Fault awareness and diagnosis method for aviation electric fuel pump based on CAN bus

Abstract

The application relates to the technical field of fuel pump fault diagnosis, and proposes an aviation electric fuel pump fault sensing and diagnosis method based on a CAN bus, which comprises the following steps: collecting working condition operation parameters and state characteristic parameters of an aviation electric fuel pump; adopting a wavelet threshold denoising algorithm to decompose the same working condition operation parameters, obtaining a detail coefficient sequence of different wavelet decomposition layers, determining whether to use a fixed threshold method to obtain a wavelet threshold, if not, calculating a detail coefficient adjustment degree of the wavelet decomposition layer, combining the wavelet threshold determined by the fixed threshold method, obtaining a detail adjustment coefficient sequence and reconstruction, obtaining denoised working condition operation parameters, and using an anomaly detection algorithm to obtain reconstructed state characteristic parameters; using a pre-trained fault sensing model, according to the denoised working condition operation parameters and the reconstructed state characteristic parameters of a preset time length, completing aviation electric fuel pump fault diagnosis. The application can improve the accuracy of aviation electric fuel pump fault diagnosis.

Description

Technical Field

[0001] This application relates to the field of fuel pump fault diagnosis technology, specifically to a method for fault detection and diagnosis of aviation electric fuel pumps based on CAN bus. Background Technology

[0002] The electric fuel pump is a core component of modern aircraft propulsion systems. Its pump control technology precisely regulates the motor speed and output flow through a controller, replacing traditional valve control systems. This significantly simplifies the engine fuel servo system structure, reduces system weight, and improves engine thrust-to-weight ratio and fuel efficiency. The safe and stable operation of the electric fuel pump is directly related to aircraft flight safety. Fault detection and assessment are crucial for the safety and reliability of aviation power control systems. Their core purpose is to achieve early fault warning and precise isolation. By capturing vibration signals and analyzing pressure pulsations, initial faults are identified, preventing sudden hardware failures from causing serious aviation accidents.

[0003] In practical fault diagnosis, the working and testing environment of fuel pumps is subject to interference such as flow-induced noise and cavitation noise. These noises are similar to early fault characteristics such as bearing pitting and initial blade friction, easily leading to confusion between faults and noise, thus affecting diagnostic accuracy. Conventional filtering algorithms tend to over-filter out fault details, while wavelet threshold denoising, although capable of layered denoising, cannot adaptively adjust its wavelet coefficients, making it difficult to effectively separate faults from noise. This severely impacts the accuracy of subsequent fault diagnosis and restricts the engineering application of fault diagnosis technology for aviation electric fuel pumps. Summary of the Invention

[0004] This application provides a fault detection and diagnosis method for aviation electric fuel pumps based on CAN bus, in order to solve the problem that wavelet threshold denoising is difficult to effectively separate faults from noise, which restricts the fault diagnosis effect of aviation electric fuel pumps. The specific technical solution adopted is as follows: One embodiment of this application provides a method for fault detection and diagnosis of aviation electric fuel pumps based on CAN bus, the method including the following steps: The operating parameters and status characteristics of the aviation electric fuel pump are collected at each collection time. The operating parameters include current, voltage, speed, pressure and flow rate, and the status characteristics include vibration and pump body temperature. A wavelet thresholding denoising algorithm is used to decompose the operating parameters of the same working condition within a preset time period to obtain the detail coefficient sequences of different wavelet decomposition layers. Based on the differences between adjacent detail coefficients in the detail coefficient sequences, it is determined whether to use the fixed threshold method to obtain the wavelet threshold. If not, a type of peak coefficient is identified and the fluctuation feature value of the wavelet decomposition layer is calculated. Feature vectors are established and clustered based on the acquisition time and time interval of adjacent peak coefficients to obtain feature vector clusters. Based on the detail coefficients corresponding to the feature vectors in all feature vector clusters corresponding to the wavelet decomposition layer, the time intervals corresponding to different feature vectors, and the fluctuation feature value, the detail coefficient adjustment degree of the wavelet decomposition layer is calculated. Combined with the wavelet threshold determined by the fixed threshold method, the detail coefficients of the wavelet decomposition layer are adjusted to obtain the detail adjustment coefficient sequence. The detail adjustment coefficient sequences of all wavelet decomposition layers of the same operating condition are reconstructed to obtain the denoised operating parameters. An anomaly detection algorithm is used to obtain the reconstructed state feature parameters. Using a pre-trained fault perception model, fault diagnosis of aviation electric fuel pumps is completed based on the denoised operating parameters and reconstructed state feature parameters after a preset duration.

[0005] Furthermore, the specific method for determining whether to use a fixed threshold method to obtain the wavelet threshold based on the difference between adjacent detail coefficients in the detail coefficient sequence includes: The detail coefficients in the detail coefficient sequence that are greater than the two adjacent detail coefficients are recorded as peak coefficients. When the number of peak coefficients in the detail coefficient sequence is less than or equal to the second preset number, the wavelet threshold of the corresponding layer is obtained by using the fixed threshold method.

[0006] Furthermore, the specific method for identifying a class of peak coefficients and calculating the wavelet decomposition layer's fluctuation characteristic values ​​includes: When the number of peak coefficients in the detail coefficient sequence is greater than the second preset number, the peak coefficients in the detail coefficient sequence that are greater than or equal to the mean of all peak coefficients are recorded as a class of peak coefficients. Peak coefficients in the detail coefficient sequence that are less than the mean of all peak coefficients are denoted as Class II peak coefficients. The difference between the mean of the normalized values ​​of all Class I peak coefficients and the mean of the normalized values ​​of all Class II peak coefficients is denoted as the fluctuation difference feature value of the corresponding wavelet decomposition layer. The ratio of the permutation entropy of the time interval of all adjacent Class I peak coefficients in the wavelet decomposition layer to the fluctuation difference feature value is denoted as the fluctuation feature value of the wavelet decomposition layer.

[0007] Furthermore, the specific method for obtaining the feature vector clusters is as follows: For any wavelet decomposition layer detail coefficient sequence under the same operating conditions and parameters, the minimum time interval between a class of peak coefficients and two adjacent class of peak coefficients is denoted as the characteristic time interval of the class of peak coefficients. The vector composed of the acquisition time corresponding to the class of peak coefficients and the characteristic time interval is denoted as the characteristic vector of the class of peak coefficients. All characteristic vectors determined by the detail coefficient sequence are clustered to obtain the characteristic vector cluster.

[0008] Furthermore, the specific calculation method for the detail coefficient adjustment degree of the wavelet decomposition layer is as follows: The square of the detail coefficients corresponding to the feature vector is denoted as the energy of the feature vector. The sum of the squares of all detail coefficients corresponding to the wavelet decomposition layer is denoted as the total energy of the wavelet decomposition layer. The first ratio of the feature vector cluster is calculated based on the energy of the feature vector within the feature vector cluster and the total energy of the corresponding wavelet decomposition layer. The ratio of the maximum time interval corresponding to different feature vectors within a feature vector cluster to a preset duration is denoted as the second ratio of the feature vector cluster. The ratio of the first ratio to the second ratio of the eigenvector clusters is denoted as the energy eigenvalue of the eigenvector clusters. When the number of eigenvector clusters is less than or equal to 1, the energy distribution eigenvalue of the wavelet decomposition layer is directly assigned to 1.0. When the number of eigenvector clusters is greater than 1, the coefficient of variation of the energy eigenvalues ​​of all eigenvector clusters corresponding to the wavelet decomposition layer is denoted as the energy distribution eigenvalue of the wavelet decomposition layer. The normalized value of the product of the energy distribution eigenvalue and the fluctuation eigenvalue of the wavelet decomposition layer is denoted as the detail coefficient adjustment degree of the wavelet decomposition layer.

[0009] Furthermore, the first ratio of the feature vector cluster is the ratio of the sum of the energies of the two feature vectors with the largest time interval within the feature vector cluster to the total energy of the corresponding wavelet decomposition layer.

[0010] Furthermore, the specific method for obtaining the detailed adjustment coefficient sequence is as follows: The product of the sum of the adjustment degree of the detail coefficients of the wavelet decomposition layer and the preset second adjustment parameter value and the preset first adjustment parameter value is recorded as the optimal adjustment parameter of the wavelet decomposition layer; the adjustment value of the detail coefficients is calculated by combining the wavelet threshold determined by the fixed threshold method and the detail coefficients of the wavelet decomposition layer. The sequence of adjustment values ​​of the detail coefficients of the same wavelet decomposition layer is denoted as the detail adjustment coefficient sequence of the same wavelet decomposition layer.

[0011] Furthermore, the formula for calculating the adjustment value of the detail coefficient is: in, Indicates the first The first operating parameter under various working conditions The first layer of wavelet decomposition layer Adjustment values ​​for each detail coefficient; Indicates the first The first operating parameter under various working conditions The first layer of wavelet decomposition layer Detail coefficients; This indicates that the wavelet threshold was determined using the fixed threshold method; Indicates the first The first operating parameter under various working conditions Optimal adjustment parameters for layered wavelet decomposition layers; Represents the natural constant.

[0012] Furthermore, the fault perception model is trained using a support vector machine.

[0013] Furthermore, the fault diagnosis results for aviation electric fuel pumps include: normal state, motor controller failure, bearing failure, cavitation failure, blockage failure, and leakage failure.

[0014] The beneficial effects of this application are: This application considers that wavelet threshold denoising can perform layered signal processing, enabling detailed and effective denoising operations for different frequency bands. Wavelet threshold denoising is used to filter each type of operating parameter. Based on the characteristic that operating parameters exhibit regular and drastic changes during pump rotation in the event of an aviation electric fuel pump failure, the fluctuation characteristic value of the wavelet decomposition layer is calculated. The more significant the fault characteristics, the smaller the fluctuation characteristic value of the wavelet decomposition layer. Considering the sparse distribution and localized large impacts of the periodic abrupt changes in detail coefficients during a fault, the adjustment degree of the detail coefficients in the wavelet decomposition layer is calculated based on the degree of energy concentration and fluctuation regularity. A larger adjustment degree indicates that the detail coefficients within the wavelet decomposition layer should be adjusted during the wavelet decomposition process. To avoid the influence of noise, the wavelet threshold determined by the fixed threshold method is combined with the adjustment of the detail coefficients of the wavelet decomposition layer to obtain the detail adjustment coefficient sequence. The detail adjustment coefficient sequences of all wavelet decomposition layers for the same operating condition parameters are reconstructed to obtain the denoised operating condition parameters. Since the features of the state feature parameters affected by noise are relatively easy to distinguish from the fault features, the anomaly detection algorithm is directly used to obtain the reconstructed state feature parameters. Finally, a pre-trained fault perception model is used to complete the fault diagnosis of the aviation electric fuel pump based on the denoised operating condition parameters and the reconstructed state feature parameters for a preset duration. This solves the problem that wavelet threshold denoising is difficult to effectively separate faults from noise, which restricts the fault diagnosis effect of aviation electric fuel pumps, and improves the accuracy of aviation electric fuel pump fault diagnosis. Attached Figure Description

[0015] Figure 1 This is a schematic flowchart of a CAN bus-based fault detection and diagnosis method for aviation electric fuel pumps, provided as an embodiment of this application. Detailed Implementation

[0016] Please see Figure 1 The diagram illustrates a flowchart of a CAN bus-based fault detection and diagnosis method for aviation electric fuel pumps according to an embodiment of this application. The method includes the following steps: Step S001: Collect the operating parameters and status characteristic parameters of the aviation electric fuel pump at each collection time. The operating parameters include current, voltage, speed, pressure and flow rate, and the status characteristic parameters include vibration and pump body temperature.

[0017] CAN bus, or Controller Area Network, is a serial communication bus that enables real-time data transmission and communication between multiple electronic control units. In motor controllers using the CAN bus, Hall effect current sensors and resistive voltage dividers collect current and voltage data, respectively. In oil pump systems using the CAN bus, Hall effect sensors, pressure sensors, turbine flow meters, and ICP-type accelerometers collect speed, pressure, flow rate, and vibration data, respectively. NTC thermistors using the CAN bus collect pump body temperature data. The real-time acquired current, voltage, speed, pressure, flow rate, vibration, and pump body temperature data are transmitted to the terminal processing unit using standardized CANopen messages via the CAN bus.

[0018] In this embodiment, the sampling frequency of current, voltage, speed, pressure, flow rate and vibration is set to 10kHz, and the sampling frequency of pump body temperature is set to 1Hz.

[0019] Current, voltage, speed, pressure, and flow rate are all recorded as operating parameters, which are easily affected by noise. Vibration and pump body temperature are recorded as state characteristic parameters, which are less affected by noise.

[0020] Thus, the operating parameters and status characteristics of the aviation electric fuel pump at each data acquisition time were obtained.

[0021] Step S002: Decompose the same operating parameters within a preset time period using a wavelet threshold denoising algorithm to obtain detail coefficient sequences of different wavelet decomposition layers. Based on the differences between adjacent detail coefficients in the detail coefficient sequence, determine whether to use the fixed threshold method to obtain the wavelet threshold. If not, identify a type of peak coefficient and calculate the fluctuation feature value of the wavelet decomposition layer. Establish feature vectors and cluster them based on the acquisition time and time interval of adjacent peak coefficients to obtain feature vector clusters. Calculate the detail coefficient adjustment degree of the wavelet decomposition layer based on the detail coefficients corresponding to the feature vectors in all feature vector clusters corresponding to the wavelet decomposition layer, the time intervals corresponding to different feature vectors, and the fluctuation feature value. Combine the wavelet threshold determined by the fixed threshold method to adjust the detail coefficients of the wavelet decomposition layer and obtain the detail adjustment coefficient sequence. Reconstruct the detail adjustment coefficient sequences of all wavelet decomposition layers for the same operating parameters to obtain the denoised operating parameters. Use an anomaly detection algorithm to obtain the reconstructed state feature parameters.

[0022] In the terminal processing unit, the real-time acquired operating parameters and status characteristic parameters are filtered and denoised.

[0023] Because aviation electric fuel pumps are subject to interference from internal flow-induced noise and cavitation noise in their operating and testing environments, and these noises are similar to early fault characteristics such as bearing pitting, initial blade friction, and gear tooth wear, conventional filtering algorithms tend to over-filter out fault details in the signal when smoothing noise, leading to a decrease in the accuracy of subsequent diagnostic models. Wavelet threshold denoising can perform layered signal processing and can carry out detailed and effective denoising operations for different frequency bands. Therefore, wavelet threshold denoising is used to filter each type of operating parameter separately.

[0024] The fundamental cause of flow-induced noise is the relative motion between the fluid and the solid boundary, and the instability within the fluid. The fluid-solid boundary includes components such as pipe walls, sensor probes, and hydrophones; the instability within the fluid is turbulence, which contains eddies of various scales. Large-scale eddies have high energy but low frequency; when they rotate or oscillate in the fluid, the resulting pressure fluctuations are concentrated in the lower frequency band, and within this lower frequency band, smaller eddies appear as the frequency increases. Due to fluid viscosity dissipation, small-scale eddies have low energy, and this energy decays rapidly with increasing frequency. Therefore, flow-induced noise is typically distributed across the entire frequency band, and due to its dispersed signal, its fluctuation amplitude is small but its fluctuation frequency is frequent, exhibiting a dense, irregular spike-like appearance. Furthermore, when an aviation electric fuel pump malfunctions, its operating parameters exhibit regular, drastic abrupt changes as the pump rotates.

[0025] Any operating condition parameter is denoted as the target operating condition parameter. The wavelet threshold denoising algorithm is used to decompose each operating condition parameter within a preset time period to obtain the detail coefficient sequence of the first preset number of wavelet decomposition layers.

[0026] In the wavelet threshold denoising algorithm, this embodiment uses the db4 wavelet basis for hierarchical decomposition. In this embodiment, the preset duration is set to 1 second and the first preset quantity is set to 7.

[0027] For any wavelet decomposition layer detail coefficient sequence of the target operating parameters, the detail coefficients in the detail coefficient sequence that are greater than the two adjacent detail coefficients are recorded as peak coefficients. When the number of peak coefficients in the detail coefficient sequence is less than or equal to a second preset number, the wavelet threshold of the corresponding layer is obtained using a fixed threshold method. When the number of peak coefficients in the detail coefficient sequence is greater than the second preset number, the peak coefficients in the detail coefficient sequence that are greater than or equal to the mean of all peak coefficients are recorded as Class I peak coefficients, and the peak coefficients in the detail coefficient sequence that are less than the mean of all peak coefficients are recorded as Class II peak coefficients. The difference between the mean of the normalized values ​​of all Class I peak coefficients and the mean of the normalized values ​​of all Class II peak coefficients is recorded as the fluctuation difference feature value of the corresponding wavelet decomposition layer. The ratio of the permutation entropy of the time interval of all adjacent Class I peak coefficients in the wavelet decomposition layer to the fluctuation difference feature value is recorded as the fluctuation feature value of the wavelet decomposition layer.

[0028] Before calculating the permutation entropy, this embodiment extracts the time interval sequence, sets the embedding dimension to 3, and the delay time to 1.

[0029] In this embodiment, the second preset quantity is set to 10; the normalized value of the first type of peak coefficient is the ratio of the first type of peak coefficient to the maximum value of all peak coefficients, and the normalized value of the second type of peak coefficient is the ratio of the second type of peak coefficient to the maximum value of all peak coefficients.

[0030] In the process of calculating the ratio, in order to avoid the denominator being zero, a preset value needs to be added to the denominator. In this example, the preset value is 0.01.

[0031] When the flow-induced noise characteristics are more significant, the fluctuation amplitude of the detail coefficients within the wavelet decomposition layer is smaller, and the difference between the first-class and second-class peak coefficients is generally smaller. Simultaneously, influenced by the vortex characteristics of flow-induced noise, the peak coefficients within the wavelet layer exhibit an irregular distribution. In this case, the fluctuation eigenvalue of the wavelet decomposition layer is larger. Conversely, when the fault characteristics are more significant, the detail coefficients within the wavelet decomposition layer exhibit regular and dramatic fluctuations. A more pronounced amplitude difference exists between the first-class and second-class peak coefficients, and the peak points within the clustered fluctuation regions show significant regularity due to fault characteristic modulation. The peak point arrangement is fixed and exhibits periodicity. In this case, the fluctuation eigenvalue of the wavelet decomposition layer is smaller.

[0032] Furthermore, when an aviation electric fuel pump experiences faults such as bearing pitting, initial blade friction, or gear tooth wear, the detail coefficients within the wavelet decomposition layer exhibit periodic abrupt changes during fuel pump operation, showing sparse distribution and localized large impacts, resulting in concentrated energy distribution in the abrupt change region; while flow-induced noise exhibits irregular and frequent fluctuations, with a more dispersed energy distribution, typically not concentrated in a specific area.

[0033] For any wavelet decomposition layer detail coefficient sequence of the target operating condition parameters, the minimum time interval between a class of peak coefficients and two adjacent class of peak coefficients is denoted as the characteristic time interval of the class of peak coefficients. The vector composed of the acquisition time corresponding to the class of peak coefficients and the characteristic time interval is denoted as the characteristic vector of the class of peak coefficients. All characteristic vectors determined by the detail coefficient sequence are clustered to obtain the characteristic vector cluster.

[0034] Clustering is a well-known technique and will not be elaborated further. In this embodiment, the DBSCAN clustering algorithm is used to implement clustering. MinPts is set to 4, and the neighborhood radius is determined by the K-distance graph. After sorting the K-distances in descending order, the point where the maximum difference between adjacent distances is located is calculated as the neighborhood radius.

[0035] As the impact of flow-induced noise becomes more severe, the energy fluctuations of the detail coefficients within the wavelet decomposition layer become more pronounced. Simultaneously, the vortices generated by flow-induced noise have a more dispersed impact on the wavelet layer energy, resulting in smaller energy differences for the detail coefficients within the affected wavelet layer, and a generally lower overall energy in the affected fluctuation region. However, when a fault occurs, the wavelet decomposition layer exhibits regular energy abrupt changes in accordance with the fault characteristics, and the energy changes in the abrupt change region are larger than those in the normal region, with the overall energy in the fault-affected region being higher.

[0036] The square of the detail coefficients corresponding to the eigenvector is denoted as the energy of the eigenvector. The sum of the squares of all detail coefficients corresponding to the wavelet decomposition layer is denoted as the total energy of the wavelet decomposition layer. The ratio of the sum of the energies of the two eigenvectors with the largest time intervals within the eigenvector cluster to the total energy of the corresponding wavelet decomposition layer is denoted as the first ratio of the eigenvector cluster. The ratio of the maximum time interval corresponding to different eigenvectors within the eigenvector cluster to a preset duration is denoted as the second ratio of the eigenvector cluster. The ratio of the first ratio to the second ratio of the eigenvector cluster is denoted as the energy characteristic value of the eigenvector cluster. When the number of eigenvector clusters is less than or equal to 1, the energy distribution characteristic value of the wavelet decomposition layer is directly assigned a value of 1.0. When the number of eigenvector clusters is greater than 1, the coefficient of variation of the energy characteristic values ​​of all eigenvector clusters corresponding to the wavelet decomposition layer is denoted as the energy distribution characteristic value of the wavelet decomposition layer. The normalized value of the product of the energy distribution characteristic value and the fluctuation characteristic value of the wavelet decomposition layer is denoted as the detail coefficient adjustment degree of the wavelet decomposition layer.

[0037] It should be noted that this embodiment uses the maximum-minimum normalization method to calculate the normalized value. The product of the energy distribution characteristic value and the fluctuation characteristic value obtained within 30 seconds before the start time of the preset duration corresponding to the wavelet decomposition layer is selected to determine the standard for the maximum and minimum values. When less than 30 seconds have passed before the start time of the preset duration corresponding to the wavelet decomposition layer, or the difference between the standard for the maximum and minimum values ​​is less than... At this point, the detail coefficient adjustment degree of the wavelet decomposition layer is assigned a value of 0.5. In practical applications, implementers can use other methods of existing technology, such as the tanh function or the sigmoid function, to calculate the normalized value, which is not limited here.

[0038] When the energy eigenvalue is larger, the energy within the corresponding interval of the eigenvector cluster is relatively larger, which better matches the energy mutation characteristics at the time of fault occurrence. When the energy distribution eigenvalue of the wavelet decomposition layer is larger, the difference in energy characteristics between the intervals corresponding to all eigenvector clusters of the wavelet decomposition layer is greater, which better matches the characteristics of dispersed and relatively small energy distribution under the influence of flow-induced noise. When the adjustment degree of the detail coefficients of the wavelet decomposition layer is larger, the detail coefficients within the wavelet decomposition layer should be adjusted during the wavelet decomposition process to avoid the influence of noise.

[0039] Based on the adjustment degree of the detail coefficients of the wavelet decomposition layer and the wavelet threshold determined by the fixed threshold method, the detail coefficients of the wavelet decomposition layer are adjusted. The specific formula for calculating the adjustment value of the detail coefficients is as follows: in, Indicates the first The first operating parameter under various working conditions The first layer of wavelet decomposition layer Adjustment values ​​for each detail coefficient; Indicates the first The first operating parameter under various working conditions The first layer of wavelet decomposition layer Detail coefficients; This indicates that the wavelet threshold was determined using the fixed threshold method; Indicates the first The first operating parameter under various working conditions Optimal adjustment parameters for layered wavelet decomposition layers; This indicates the preset first adjustment parameter value. In this embodiment, the first adjustment parameter value is 0.9. This indicates a preset second adjustment parameter value; in this embodiment, the second adjustment parameter value is 0.5. Indicates the first The first operating parameter under various working conditions The degree of detail coefficient adjustment of the layer wavelet decomposition layer; Represents the natural constant.

[0040] Existing technologies typically rely on empirical values ​​to set the wavelet coefficients for wavelet threshold denoising algorithms, resulting in fixed parameters that cannot be adaptively adjusted. However, in applications relying on bus transmission for fault detection and diagnosis of aviation electric fuel pumps, the operating and testing environments of fuel pumps are generally plagued by interference such as flow-induced noise and cavitation noise. These interference signals are highly similar to early fault characteristics such as bearing pitting, initial blade friction, and gear tooth wear, easily causing confusion between noise and fault features, leading to significant deviations in fuel pump fault detection and diagnosis results. Therefore, to address these technical shortcomings, the above process, based on prior knowledge, fully utilizes the differentiated characteristics of the detail coefficients at each wavelet decomposition level. It combines the peak fluctuation amplitude, variation pattern, and energy distribution characteristics of the detail coefficients at each wavelet level under flow-induced noise interference and early fault states of the aviation electric fuel pump to adaptively adjust the wavelet coefficients at different wavelet levels. This allows for precise filtering of various operating signals collected by sensors, effectively avoiding identification bias problems in the fault detection and diagnosis process.

[0041] At this point, a sequence of adjusted values ​​of detail coefficients of the first preset number of wavelet decomposition layers for each type of operating parameter is obtained. The sequence of adjusted values ​​of detail coefficients of the same wavelet decomposition layer is denoised as the detail adjustment coefficient sequence of the same wavelet decomposition layer. The wavelet threshold denoising algorithm is used to denoise and reconstruct the detail adjustment coefficient sequence of the first preset number of wavelet decomposition layers for the same type of operating parameter to obtain the denoised operating parameter.

[0042] Anomaly detection algorithms are used to identify and process outliers of any state feature parameter to obtain the reconstructed state feature parameter.

[0043] Among them, the wavelet threshold denoising algorithm is a well-known technique and will not be elaborated further; the anomaly detection algorithm can use either Z-score or 3sigma principle, without any specific restrictions, and the median filling method is used to fill the identified outliers.

[0044] At this point, the denoised operating parameters and the reconstructed state characteristic parameters are obtained.

[0045] Step S003: Using a pre-trained fault perception model, based on the denoised operating parameters and reconstructed state feature parameters after a preset duration, complete the fault diagnosis of the aviation electric fuel pump.

[0046] Aviation electric fuel pumps in normal and different types of fault states are manually selected by those skilled in the art. Operating parameters and state characteristic parameters are collected 500 times per second under each state. Noise-reduced operating parameters and reconstructed state characteristic parameters are obtained. The kurtosis, root mean square, and peak value of the noise-reduced operating parameters collected within one second are calculated. The reconstructed state characteristic parameters are then arranged sequentially to obtain data feature vectors. All data feature vectors form a dataset. Labels are manually assigned to the data feature vectors, including normal state, motor controller fault, bearing fault, cavitation fault, blockage fault, and leakage fault. A support vector machine is used to train a fault perception model based on the dataset, resulting in a trained fault perception model.

[0047] In the process of training the model using support vector machines, this embodiment selects the radial basis function (RBF) as the kernel function and sets the penalty factor and Gamma value through grid search. The number and specific types of labels can be set according to the actual situation. The model training process is a well-known technique and will not be described in detail here.

[0048] The denoised operating parameters and reconstructed state feature parameters after a preset duration are input into the fault perception model to obtain the fault diagnosis results for each corresponding label. Specifically, the fault diagnosis results include: normal state, motor controller fault, bearing fault, cavitation fault, blockage fault, and leakage fault.

[0049] This completes the fault diagnosis of the aviation electric fuel pump.

Claims

1. A method for fault detection and diagnosis of aviation electric fuel pumps based on CAN bus, characterized in that, The method includes the following steps: The operating parameters and status characteristics of the aviation electric fuel pump are collected at each collection time. The operating parameters include current, voltage, speed, pressure and flow rate, and the status characteristics include vibration and pump body temperature. A wavelet thresholding denoising algorithm is used to decompose the same operating parameters within a preset time period to obtain the detail coefficient sequences of different wavelet decomposition layers. Based on the differences between adjacent detail coefficients in the detail coefficient sequences, it is determined whether to use the fixed threshold method to obtain the wavelet threshold. If not, a type of peak coefficient is identified and the fluctuation feature value of the wavelet decomposition layer is calculated. Feature vectors are established and clustered based on the acquisition time and time interval of adjacent peak coefficients to obtain feature vector clusters. Based on the detail coefficients corresponding to the feature vectors in all feature vector clusters corresponding to the wavelet decomposition layer, the time intervals corresponding to different feature vectors, and the fluctuation feature value, the detail coefficient adjustment degree of the wavelet decomposition layer is calculated. Combined with the wavelet threshold determined by the fixed threshold method, the detail coefficients of the wavelet decomposition layer are adjusted to obtain the detail adjustment coefficient sequence. The detail adjustment coefficient sequences of all wavelet decomposition layers of the same operating parameters are reconstructed to obtain the denoised operating parameters. An anomaly detection algorithm is used to obtain the reconstructed state feature parameters. Using a pre-trained fault perception model, fault diagnosis of aviation electric fuel pumps is completed based on the denoised operating parameters and reconstructed state feature parameters after a preset duration. The specific method for identifying a class of peak coefficients and calculating the wavelet decomposition layer's fluctuation characteristic values ​​includes: The detail coefficients in the detail coefficient sequence that are greater than the two adjacent detail coefficients are recorded as peak coefficients. When the number of peak coefficients in the detail coefficient sequence is greater than the second preset number, the peak coefficients in the detail coefficient sequence that are greater than or equal to the mean of all peak coefficients are recorded as a type of peak coefficient. Peak coefficients in the detail coefficient sequence that are less than the mean of all peak coefficients are denoted as Class II peak coefficients. The difference between the mean of the normalized values ​​of all Class I peak coefficients and the mean of the normalized values ​​of all Class II peak coefficients is denoted as the fluctuation difference feature value of the corresponding wavelet decomposition layer. The ratio of the permutation entropy of the time interval of all adjacent Class I peak coefficients in the wavelet decomposition layer to the fluctuation difference feature value is denoted as the fluctuation feature value of the wavelet decomposition layer.

2. The method for fault detection and diagnosis of aviation electric fuel pump based on CAN bus according to claim 1, characterized in that, The specific method for determining whether to use a fixed threshold method to obtain the wavelet threshold based on the difference between adjacent detail coefficients in the detail coefficient sequence is as follows: When the number of peak coefficients in the detail coefficient sequence is less than or equal to the second preset number, the wavelet threshold of the corresponding layer is obtained by using the fixed threshold method.

3. The method for fault detection and diagnosis of aviation electric fuel pump based on CAN bus according to claim 1, characterized in that, The specific method for obtaining the feature vector clusters is as follows: For any wavelet decomposition layer detail coefficient sequence under the same operating conditions and parameters, the minimum time interval between a class of peak coefficients and two adjacent class of peak coefficients is denoted as the characteristic time interval of the class of peak coefficients. The vector composed of the acquisition time corresponding to the class of peak coefficients and the characteristic time interval is denoted as the characteristic vector of the class of peak coefficients. All characteristic vectors determined by the detail coefficient sequence are clustered to obtain the characteristic vector cluster.

4. The method for fault detection and diagnosis of aviation electric fuel pump based on CAN bus according to claim 3, characterized in that, The specific calculation method for the detail coefficient adjustment degree of the wavelet decomposition layer is as follows: The square of the detail coefficients corresponding to the feature vector is denoted as the energy of the feature vector. The sum of the squares of all detail coefficients corresponding to the wavelet decomposition layer is denoted as the total energy of the wavelet decomposition layer. The first ratio of the feature vector cluster is calculated based on the energy of the feature vector within the feature vector cluster and the total energy of the corresponding wavelet decomposition layer. The ratio of the maximum time interval corresponding to different feature vectors within a feature vector cluster to a preset duration is denoted as the second ratio of the feature vector cluster. The ratio of the first ratio to the second ratio of the eigenvector clusters is denoted as the energy eigenvalue of the eigenvector clusters. When the number of eigenvector clusters is less than or equal to 1, the energy distribution eigenvalue of the wavelet decomposition layer is directly assigned to 1.

0. When the number of eigenvector clusters is greater than 1, the coefficient of variation of the energy eigenvalues ​​of all eigenvector clusters corresponding to the wavelet decomposition layer is denoted as the energy distribution eigenvalue of the wavelet decomposition layer. The normalized value of the product of the energy distribution eigenvalue and the fluctuation eigenvalue of the wavelet decomposition layer is denoted as the detail coefficient adjustment degree of the wavelet decomposition layer.

5. The method for fault detection and diagnosis of aviation electric fuel pump based on CAN bus according to claim 4, characterized in that, The first ratio of the feature vector cluster is the ratio of the energy corresponding to the two feature vectors with the largest time interval within the feature vector cluster to the total energy of the corresponding wavelet decomposition layer.

6. The method for fault detection and diagnosis of aviation electric fuel pump based on CAN bus according to claim 1, characterized in that, The specific method for obtaining the detailed adjustment coefficient sequence is as follows: The product of the sum of the adjustment degree of the detail coefficients of the wavelet decomposition layer and the preset second adjustment parameter value and the preset first adjustment parameter value is recorded as the optimal adjustment parameter of the wavelet decomposition layer; the adjustment value of the detail coefficients is calculated by combining the wavelet threshold determined by the fixed threshold method and the detail coefficients of the wavelet decomposition layer. The sequence of adjustment values ​​of the detail coefficients of the same wavelet decomposition layer is denoted as the detail adjustment coefficient sequence of the same wavelet decomposition layer.

7. The method for fault detection and diagnosis of aviation electric fuel pump based on CAN bus according to claim 6, characterized in that, The formula for calculating the adjustment value of the detail coefficient is: in, Indicates the first The first operating parameter under various working conditions The first layer of wavelet decomposition layer Adjustment values ​​for each detail coefficient; Indicates the first The first operating parameter under various working conditions The first layer of wavelet decomposition layer Detail coefficients; This indicates that the wavelet threshold was determined using the fixed threshold method; Indicates the first The first operating parameter under various working conditions Optimal adjustment parameters for layered wavelet decomposition layers; Represents the natural constant.

8. The method for fault detection and diagnosis of aviation electric fuel pump based on CAN bus according to claim 1, characterized in that, The fault perception model is trained using a support vector machine.

9. The method for fault detection and diagnosis of aviation electric fuel pump based on CAN bus according to claim 1, characterized in that, Fault diagnosis results for aviation electric fuel pumps include: normal state, motor controller failure, bearing failure, cavitation failure, blockage failure, and leakage failure.

Images