A method and device for adaptive filtering processing of hydraulic pump monitoring signals

Abstract

The present application provides a method and device for adaptive filtering and processing of hydraulic pump monitoring signals. The method provided in the present application includes: obtaining a pure monitoring signal data set of the hydraulic pump in different health states and a noisy monitoring signal data set of the hydraulic pump in different airborne working scenarios; performing feature extraction on the data sets respectively to determine the first feature corresponding to the pure monitoring signal data set and the second feature corresponding to the noisy monitoring signal data set; training a filtering model based on the distribution difference between the first feature and the second feature; inputting the hydraulic pump monitoring signal to be filtered into the filtering model, filtering the hydraulic pump monitoring signal based on the filtering processing method output by the filtering model, and obtaining a filtered hydraulic pump monitoring signal. The method and device provided in the present application are used to adaptively adjust the filtering method according to the working state of the hydraulic pump and the monitoring signal, without the need to manually adjust the filtering parameters.

Description

Technical Field

[0001] The present application relates to the field of signal processing technology, and in particular to a method and device for adaptive filtering processing of hydraulic pump monitoring signals. Background Art

[0002] Filtering is a key signal processing step in the monitoring and fault diagnosis of aircraft onboard electromechanical systems. Its primary purpose is to extract target signal features from complex backgrounds, enabling accurate health assessment and early detection of potential faults. Hydraulic pumps, as a crucial component of airborne electromechanical systems, operate in a complex environment. Vibration signals are often affected by coupling interference from nearby equipment, such as the engine and casing, and are also subject to high levels of noise pollution. These interference and noise signals not only have wide spectral distributions and diverse statistical characteristics, but also exhibit significant nonlinearity and dynamic variations, making accurate extraction of target signals extremely challenging.

[0003] Commonly used signal filtering methods currently include median filtering, mean filtering, frequency-band-based low-pass and high-pass filtering, adaptive filtering, and neural network-based filtering. Median filtering performs well in removing impulse noise, but its effectiveness is limited when dealing with Gaussian noise or other complex noise types. Mean filtering has some suppressive effect on Gaussian noise, but it can introduce signal blurring and result in the loss of important features. Frequency-band filtering, based on prior knowledge of the interfering frequency components, can effectively filter out-of-band noise but is ineffective against complex in-band noise. These methods all rely heavily on the statistical properties of both the signal and the noise. In practical applications, they are often limited to specific scenarios and lack the flexibility to cope with changing environments. In particular, under the complex and variable operating conditions of airborne hydraulic pumps, these methods struggle to meet the filtering requirements of time-varying signals, and their filtering performance is easily severely limited by environmental noise and nonlinear interference.

[0004] Therefore, there is an urgent need for an adaptive filtering processing method for hydraulic pump monitoring signals, which can adaptively adjust the filtering method according to the working status of the hydraulic pump and the monitoring signal. Moreover, when the working status and noise characteristics of the hydraulic pump change, there is no need to manually adjust the filtering parameters, and it has stronger adaptability to complex nonlinear and time-varying noise environments. Summary of the Invention

[0005] In view of this, the present application provides a method and device for adaptive filtering processing of hydraulic pump monitoring signals, which is used to adaptively adjust the filtering method according to the working state of the hydraulic pump and the monitoring signal. Moreover, when the working state and noise characteristics of the hydraulic pump change, there is no need to manually adjust the filtering parameters, and it has stronger adaptability to complex nonlinear and time-varying noise environments.

[0006] Specifically, this application is implemented through the following technical solutions:

[0007] A first aspect of the present application provides a method for adaptive filtering and processing a hydraulic pump monitoring signal, the method comprising:

[0008] Obtain pure monitoring signals of the hydraulic pump in different health states and noisy monitoring signals of the hydraulic pump in different airborne working scenarios to obtain pure monitoring signal datasets and noisy monitoring signal datasets;

[0009] Performing feature extraction on the pure monitoring signal dataset and the noisy monitoring signal dataset respectively, screening a first feature corresponding to the pure monitoring signal dataset and a second feature corresponding to the noisy monitoring signal dataset; the feature values of the first feature and the second feature in a fault state and a healthy state are greater than feature values corresponding to other unselected features;

[0010] Training a filtering model based on a distribution difference between the first feature and the second feature, the filtering model outputting a filtering processing mode corresponding to the distribution difference, wherein the distribution difference is determined based on a degree of overlap and a frequency distribution consistency between the first feature and the second feature;

[0011] The hydraulic pump monitoring signal to be filtered is input into the filtering model, and the hydraulic pump monitoring signal is filtered based on the filtering processing method output by the filtering model to obtain a filtered hydraulic pump monitoring signal.

[0012] A second aspect of the present application provides a device for adaptive filtering and processing a hydraulic pump monitoring signal, the device comprising an acquisition module, a determination module, a training module, and a filtering module;

[0013] The acquisition module is used to acquire the pure monitoring signals of the hydraulic pump in different health states and the noisy monitoring signals of the hydraulic pump in different airborne working scenarios, to obtain the pure monitoring signal data set and the noisy monitoring signal data set;

[0014] The determination module is used to perform feature extraction on the pure monitoring signal data set and the noisy monitoring signal data set, respectively, to screen a first feature corresponding to the pure monitoring signal data set and a second feature corresponding to the noisy monitoring signal data set; the feature values of the first feature and the second feature in the fault state and the healthy state are greater than the feature values corresponding to other unselected features;

[0015] The training module is configured to train a filtering model based on a distribution difference between the first feature and the second feature, wherein the filtering model outputs a filtering processing mode corresponding to the distribution difference, wherein the distribution difference is determined based on a degree of overlap and a frequency distribution consistency between the first feature and the second feature;

[0016] The filtering module is used to input the hydraulic pump monitoring signal to be filtered into the filtering model, and filter the hydraulic pump monitoring signal based on the filtering processing method output by the filtering model to obtain a filtered hydraulic pump monitoring signal.

[0017] The present application provides a method and apparatus for adaptive filtering of hydraulic pump monitoring signals. First, a comprehensive dataset is established by acquiring datasets of pure monitoring signals from the hydraulic pump in different health states and datasets of noisy monitoring signals in different airborne scenarios. This multi-state, multi-scenario data coverage lays the foundation for subsequent feature extraction, enabling the filtering model to adapt to diverse signal inputs. Furthermore, through feature extraction and feature screening, the extracted first features (pure monitoring signal features) and second features (noise-containing monitoring signal features) are more pronounced in healthy and faulty states, providing highly discriminative input variables. Based on the distribution differences between the two types of features (including overlap and frequency distribution consistency), the trained filtering model can dynamically adjust the filtering strategy, achieving adaptive filtering processing for different noise environments and health states. Secondly, after screening the first and second features, the present application utilizes the distribution differences between the first and second features to train the filtering model, ensuring that the filtering strategy is highly correlated with the feature distribution differences. Because the distributions of the first and second features differ significantly in noisy environments and pure monitoring signals, the filtering model can focus on specific frequency bands or signal patterns, effectively avoiding interference from noise or non-target signals. At the same time, the filtering strategy based on distribution differences can be dynamically adjusted to adaptively optimize filtering parameters for different input signals. This rapid response mechanism based on feature distribution differences significantly improves filtering speed and accuracy, achieving fast, accurate, and interference-free filtering processing. Thirdly, when performing feature screening, the present application ensures that the selected first and second features have strong distinguishing and representativeness by following the principle that the eigenvalues of the first and second features are higher than the eigenvalues of unselected features. When training the filtering model, the overlap of features and the consistency of frequency distribution are used as judgment criteria, enabling the filtering model to more accurately identify and process noise signals. At the same time, the filtering processing method output by the filtering model is directly linked to the feature distribution differences, ensuring that the filtering strategy always matches the current signal state. Therefore, the present application can dynamically adapt to different noise environments and health states, maintaining the stability and high precision of the filtering effect. Moreover, when the operating state of the hydraulic pump and the interference signal change dynamically, the intelligent filtering model training and adaptive adjustment reduce the need for manual intervention and the complexity of manually adjusting the filter parameters. This can make the processing of hydraulic pump monitoring signals more automated and efficient, with greater adaptability, and can process and optimize signals under different working conditions in real time. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 Flowchart of the method for adaptive filtering of hydraulic pump monitoring signals provided in Example 1 of the present application;

[0019] Figure 2 This is a structural diagram of the hydraulic pump monitoring signal adaptive filtering processing device provided in Example 2 of the present application. DETAILED DESCRIPTION

[0020] Exemplary embodiments are described in detail herein, with examples illustrated in the accompanying drawings. When the following description refers to the drawings, identical numerals in different drawings represent identical or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all embodiments consistent with this application.

[0021] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. The singular forms "a," "the," and "the" used in this application are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0022] It should be understood that although the terms first, second, third, etc. may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from each other. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0023] Specific embodiments are given below to introduce the technical solutions of the present application in detail.

[0024] Figure 1 This is a flow chart of the method for adaptive filtering of hydraulic pump monitoring signals provided in Example 1 of this application. Figure 1 The method provided in this embodiment may include:

[0025] S101. Acquire pure monitoring signals of a hydraulic pump in different health states and noisy monitoring signals of the hydraulic pump in different airborne working scenarios to obtain a pure monitoring signal dataset and a noisy monitoring signal dataset.

[0026] Specifically, a pure monitoring signal refers to the signal collected when the hydraulic pump is operating under ideal conditions. It is not interfered with by external noise and only contains the vibration characteristics and operating status information of the hydraulic pump itself. Different health states refer to the working states of the hydraulic pump at different stages of its operating life or at different levels of fault, including healthy state (normal state), slightly degraded state, severely degraded state, fault state, etc. Noise-containing monitoring signals refer to the signals collected in a real airborne environment during the operation of the hydraulic pump. In addition to the vibration characteristics of the hydraulic pump itself, it also contains various external noises and interferences, such as coupled vibrations of adjacent equipment (the impact of vibrations of the engine, casing, etc. on the signal), environmental noise, and system noise. Different airborne working scenarios refer to the various typical working conditions and environmental conditions that the hydraulic pump may encounter during aircraft operation, including the startup and shutdown stages, stable flight stages, takeoff and landing stages, and sudden working conditions.

[0027] It should be noted that pure monitoring signals and noisy monitoring signals can be obtained through experimental means or through simulation means.

[0028] In specific implementation, the acquisition of pure monitoring signals of the hydraulic pump in different health states and noisy monitoring signals of the hydraulic pump in different airborne working scenarios includes: acquiring pure monitoring signals of the hydraulic pump in different health states on a standard test bench; the pure monitoring signals cover all working conditions of the hydraulic pump within the actual working envelope; acquiring environmental interference noise signals of the hydraulic pump when it is not working in an airborne environment on a vibration test bench; acquiring noisy monitoring signals of the hydraulic pump when it is working under different working conditions in an airborne environment on a vibration test bench.

[0029] Specifically, a standard hydraulic pump test bench was used to inspect the test environment to ensure it was free of additional vibration interference or electromagnetic noise. High-precision sensors (such as accelerometers, pressure sensors, and flow sensors) were installed in key hydraulic pump locations, such as the pump housing and bearing housing. Furthermore, multiple typical operating conditions were set within the hydraulic pump's operating envelope. These included speeds ranging from low, medium, and high; loads ranging from light, medium, and full load; and pressures ranging from low, medium, and high. Under each operating condition, the hydraulic pump's internal components were artificially adjusted (e.g., by creating localized wear or adding leakage channels) to simulate different health states. For example, in the simulated normal state, the hydraulic pump exhibited no faults or performance degradation. In the simulated mild degradation state, the hydraulic pump's slippers and bearings exhibited minor wear, or the pump chamber exhibited small amounts of leakage. In the simulated severe degradation state, the hydraulic pump's performance significantly decreased, such as severe leakage or significantly insufficient flow. In the simulated fault state, key hydraulic pump components exhibited damage, such as ruptured seals or failed bearings. In each health state, pure monitoring signals are collected to obtain pure monitoring signals in different health states under all working conditions.

[0030] Furthermore, a vibration test bench was used to simulate the vibration environment of the aircraft casing and external noise sources (such as engine operating vibration). The hydraulic pump and sensor were fixed to the vibration test bench, but the hydraulic pump was not running. Various vibration frequencies and amplitudes were set. For example, the frequency range was from 10 Hz to 10 kHz, covering the typical frequency range of airborne equipment vibration; the amplitude range included low, medium, and high vibration intensities to simulate different flight conditions (such as cruise, takeoff, and landing). The environmental interference noise signals of the hydraulic pump under different vibration conditions were recorded.

[0031] Furthermore, the hydraulic pump was operated on a vibration test rig, subjected to different health states and operating conditions. As in the previous embodiment, health states include normal, slightly degraded, severely degraded, and faulty. Operating conditions included various combinations, such as low speed and low load, and high speed and high load, covering the actual operating envelope. The vibration test rig applied vibration conditions consistent with the airborne environment, and sensors simultaneously recorded the hydraulic pump's noise-containing monitoring signals (including the hydraulic pump's operating characteristics and external vibration and noise signals).

[0032] Optionally, obtaining the pure monitoring signal of the hydraulic pump in different health states and the noisy monitoring signal of the hydraulic pump in different airborne working scenarios includes: modeling the noise environment according to the working scenario of the hydraulic pump, generating the noise signal based on the vibration load spectrum, and obtaining the environmental interference noise signal; fusing the environmental interference noise signal with the pure monitoring signal to obtain the noisy monitoring signal of the hydraulic pump when working under different working conditions.

[0033] In specific implementation, the hydraulic pump's operating environment is analyzed to identify possible noise sources, such as engine vibration, casing-conducted vibration, structure-borne noise caused by flight operations, and vibration from other onboard equipment. A mathematical model of the noise environment is established based on the actual airborne environment. Vibration data from the external environment is used as input, and appropriate modeling is performed based on the hydraulic pump's operating conditions. Spectral analysis is used to extract the frequency components and power spectrum of the vibration signal. Based on different operating conditions, the noise characteristics of different frequency bands are considered to simulate the interference of the airborne environment on the hydraulic pump. Furthermore, a load spectrum tailored to the hydraulic pump's specific operating conditions (such as speed and load) is designed in conjunction with the noise sources in the airborne environment. The load spectrum can contain vibration components in multiple frequency bands, typically requiring a comprehensive consideration of both low-frequency (such as engine vibration) and high-frequency (such as sensor noise and mechanical resonance) signals. A vibration simulation tool is used to generate a signal that matches the specified load spectrum, simulating the noise's time and frequency domain characteristics. The frequency components, amplitude, and phase of the ambient noise are combined with the vibration load spectrum, and an appropriate noise model (such as white noise or colored noise) is used to generate an environmental interference noise signal consistent with the airborne environment. Furthermore, a simulation tool is used to perform weighted fusion of the generated environmental interference noise signal and the pure monitoring signal collected from the healthy state of the hydraulic pump to obtain a noisy monitoring signal.

[0034] It's important to note that the fusion process requires consideration of the strength and frequency content of the environmental noise signal relative to the pure monitoring signal to ensure that the noisy monitoring signal closely matches the actual operating conditions. Furthermore, the noise intensity can be adjusted under different operating conditions to simulate the noisy monitoring signal of the hydraulic pump under different airborne environments. For example, under high load or high speed, the signal generated by the hydraulic pump is more susceptible to external noise, so the noise component can be appropriately enhanced.

[0035] S102. Perform feature extraction on the pure monitoring signal dataset and the noisy monitoring signal dataset respectively, and screen the first feature corresponding to the pure monitoring signal dataset and the second feature corresponding to the noisy monitoring signal dataset; the feature values of the first feature and the second feature in the fault state and the healthy state are greater than the feature values corresponding to other unselected features.

[0036] Specifically, the first feature corresponding to the pure monitoring signal data set and the second feature corresponding to the noisy monitoring signal data set are different, but the process and implementation principle of obtaining the first feature and obtaining the second feature are similar. In this embodiment, only the acquisition of the first feature is introduced. The second feature is similar and will not be repeated in this embodiment.

[0037] In a specific implementation, the feature extraction of the pure monitoring signal data set and the screening of the first feature corresponding to the pure monitoring signal data set include: denoising and standardizing the pure monitoring signal data set; extracting features of the pure monitoring signal data set based on Fourier transform to obtain time domain features and frequency domain features of the pure monitoring signal under different health states; analyzing the time domain features and frequency domain features to determine the first feature of the mutation from the healthy state to the fault state.

[0038] Specifically, filtering methods (such as wavelet denoising, median filtering, or low-pass filtering) are used to remove high-frequency noise or abnormal spikes from the clean monitoring signal dataset. A denoising method suitable for hydraulic pump monitoring signals is selected to ensure that the main signal components are preserved without excessive smoothing. The signal amplitudes of the clean monitoring signal dataset are normalized to a uniform range (such as [0, 1] or a standard normal distribution with a mean of 0 and a variance of 1) to eliminate the effects of dimensional differences. Furthermore, statistical features such as mean, variance, standard deviation, skewness, kurtosis, and crest factor are directly extracted from the time domain signal. The time domain signal is converted to the frequency domain using a Fourier transform, and key frequency domain features such as the main frequency component, bandwidth, harmonic components, and energy distribution are extracted. The distribution differences between the time domain and frequency domain features under different health states are compared to identify significant change points from the healthy state to the faulty state. Statistical analysis methods (such as analysis of variance and eigenvalue curve plotting) are used to identify the feature with the most significant change as the first feature.

[0039] It should be noted that the first feature is determined based on the variation trend of the pure monitoring signal between different health states (faulty and healthy). Specifically, the characteristic values of the first feature in both the faulty and healthy states are greater than the characteristic values corresponding to the other unselected features. However, the first feature may or may not vary with the operating conditions of the hydraulic pump. This step first describes the variation of the first feature with the operating conditions of the hydraulic pump, indicating that the variation of the first feature with the operating conditions of the hydraulic pump is greater than the variation of the other unselected features.

[0040] Optionally, after filtering the first feature corresponding to the pure monitoring signal data set and the second feature corresponding to the noisy monitoring signal data set, it also includes: when the changing trend of the first feature and the second feature with the hydraulic pump working condition is higher than a first preset value, constructing a mapping relationship between the first feature, the second feature and the working condition based on a neural network as a sign of the first feature and the second feature.

[0041] Specifically, in combination with the above description, the first preset value is set according to actual needs, which is not limited in this embodiment.

[0042] In specific implementations, a curve is fitted between the first feature and the hydraulic pump operating condition to calculate the gradient or rate of change of the first feature as it changes with the hydraulic pump operating condition. The trend of change is measured using slope, amplitude, or standard deviation. Furthermore, the trend of change of each first feature as it changes with the hydraulic pump operating condition is traversed. When the trend is determined to be greater than a first preset value, the first feature is determined to be sensitive to operating condition changes. A neural network (such as a multi-layer perceptron or convolutional neural network) is then used to construct a mapping relationship between the first feature and the hydraulic pump operating condition. The input is hydraulic pump operating condition data (such as speed and pressure), and the output is the first feature value. The neural network is trained and validated using a large amount of data to ensure the accuracy of the mapping between the first feature and the operating condition.

[0043] The method provided in this embodiment uses the mutation of different signals from a healthy to a faulty state as a criterion to accurately select features sensitive to healthy and faulty states, highlighting the distinguishability and representativeness of the features. The selected first and second features not only significantly reduce the interference of redundant features but also reduce data dimensionality, improving feature extraction efficiency and the processing speed and accuracy of the filtering model. Subsequently, for features that have a high tendency to change with the hydraulic pump's operating conditions, a neural network is used to establish a mapping relationship between the features and the hydraulic pump's operating conditions, dynamically associating the features with the hydraulic pump's operating conditions. This enables feature selection with greater adaptability and real-time adjustment capabilities. This feature signature based on operating condition changes gives the features greater stability and generalization capabilities, ensuring consistency across different scenarios, thereby enhancing the robustness of the filtering model to complex hydraulic pump operating conditions. Furthermore, by combining the correlation between feature change trends and operating conditions, the selected features not only optimize filtering performance but also better meet practical engineering requirements, demonstrating excellent application value. Therefore, by highlighting the operating condition relevance and signature properties of features, this feature selection method significantly improves the accuracy, adaptability, and practical application of the filtering model.

[0044] Furthermore, in conjunction with the above description, the first feature may or may not vary with the operating conditions of the hydraulic pump. The above embodiment first describes the case where the first feature varies with the operating conditions of the hydraulic pump. In this step, the case where the first feature does not vary with the operating conditions of the hydraulic pump is described, i.e., the magnitude of the first feature's variation with the operating conditions of the hydraulic pump is lower than the magnitude of the variation corresponding to other unselected features.

[0045] In a specific implementation, the gradient or rate of change of the first feature as it changes with the hydraulic pump operating condition is calculated by fitting a curve corresponding to the first feature and the hydraulic pump operating condition. The trend of change is measured using slope, amplitude, or standard deviation. Furthermore, the trend of change of each first feature as it changes with the hydraulic pump operating condition is traversed. When the trend is determined to be less than a second preset value, the first feature is determined to have good operating condition invariance, is suitable for independent health status assessment, and can stably represent the health status. The feature is then retained as the first feature.

[0046] S103: Training a filtering model based on the distribution difference between the first feature and the second feature, and the filtering model outputs a filtering processing method corresponding to the distribution difference.

[0047] Specifically, the input of the filtering model is the first feature and the second feature, and the output is a filtering processing method. The filtering processing method includes a frequency band filtering method, a signal adaptive decomposition method, an adaptive filtering method, a neural network filtering method, and the like.

[0048] Training a filtering model based on the distribution difference between the first and second features includes: comparing the frequency band overlap of the first and second features; if the frequency band overlap is less than a threshold, filtering the noise from the noisy monitoring signal dataset based on the non-overlapping frequency bands; and if the frequency band overlap is greater than or equal to the threshold, determining the difference between the fault feature in the second feature and the noise feature in the second feature, and the degree of similarity between the feature distribution and a known distribution, to determine a target filtering method. Different target filtering methods have different feature recognition capabilities. In other words, if the overlap is low, direct distinction can be made using frequency bands, and frequency band-based filtering is the most direct, rapid, and accurate. However, if the overlap is high, further investigation is needed to determine whether the noise and valid signal features differ significantly. If so, direct noise filtering can still be performed using a signal adaptive decomposition method. However, if the difference is not significant, signal decomposition cannot distinguish the two signals. In this case, the noise and valid signal feature distribution characteristics are compared to determine whether they are consistent with the known signal type. If so, adaptive filtering can be used again based on known information to achieve rapid differentiation and filtering. If not, the signal is an unknown signal, a mixture of the two, and the difference is not significant, in which case a neural network recognition method can be used. The method provided by the present invention adopts filtering methods of different precisions and efficiencies according to the differences and known characteristics of the mixed contents in the mixed signals. For signals that can be clearly distinguished, a fast and simple filtering method is directly adopted, which ensures the accuracy of filtering while saving filtering time.

[0049] In specific implementation, the distribution difference training filtering model based on the first feature and the second feature includes: performing spectral analysis on the first feature and the second feature to determine the spectral components affected by noise; when the spectral components are not within the operating frequency range of the pure monitoring signal, determining that the distribution difference is out-of-band noise; when the spectral components are within the operating frequency range of the pure monitoring signal, determining that the distribution difference is in-band noise.

[0050] Specifically, the first and second features are Fourier transformed to obtain a spectrum. These are then converted into spectral expressions to obtain amplitude and frequency distribution. The energy concentration and noise characteristics within the spectrum are analyzed to determine the spectral components of the first and second features and their relationship to the frequency distribution of the pure monitoring signal. Furthermore, the spectral components of the first and second features are compared with the spectrum of the pure monitoring signal to identify newly added or abnormally enhanced frequency components in the characteristic spectrum, such as abnormally increased amplitudes or sudden increases in high-frequency energy. If the abnormal frequency components are outside the operating frequency range of the pure monitoring signal, this indicates that the noise signature and the fault signature have no cross-influence within the frequency band and are therefore identified as out-of-band noise. (For example, if a hydraulic pump is operating at a specific speed and its main frequency range is known, any frequency component within this range is considered out-of-band noise.) Out-of-band noise may arise from external equipment vibration, environmental vibration, or other sources. When the abnormal frequency component is within the operating frequency range of the pure monitoring signal, it indicates that the noise characteristics and fault characteristics have cross-influences within the frequency band, and it is determined to be in-band noise. In-band noise may be caused by internal faults or coupled interference (such as harmonic distortion or bearing failure) of the hydraulic pump, and manifests as abnormal increase in amplitude, harmonic distortion, etc.

[0051] Optionally, the filtering model outputs a filtering processing method corresponding to the distribution difference, including:

[0052] (1) When it is determined that the distribution difference is out-of-band noise, the filtering model determines that the corresponding filtering processing method is a frequency band filtering method.

[0053] Specifically, adaptive filtering methods need to be designed based on the different characteristics of out-of-band noise and in-band noise. When out-of-band noise is determined, frequency band filtering methods are usually used to selectively retain signals within the hydraulic pump's operating frequency range and suppress noise in other frequency bands.

[0054] (2) When it is determined that the distribution difference is in-band noise and the second feature is smaller than the first feature, the filtering model determines that the corresponding filtering processing method is a signal adaptive decomposition method.

[0055] Specifically, if the signal is determined to be in-band noise and the second characteristic is smaller than the first characteristic, this indicates that the noise and fault characteristics partially overlap within the frequency band, and that the amplitude or energy of the fault characteristic is significantly higher than that of the noise characteristic. Therefore, adaptive signal decomposition methods can be used to extract the characteristic frequencies associated with the fault while simultaneously removing the frequency components associated with the noise.

[0056] (3) When it is determined that the distribution difference is in-band noise and the distribution of the first feature is determined or the distribution of the second feature is determined, the filtering model determines that the corresponding filtering processing method is an adaptive filtering method or a neural network-based filtering method.

[0057] Specifically, if in-band noise is determined and the distribution of the first or second characteristics is determined, this indicates that the noise and fault characteristics partially overlap within the frequency band. Furthermore, if the statistical characteristics of the noise or fault characteristics (such as distribution shape, energy distribution, or spectral characteristics) are known, a more accurate filtering method can be selected based on this distribution information, namely, an adaptive filtering method or a neural network-based filtering method. The adaptive filtering method can adjust the filtering parameters in real time based on the signal characteristics, while the neural network-based filtering method can accurately separate the noise and fault characteristics when the distribution characteristics are known.

[0058] Optionally, after the filtering model is trained based on the distribution difference between the first feature and the second feature, and the filtering model outputs a filtering processing method corresponding to the distribution difference, the method further includes: obtaining actual onboard data of the hydraulic pump under an actual working environment based on a sensor; comparing the actual onboard data with the hydraulic pump monitoring signal after filtering processing by the filtering model, and adjusting the filtering model based on the comparison result.

[0059] Specifically, actual onboard data from the hydraulic pump under actual operating conditions is acquired. This data is typically collected in real time by sensors and reflects the actual operating status of the hydraulic pump under different operating conditions. This data includes the operating characteristics of the hydraulic pump under complex operating conditions, including equipment health, vibration, pressure, temperature, and other signals. These signals are often affected by noise and may include interference from the onboard environment and adjacent equipment.

[0060] Furthermore, these actual airborne data are compared with hydraulic pump monitoring signals processed by the filtering model. Specifically, the signals optimized by the filtering model are compared with the actual airborne signals directly collected by the sensor. The focus is on detecting differences between the two, particularly in noise removal, feature extraction accuracy, and signal clarity. This comparison can identify potential problems in the filtering process, such as poor filter processing of certain noise signals or failure to fully restore the health characteristics of the hydraulic pump. Based on the comparison results, if the filtering model's processing performance is unsatisfactory, adjustments to the filtering model's parameters or training process may be necessary. These adjustments can include optimizing filter weights, reselecting features, or modifying the filtering model's structure to improve its adaptability and accuracy. In this way, the filtering model can be continuously adjusted and optimized based on feedback from actual data to achieve better signal processing performance, particularly in noise processing and fault feature extraction under real-world operating conditions.

[0061] In specific implementations, sensors installed on the hydraulic pump (such as pressure sensors, vibration sensors, or acoustic sensors) collect real-time data from the hydraulic pump under actual operating conditions. This data is typically uploaded to a central control unit or storage device via a data acquisition module. Furthermore, the collected real-time data is fed into a pre-trained filtering model. The filtering model applies an optimal filtering strategy based on the differences in feature distribution within the real-time data, outputting a filtered, denoised monitoring signal. The real-time data is compared with the denoised monitoring signal, extracting common features such as spectral distribution, time-domain waveform, and noise power. Error metrics (such as mean square error, signal correlation coefficient, and signal-to-noise ratio improvement) are used to quantify the difference between the denoised monitoring signal and the real-time data. This determines whether the denoised monitoring signal has reduced noise while preserving key hydraulic pump operational characteristics. If the denoised monitoring signal loses important features, the filtering model needs adjustment. Based on the comparison results, the filter model is dynamically optimized, specifically including: (1) analyzing the source of the difference. If the denoising is insufficient (too much noise remains), the filter model's noise recognition ability for specific frequencies or features is analyzed, and the filter model parameters are adjusted. If the feature is lost (useful signals are filtered out): check whether the weight distribution of the filter model to the feature is reasonable, and retrain the filter model. (2) Fine-tune the filter model. Using the features of the actual airborne data as a supplementary data set, the filter model is optimized through incremental training or online learning to make it more suitable for the current operating conditions. (3) Based on the difference analysis results, the processing logic of the filter model is dynamically adjusted, such as changing the weight and frequency range of the filter algorithm, or adding a processing module for specific interference.

[0062] The method provided in this embodiment further enhances the adaptability and accuracy of the filtering process by dynamically adjusting the filtering model. First, during the training phase, the filtering model is trained based on the distribution difference between the first and second features, and the filtering process is adapted to match the distribution difference. This provides a strong foundation for the initial filtering effect. In practical applications, onboard data from the hydraulic pump in the actual operating environment is collected by sensors and compared with the filtered monitoring signal to assess the deviation in the filtering effect. This comparison method can dynamically identify deficiencies in the filtering model under different environments or operating conditions, providing an accurate basis for adjusting model parameters. Through this feedback mechanism, the filtering model can be continuously optimized, gradually adapting to the characteristics of the actual onboard data, and reducing errors caused by varying operating conditions or signal characteristics. Furthermore, this filter model adjustment based on actual data can effectively compensate for the distribution deviation between the training data and the actual operating environment, ensuring that the filtering process not only performs well under ideal conditions but also operates stably under complex and variable actual operating conditions. By introducing a closed-loop mechanism for actual data comparison and filter model adjustment, this application achieves dynamic adaptive optimization of the filter model, significantly improving the model's robustness, generalization ability, and practical application effectiveness.

[0063] S104 , inputting the hydraulic pump monitoring signal to be filtered into the filtering model, filtering the hydraulic pump monitoring signal based on the filtering processing method output by the filtering model, and obtaining a filtered hydraulic pump monitoring signal.

[0064] Specifically, in combination with the above description, the filtering model is based on a filtering processing method corresponding to the characteristics of the input hydraulic pump monitoring signal and the characteristic output of the pure monitoring signal. When the filtering processing method is a frequency band filtering method, in this step, a frequency band filter is used to filter the hydraulic pump monitoring signal, wherein the types of frequency band filters include bandpass filters and low-pass / high-pass filters. The bandpass filter is suitable for effectively isolating the out-of-band noise in the hydraulic pump monitoring signal within a known operating frequency range (such as the main frequency and harmonic range). The low-pass / high-pass filter is suitable for specific situations where only high-frequency or low-frequency noise in the hydraulic pump monitoring signal needs to be removed.

[0065] For example, for the following hydraulic pump monitoring signal: target frequency: the hydraulic pump's main frequency is 100 Hz, with harmonics ranging from 100-500 Hz; interference frequency: low-frequency noise is between 0-50 Hz, and high-frequency noise is above 600 Hz. The following frequency band filters can be used: a low-pass filter: retains signals below 500 Hz and suppresses high-frequency noise; a band-pass filter: retains signals in the 100-500 Hz range and suppresses both low- and high-frequency noise.

[0066] Optionally, when the filtering processing method is the signal adaptive decomposition method, in this step, when the variational modal decomposition method in the signal adaptive decomposition method is selected to filter the hydraulic pump monitoring signal, the hydraulic pump monitoring signal is first subjected to spectral analysis to identify frequency characteristics related to the fault state: the hydraulic pump monitoring signal is subjected to a fast Fourier transform to obtain its frequency distribution. The fault characteristic frequency range (such as certain frequency harmonic components or modulation frequencies) is identified, and the center frequency of these frequency bands is determined. When determining the fault characteristic frequency range, the frequency component with significant mutation (i.e., the frequency with the largest energy change from the healthy state to the fault state) is selected, and specific mechanical characteristic frequencies (such as the fundamental frequency, harmonic frequency, or modulation frequency of the hydraulic pump) are considered. Furthermore, the variational modal decomposition method is used to decompose the hydraulic pump monitoring signal and extract the narrowband signal components corresponding to each center frequency: the variational modal decomposition method decomposes the hydraulic pump monitoring signal into several modes, each of which is a narrowband signal in the frequency domain. Reconstruct the components of each modal signal that match the fault characteristic frequency range to form a new signal containing only the fault characteristic distribution frequency band, namely the filtered hydraulic pump monitoring signal. The spectral distribution of each mode is analyzed and matched with the set of fault center frequencies. Noise-related modes (such as modes whose frequencies do not overlap with the fault characteristic frequency or modes with low energy) are excluded. The filtered modal signals are superimposed to reconstruct the new signal.

[0067] Optionally, when the filtering processing method is an adaptive filtering method, in this step, the hydraulic pump monitoring signal is filtered based on the adaptive filtering method. The adaptive filtering method is to generate a filtered output signal after the hydraulic pump monitoring signal passes through a parameter-adjustable digital filter, compare it with the pure monitoring signal, form an error signal, and adjust the filter parameters through an adaptive algorithm to ultimately minimize the mean square value of the error signal. Adaptive filtering can use the results of the filter parameters obtained at the previous moment to automatically adjust the filter parameters at the current moment to adapt to the unknown or time-varying statistical characteristics of the signal and noise, thereby achieving optimal filtering.

[0068] In practice, the hydraulic pump monitoring signal is processed to approximate a pure monitoring signal, thereby removing noise. Pure monitoring signals are typically obtained through experimentation or simulation. Filter design utilizes digital filters with adjustable parameters, such as finite impulse response (FIR) or infinite impulse response (IIR) filters. During initialization, the filter parameters are typically set to zero or random values. The filter's output signal is the result of weighted processing of the hydraulic pump monitoring signal.

[0069] Furthermore, the error between the filter output and the pure monitoring signal is calculated to form an error signal. This error represents the difference between the filter output and the ideal target signal. The error signal is used to guide the update of the filter parameters. By minimizing the error signal, the filter output is brought closer to the pure monitoring signal. An adaptive algorithm is used to update the filter parameters. Common adaptive algorithms include the least mean square (LMS) algorithm and the recursive least squares (RLS) algorithm. The filter uses error feedback at each moment to adjust its weights so that the output signal (the hydraulic pump monitoring signal) gradually approaches the target signal (the pure monitoring signal). This process is iterative, with filter parameters adjusted and updated repeatedly until the error reaches a stable minimum or meets the convergence criteria. At each moment, the filter adjusts its parameters based on the parameter values at the previous moment, continuously optimizing the filtering effect. Ultimately, after adaptive adjustments, the filter output signal is closest to the target signal, and the mean square value of the error signal is minimized, achieving optimal filtering effect.

[0070] Optionally, when the filtering processing method is a neural network filtering-based method, in this step, the hydraulic pump monitoring signal is filtered based on the neural network filtering-based method. The neural network filtering-based method is usually implemented through the generative adversarial network (GAN) framework, involving two main components: the generator and the discriminator. First, in the training phase, the neural network is trained by providing a large number of noisy monitoring signals and corresponding clean monitoring signals. The task of the generator is to receive the noisy monitoring signal as input and generate a filtered signal, which should be as close as possible to the target clean monitoring signal. The task of the discriminator is to determine whether the input signal comes from a real clean monitoring signal dataset or a filtered signal generated by the generator.

[0071] The generator and discriminator are optimized through competitive training. The generator attempts to generate signals that are as "realistic" as possible, while the discriminator attempts to accurately distinguish between real and generated signals. This process is evaluated using a loss function, a common one being the residual loss function, which calculates the difference between the generated signal and the pure monitoring signal. Through alternating optimization of the generator and discriminator, the generator gradually learns to generate higher-quality signals in different noise environments, and is able to remove noise while retaining the main features of the signal. After multiple rounds of iterative optimization, a trained filtering network is obtained, which has strong adaptability and can adapt to different types of noise and effectively denoise.

[0072] After training is complete, the trained adaptive filtering network directly processes the hydraulic pump monitoring signal, generating a filtered output signal to achieve feature extraction and noise reduction. Compared to traditional filtering methods, this process has stronger nonlinear modeling capabilities and is particularly capable of handling complex and unknown noise environments, thereby providing more accurate and stable noise suppression.

[0073] In specific implementation, an appropriate filtering method (such as frequency band filtering, adaptive signal decomposition, adaptive filtering, or neural network-based filtering) is selected based on the noise and fault characteristics of the hydraulic pump monitoring signal to be filtered. This selection determines the basic filter type, but the filter parameters require further adjustment. Furthermore, actual onboard data from the hydraulic pump under different operating conditions is collected, and signal features are extracted to analyze the noise and fault characteristics. Based on the selected filtering method, the filtering model is initialized, and basic parameters (such as frequency range, filter order, and gain) are set. These parameters can be determined using general rules or prior experience. The filtering model is trained using labeled training data (including noise and fault signals), and the filter parameters are adjusted to adaptively optimize the noise suppression effect based on the different operating conditions and fault states of the hydraulic pump. During the training process, the filter parameters are continuously optimized based on the performance of the filtering model on the validation dataset. Error analysis (such as mean square error and signal-to-noise ratio) is used to determine the effectiveness of the filtering model in denoising and retaining key signals. If the filtering effect is unsatisfactory, the filter parameters can be fine-tuned or a combination of multiple filtering methods can be used for optimization. The filter model can further adjust the filter response based on real-time data feedback. For example, the filter can dynamically adjust its parameters based on the hydraulic pump's operating conditions (such as load changes and speed changes) to ensure optimal filtering performance. After obtaining a trained filter model, an appropriate filtering method is reselected based on the trained filter model. The selected filtering method is then used to filter the hydraulic pump monitoring signal to obtain the filtered hydraulic pump monitoring signal.

[0074] This embodiment provides an adaptive filtering method for hydraulic pump monitoring signals. First, a comprehensive dataset is established by acquiring datasets of pure monitoring signals from the hydraulic pump in different health states and datasets of noisy monitoring signals in different airborne scenarios. This multi-state, multi-scenario data coverage lays the foundation for subsequent feature extraction, enabling the filtering model to adapt to diverse signal inputs. Furthermore, through feature extraction and feature screening, the extracted first feature (pure monitoring signal feature) and second feature (noise-containing monitoring signal feature) are more pronounced in healthy and faulty states, providing highly discriminative input variables. Based on the distribution differences between the two types of features (including overlap and frequency distribution consistency), the trained filtering model can dynamically adjust the filtering strategy, achieving adaptive filtering processing for different noise environments and health states. Secondly, after screening the first and second features, the present application utilizes the distribution differences between the first and second features to train the filtering model, ensuring that the filtering strategy is highly correlated with the feature distribution differences. Because the distributions of the first and second features differ significantly in noisy environments and pure monitoring signals, the filtering model can focus on specific frequency bands or signal patterns, effectively avoiding interference from noise or non-target signals. At the same time, the filtering strategy based on distribution differences can dynamically adjust and adaptively optimize filtering parameters for different input signals. This rapid response mechanism based on feature distribution differences significantly improves filtering speed and accuracy, achieving fast, accurate, and interference-free filtering. Thirdly, when performing feature screening, the present application ensures that the selected features have strong discriminative and representative characteristics by ensuring that the eigenvalues of selected features are higher than those of unselected features. When training the filtering model, the overlap and frequency distribution consistency of the features are used as judgment criteria, enabling the filtering model to more accurately identify and process noise signals. Furthermore, the filtering processing method output by the filtering model is directly linked to the feature distribution differences, ensuring that the filtering strategy always matches the current signal state. Therefore, the present application can dynamically adapt to different noise environments and health conditions, maintaining the stability and high precision of the filtering effect. Furthermore, when the operating state of the hydraulic pump and the interference signal dynamically change, the intelligent filtering model training and adaptive adjustment reduce the need for manual intervention and the complexity of manually adjusting the filter parameters. This makes the processing of hydraulic pump monitoring signals more automated and efficient, with greater adaptability, and can process and optimize signals under different operating conditions in real time. Fourthly, when determining features, feature selection not only relies on the transition from a healthy state to a faulty state but also considers the changing trends of the hydraulic pump's operating conditions. Hydraulic pumps exhibit varying signal characteristics under different operating conditions, so features must be adaptable to these variations. By analyzing how these features vary with operating conditions, we can select features with strong recognition capabilities across a wide range of operating conditions, thereby ensuring the stability and accuracy of the filtering model in various practical application scenarios.Through this screening process, the final features identified are those with significant discriminative power, and these features perform well across different hydraulic pump operating conditions and health states. This enhances the robustness and generalization of the subsequent fault diagnosis model, ensuring accurate identification of the hydraulic pump's health state under diverse operating environments. Fifthly, dynamic adjustment of the filtering model further enhances the adaptability and accuracy of the filtering process. First, during the training phase, the filtering model is trained based on the distribution difference between the first and second features, and its output is matched to this distribution difference. This provides a strong foundation for the initial filtering effect. In practical applications, onboard data from the hydraulic pump in actual operating conditions is collected by sensors and compared with the filtered monitoring signal to assess deviations in the filtering effect. This comparison method dynamically identifies deficiencies in the filtering model under different environments or operating conditions, providing a precise basis for adjusting model parameters. Through this feedback mechanism, the filtering model can be continuously optimized, gradually adapting to the characteristics of actual onboard data, reducing errors caused by varying operating conditions or varying signal characteristics. Furthermore, this real-data-based filter model adjustment can effectively compensate for the distribution deviation between the training data and the actual working environment, ensuring that the filtering process not only performs well under ideal conditions but also operates stably under complex and changing actual working conditions. By introducing a closed-loop mechanism for comparing real-data and adjusting the filter model, this application achieves dynamic adaptive optimization of the filter model, significantly improving the model's robustness, generalization ability, and practical application effects.

[0075] Corresponding to the aforementioned embodiment of a method for adaptive filtering and processing a hydraulic pump monitoring signal, the present application also provides an embodiment of a device for adaptive filtering and processing a hydraulic pump monitoring signal.

[0076] Figure 2 This is a structural diagram of the hydraulic pump monitoring signal adaptive filtering processing device provided in Example 2 of this application. Figure 2 The device provided in this embodiment includes an acquisition module 210, a determination module 220, a training module 230 and a filtering module 240;

[0077] The acquisition module is used to acquire the pure monitoring signals of the hydraulic pump in different health states and the noisy monitoring signals of the hydraulic pump in different airborne working scenarios, to obtain the pure monitoring signal data set and the noisy monitoring signal data set;

[0078] The determination module is used to perform feature extraction on the pure monitoring signal data set and the noisy monitoring signal data set, respectively, to screen a first feature corresponding to the pure monitoring signal data set and a second feature corresponding to the noisy monitoring signal data set; the feature values of the first feature and the second feature in the fault state and the healthy state are greater than the feature values corresponding to other unselected features;

[0079] The training module is configured to train a filtering model based on a distribution difference between the first feature and the second feature, wherein the filtering model outputs a filtering processing mode corresponding to the distribution difference, wherein the distribution difference is determined based on a degree of overlap and a frequency distribution consistency between the first feature and the second feature;

[0080] The filtering module is used to input the hydraulic pump monitoring signal to be filtered into the filtering model, and filter the hydraulic pump monitoring signal based on the filtering processing method output by the filtering model to obtain a filtered hydraulic pump monitoring signal.

[0081] The device of this embodiment can be used to perform Figure 1 The steps, specific implementation principles and implementation processes of the method embodiment shown are similar and will not be repeated here.

[0082] The implementation process of the functions and effects of each unit in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, and will not be repeated here.

[0083] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial description of the method embodiments. The device embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present application scheme. A person of ordinary skill in the art can understand and implement it without paying any creative work.

[0084] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A method for adaptive filtering of hydraulic pump monitoring signals, characterized in that: The method comprises: Obtain pure monitoring signals of the hydraulic pump in different health states and noisy monitoring signals of the hydraulic pump in different airborne working scenarios to obtain pure monitoring signal datasets and noisy monitoring signal datasets; Performing feature extraction on the pure monitoring signal dataset and the noisy monitoring signal dataset respectively, screening a first feature corresponding to the pure monitoring signal dataset and a second feature corresponding to the noisy monitoring signal dataset; the feature values of the first feature and the second feature in a fault state and a healthy state are greater than feature values corresponding to other unselected features; A filtering model is trained based on the distribution difference between the first feature and the second feature, and the filtering model outputs a filtering processing mode corresponding to the distribution difference, wherein the distribution difference is determined based on the overlap degree and frequency distribution consistency of the first feature and the second feature; different filtering processing modes have different filtering accuracy and efficiency; The filtering model is trained based on the distribution difference between the first feature and the second feature, including: comparing the frequency band overlap between the first feature and the second feature; If the frequency band overlap is less than a threshold, filtering the noise of the noisy monitoring signal dataset based on non-overlapping frequency bands; If the frequency band overlap is greater than or equal to a threshold, the difference between the fault feature in the second feature and the noise feature in the second feature and the similarity between the feature distribution and the known distribution are determined to determine the target filtering method, and different target filtering methods have different feature recognition capabilities; After filtering the first feature corresponding to the pure monitoring signal data set and the second feature corresponding to the noisy monitoring signal data set, the method further includes: when the changing trend of the first feature and the second feature with the hydraulic pump working condition is higher than a first preset value, constructing a mapping relationship between the first feature, the second feature and the working condition based on a neural network as a sign of the first feature and the second feature; when the changing trend of the first feature and the second feature with the hydraulic pump working condition is less than a second preset value, retaining the first feature and the second feature; wherein, fitting a relationship curve between the first feature and the hydraulic pump working condition, and calculating the gradient or rate of change of the first feature with the hydraulic pump working condition; traversing the changing trend of each first feature with the hydraulic pump working condition, and when it is determined that the changing trend is greater than the first preset value, determining that the first feature is sensitive to the working condition change, and using a neural network to construct a mapping relationship between the first feature and the hydraulic pump working condition; The hydraulic pump monitoring signal to be filtered is input into the filtering model, and the hydraulic pump monitoring signal is filtered based on the filtering processing method output by the filtering model to obtain a filtered hydraulic pump monitoring signal.

2. The method according to claim 1, characterized in that The extracting features of the clean monitoring signal data set and screening the first feature corresponding to the clean monitoring signal data set includes: Performing denoising and standardization on the clean monitoring signal data set; Performing feature extraction on the pure monitoring signal data set based on Fourier transform to obtain time domain features and frequency domain features of the pure monitoring signals under different health states; The time domain features and frequency domain features are analyzed to determine a first feature of a sudden change from a healthy state to a fault state.

3. The method according to claim 1, characterized in that The filtering model is trained based on the distribution difference between the first feature and the second feature, including: performing spectrum analysis on the first feature and the second feature to determine spectrum components affected by noise; When the spectral component is not within the operating frequency range of the pure monitoring signal, determining that the distribution difference is out-of-band noise; When the spectrum component is within the operating frequency range of the pure monitoring signal, the distribution difference is determined to be in-band noise.

4. The method according to claim 3, characterized in that The filtering model output and the filtering processing method corresponding to the distribution difference include: When it is determined that the distribution difference is out-of-band noise, the filtering model determines that the corresponding filtering processing method is a frequency band filtering method; When it is determined that the distribution difference is in-band noise and the second feature is smaller than the first feature, the filtering model determines that the corresponding filtering processing method is a signal adaptive decomposition method; When it is determined that the distribution difference is in-band noise and the distribution of the first feature is determined or the distribution of the second feature is determined, the filtering model determines that the corresponding filtering processing method is an adaptive filtering method or a neural network filtering-based method.

5. The method according to claim 1, wherein The obtaining of pure monitoring signals of the hydraulic pump in different health states and noisy monitoring signals of the hydraulic pump in different airborne working scenarios includes: Acquire pure monitoring signals of the hydraulic pump under different health conditions on a standard test bench; the pure monitoring signals cover all operating conditions of the hydraulic pump within the actual operating envelope; Acquire the environmental interference noise signal of the hydraulic pump when it is not working in an airborne environment on a vibration test bench; The noise-containing monitoring signals of the hydraulic pump operating under different working conditions in an airborne environment are obtained on a vibration test bench.

6. The method according to claim 1, characterized in that The obtaining of pure monitoring signals of the hydraulic pump in different health states and noisy monitoring signals of the hydraulic pump in different airborne working scenarios includes: Noise environment modeling is performed based on the working scenario of the hydraulic pump, noise signals are generated based on the vibration load spectrum, and environmental interference noise signals are obtained; The environmental interference noise signal is fused with the pure monitoring signal to obtain the noise-containing monitoring signal when the hydraulic pump operates under different working conditions.

7. The method according to claim 1, characterized in that After the filtering model is trained based on the distribution difference between the first feature and the second feature, and the filtering model outputs a filtering processing mode corresponding to the distribution difference, the method further includes: Acquire actual onboard data of the hydraulic pump in actual working environment based on sensors; The actual onboard data is compared with the hydraulic pump monitoring signal after filtering by the filtering model, and the filtering model is adjusted based on the comparison result.

8. A hydraulic pump monitoring signal adaptive filtering processing device, characterized in that: The device includes an acquisition module, a determination module, a training module and a filtering module; The acquisition module is used to acquire the pure monitoring signals of the hydraulic pump in different health states and the noisy monitoring signals of the hydraulic pump in different airborne working scenarios, to obtain the pure monitoring signal data set and the noisy monitoring signal data set; The determination module is used to perform feature extraction on the pure monitoring signal data set and the noisy monitoring signal data set, respectively, to screen a first feature corresponding to the pure monitoring signal data set and a second feature corresponding to the noisy monitoring signal data set; the feature values of the first feature and the second feature in the fault state and the healthy state are greater than the feature values corresponding to other unselected features; The training module is configured to train a filtering model based on a distribution difference between the first feature and the second feature, the filtering model outputting a filtering processing mode corresponding to the distribution difference, wherein the distribution difference is determined based on a degree of overlap and a frequency distribution consistency between the first feature and the second feature; different filtering processing modes have different filtering accuracy and efficiency; The filtering model is trained based on the distribution difference between the first feature and the second feature, including: comparing the frequency band overlap between the first feature and the second feature; If the frequency band overlap is less than a threshold, filtering the noise of the noisy monitoring signal dataset based on non-overlapping frequency bands; If the frequency band overlap is greater than or equal to a threshold, the difference between the fault feature in the second feature and the noise feature in the second feature and the similarity between the feature distribution and the known distribution are determined to determine the target filtering method, and different target filtering methods have different feature recognition capabilities; After filtering the first feature corresponding to the pure monitoring signal data set and the second feature corresponding to the noisy monitoring signal data set, the method further includes: when the changing trend of the first feature and the second feature with the hydraulic pump working condition is higher than a first preset value, constructing a mapping relationship between the first feature, the second feature and the working condition based on a neural network as a sign of the first feature and the second feature; when the changing trend of the first feature and the second feature with the hydraulic pump working condition is less than a second preset value, retaining the first feature and the second feature; wherein, fitting a relationship curve between the first feature and the hydraulic pump working condition, and calculating the gradient or rate of change of the first feature with the hydraulic pump working condition; traversing the changing trend of each first feature with the hydraulic pump working condition, and when it is determined that the changing trend is greater than the first preset value, determining that the first feature is sensitive to the working condition change, and using a neural network to construct a mapping relationship between the first feature and the hydraulic pump working condition; The filtering module is used to input the hydraulic pump monitoring signal to be filtered into the filtering model, and filter the hydraulic pump monitoring signal based on the filtering processing method output by the filtering model to obtain a filtered hydraulic pump monitoring signal.

Images