In step S2, the temporal convolutional network model is a multi-scale temporal convolutional network model.

5. The method for diagnosing faults of an electro-hydraulic servo valve of an aircraft brake system according to claim 1, characterized in that: The step S2 specifically includes: S21, the sequence processed by FFT and VMD Send it to the embedding layer, where M represents the number of variables and L represents the length of the input sequence; To perform embedding operation: ； In the formula, is the time series after the embedding operation, D is the data feature dimension size, and N is the time series dimension size; S22, use convolution operation to X emb Perform feature extraction; Among them, the large kernel convolution captures the long-term dependency of the sequence by enhancing the effective receptive field of the model, and the small kernel convolution is used to capture fine-grained features; both convolutions use grouped convolution, that is, each input channel is independently convolved with the corresponding convolution kernel: ； In the formula, the mth convolution kernel And the feature map F of the corresponding channel l+k∙D,m Calculate the output feature map ; Where m represents the number of channels of the convolution kernel and feature map, l represents the spatial position of the input and output feature maps, and k represents the position index of the convolution kernel; S23, sending the feature map after feature extraction to two feature fusion modules for feature fusion; Each feature fusion module consists of two point-by-point convolutions and a GeLU activation function, wherein the point-by-point convolution performs linear combinations on the variable and feature dimensions, and the GeLU activation function adds nonlinear factors to the feature fusion; wherein the point-by-point convolution is calculated as follows: ； Where W m,n is the mapping weight from the mth input channel to the nth output channel, where M is the number of input channels; The first feature fusion module is used to recombine features in the variable dimension, and the second feature fusion module is used to capture the temporal dependencies across variables. Both feature fusion modules use separable convolution to ensure the independence of variable dimension and feature dimension.

6. The method for diagnosing faults of an electro-hydraulic servo valve of an aircraft brake system according to claim 1, characterized in that: In the step S2, the fault classification results include: normal operation, nozzle blockage and degradation, nozzle blockage fault, valve core wear and degradation, valve core wear fault, and electromagnet degradation.

7. The method for diagnosing faults of an electro-hydraulic servo valve in an aircraft brake system according to claim 1, characterized in that: The step S3 specifically includes: S31. Cross entropy CE and normalized cross entropy NCE are used as active loss and passive loss to improve the robustness and generalization ability of the model. NCE is expressed as follows: ； In the formula, is a sample of the data set, is the corresponding noise label, is the output probability distribution of the neural network model, p(y | x) is the probability distribution when the predicted label is consistent with the actual label, q(k | x) is the label probability distribution of x, and q(y = j | x) is the probability distribution of the label value j, and satisfies , ; The variance V of the predicted label probability distribution is calculated by the following formula, and the weight of the CE item is dynamically adjusted: ； In the formula, for The mean of , then the adaptive loss function is: ； In the formula, α, β>0 are weight parameters, and f(x) is the function expression of the neural network; S32, adopting an adaptive loss without sample screening in the early stage of training, and using the prediction variance V as a performance indicator of classification effectiveness in the late stage of training to screen the training samples; The adaptive loss function with sample screening is expressed as: ； In the formula, refers to the selected training samples. Refers to based on Compute the sample variance.

8. A fault diagnosis device for an electro-hydraulic servo valve of an aircraft brake system, the device being used to implement the method according to any one of claims 1 to 7, characterized in that: The device comprises: A signal acquisition and processing module is used to acquire the vibration signal of the electro-hydraulic servo valve, perform fast Fourier transform and variational mode decomposition processing on the vibration signal, and generate multi-scale time-frequency features; A temporal convolutional network model, used for extracting, fusing and classifying the multi-scale time-frequency features; The training module is used to train the temporal convolutional network model using an adaptive loss function with sample screening to enhance the model's ability to resist interference from noise labels.

9. An electronic device, characterized in that: The electronic device comprises: processor; A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are loaded and executed by the processor, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program code, which can be called by a processor to execute the method according to any one of claims 1 to 7.

Images